1 /*
2 * Copyright (c) 1999, 2019, 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 // Bail out if 'n' is a Div or Mod node whose zero check was removed earlier (i.e. control is NULL) and its divisor is an induction variable
72 // phi p of a trip-counted (integer) loop whose inputs could be zero (include zero in their type range). p could have a more precise type
73 // range that does not necessarily include all values of its inputs. Since each of these inputs will be a divisor of the newly cloned nodes
74 // of 'n', we need to bail out of one of these divisors could be zero (zero in its type range).
75 if ((n->Opcode() == Op_DivI || n->Opcode() == Op_ModI) && n->in(0) == NULL
76 && region->is_CountedLoop() && n->in(2) == region->as_CountedLoop()->phi()) {
77 Node* phi = region->as_CountedLoop()->phi();
78 for (uint i = 1; i < phi->req(); i++) {
79 if (_igvn.type(phi->in(i))->filter_speculative(TypeInt::ZERO) != Type::TOP) {
80 // Zero could be a possible value but we already removed the zero check. Bail out to avoid a possible division by zero at a later point.
81 return NULL;
82 }
83 }
84 }
85
86 int wins = 0;
87 assert(!n->is_CFG(), "");
88 assert(region->is_Region(), "");
89
90 const Type* type = n->bottom_type();
91 const TypeOopPtr* t_oop = _igvn.type(n)->isa_oopptr();
92 Node* phi;
93 if (t_oop != NULL && t_oop->is_known_instance_field()) {
94 int iid = t_oop->instance_id();
95 int index = C->get_alias_index(t_oop);
96 int offset = t_oop->offset();
97 phi = new PhiNode(region, type, NULL, iid, index, offset);
98 } else {
99 phi = PhiNode::make_blank(region, n);
100 }
101 uint old_unique = C->unique();
102 for (uint i = 1; i < region->req(); i++) {
103 Node* x;
104 Node* the_clone = NULL;
105 if (region->in(i) == C->top()) {
106 x = C->top(); // Dead path? Use a dead data op
107 } else {
108 x = n->clone(); // Else clone up the data op
109 the_clone = x; // Remember for possible deletion.
110 // Alter data node to use pre-phi inputs
111 if (n->in(0) == region)
112 x->set_req( 0, region->in(i) );
113 for (uint j = 1; j < n->req(); j++) {
114 Node* in = n->in(j);
115 if (in->is_Phi() && in->in(0) == region)
116 x->set_req(j, in->in(i)); // Use pre-Phi input for the clone
117 }
118 }
119 // Check for a 'win' on some paths
120 const Type* t = x->Value(&_igvn);
121
122 bool singleton = t->singleton();
123
124 // A TOP singleton indicates that there are no possible values incoming
125 // along a particular edge. In most cases, this is OK, and the Phi will
126 // be eliminated later in an Ideal call. However, we can't allow this to
127 // happen if the singleton occurs on loop entry, as the elimination of
128 // the PhiNode may cause the resulting node to migrate back to a previous
129 // loop iteration.
130 if (singleton && t == Type::TOP) {
131 // Is_Loop() == false does not confirm the absence of a loop (e.g., an
132 // irreducible loop may not be indicated by an affirmative is_Loop());
133 // therefore, the only top we can split thru a phi is on a backedge of
134 // a loop.
135 singleton &= region->is_Loop() && (i != LoopNode::EntryControl);
136 }
137
138 if (singleton) {
139 wins++;
140 x = ((PhaseGVN&)_igvn).makecon(t);
141 } else {
142 // We now call Identity to try to simplify the cloned node.
143 // Note that some Identity methods call phase->type(this).
144 // Make sure that the type array is big enough for
145 // our new node, even though we may throw the node away.
146 // (Note: This tweaking with igvn only works because x is a new node.)
147 _igvn.set_type(x, t);
148 // If x is a TypeNode, capture any more-precise type permanently into Node
149 // otherwise it will be not updated during igvn->transform since
150 // igvn->type(x) is set to x->Value() already.
151 x->raise_bottom_type(t);
152 Node* y = x->Identity(&_igvn);
153 if (y != x) {
154 wins++;
155 x = y;
156 } else {
157 y = _igvn.hash_find(x);
158 if (y) {
159 wins++;
160 x = y;
161 } else {
162 // Else x is a new node we are keeping
163 // We do not need register_new_node_with_optimizer
164 // because set_type has already been called.
165 _igvn._worklist.push(x);
166 }
167 }
168 }
169 if (x != the_clone && the_clone != NULL)
170 _igvn.remove_dead_node(the_clone);
171 phi->set_req( i, x );
172 }
173 // Too few wins?
174 if (wins <= policy) {
175 _igvn.remove_dead_node(phi);
176 return NULL;
177 }
178
179 // Record Phi
180 register_new_node( phi, region );
181
182 for (uint i2 = 1; i2 < phi->req(); i2++) {
183 Node *x = phi->in(i2);
184 // If we commoned up the cloned 'x' with another existing Node,
185 // the existing Node picks up a new use. We need to make the
186 // existing Node occur higher up so it dominates its uses.
187 Node *old_ctrl;
188 IdealLoopTree *old_loop;
189
190 if (x->is_Con()) {
191 // Constant's control is always root.
192 set_ctrl(x, C->root());
193 continue;
194 }
195 // The occasional new node
196 if (x->_idx >= old_unique) { // Found a new, unplaced node?
197 old_ctrl = NULL;
198 old_loop = NULL; // Not in any prior loop
199 } else {
200 old_ctrl = get_ctrl(x);
201 old_loop = get_loop(old_ctrl); // Get prior loop
202 }
203 // New late point must dominate new use
204 Node *new_ctrl = dom_lca(old_ctrl, region->in(i2));
205 if (new_ctrl == old_ctrl) // Nothing is changed
206 continue;
207
208 IdealLoopTree *new_loop = get_loop(new_ctrl);
209
210 // Don't move x into a loop if its uses are
211 // outside of loop. Otherwise x will be cloned
212 // for each use outside of this loop.
213 IdealLoopTree *use_loop = get_loop(region);
214 if (!new_loop->is_member(use_loop) &&
215 (old_loop == NULL || !new_loop->is_member(old_loop))) {
216 // Take early control, later control will be recalculated
217 // during next iteration of loop optimizations.
218 new_ctrl = get_early_ctrl(x);
219 new_loop = get_loop(new_ctrl);
220 }
221 // Set new location
222 set_ctrl(x, new_ctrl);
223 // If changing loop bodies, see if we need to collect into new body
224 if (old_loop != new_loop) {
225 if (old_loop && !old_loop->_child)
226 old_loop->_body.yank(x);
227 if (!new_loop->_child)
228 new_loop->_body.push(x); // Collect body info
229 }
230 }
231
232 return phi;
233 }
234
235 //------------------------------dominated_by------------------------------------
236 // Replace the dominated test with an obvious true or false. Place it on the
237 // IGVN worklist for later cleanup. Move control-dependent data Nodes on the
238 // live path up to the dominating control.
239 void PhaseIdealLoop::dominated_by(IfProjNode* prevdom, IfNode* iff, bool flip, bool exclude_loop_predicate) {
240 if (VerifyLoopOptimizations && PrintOpto) { tty->print_cr("dominating test"); }
241
242 // prevdom is the dominating projection of the dominating test.
243 assert(iff->Opcode() == Op_If ||
244 iff->Opcode() == Op_CountedLoopEnd ||
245 iff->Opcode() == Op_LongCountedLoopEnd ||
246 iff->Opcode() == Op_RangeCheck,
247 "Check this code when new subtype is added");
248
249 int pop = prevdom->Opcode();
250 assert( pop == Op_IfFalse || pop == Op_IfTrue, "" );
251 if (flip) {
252 if (pop == Op_IfTrue)
253 pop = Op_IfFalse;
254 else
255 pop = Op_IfTrue;
256 }
257 // 'con' is set to true or false to kill the dominated test.
258 Node *con = _igvn.makecon(pop == Op_IfTrue ? TypeInt::ONE : TypeInt::ZERO);
259 set_ctrl(con, C->root()); // Constant gets a new use
260 // Hack the dominated test
261 _igvn.replace_input_of(iff, 1, con);
262
263 // If I dont have a reachable TRUE and FALSE path following the IfNode then
264 // I can assume this path reaches an infinite loop. In this case it's not
265 // important to optimize the data Nodes - either the whole compilation will
266 // be tossed or this path (and all data Nodes) will go dead.
267 if (iff->outcnt() != 2) return;
268
269 // Make control-dependent data Nodes on the live path (path that will remain
270 // once the dominated IF is removed) become control-dependent on the
271 // dominating projection.
272 Node* dp = iff->proj_out_or_null(pop == Op_IfTrue);
273
274 // Loop predicates may have depending checks which should not
275 // be skipped. For example, range check predicate has two checks
276 // for lower and upper bounds.
277 if (dp == NULL)
278 return;
279
280 ProjNode* dp_proj = dp->as_Proj();
281 ProjNode* unc_proj = iff->proj_out(1 - dp_proj->_con)->as_Proj();
282 if (exclude_loop_predicate &&
283 (unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_predicate) != NULL ||
284 unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_profile_predicate) != NULL ||
285 unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_range_check) != NULL)) {
286 // If this is a range check (IfNode::is_range_check), do not
287 // reorder because Compile::allow_range_check_smearing might have
288 // changed the check.
289 return; // Let IGVN transformation change control dependence.
290 }
291
292 IdealLoopTree* old_loop = get_loop(dp);
293
294 for (DUIterator_Fast imax, i = dp->fast_outs(imax); i < imax; i++) {
295 Node* cd = dp->fast_out(i); // Control-dependent node
296 // Do not rewire Div and Mod nodes which could have a zero divisor to avoid skipping their zero check.
297 if (cd->depends_only_on_test() && _igvn.no_dependent_zero_check(cd)) {
298 assert(cd->in(0) == dp, "");
299 _igvn.replace_input_of(cd, 0, prevdom);
300 set_early_ctrl(cd, false);
301 IdealLoopTree* new_loop = get_loop(get_ctrl(cd));
302 if (old_loop != new_loop) {
303 if (!old_loop->_child) {
304 old_loop->_body.yank(cd);
305 }
306 if (!new_loop->_child) {
307 new_loop->_body.push(cd);
308 }
309 }
310 --i;
311 --imax;
312 }
313 }
314 }
315
316 //------------------------------has_local_phi_input----------------------------
317 // Return TRUE if 'n' has Phi inputs from its local block and no other
318 // block-local inputs (all non-local-phi inputs come from earlier blocks)
319 Node *PhaseIdealLoop::has_local_phi_input( Node *n ) {
320 Node *n_ctrl = get_ctrl(n);
321 // See if some inputs come from a Phi in this block, or from before
322 // this block.
323 uint i;
324 for( i = 1; i < n->req(); i++ ) {
325 Node *phi = n->in(i);
326 if( phi->is_Phi() && phi->in(0) == n_ctrl )
327 break;
328 }
329 if( i >= n->req() )
330 return NULL; // No Phi inputs; nowhere to clone thru
331
332 // Check for inputs created between 'n' and the Phi input. These
333 // must split as well; they have already been given the chance
334 // (courtesy of a post-order visit) and since they did not we must
335 // recover the 'cost' of splitting them by being very profitable
336 // when splitting 'n'. Since this is unlikely we simply give up.
337 for( i = 1; i < n->req(); i++ ) {
338 Node *m = n->in(i);
339 if( get_ctrl(m) == n_ctrl && !m->is_Phi() ) {
340 // We allow the special case of AddP's with no local inputs.
341 // This allows us to split-up address expressions.
342 if (m->is_AddP() &&
343 get_ctrl(m->in(2)) != n_ctrl &&
344 get_ctrl(m->in(3)) != n_ctrl) {
345 // Move the AddP up to dominating point
346 Node* c = find_non_split_ctrl(idom(n_ctrl));
347 if (c->is_OuterStripMinedLoop()) {
348 c->as_Loop()->verify_strip_mined(1);
349 c = c->in(LoopNode::EntryControl);
350 }
351 set_ctrl_and_loop(m, c);
352 continue;
353 }
354 return NULL;
355 }
356 assert(n->is_Phi() || m->is_Phi() || is_dominator(get_ctrl(m), n_ctrl), "m has strange control");
357 }
358
359 return n_ctrl;
360 }
361
362 // Replace expressions like ((V+I) << 2) with (V<<2 + I<<2).
363 Node* PhaseIdealLoop::remix_address_expressions_add_left_shift(Node* n, IdealLoopTree* n_loop, Node* n_ctrl, BasicType bt) {
364 assert(bt == T_INT || bt == T_LONG, "only for integers");
365 int n_op = n->Opcode();
366
367 if (n_op == Op_LShift(bt)) {
368 // Scale is loop invariant
369 Node* scale = n->in(2);
370 Node* scale_ctrl = get_ctrl(scale);
371 IdealLoopTree* scale_loop = get_loop(scale_ctrl);
372 if (n_loop == scale_loop || !scale_loop->is_member(n_loop)) {
373 return NULL;
374 }
375 const TypeInt* scale_t = scale->bottom_type()->isa_int();
376 if (scale_t != NULL && scale_t->is_con() && scale_t->get_con() >= 16) {
377 return NULL; // Dont bother with byte/short masking
378 }
379 // Add must vary with loop (else shift would be loop-invariant)
380 Node* add = n->in(1);
381 Node* add_ctrl = get_ctrl(add);
382 IdealLoopTree* add_loop = get_loop(add_ctrl);
383 if (n_loop != add_loop) {
384 return NULL; // happens w/ evil ZKM loops
385 }
386
387 // Convert I-V into I+ (0-V); same for V-I
388 if (add->Opcode() == Op_Sub(bt) &&
389 _igvn.type(add->in(1)) != TypeInteger::zero(bt)) {
390 assert(add->Opcode() == Op_SubI || add->Opcode() == Op_SubL, "");
391 Node* zero = _igvn.integercon(0, bt);
392 set_ctrl(zero, C->root());
393 Node* neg = SubNode::make(zero, add->in(2), bt);
394 register_new_node(neg, get_ctrl(add->in(2)));
395 add = AddNode::make(add->in(1), neg, bt);
396 register_new_node(add, add_ctrl);
397 }
398 if (add->Opcode() != Op_Add(bt)) return NULL;
399 assert(add->Opcode() == Op_AddI || add->Opcode() == Op_AddL, "");
400 // See if one add input is loop invariant
401 Node* add_var = add->in(1);
402 Node* add_var_ctrl = get_ctrl(add_var);
403 IdealLoopTree* add_var_loop = get_loop(add_var_ctrl);
404 Node* add_invar = add->in(2);
405 Node* add_invar_ctrl = get_ctrl(add_invar);
406 IdealLoopTree* add_invar_loop = get_loop(add_invar_ctrl);
407 if (add_invar_loop == n_loop) {
408 // Swap to find the invariant part
409 add_invar = add_var;
410 add_invar_ctrl = add_var_ctrl;
411 add_invar_loop = add_var_loop;
412 add_var = add->in(2);
413 } else if (add_var_loop != n_loop) { // Else neither input is loop invariant
414 return NULL;
415 }
416 if (n_loop == add_invar_loop || !add_invar_loop->is_member(n_loop)) {
417 return NULL; // No invariant part of the add?
418 }
419
420 // Yes! Reshape address expression!
421 Node* inv_scale = LShiftNode::make(add_invar, scale, bt);
422 Node* inv_scale_ctrl =
423 dom_depth(add_invar_ctrl) > dom_depth(scale_ctrl) ?
424 add_invar_ctrl : scale_ctrl;
425 register_new_node(inv_scale, inv_scale_ctrl);
426 Node* var_scale = LShiftNode::make(add_var, scale, bt);
427 register_new_node(var_scale, n_ctrl);
428 Node* var_add = AddNode::make(var_scale, inv_scale, bt);
429 register_new_node(var_add, n_ctrl);
430 _igvn.replace_node(n, var_add);
431 return var_add;
432 }
433 return NULL;
434 }
435
436 //------------------------------remix_address_expressions----------------------
437 // Rework addressing expressions to get the most loop-invariant stuff
438 // moved out. We'd like to do all associative operators, but it's especially
439 // important (common) to do address expressions.
440 Node* PhaseIdealLoop::remix_address_expressions(Node* n) {
441 if (!has_ctrl(n)) return NULL;
442 Node* n_ctrl = get_ctrl(n);
443 IdealLoopTree* n_loop = get_loop(n_ctrl);
444
445 // See if 'n' mixes loop-varying and loop-invariant inputs and
446 // itself is loop-varying.
447
448 // Only interested in binary ops (and AddP)
449 if (n->req() < 3 || n->req() > 4) return NULL;
450
451 Node* n1_ctrl = get_ctrl(n->in( 1));
452 Node* n2_ctrl = get_ctrl(n->in( 2));
453 Node* n3_ctrl = get_ctrl(n->in(n->req() == 3 ? 2 : 3));
454 IdealLoopTree* n1_loop = get_loop(n1_ctrl);
455 IdealLoopTree* n2_loop = get_loop(n2_ctrl);
456 IdealLoopTree* n3_loop = get_loop(n3_ctrl);
457
458 // Does one of my inputs spin in a tighter loop than self?
459 if ((n_loop->is_member(n1_loop) && n_loop != n1_loop) ||
460 (n_loop->is_member(n2_loop) && n_loop != n2_loop) ||
461 (n_loop->is_member(n3_loop) && n_loop != n3_loop)) {
462 return NULL; // Leave well enough alone
463 }
464
465 // Is at least one of my inputs loop-invariant?
466 if (n1_loop == n_loop &&
467 n2_loop == n_loop &&
468 n3_loop == n_loop) {
469 return NULL; // No loop-invariant inputs
470 }
471
472 Node* res = remix_address_expressions_add_left_shift(n, n_loop, n_ctrl, T_INT);
473 if (res != NULL) {
474 return res;
475 }
476 res = remix_address_expressions_add_left_shift(n, n_loop, n_ctrl, T_LONG);
477 if (res != NULL) {
478 return res;
479 }
480
481 int n_op = n->Opcode();
482 // Replace (I+V) with (V+I)
483 if (n_op == Op_AddI ||
484 n_op == Op_AddL ||
485 n_op == Op_AddF ||
486 n_op == Op_AddD ||
487 n_op == Op_MulI ||
488 n_op == Op_MulL ||
489 n_op == Op_MulF ||
490 n_op == Op_MulD) {
491 if (n2_loop == n_loop) {
492 assert(n1_loop != n_loop, "");
493 n->swap_edges(1, 2);
494 }
495 }
496
497 // Replace ((I1 +p V) +p I2) with ((I1 +p I2) +p V),
498 // but not if I2 is a constant.
499 if (n_op == Op_AddP) {
500 if (n2_loop == n_loop && n3_loop != n_loop) {
501 if (n->in(2)->Opcode() == Op_AddP && !n->in(3)->is_Con()) {
502 Node* n22_ctrl = get_ctrl(n->in(2)->in(2));
503 Node* n23_ctrl = get_ctrl(n->in(2)->in(3));
504 IdealLoopTree* n22loop = get_loop(n22_ctrl);
505 IdealLoopTree* n23_loop = get_loop(n23_ctrl);
506 if (n22loop != n_loop && n22loop->is_member(n_loop) &&
507 n23_loop == n_loop) {
508 Node* add1 = new AddPNode(n->in(1), n->in(2)->in(2), n->in(3));
509 // Stuff new AddP in the loop preheader
510 register_new_node(add1, n_loop->_head->in(LoopNode::EntryControl));
511 Node* add2 = new AddPNode(n->in(1), add1, n->in(2)->in(3));
512 register_new_node(add2, n_ctrl);
513 _igvn.replace_node(n, add2);
514 return add2;
515 }
516 }
517 }
518
519 // Replace (I1 +p (I2 + V)) with ((I1 +p I2) +p V)
520 if (n2_loop != n_loop && n3_loop == n_loop) {
521 if (n->in(3)->Opcode() == Op_AddX) {
522 Node* V = n->in(3)->in(1);
523 Node* I = n->in(3)->in(2);
524 if (is_member(n_loop,get_ctrl(V))) {
525 } else {
526 Node *tmp = V; V = I; I = tmp;
527 }
528 if (!is_member(n_loop,get_ctrl(I))) {
529 Node* add1 = new AddPNode(n->in(1), n->in(2), I);
530 // Stuff new AddP in the loop preheader
531 register_new_node(add1, n_loop->_head->in(LoopNode::EntryControl));
532 Node* add2 = new AddPNode(n->in(1), add1, V);
533 register_new_node(add2, n_ctrl);
534 _igvn.replace_node(n, add2);
535 return add2;
536 }
537 }
538 }
539 }
540
541 return NULL;
542 }
543
544 // Optimize ((in1[2*i] * in2[2*i]) + (in1[2*i+1] * in2[2*i+1]))
545 Node *PhaseIdealLoop::convert_add_to_muladd(Node* n) {
546 assert(n->Opcode() == Op_AddI, "sanity");
547 Node * nn = NULL;
548 Node * in1 = n->in(1);
549 Node * in2 = n->in(2);
550 if (in1->Opcode() == Op_MulI && in2->Opcode() == Op_MulI) {
551 IdealLoopTree* loop_n = get_loop(get_ctrl(n));
552 if (loop_n->is_counted() &&
553 loop_n->_head->as_Loop()->is_valid_counted_loop(T_INT) &&
554 Matcher::match_rule_supported(Op_MulAddVS2VI) &&
555 Matcher::match_rule_supported(Op_MulAddS2I)) {
556 Node* mul_in1 = in1->in(1);
557 Node* mul_in2 = in1->in(2);
558 Node* mul_in3 = in2->in(1);
559 Node* mul_in4 = in2->in(2);
560 if (mul_in1->Opcode() == Op_LoadS &&
561 mul_in2->Opcode() == Op_LoadS &&
562 mul_in3->Opcode() == Op_LoadS &&
563 mul_in4->Opcode() == Op_LoadS) {
564 IdealLoopTree* loop1 = get_loop(get_ctrl(mul_in1));
565 IdealLoopTree* loop2 = get_loop(get_ctrl(mul_in2));
566 IdealLoopTree* loop3 = get_loop(get_ctrl(mul_in3));
567 IdealLoopTree* loop4 = get_loop(get_ctrl(mul_in4));
568 IdealLoopTree* loop5 = get_loop(get_ctrl(in1));
569 IdealLoopTree* loop6 = get_loop(get_ctrl(in2));
570 // All nodes should be in the same counted loop.
571 if (loop_n == loop1 && loop_n == loop2 && loop_n == loop3 &&
572 loop_n == loop4 && loop_n == loop5 && loop_n == loop6) {
573 Node* adr1 = mul_in1->in(MemNode::Address);
574 Node* adr2 = mul_in2->in(MemNode::Address);
575 Node* adr3 = mul_in3->in(MemNode::Address);
576 Node* adr4 = mul_in4->in(MemNode::Address);
577 if (adr1->is_AddP() && adr2->is_AddP() && adr3->is_AddP() && adr4->is_AddP()) {
578 if ((adr1->in(AddPNode::Base) == adr3->in(AddPNode::Base)) &&
579 (adr2->in(AddPNode::Base) == adr4->in(AddPNode::Base))) {
580 nn = new MulAddS2INode(mul_in1, mul_in2, mul_in3, mul_in4);
581 register_new_node(nn, get_ctrl(n));
582 _igvn.replace_node(n, nn);
583 return nn;
584 } else if ((adr1->in(AddPNode::Base) == adr4->in(AddPNode::Base)) &&
585 (adr2->in(AddPNode::Base) == adr3->in(AddPNode::Base))) {
586 nn = new MulAddS2INode(mul_in1, mul_in2, mul_in4, mul_in3);
587 register_new_node(nn, get_ctrl(n));
588 _igvn.replace_node(n, nn);
589 return nn;
590 }
591 }
592 }
593 }
594 }
595 }
596 return nn;
597 }
598
599 //------------------------------conditional_move-------------------------------
600 // Attempt to replace a Phi with a conditional move. We have some pretty
601 // strict profitability requirements. All Phis at the merge point must
602 // be converted, so we can remove the control flow. We need to limit the
603 // number of c-moves to a small handful. All code that was in the side-arms
604 // of the CFG diamond is now speculatively executed. This code has to be
605 // "cheap enough". We are pretty much limited to CFG diamonds that merge
606 // 1 or 2 items with a total of 1 or 2 ops executed speculatively.
607 Node *PhaseIdealLoop::conditional_move( Node *region ) {
608
609 assert(region->is_Region(), "sanity check");
610 if (region->req() != 3) return NULL;
611
612 // Check for CFG diamond
613 Node *lp = region->in(1);
614 Node *rp = region->in(2);
615 if (!lp || !rp) return NULL;
616 Node *lp_c = lp->in(0);
617 if (lp_c == NULL || lp_c != rp->in(0) || !lp_c->is_If()) return NULL;
618 IfNode *iff = lp_c->as_If();
619
620 // Check for ops pinned in an arm of the diamond.
621 // Can't remove the control flow in this case
622 if (lp->outcnt() > 1) return NULL;
623 if (rp->outcnt() > 1) return NULL;
624
625 IdealLoopTree* r_loop = get_loop(region);
626 assert(r_loop == get_loop(iff), "sanity");
627 // Always convert to CMOVE if all results are used only outside this loop.
628 bool used_inside_loop = (r_loop == _ltree_root);
629
630 // Check profitability
631 int cost = 0;
632 int phis = 0;
633 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
634 Node *out = region->fast_out(i);
635 if (!out->is_Phi()) continue; // Ignore other control edges, etc
636 phis++;
637 PhiNode* phi = out->as_Phi();
638 BasicType bt = phi->type()->basic_type();
639 switch (bt) {
640 case T_DOUBLE:
641 case T_FLOAT:
642 if (C->use_cmove()) {
643 continue; //TODO: maybe we want to add some cost
644 }
645 cost += Matcher::float_cmove_cost(); // Could be very expensive
646 break;
647 case T_LONG: {
648 cost += Matcher::long_cmove_cost(); // May encodes as 2 CMOV's
649 }
650 case T_INT: // These all CMOV fine
651 case T_ADDRESS: { // (RawPtr)
652 cost++;
653 break;
654 }
655 case T_NARROWOOP: // Fall through
656 case T_OBJECT: { // Base oops are OK, but not derived oops
657 const TypeOopPtr *tp = phi->type()->make_ptr()->isa_oopptr();
658 // Derived pointers are Bad (tm): what's the Base (for GC purposes) of a
659 // CMOVE'd derived pointer? It's a CMOVE'd derived base. Thus
660 // CMOVE'ing a derived pointer requires we also CMOVE the base. If we
661 // have a Phi for the base here that we convert to a CMOVE all is well
662 // and good. But if the base is dead, we'll not make a CMOVE. Later
663 // the allocator will have to produce a base by creating a CMOVE of the
664 // relevant bases. This puts the allocator in the business of
665 // manufacturing expensive instructions, generally a bad plan.
666 // Just Say No to Conditionally-Moved Derived Pointers.
667 if (tp && tp->offset() != 0)
668 return NULL;
669 cost++;
670 break;
671 }
672 default:
673 return NULL; // In particular, can't do memory or I/O
674 }
675 // Add in cost any speculative ops
676 for (uint j = 1; j < region->req(); j++) {
677 Node *proj = region->in(j);
678 Node *inp = phi->in(j);
679 if (get_ctrl(inp) == proj) { // Found local op
680 cost++;
681 // Check for a chain of dependent ops; these will all become
682 // speculative in a CMOV.
683 for (uint k = 1; k < inp->req(); k++)
684 if (get_ctrl(inp->in(k)) == proj)
685 cost += ConditionalMoveLimit; // Too much speculative goo
686 }
687 }
688 // See if the Phi is used by a Cmp or Narrow oop Decode/Encode.
689 // This will likely Split-If, a higher-payoff operation.
690 for (DUIterator_Fast kmax, k = phi->fast_outs(kmax); k < kmax; k++) {
691 Node* use = phi->fast_out(k);
692 if (use->is_Cmp() || use->is_DecodeNarrowPtr() || use->is_EncodeNarrowPtr())
693 cost += ConditionalMoveLimit;
694 // Is there a use inside the loop?
695 // Note: check only basic types since CMoveP is pinned.
696 if (!used_inside_loop && is_java_primitive(bt)) {
697 IdealLoopTree* u_loop = get_loop(has_ctrl(use) ? get_ctrl(use) : use);
698 if (r_loop == u_loop || r_loop->is_member(u_loop)) {
699 used_inside_loop = true;
700 }
701 }
702 }
703 }//for
704 Node* bol = iff->in(1);
705 if (bol->Opcode() == Op_Opaque4) {
706 return NULL; // Ignore loop predicate checks (the Opaque4 ensures they will go away)
707 }
708 assert(bol->Opcode() == Op_Bool, "Unexpected node");
709 int cmp_op = bol->in(1)->Opcode();
710 if (cmp_op == Op_SubTypeCheck) { // SubTypeCheck expansion expects an IfNode
711 return NULL;
712 }
713 // It is expensive to generate flags from a float compare.
714 // Avoid duplicated float compare.
715 if (phis > 1 && (cmp_op == Op_CmpF || cmp_op == Op_CmpD)) return NULL;
716
717 float infrequent_prob = PROB_UNLIKELY_MAG(3);
718 // Ignore cost and blocks frequency if CMOVE can be moved outside the loop.
719 if (used_inside_loop) {
720 if (cost >= ConditionalMoveLimit) return NULL; // Too much goo
721
722 // BlockLayoutByFrequency optimization moves infrequent branch
723 // from hot path. No point in CMOV'ing in such case (110 is used
724 // instead of 100 to take into account not exactness of float value).
725 if (BlockLayoutByFrequency) {
726 infrequent_prob = MAX2(infrequent_prob, (float)BlockLayoutMinDiamondPercentage/110.0f);
727 }
728 }
729 // Check for highly predictable branch. No point in CMOV'ing if
730 // we are going to predict accurately all the time.
731 if (C->use_cmove() && (cmp_op == Op_CmpF || cmp_op == Op_CmpD)) {
732 //keep going
733 } else if (iff->_prob < infrequent_prob ||
734 iff->_prob > (1.0f - infrequent_prob))
735 return NULL;
736
737 // --------------
738 // Now replace all Phis with CMOV's
739 Node *cmov_ctrl = iff->in(0);
740 uint flip = (lp->Opcode() == Op_IfTrue);
741 Node_List wq;
742 while (1) {
743 PhiNode* phi = NULL;
744 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
745 Node *out = region->fast_out(i);
746 if (out->is_Phi()) {
747 phi = out->as_Phi();
748 break;
749 }
750 }
751 if (phi == NULL || _igvn.type(phi) == Type::TOP) {
752 break;
753 }
754 if (PrintOpto && VerifyLoopOptimizations) { tty->print_cr("CMOV"); }
755 // Move speculative ops
756 wq.push(phi);
757 while (wq.size() > 0) {
758 Node *n = wq.pop();
759 for (uint j = 1; j < n->req(); j++) {
760 Node* m = n->in(j);
761 if (m != NULL && !is_dominator(get_ctrl(m), cmov_ctrl)) {
762 #ifndef PRODUCT
763 if (PrintOpto && VerifyLoopOptimizations) {
764 tty->print(" speculate: ");
765 m->dump();
766 }
767 #endif
768 set_ctrl(m, cmov_ctrl);
769 wq.push(m);
770 }
771 }
772 }
773 Node *cmov = CMoveNode::make(cmov_ctrl, iff->in(1), phi->in(1+flip), phi->in(2-flip), _igvn.type(phi));
774 register_new_node( cmov, cmov_ctrl );
775 _igvn.replace_node( phi, cmov );
776 #ifndef PRODUCT
777 if (TraceLoopOpts) {
778 tty->print("CMOV ");
779 r_loop->dump_head();
780 if (Verbose) {
781 bol->in(1)->dump(1);
782 cmov->dump(1);
783 }
784 }
785 if (VerifyLoopOptimizations) verify();
786 #endif
787 }
788
789 // The useless CFG diamond will fold up later; see the optimization in
790 // RegionNode::Ideal.
791 _igvn._worklist.push(region);
792
793 return iff->in(1);
794 }
795
796 static void enqueue_cfg_uses(Node* m, Unique_Node_List& wq) {
797 for (DUIterator_Fast imax, i = m->fast_outs(imax); i < imax; i++) {
798 Node* u = m->fast_out(i);
799 if (u->is_CFG()) {
800 if (u->Opcode() == Op_NeverBranch) {
801 u = ((NeverBranchNode*)u)->proj_out(0);
802 enqueue_cfg_uses(u, wq);
803 } else {
804 wq.push(u);
805 }
806 }
807 }
808 }
809
810 // Try moving a store out of a loop, right before the loop
811 Node* PhaseIdealLoop::try_move_store_before_loop(Node* n, Node *n_ctrl) {
812 // Store has to be first in the loop body
813 IdealLoopTree *n_loop = get_loop(n_ctrl);
814 if (n->is_Store() && n_loop != _ltree_root &&
815 n_loop->is_loop() && n_loop->_head->is_Loop() &&
816 n->in(0) != NULL) {
817 Node* address = n->in(MemNode::Address);
818 Node* value = n->in(MemNode::ValueIn);
819 Node* mem = n->in(MemNode::Memory);
820 IdealLoopTree* address_loop = get_loop(get_ctrl(address));
821 IdealLoopTree* value_loop = get_loop(get_ctrl(value));
822
823 // - address and value must be loop invariant
824 // - memory must be a memory Phi for the loop
825 // - Store must be the only store on this memory slice in the
826 // loop: if there's another store following this one then value
827 // written at iteration i by the second store could be overwritten
828 // at iteration i+n by the first store: it's not safe to move the
829 // first store out of the loop
830 // - nothing must observe the memory Phi: it guarantees no read
831 // before the store, we are also guaranteed the store post
832 // dominates the loop head (ignoring a possible early
833 // exit). Otherwise there would be extra Phi involved between the
834 // loop's Phi and the store.
835 // - there must be no early exit from the loop before the Store
836 // (such an exit most of the time would be an extra use of the
837 // memory Phi but sometimes is a bottom memory Phi that takes the
838 // store as input).
839
840 if (!n_loop->is_member(address_loop) &&
841 !n_loop->is_member(value_loop) &&
842 mem->is_Phi() && mem->in(0) == n_loop->_head &&
843 mem->outcnt() == 1 &&
844 mem->in(LoopNode::LoopBackControl) == n) {
845
846 assert(n_loop->_tail != NULL, "need a tail");
847 assert(is_dominator(n_ctrl, n_loop->_tail), "store control must not be in a branch in the loop");
848
849 // Verify that there's no early exit of the loop before the store.
850 bool ctrl_ok = false;
851 {
852 // Follow control from loop head until n, we exit the loop or
853 // we reach the tail
854 ResourceMark rm;
855 Unique_Node_List wq;
856 wq.push(n_loop->_head);
857
858 for (uint next = 0; next < wq.size(); ++next) {
859 Node *m = wq.at(next);
860 if (m == n->in(0)) {
861 ctrl_ok = true;
862 continue;
863 }
864 assert(!has_ctrl(m), "should be CFG");
865 if (!n_loop->is_member(get_loop(m)) || m == n_loop->_tail) {
866 ctrl_ok = false;
867 break;
868 }
869 enqueue_cfg_uses(m, wq);
870 if (wq.size() > 10) {
871 ctrl_ok = false;
872 break;
873 }
874 }
875 }
876 if (ctrl_ok) {
877 // move the Store
878 _igvn.replace_input_of(mem, LoopNode::LoopBackControl, mem);
879 _igvn.replace_input_of(n, 0, n_loop->_head->as_Loop()->skip_strip_mined()->in(LoopNode::EntryControl));
880 _igvn.replace_input_of(n, MemNode::Memory, mem->in(LoopNode::EntryControl));
881 // Disconnect the phi now. An empty phi can confuse other
882 // optimizations in this pass of loop opts.
883 _igvn.replace_node(mem, mem->in(LoopNode::EntryControl));
884 n_loop->_body.yank(mem);
885
886 set_ctrl_and_loop(n, n->in(0));
887
888 return n;
889 }
890 }
891 }
892 return NULL;
893 }
894
895 // Try moving a store out of a loop, right after the loop
896 void PhaseIdealLoop::try_move_store_after_loop(Node* n) {
897 if (n->is_Store() && n->in(0) != NULL) {
898 Node *n_ctrl = get_ctrl(n);
899 IdealLoopTree *n_loop = get_loop(n_ctrl);
900 // Store must be in a loop
901 if (n_loop != _ltree_root && !n_loop->_irreducible) {
902 Node* address = n->in(MemNode::Address);
903 Node* value = n->in(MemNode::ValueIn);
904 IdealLoopTree* address_loop = get_loop(get_ctrl(address));
905 // address must be loop invariant
906 if (!n_loop->is_member(address_loop)) {
907 // Store must be last on this memory slice in the loop and
908 // nothing in the loop must observe it
909 Node* phi = NULL;
910 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
911 Node* u = n->fast_out(i);
912 if (has_ctrl(u)) { // control use?
913 IdealLoopTree *u_loop = get_loop(get_ctrl(u));
914 if (!n_loop->is_member(u_loop)) {
915 continue;
916 }
917 if (u->is_Phi() && u->in(0) == n_loop->_head) {
918 assert(_igvn.type(u) == Type::MEMORY, "bad phi");
919 // multiple phis on the same slice are possible
920 if (phi != NULL) {
921 return;
922 }
923 phi = u;
924 continue;
925 }
926 }
927 return;
928 }
929 if (phi != NULL) {
930 // Nothing in the loop before the store (next iteration)
931 // must observe the stored value
932 bool mem_ok = true;
933 {
934 ResourceMark rm;
935 Unique_Node_List wq;
936 wq.push(phi);
937 for (uint next = 0; next < wq.size() && mem_ok; ++next) {
938 Node *m = wq.at(next);
939 for (DUIterator_Fast imax, i = m->fast_outs(imax); i < imax && mem_ok; i++) {
940 Node* u = m->fast_out(i);
941 if (u->is_Store() || u->is_Phi()) {
942 if (u != n) {
943 wq.push(u);
944 mem_ok = (wq.size() <= 10);
945 }
946 } else {
947 mem_ok = false;
948 break;
949 }
950 }
951 }
952 }
953 if (mem_ok) {
954 // Move the store out of the loop if the LCA of all
955 // users (except for the phi) is outside the loop.
956 Node* hook = new Node(1);
957 hook->init_req(0, n_ctrl); // Add an input to prevent hook from being dead
958 _igvn.rehash_node_delayed(phi);
959 int count = phi->replace_edge(n, hook, &_igvn);
960 assert(count > 0, "inconsistent phi");
961
962 // Compute latest point this store can go
963 Node* lca = get_late_ctrl(n, get_ctrl(n));
964 if (lca->is_OuterStripMinedLoop()) {
965 lca = lca->in(LoopNode::EntryControl);
966 }
967 if (n_loop->is_member(get_loop(lca))) {
968 // LCA is in the loop - bail out
969 _igvn.replace_node(hook, n);
970 return;
971 }
972 #ifdef ASSERT
973 if (n_loop->_head->is_Loop() && n_loop->_head->as_Loop()->is_strip_mined()) {
974 assert(n_loop->_head->Opcode() == Op_CountedLoop, "outer loop is a strip mined");
975 n_loop->_head->as_Loop()->verify_strip_mined(1);
976 Node* outer = n_loop->_head->as_CountedLoop()->outer_loop();
977 IdealLoopTree* outer_loop = get_loop(outer);
978 assert(n_loop->_parent == outer_loop, "broken loop tree");
979 assert(get_loop(lca) == outer_loop, "safepoint in outer loop consume all memory state");
980 }
981 #endif
982 lca = place_outside_loop(lca, n_loop);
983 assert(!n_loop->is_member(get_loop(lca)), "control must not be back in the loop");
984 assert(get_loop(lca)->_nest < n_loop->_nest || lca->in(0)->Opcode() == Op_NeverBranch, "must not be moved into inner loop");
985
986 // Move store out of the loop
987 _igvn.replace_node(hook, n->in(MemNode::Memory));
988 _igvn.replace_input_of(n, 0, lca);
989 set_ctrl_and_loop(n, lca);
990
991 // Disconnect the phi now. An empty phi can confuse other
992 // optimizations in this pass of loop opts..
993 if (phi->in(LoopNode::LoopBackControl) == phi) {
994 _igvn.replace_node(phi, phi->in(LoopNode::EntryControl));
995 n_loop->_body.yank(phi);
996 }
997 }
998 }
999 }
1000 }
1001 }
1002 }
1003
1004 // If UseArrayMarkWordCheck is enabled, we can't use immutable memory for the flat array check
1005 // because we are loading the mark word which is mutable. Although the bits we are interested in
1006 // are immutable (we check for markWord::unlocked_value), we need to use raw memory to not break
1007 // anti dependency analysis. Below code will attempt to still move flat array checks out of loops,
1008 // mainly to enable loop unswitching.
1009 void PhaseIdealLoop::move_flat_array_check_out_of_loop(Node* n) {
1010 // Skip checks for more than one array
1011 if (n->req() > 3) {
1012 return;
1013 }
1014 Node* mem = n->in(FlatArrayCheckNode::Memory);
1015 Node* array = n->in(FlatArrayCheckNode::ArrayOrKlass)->uncast();
1016 IdealLoopTree* check_loop = get_loop(get_ctrl(n));
1017 IdealLoopTree* ary_loop = get_loop(get_ctrl(array));
1018
1019 // Check if array is loop invariant
1020 if (!check_loop->is_member(ary_loop)) {
1021 // Walk up memory graph from the check until we leave the loop
1022 VectorSet wq;
1023 wq.set(mem->_idx);
1024 while (check_loop->is_member(get_loop(ctrl_or_self(mem)))) {
1025 if (mem->is_Phi()) {
1026 mem = mem->in(1);
1027 } else if (mem->is_MergeMem()) {
1028 mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw);
1029 } else if (mem->is_Proj()) {
1030 mem = mem->in(0);
1031 } else if (mem->is_MemBar() || mem->is_SafePoint()) {
1032 mem = mem->in(TypeFunc::Memory);
1033 } else if (mem->is_Store() || mem->is_LoadStore() || mem->is_ClearArray()) {
1034 mem = mem->in(MemNode::Memory);
1035 } else {
1036 #ifdef ASSERT
1037 mem->dump();
1038 #endif
1039 ShouldNotReachHere();
1040 }
1041 if (wq.test_set(mem->_idx)) {
1042 return;
1043 }
1044 }
1045 // Replace memory input and re-compute ctrl to move the check out of the loop
1046 _igvn.replace_input_of(n, 1, mem);
1047 set_ctrl_and_loop(n, get_early_ctrl(n));
1048 Node* bol = n->unique_out();
1049 set_ctrl_and_loop(bol, get_early_ctrl(bol));
1050 }
1051 }
1052
1053 //------------------------------split_if_with_blocks_pre-----------------------
1054 // Do the real work in a non-recursive function. Data nodes want to be
1055 // cloned in the pre-order so they can feed each other nicely.
1056 Node *PhaseIdealLoop::split_if_with_blocks_pre( Node *n ) {
1057 // Cloning these guys is unlikely to win
1058 int n_op = n->Opcode();
1059 if (n_op == Op_MergeMem) {
1060 return n;
1061 }
1062 if (n->is_Proj()) {
1063 return n;
1064 }
1065
1066 if (UseArrayMarkWordCheck && n->isa_FlatArrayCheck()) {
1067 move_flat_array_check_out_of_loop(n);
1068 return n;
1069 }
1070
1071 // Do not clone-up CmpFXXX variations, as these are always
1072 // followed by a CmpI
1073 if (n->is_Cmp()) {
1074 return n;
1075 }
1076 // Attempt to use a conditional move instead of a phi/branch
1077 if (ConditionalMoveLimit > 0 && n_op == Op_Region) {
1078 Node *cmov = conditional_move( n );
1079 if (cmov) {
1080 return cmov;
1081 }
1082 }
1083 if (n->is_CFG() || n->is_LoadStore()) {
1084 return n;
1085 }
1086 if (n->is_Opaque1() || // Opaque nodes cannot be mod'd
1087 n_op == Op_Opaque2) {
1088 if (!C->major_progress()) { // If chance of no more loop opts...
1089 _igvn._worklist.push(n); // maybe we'll remove them
1090 }
1091 return n;
1092 }
1093
1094 if (n->is_Con()) {
1095 return n; // No cloning for Con nodes
1096 }
1097
1098 Node *n_ctrl = get_ctrl(n);
1099 if (!n_ctrl) {
1100 return n; // Dead node
1101 }
1102
1103 Node* res = try_move_store_before_loop(n, n_ctrl);
1104 if (res != NULL) {
1105 return n;
1106 }
1107
1108 // Attempt to remix address expressions for loop invariants
1109 Node *m = remix_address_expressions( n );
1110 if( m ) return m;
1111
1112 if (n_op == Op_AddI) {
1113 Node *nn = convert_add_to_muladd( n );
1114 if ( nn ) return nn;
1115 }
1116
1117 if (n->is_ConstraintCast()) {
1118 Node* dom_cast = n->as_ConstraintCast()->dominating_cast(&_igvn, this);
1119 // ConstraintCastNode::dominating_cast() uses node control input to determine domination.
1120 // Node control inputs don't necessarily agree with loop control info (due to
1121 // transformations happened in between), thus additional dominance check is needed
1122 // to keep loop info valid.
1123 if (dom_cast != NULL && is_dominator(get_ctrl(dom_cast), get_ctrl(n))) {
1124 _igvn.replace_node(n, dom_cast);
1125 return dom_cast;
1126 }
1127 }
1128
1129 // Determine if the Node has inputs from some local Phi.
1130 // Returns the block to clone thru.
1131 Node *n_blk = has_local_phi_input( n );
1132 if( !n_blk ) return n;
1133
1134 // Do not clone the trip counter through on a CountedLoop
1135 // (messes up the canonical shape).
1136 if (((n_blk->is_CountedLoop() || (n_blk->is_Loop() && n_blk->as_Loop()->is_loop_nest_inner_loop())) && n->Opcode() == Op_AddI) ||
1137 (n_blk->is_LongCountedLoop() && n->Opcode() == Op_AddL)) {
1138 return n;
1139 }
1140
1141 // Check for having no control input; not pinned. Allow
1142 // dominating control.
1143 if (n->in(0)) {
1144 Node *dom = idom(n_blk);
1145 if (dom_lca(n->in(0), dom) != n->in(0)) {
1146 return n;
1147 }
1148 }
1149 // Policy: when is it profitable. You must get more wins than
1150 // policy before it is considered profitable. Policy is usually 0,
1151 // so 1 win is considered profitable. Big merges will require big
1152 // cloning, so get a larger policy.
1153 int policy = n_blk->req() >> 2;
1154
1155 // If the loop is a candidate for range check elimination,
1156 // delay splitting through it's phi until a later loop optimization
1157 if (n_blk->is_BaseCountedLoop()) {
1158 IdealLoopTree *lp = get_loop(n_blk);
1159 if (lp && lp->_rce_candidate) {
1160 return n;
1161 }
1162 }
1163
1164 if (must_throttle_split_if()) return n;
1165
1166 // Split 'n' through the merge point if it is profitable
1167 Node *phi = split_thru_phi( n, n_blk, policy );
1168 if (!phi) return n;
1169
1170 // Found a Phi to split thru!
1171 // Replace 'n' with the new phi
1172 _igvn.replace_node( n, phi );
1173 // Moved a load around the loop, 'en-registering' something.
1174 if (n_blk->is_Loop() && n->is_Load() &&
1175 !phi->in(LoopNode::LoopBackControl)->is_Load())
1176 C->set_major_progress();
1177
1178 return phi;
1179 }
1180
1181 static bool merge_point_too_heavy(Compile* C, Node* region) {
1182 // Bail out if the region and its phis have too many users.
1183 int weight = 0;
1184 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
1185 weight += region->fast_out(i)->outcnt();
1186 }
1187 int nodes_left = C->max_node_limit() - C->live_nodes();
1188 if (weight * 8 > nodes_left) {
1189 if (PrintOpto) {
1190 tty->print_cr("*** Split-if bails out: %d nodes, region weight %d", C->unique(), weight);
1191 }
1192 return true;
1193 } else {
1194 return false;
1195 }
1196 }
1197
1198 static bool merge_point_safe(Node* region) {
1199 // 4799512: Stop split_if_with_blocks from splitting a block with a ConvI2LNode
1200 // having a PhiNode input. This sidesteps the dangerous case where the split
1201 // ConvI2LNode may become TOP if the input Value() does not
1202 // overlap the ConvI2L range, leaving a node which may not dominate its
1203 // uses.
1204 // A better fix for this problem can be found in the BugTraq entry, but
1205 // expediency for Mantis demands this hack.
1206 #ifdef _LP64
1207 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
1208 Node* n = region->fast_out(i);
1209 if (n->is_Phi()) {
1210 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
1211 Node* m = n->fast_out(j);
1212 if (m->Opcode() == Op_ConvI2L)
1213 return false;
1214 if (m->is_CastII()) {
1215 return false;
1216 }
1217 }
1218 }
1219 }
1220 #endif
1221 return true;
1222 }
1223
1224
1225 //------------------------------place_outside_loop---------------------------------
1226 // Place some computation outside of this loop on the path to the use passed as argument
1227 Node* PhaseIdealLoop::place_outside_loop(Node* useblock, IdealLoopTree* loop) const {
1228 Node* head = loop->_head;
1229 assert(!loop->is_member(get_loop(useblock)), "must be outside loop");
1230 if (head->is_Loop() && head->as_Loop()->is_strip_mined()) {
1231 loop = loop->_parent;
1232 assert(loop->_head->is_OuterStripMinedLoop(), "malformed strip mined loop");
1233 }
1234
1235 // Pick control right outside the loop
1236 for (;;) {
1237 Node* dom = idom(useblock);
1238 if (loop->is_member(get_loop(dom)) ||
1239 // NeverBranch nodes are not assigned to the loop when constructed
1240 (dom->Opcode() == Op_NeverBranch && loop->is_member(get_loop(dom->in(0))))) {
1241 break;
1242 }
1243 useblock = dom;
1244 }
1245 assert(find_non_split_ctrl(useblock) == useblock, "should be non split control");
1246 return useblock;
1247 }
1248
1249
1250 bool PhaseIdealLoop::identical_backtoback_ifs(Node *n) {
1251 if (!n->is_If() || n->is_BaseCountedLoopEnd()) {
1252 return false;
1253 }
1254 if (!n->in(0)->is_Region()) {
1255 return false;
1256 }
1257
1258 Node* region = n->in(0);
1259 Node* dom = idom(region);
1260 if (!dom->is_If() || dom->in(1) != n->in(1)) {
1261 return false;
1262 }
1263 IfNode* dom_if = dom->as_If();
1264 Node* proj_true = dom_if->proj_out(1);
1265 Node* proj_false = dom_if->proj_out(0);
1266
1267 for (uint i = 1; i < region->req(); i++) {
1268 if (is_dominator(proj_true, region->in(i))) {
1269 continue;
1270 }
1271 if (is_dominator(proj_false, region->in(i))) {
1272 continue;
1273 }
1274 return false;
1275 }
1276
1277 return true;
1278 }
1279
1280
1281 bool PhaseIdealLoop::can_split_if(Node* n_ctrl) {
1282 if (must_throttle_split_if()) {
1283 return false;
1284 }
1285
1286 // Do not do 'split-if' if irreducible loops are present.
1287 if (_has_irreducible_loops) {
1288 return false;
1289 }
1290
1291 if (merge_point_too_heavy(C, n_ctrl)) {
1292 return false;
1293 }
1294
1295 // Do not do 'split-if' if some paths are dead. First do dead code
1296 // elimination and then see if its still profitable.
1297 for (uint i = 1; i < n_ctrl->req(); i++) {
1298 if (n_ctrl->in(i) == C->top()) {
1299 return false;
1300 }
1301 }
1302
1303 // If trying to do a 'Split-If' at the loop head, it is only
1304 // profitable if the cmp folds up on BOTH paths. Otherwise we
1305 // risk peeling a loop forever.
1306
1307 // CNC - Disabled for now. Requires careful handling of loop
1308 // body selection for the cloned code. Also, make sure we check
1309 // for any input path not being in the same loop as n_ctrl. For
1310 // irreducible loops we cannot check for 'n_ctrl->is_Loop()'
1311 // because the alternative loop entry points won't be converted
1312 // into LoopNodes.
1313 IdealLoopTree *n_loop = get_loop(n_ctrl);
1314 for (uint j = 1; j < n_ctrl->req(); j++) {
1315 if (get_loop(n_ctrl->in(j)) != n_loop) {
1316 return false;
1317 }
1318 }
1319
1320 // Check for safety of the merge point.
1321 if (!merge_point_safe(n_ctrl)) {
1322 return false;
1323 }
1324
1325 return true;
1326 }
1327
1328 // Detect if the node is the inner strip-mined loop
1329 // Return: NULL if it's not the case, or the exit of outer strip-mined loop
1330 static Node* is_inner_of_stripmined_loop(const Node* out) {
1331 Node* out_le = NULL;
1332
1333 if (out->is_CountedLoopEnd()) {
1334 const CountedLoopNode* loop = out->as_CountedLoopEnd()->loopnode();
1335
1336 if (loop != NULL && loop->is_strip_mined()) {
1337 out_le = loop->in(LoopNode::EntryControl)->as_OuterStripMinedLoop()->outer_loop_exit();
1338 }
1339 }
1340
1341 return out_le;
1342 }
1343
1344 bool PhaseIdealLoop::flatten_array_element_type_check(Node *n) {
1345 // If the CmpP is a subtype check for a value that has just been
1346 // loaded from an array, the subtype check guarantees the value
1347 // can't be stored in a flattened array and the load of the value
1348 // happens with a flattened array check then: push the type check
1349 // through the phi of the flattened array check. This needs special
1350 // logic because the subtype check's input is not a phi but a
1351 // LoadKlass that must first be cloned through the phi.
1352 if (n->Opcode() != Op_CmpP) {
1353 return false;
1354 }
1355
1356 Node* klassptr = n->in(1);
1357 Node* klasscon = n->in(2);
1358
1359 if (klassptr->is_DecodeNarrowPtr()) {
1360 klassptr = klassptr->in(1);
1361 }
1362
1363 if (klassptr->Opcode() != Op_LoadKlass && klassptr->Opcode() != Op_LoadNKlass) {
1364 return false;
1365 }
1366
1367 if (!klasscon->is_Con()) {
1368 return false;
1369 }
1370
1371 Node* addr = klassptr->in(MemNode::Address);
1372
1373 if (!addr->is_AddP()) {
1374 return false;
1375 }
1376
1377 intptr_t offset;
1378 Node* obj = AddPNode::Ideal_base_and_offset(addr, &_igvn, offset);
1379
1380 if (obj == NULL) {
1381 return false;
1382 }
1383
1384 assert(obj != NULL && addr->in(AddPNode::Base) == addr->in(AddPNode::Address), "malformed AddP?");
1385 if (obj->Opcode() == Op_CastPP) {
1386 obj = obj->in(1);
1387 }
1388
1389 if (!obj->is_Phi()) {
1390 return false;
1391 }
1392
1393 Node* region = obj->in(0);
1394
1395 Node* phi = PhiNode::make_blank(region, n->in(1));
1396 for (uint i = 1; i < region->req(); i++) {
1397 Node* in = obj->in(i);
1398 Node* ctrl = region->in(i);
1399 if (addr->in(AddPNode::Base) != obj) {
1400 Node* cast = addr->in(AddPNode::Base);
1401 assert(cast->Opcode() == Op_CastPP && cast->in(0) != NULL, "inconsistent subgraph");
1402 Node* cast_clone = cast->clone();
1403 cast_clone->set_req(0, ctrl);
1404 cast_clone->set_req(1, in);
1405 register_new_node(cast_clone, ctrl);
1406 _igvn.set_type(cast_clone, cast_clone->Value(&_igvn));
1407 in = cast_clone;
1408 }
1409 Node* addr_clone = addr->clone();
1410 addr_clone->set_req(AddPNode::Base, in);
1411 addr_clone->set_req(AddPNode::Address, in);
1412 register_new_node(addr_clone, ctrl);
1413 _igvn.set_type(addr_clone, addr_clone->Value(&_igvn));
1414 Node* klassptr_clone = klassptr->clone();
1415 klassptr_clone->set_req(2, addr_clone);
1416 register_new_node(klassptr_clone, ctrl);
1417 _igvn.set_type(klassptr_clone, klassptr_clone->Value(&_igvn));
1418 if (klassptr != n->in(1)) {
1419 Node* decode = n->in(1);
1420 assert(decode->is_DecodeNarrowPtr(), "inconsistent subgraph");
1421 Node* decode_clone = decode->clone();
1422 decode_clone->set_req(1, klassptr_clone);
1423 register_new_node(decode_clone, ctrl);
1424 _igvn.set_type(decode_clone, decode_clone->Value(&_igvn));
1425 klassptr_clone = decode_clone;
1426 }
1427 phi->set_req(i, klassptr_clone);
1428 }
1429 register_new_node(phi, region);
1430 Node* orig = n->in(1);
1431 _igvn.replace_input_of(n, 1, phi);
1432 split_if_with_blocks_post(n);
1433 if (n->outcnt() != 0) {
1434 _igvn.replace_input_of(n, 1, orig);
1435 _igvn.remove_dead_node(phi);
1436 }
1437 return true;
1438 }
1439
1440 //------------------------------split_if_with_blocks_post----------------------
1441 // Do the real work in a non-recursive function. CFG hackery wants to be
1442 // in the post-order, so it can dirty the I-DOM info and not use the dirtied
1443 // info.
1444 void PhaseIdealLoop::split_if_with_blocks_post(Node *n) {
1445
1446 if (flatten_array_element_type_check(n)) {
1447 return;
1448 }
1449
1450 // Cloning Cmp through Phi's involves the split-if transform.
1451 // FastLock is not used by an If
1452 if (n->is_Cmp() && !n->is_FastLock()) {
1453 Node *n_ctrl = get_ctrl(n);
1454 // Determine if the Node has inputs from some local Phi.
1455 // Returns the block to clone thru.
1456 Node *n_blk = has_local_phi_input(n);
1457 if (n_blk != n_ctrl) {
1458 return;
1459 }
1460
1461 if (!can_split_if(n_ctrl)) {
1462 return;
1463 }
1464
1465 if (n->outcnt() != 1) {
1466 return; // Multiple bool's from 1 compare?
1467 }
1468 Node *bol = n->unique_out();
1469 assert(bol->is_Bool(), "expect a bool here");
1470 if (bol->outcnt() != 1) {
1471 return;// Multiple branches from 1 compare?
1472 }
1473 Node *iff = bol->unique_out();
1474
1475 // Check some safety conditions
1476 if (iff->is_If()) { // Classic split-if?
1477 if (iff->in(0) != n_ctrl) {
1478 return; // Compare must be in same blk as if
1479 }
1480 } else if (iff->is_CMove()) { // Trying to split-up a CMOVE
1481 // Can't split CMove with different control edge.
1482 if (iff->in(0) != NULL && iff->in(0) != n_ctrl ) {
1483 return;
1484 }
1485 if (get_ctrl(iff->in(2)) == n_ctrl ||
1486 get_ctrl(iff->in(3)) == n_ctrl) {
1487 return; // Inputs not yet split-up
1488 }
1489 if (get_loop(n_ctrl) != get_loop(get_ctrl(iff))) {
1490 return; // Loop-invar test gates loop-varying CMOVE
1491 }
1492 } else {
1493 return; // some other kind of node, such as an Allocate
1494 }
1495
1496 // When is split-if profitable? Every 'win' on means some control flow
1497 // goes dead, so it's almost always a win.
1498 int policy = 0;
1499 // Split compare 'n' through the merge point if it is profitable
1500 Node *phi = split_thru_phi( n, n_ctrl, policy);
1501 if (!phi) {
1502 return;
1503 }
1504
1505 // Found a Phi to split thru!
1506 // Replace 'n' with the new phi
1507 _igvn.replace_node(n, phi);
1508
1509 // Now split the bool up thru the phi
1510 Node *bolphi = split_thru_phi(bol, n_ctrl, -1);
1511 guarantee(bolphi != NULL, "null boolean phi node");
1512
1513 _igvn.replace_node(bol, bolphi);
1514 assert(iff->in(1) == bolphi, "");
1515
1516 if (bolphi->Value(&_igvn)->singleton()) {
1517 return;
1518 }
1519
1520 // Conditional-move? Must split up now
1521 if (!iff->is_If()) {
1522 Node *cmovphi = split_thru_phi(iff, n_ctrl, -1);
1523 _igvn.replace_node(iff, cmovphi);
1524 return;
1525 }
1526
1527 // Now split the IF
1528 do_split_if(iff);
1529 return;
1530 }
1531
1532 // Two identical ifs back to back can be merged
1533 if (try_merge_identical_ifs(n)) {
1534 return;
1535 }
1536
1537 // Check for an IF ready to split; one that has its
1538 // condition codes input coming from a Phi at the block start.
1539 int n_op = n->Opcode();
1540
1541 // Check for an IF being dominated by another IF same test
1542 if (n_op == Op_If ||
1543 n_op == Op_RangeCheck) {
1544 Node *bol = n->in(1);
1545 uint max = bol->outcnt();
1546 // Check for same test used more than once?
1547 if (max > 1 && bol->is_Bool()) {
1548 // Search up IDOMs to see if this IF is dominated.
1549 Node *cutoff = get_ctrl(bol);
1550
1551 // Now search up IDOMs till cutoff, looking for a dominating test
1552 Node *prevdom = n;
1553 Node *dom = idom(prevdom);
1554 while (dom != cutoff) {
1555 if (dom->req() > 1 && dom->in(1) == bol && prevdom->in(0) == dom &&
1556 safe_for_if_replacement(dom)) {
1557 // It's invalid to move control dependent data nodes in the inner
1558 // strip-mined loop, because:
1559 // 1) break validation of LoopNode::verify_strip_mined()
1560 // 2) move code with side-effect in strip-mined loop
1561 // Move to the exit of outer strip-mined loop in that case.
1562 Node* out_le = is_inner_of_stripmined_loop(dom);
1563 if (out_le != NULL) {
1564 prevdom = out_le;
1565 }
1566 // Replace the dominated test with an obvious true or false.
1567 // Place it on the IGVN worklist for later cleanup.
1568 C->set_major_progress();
1569 dominated_by(prevdom->as_IfProj(), n->as_If(), false, true);
1570 #ifndef PRODUCT
1571 if( VerifyLoopOptimizations ) verify();
1572 #endif
1573 return;
1574 }
1575 prevdom = dom;
1576 dom = idom(prevdom);
1577 }
1578 }
1579 }
1580
1581 try_sink_out_of_loop(n);
1582
1583 try_move_store_after_loop(n);
1584
1585 // Remove multiple allocations of the same inline type
1586 if (n->is_InlineType()) {
1587 n->as_InlineType()->remove_redundant_allocations(&_igvn, this);
1588 }
1589
1590 // Check for Opaque2's who's loop has disappeared - who's input is in the
1591 // same loop nest as their output. Remove 'em, they are no longer useful.
1592 if( n_op == Op_Opaque2 &&
1593 n->in(1) != NULL &&
1594 get_loop(get_ctrl(n)) == get_loop(get_ctrl(n->in(1))) ) {
1595 _igvn.replace_node( n, n->in(1) );
1596 }
1597 }
1598
1599 // Tranform:
1600 //
1601 // if (some_condition) {
1602 // // body 1
1603 // } else {
1604 // // body 2
1605 // }
1606 // if (some_condition) {
1607 // // body 3
1608 // } else {
1609 // // body 4
1610 // }
1611 //
1612 // into:
1613 //
1614 //
1615 // if (some_condition) {
1616 // // body 1
1617 // // body 3
1618 // } else {
1619 // // body 2
1620 // // body 4
1621 // }
1622 bool PhaseIdealLoop::try_merge_identical_ifs(Node* n) {
1623 if (identical_backtoback_ifs(n) && can_split_if(n->in(0))) {
1624 Node *n_ctrl = n->in(0);
1625 IfNode* dom_if = idom(n_ctrl)->as_If();
1626 ProjNode* dom_proj_true = dom_if->proj_out(1);
1627 ProjNode* dom_proj_false = dom_if->proj_out(0);
1628
1629 // Now split the IF
1630 RegionNode* new_false_region;
1631 RegionNode* new_true_region;
1632 do_split_if(n, &new_false_region, &new_true_region);
1633 assert(new_false_region->req() == new_true_region->req(), "");
1634 #ifdef ASSERT
1635 for (uint i = 1; i < new_false_region->req(); ++i) {
1636 assert(new_false_region->in(i)->in(0) == new_true_region->in(i)->in(0), "unexpected shape following split if");
1637 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");
1638 }
1639 #endif
1640 assert(new_false_region->in(1)->in(0)->in(1) == dom_if->in(1), "dominating if and dominated if after split must share test");
1641
1642 // We now have:
1643 // if (some_condition) {
1644 // // body 1
1645 // if (some_condition) {
1646 // body3: // new_true_region
1647 // // body3
1648 // } else {
1649 // goto body4;
1650 // }
1651 // } else {
1652 // // body 2
1653 // if (some_condition) {
1654 // goto body3;
1655 // } else {
1656 // body4: // new_false_region
1657 // // body4;
1658 // }
1659 // }
1660 //
1661
1662 // clone pinned nodes thru the resulting regions
1663 push_pinned_nodes_thru_region(dom_if, new_true_region);
1664 push_pinned_nodes_thru_region(dom_if, new_false_region);
1665
1666 // Optimize out the cloned ifs. Because pinned nodes were cloned, this also allows a CastPP that would be dependent
1667 // 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
1668 // unrelated control dependency.
1669 for (uint i = 1; i < new_false_region->req(); i++) {
1670 if (is_dominator(dom_proj_true, new_false_region->in(i))) {
1671 dominated_by(dom_proj_true->as_IfProj(), new_false_region->in(i)->in(0)->as_If(), false, false);
1672 } else {
1673 assert(is_dominator(dom_proj_false, new_false_region->in(i)), "bad if");
1674 dominated_by(dom_proj_false->as_IfProj(), new_false_region->in(i)->in(0)->as_If(), false, false);
1675 }
1676 }
1677 return true;
1678 }
1679 return false;
1680 }
1681
1682 void PhaseIdealLoop::push_pinned_nodes_thru_region(IfNode* dom_if, Node* region) {
1683 for (DUIterator i = region->outs(); region->has_out(i); i++) {
1684 Node* u = region->out(i);
1685 if (!has_ctrl(u) || u->is_Phi() || !u->depends_only_on_test() || !_igvn.no_dependent_zero_check(u)) {
1686 continue;
1687 }
1688 assert(u->in(0) == region, "not a control dependent node?");
1689 uint j = 1;
1690 for (; j < u->req(); ++j) {
1691 Node* in = u->in(j);
1692 if (!is_dominator(ctrl_or_self(in), dom_if)) {
1693 break;
1694 }
1695 }
1696 if (j == u->req()) {
1697 Node *phi = PhiNode::make_blank(region, u);
1698 for (uint k = 1; k < region->req(); ++k) {
1699 Node* clone = u->clone();
1700 clone->set_req(0, region->in(k));
1701 register_new_node(clone, region->in(k));
1702 phi->init_req(k, clone);
1703 }
1704 register_new_node(phi, region);
1705 _igvn.replace_node(u, phi);
1706 --i;
1707 }
1708 }
1709 }
1710
1711 bool PhaseIdealLoop::safe_for_if_replacement(const Node* dom) const {
1712 if (!dom->is_CountedLoopEnd()) {
1713 return true;
1714 }
1715 CountedLoopEndNode* le = dom->as_CountedLoopEnd();
1716 CountedLoopNode* cl = le->loopnode();
1717 if (cl == NULL) {
1718 return true;
1719 }
1720 if (!cl->is_main_loop()) {
1721 return true;
1722 }
1723 if (cl->is_canonical_loop_entry() == NULL) {
1724 return true;
1725 }
1726 // Further unrolling is possible so loop exit condition might change
1727 return false;
1728 }
1729
1730 // See if a shared loop-varying computation has no loop-varying uses.
1731 // Happens if something is only used for JVM state in uncommon trap exits,
1732 // like various versions of induction variable+offset. Clone the
1733 // computation per usage to allow it to sink out of the loop.
1734 void PhaseIdealLoop::try_sink_out_of_loop(Node* n) {
1735 bool is_raw_to_oop_cast = n->is_ConstraintCast() &&
1736 n->in(1)->bottom_type()->isa_rawptr() &&
1737 !n->bottom_type()->isa_rawptr();
1738 if (has_ctrl(n) &&
1739 !n->is_Phi() &&
1740 !n->is_Bool() &&
1741 !n->is_Proj() &&
1742 !n->is_MergeMem() &&
1743 !n->is_CMove() &&
1744 !is_raw_to_oop_cast && // don't extend live ranges of raw oops
1745 n->Opcode() != Op_Opaque4 &&
1746 !n->is_Type()) {
1747 Node *n_ctrl = get_ctrl(n);
1748 IdealLoopTree *n_loop = get_loop(n_ctrl);
1749
1750 if (n->in(0) != NULL) {
1751 IdealLoopTree* loop_ctrl = get_loop(n->in(0));
1752 if (n_loop != loop_ctrl && n_loop->is_member(loop_ctrl)) {
1753 // n has a control input inside a loop but get_ctrl() is member of an outer loop. This could happen, for example,
1754 // for Div nodes inside a loop (control input inside loop) without a use except for an UCT (outside the loop).
1755 // Rewire control of n to right outside of the loop, regardless if its input(s) are later sunk or not.
1756 _igvn.replace_input_of(n, 0, place_outside_loop(n_ctrl, loop_ctrl));
1757 }
1758 }
1759 if (n_loop != _ltree_root && n->outcnt() > 1) {
1760 // Compute early control: needed for anti-dependence analysis. It's also possible that as a result of
1761 // previous transformations in this loop opts round, the node can be hoisted now: early control will tell us.
1762 Node* early_ctrl = compute_early_ctrl(n, n_ctrl);
1763 if (n_loop->is_member(get_loop(early_ctrl)) && // check that this one can't be hoisted now
1764 ctrl_of_all_uses_out_of_loop(n, early_ctrl, n_loop)) { // All uses in outer loops!
1765 assert(!n->is_Store() && !n->is_LoadStore(), "no node with a side effect");
1766 Node* outer_loop_clone = NULL;
1767 for (DUIterator_Last jmin, j = n->last_outs(jmin); j >= jmin;) {
1768 Node* u = n->last_out(j); // Clone private computation per use
1769 _igvn.rehash_node_delayed(u);
1770 Node* x = n->clone(); // Clone computation
1771 Node* x_ctrl = NULL;
1772 if (u->is_Phi()) {
1773 // Replace all uses of normal nodes. Replace Phi uses
1774 // individually, so the separate Nodes can sink down
1775 // different paths.
1776 uint k = 1;
1777 while (u->in(k) != n) k++;
1778 u->set_req(k, x);
1779 // x goes next to Phi input path
1780 x_ctrl = u->in(0)->in(k);
1781 // Find control for 'x' next to use but not inside inner loops.
1782 x_ctrl = place_outside_loop(x_ctrl, n_loop);
1783 --j;
1784 } else { // Normal use
1785 if (has_ctrl(u)) {
1786 x_ctrl = get_ctrl(u);
1787 } else {
1788 x_ctrl = u->in(0);
1789 }
1790 // Find control for 'x' next to use but not inside inner loops.
1791 x_ctrl = place_outside_loop(x_ctrl, n_loop);
1792 // Replace all uses
1793 if (u->is_ConstraintCast() && u->bottom_type()->higher_equal(_igvn.type(n)) && u->in(0) == x_ctrl) {
1794 // If we're sinking a chain of data nodes, we might have inserted a cast to pin the use which is not necessary
1795 // anymore now that we're going to pin n as well
1796 _igvn.replace_node(u, x);
1797 --j;
1798 } else {
1799 int nb = u->replace_edge(n, x, &_igvn);
1800 j -= nb;
1801 }
1802 }
1803
1804 if (n->is_Load()) {
1805 // For loads, add a control edge to a CFG node outside of the loop
1806 // to force them to not combine and return back inside the loop
1807 // during GVN optimization (4641526).
1808 assert(x_ctrl == get_late_ctrl_with_anti_dep(x->as_Load(), early_ctrl, x_ctrl), "anti-dependences were already checked");
1809
1810 IdealLoopTree* x_loop = get_loop(x_ctrl);
1811 Node* x_head = x_loop->_head;
1812 if (x_head->is_Loop() && x_head->is_OuterStripMinedLoop()) {
1813 // Do not add duplicate LoadNodes to the outer strip mined loop
1814 if (outer_loop_clone != NULL) {
1815 _igvn.replace_node(x, outer_loop_clone);
1816 continue;
1817 }
1818 outer_loop_clone = x;
1819 }
1820 x->set_req(0, x_ctrl);
1821 } else if (n->in(0) != NULL){
1822 x->set_req(0, x_ctrl);
1823 }
1824 assert(dom_depth(n_ctrl) <= dom_depth(x_ctrl), "n is later than its clone");
1825 assert(!n_loop->is_member(get_loop(x_ctrl)), "should have moved out of loop");
1826 register_new_node(x, x_ctrl);
1827
1828 // Chain of AddP: (AddP base (AddP base )) must keep the same base after sinking so:
1829 // 1- We don't add a CastPP here when the first one is sunk so if the second one is not, their bases remain
1830 // the same.
1831 // (see 2- below)
1832 assert(!x->is_AddP() || !x->in(AddPNode::Address)->is_AddP() ||
1833 x->in(AddPNode::Address)->in(AddPNode::Base) == x->in(AddPNode::Base) ||
1834 !x->in(AddPNode::Address)->in(AddPNode::Base)->eqv_uncast(x->in(AddPNode::Base)), "unexpected AddP shape");
1835 if (x->in(0) == NULL && !x->is_DecodeNarrowPtr() &&
1836 !(x->is_AddP() && x->in(AddPNode::Address)->is_AddP() && x->in(AddPNode::Address)->in(AddPNode::Base) == x->in(AddPNode::Base))) {
1837 assert(!x->is_Load(), "load should be pinned");
1838 // Use a cast node to pin clone out of loop
1839 Node* cast = NULL;
1840 for (uint k = 0; k < x->req(); k++) {
1841 Node* in = x->in(k);
1842 if (in != NULL && n_loop->is_member(get_loop(get_ctrl(in)))) {
1843 const Type* in_t = _igvn.type(in);
1844 cast = ConstraintCastNode::make_cast_for_type(x_ctrl, in, in_t, ConstraintCastNode::UnconditionalDependency);
1845 }
1846 if (cast != NULL) {
1847 register_new_node(cast, x_ctrl);
1848 x->replace_edge(in, cast);
1849 // Chain of AddP:
1850 // 2- A CastPP of the base is only added now that both AddP nodes are sunk
1851 if (x->is_AddP() && k == AddPNode::Base) {
1852 for (DUIterator_Fast imax, i = x->fast_outs(imax); i < imax; i++) {
1853 Node* u = x->fast_out(i);
1854 if (u->is_AddP() && u->in(AddPNode::Base) == n->in(AddPNode::Base)) {
1855 _igvn.replace_input_of(u, AddPNode::Base, cast);
1856 assert(u->find_out_with(Op_AddP) == NULL, "more than 2 chained AddP nodes?");
1857 }
1858 }
1859 }
1860 break;
1861 }
1862 }
1863 assert(cast != NULL, "must have added a cast to pin the node");
1864 }
1865 }
1866 _igvn.remove_dead_node(n);
1867 }
1868 _dom_lca_tags_round = 0;
1869 }
1870 }
1871 }
1872
1873 Node* PhaseIdealLoop::compute_early_ctrl(Node* n, Node* n_ctrl) {
1874 Node* early_ctrl = NULL;
1875 ResourceMark rm;
1876 Unique_Node_List wq;
1877 wq.push(n);
1878 for (uint i = 0; i < wq.size(); i++) {
1879 Node* m = wq.at(i);
1880 Node* c = NULL;
1881 if (m->is_CFG()) {
1882 c = m;
1883 } else if (m->pinned()) {
1884 c = m->in(0);
1885 } else {
1886 for (uint j = 0; j < m->req(); j++) {
1887 Node* in = m->in(j);
1888 if (in == NULL) {
1889 continue;
1890 }
1891 wq.push(in);
1892 }
1893 }
1894 if (c != NULL) {
1895 assert(is_dominator(c, n_ctrl), "");
1896 if (early_ctrl == NULL) {
1897 early_ctrl = c;
1898 } else if (is_dominator(early_ctrl, c)) {
1899 early_ctrl = c;
1900 }
1901 }
1902 }
1903 assert(is_dominator(early_ctrl, n_ctrl), "early control must dominate current control");
1904 return early_ctrl;
1905 }
1906
1907 bool PhaseIdealLoop::ctrl_of_all_uses_out_of_loop(const Node* n, Node* n_ctrl, IdealLoopTree* n_loop) {
1908 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1909 Node* u = n->fast_out(i);
1910 if (u->Opcode() == Op_Opaque1) {
1911 return false; // Found loop limit, bugfix for 4677003
1912 }
1913 // We can't reuse tags in PhaseIdealLoop::dom_lca_for_get_late_ctrl_internal() so make sure calls to
1914 // get_late_ctrl_with_anti_dep() use their own tag
1915 _dom_lca_tags_round++;
1916 assert(_dom_lca_tags_round != 0, "shouldn't wrap around");
1917
1918 if (u->is_Phi()) {
1919 for (uint j = 1; j < u->req(); ++j) {
1920 if (u->in(j) == n && !ctrl_of_use_out_of_loop(n, n_ctrl, n_loop, u->in(0)->in(j))) {
1921 return false;
1922 }
1923 }
1924 } else {
1925 Node* ctrl = has_ctrl(u) ? get_ctrl(u) : u->in(0);
1926 if (!ctrl_of_use_out_of_loop(n, n_ctrl, n_loop, ctrl)) {
1927 return false;
1928 }
1929 }
1930 }
1931 return true;
1932 }
1933
1934 bool PhaseIdealLoop::ctrl_of_use_out_of_loop(const Node* n, Node* n_ctrl, IdealLoopTree* n_loop, Node* ctrl) {
1935 if (n->is_Load()) {
1936 ctrl = get_late_ctrl_with_anti_dep(n->as_Load(), n_ctrl, ctrl);
1937 }
1938 IdealLoopTree *u_loop = get_loop(ctrl);
1939 if (u_loop == n_loop) {
1940 return false; // Found loop-varying use
1941 }
1942 if (n_loop->is_member(u_loop)) {
1943 return false; // Found use in inner loop
1944 }
1945 return true;
1946 }
1947
1948 //------------------------------split_if_with_blocks---------------------------
1949 // Check for aggressive application of 'split-if' optimization,
1950 // using basic block level info.
1951 void PhaseIdealLoop::split_if_with_blocks(VectorSet &visited, Node_Stack &nstack) {
1952 Node* root = C->root();
1953 visited.set(root->_idx); // first, mark root as visited
1954 // Do pre-visit work for root
1955 Node* n = split_if_with_blocks_pre(root);
1956 uint cnt = n->outcnt();
1957 uint i = 0;
1958
1959 while (true) {
1960 // Visit all children
1961 if (i < cnt) {
1962 Node* use = n->raw_out(i);
1963 ++i;
1964 if (use->outcnt() != 0 && !visited.test_set(use->_idx)) {
1965 // Now do pre-visit work for this use
1966 use = split_if_with_blocks_pre(use);
1967 nstack.push(n, i); // Save parent and next use's index.
1968 n = use; // Process all children of current use.
1969 cnt = use->outcnt();
1970 i = 0;
1971 }
1972 }
1973 else {
1974 // All of n's children have been processed, complete post-processing.
1975 if (cnt != 0 && !n->is_Con()) {
1976 assert(has_node(n), "no dead nodes");
1977 split_if_with_blocks_post(n);
1978 }
1979 if (must_throttle_split_if()) {
1980 nstack.clear();
1981 }
1982 if (nstack.is_empty()) {
1983 // Finished all nodes on stack.
1984 break;
1985 }
1986 // Get saved parent node and next use's index. Visit the rest of uses.
1987 n = nstack.node();
1988 cnt = n->outcnt();
1989 i = nstack.index();
1990 nstack.pop();
1991 }
1992 }
1993 }
1994
1995
1996 //=============================================================================
1997 //
1998 // C L O N E A L O O P B O D Y
1999 //
2000
2001 //------------------------------clone_iff--------------------------------------
2002 // Passed in a Phi merging (recursively) some nearly equivalent Bool/Cmps.
2003 // "Nearly" because all Nodes have been cloned from the original in the loop,
2004 // but the fall-in edges to the Cmp are different. Clone bool/Cmp pairs
2005 // through the Phi recursively, and return a Bool.
2006 Node* PhaseIdealLoop::clone_iff(PhiNode *phi, IdealLoopTree *loop) {
2007
2008 // Convert this Phi into a Phi merging Bools
2009 uint i;
2010 for (i = 1; i < phi->req(); i++) {
2011 Node *b = phi->in(i);
2012 if (b->is_Phi()) {
2013 _igvn.replace_input_of(phi, i, clone_iff(b->as_Phi(), loop));
2014 } else {
2015 assert(b->is_Bool() || b->Opcode() == Op_Opaque4, "");
2016 }
2017 }
2018
2019 Node* n = phi->in(1);
2020 Node* sample_opaque = NULL;
2021 Node *sample_bool = NULL;
2022 if (n->Opcode() == Op_Opaque4) {
2023 sample_opaque = n;
2024 sample_bool = n->in(1);
2025 assert(sample_bool->is_Bool(), "wrong type");
2026 } else {
2027 sample_bool = n;
2028 }
2029 Node* sample_cmp = sample_bool->in(1);
2030 const Type* t = Type::TOP;
2031 const TypePtr* at = NULL;
2032 if (sample_cmp->is_FlatArrayCheck()) {
2033 // Left input of a FlatArrayCheckNode is memory, set the (adr) type of the phi accordingly
2034 assert(sample_cmp->in(1)->bottom_type() == Type::MEMORY, "unexpected input type");
2035 t = Type::MEMORY;
2036 at = TypeRawPtr::BOTTOM;
2037 }
2038
2039 // Make Phis to merge the Cmp's inputs.
2040 PhiNode *phi1 = new PhiNode(phi->in(0), t, at);
2041 PhiNode *phi2 = new PhiNode(phi->in(0), Type::TOP);
2042 for (i = 1; i < phi->req(); i++) {
2043 Node *n1 = sample_opaque == NULL ? phi->in(i)->in(1)->in(1) : phi->in(i)->in(1)->in(1)->in(1);
2044 Node *n2 = sample_opaque == NULL ? phi->in(i)->in(1)->in(2) : phi->in(i)->in(1)->in(1)->in(2);
2045 phi1->set_req(i, n1);
2046 phi2->set_req(i, n2);
2047 phi1->set_type(phi1->type()->meet_speculative(n1->bottom_type()));
2048 phi2->set_type(phi2->type()->meet_speculative(n2->bottom_type()));
2049 }
2050 // See if these Phis have been made before.
2051 // Register with optimizer
2052 Node *hit1 = _igvn.hash_find_insert(phi1);
2053 if (hit1) { // Hit, toss just made Phi
2054 _igvn.remove_dead_node(phi1); // Remove new phi
2055 assert(hit1->is_Phi(), "" );
2056 phi1 = (PhiNode*)hit1; // Use existing phi
2057 } else { // Miss
2058 _igvn.register_new_node_with_optimizer(phi1);
2059 }
2060 Node *hit2 = _igvn.hash_find_insert(phi2);
2061 if (hit2) { // Hit, toss just made Phi
2062 _igvn.remove_dead_node(phi2); // Remove new phi
2063 assert(hit2->is_Phi(), "" );
2064 phi2 = (PhiNode*)hit2; // Use existing phi
2065 } else { // Miss
2066 _igvn.register_new_node_with_optimizer(phi2);
2067 }
2068 // Register Phis with loop/block info
2069 set_ctrl(phi1, phi->in(0));
2070 set_ctrl(phi2, phi->in(0));
2071 // Make a new Cmp
2072 Node *cmp = sample_cmp->clone();
2073 cmp->set_req(1, phi1);
2074 cmp->set_req(2, phi2);
2075 _igvn.register_new_node_with_optimizer(cmp);
2076 set_ctrl(cmp, phi->in(0));
2077
2078 // Make a new Bool
2079 Node *b = sample_bool->clone();
2080 b->set_req(1,cmp);
2081 _igvn.register_new_node_with_optimizer(b);
2082 set_ctrl(b, phi->in(0));
2083
2084 if (sample_opaque != NULL) {
2085 Node* opaque = sample_opaque->clone();
2086 opaque->set_req(1, b);
2087 _igvn.register_new_node_with_optimizer(opaque);
2088 set_ctrl(opaque, phi->in(0));
2089 return opaque;
2090 }
2091
2092 assert(b->is_Bool(), "");
2093 return b;
2094 }
2095
2096 //------------------------------clone_bool-------------------------------------
2097 // Passed in a Phi merging (recursively) some nearly equivalent Bool/Cmps.
2098 // "Nearly" because all Nodes have been cloned from the original in the loop,
2099 // but the fall-in edges to the Cmp are different. Clone bool/Cmp pairs
2100 // through the Phi recursively, and return a Bool.
2101 CmpNode *PhaseIdealLoop::clone_bool( PhiNode *phi, IdealLoopTree *loop ) {
2102 uint i;
2103 // Convert this Phi into a Phi merging Bools
2104 for( i = 1; i < phi->req(); i++ ) {
2105 Node *b = phi->in(i);
2106 if( b->is_Phi() ) {
2107 _igvn.replace_input_of(phi, i, clone_bool( b->as_Phi(), loop ));
2108 } else {
2109 assert( b->is_Cmp() || b->is_top(), "inputs are all Cmp or TOP" );
2110 }
2111 }
2112
2113 Node *sample_cmp = phi->in(1);
2114
2115 // Make Phis to merge the Cmp's inputs.
2116 PhiNode *phi1 = new PhiNode( phi->in(0), Type::TOP );
2117 PhiNode *phi2 = new PhiNode( phi->in(0), Type::TOP );
2118 for( uint j = 1; j < phi->req(); j++ ) {
2119 Node *cmp_top = phi->in(j); // Inputs are all Cmp or TOP
2120 Node *n1, *n2;
2121 if( cmp_top->is_Cmp() ) {
2122 n1 = cmp_top->in(1);
2123 n2 = cmp_top->in(2);
2124 } else {
2125 n1 = n2 = cmp_top;
2126 }
2127 phi1->set_req( j, n1 );
2128 phi2->set_req( j, n2 );
2129 phi1->set_type(phi1->type()->meet_speculative(n1->bottom_type()));
2130 phi2->set_type(phi2->type()->meet_speculative(n2->bottom_type()));
2131 }
2132
2133 // See if these Phis have been made before.
2134 // Register with optimizer
2135 Node *hit1 = _igvn.hash_find_insert(phi1);
2136 if( hit1 ) { // Hit, toss just made Phi
2137 _igvn.remove_dead_node(phi1); // Remove new phi
2138 assert( hit1->is_Phi(), "" );
2139 phi1 = (PhiNode*)hit1; // Use existing phi
2140 } else { // Miss
2141 _igvn.register_new_node_with_optimizer(phi1);
2142 }
2143 Node *hit2 = _igvn.hash_find_insert(phi2);
2144 if( hit2 ) { // Hit, toss just made Phi
2145 _igvn.remove_dead_node(phi2); // Remove new phi
2146 assert( hit2->is_Phi(), "" );
2147 phi2 = (PhiNode*)hit2; // Use existing phi
2148 } else { // Miss
2149 _igvn.register_new_node_with_optimizer(phi2);
2150 }
2151 // Register Phis with loop/block info
2152 set_ctrl(phi1, phi->in(0));
2153 set_ctrl(phi2, phi->in(0));
2154 // Make a new Cmp
2155 Node *cmp = sample_cmp->clone();
2156 cmp->set_req( 1, phi1 );
2157 cmp->set_req( 2, phi2 );
2158 _igvn.register_new_node_with_optimizer(cmp);
2159 set_ctrl(cmp, phi->in(0));
2160
2161 assert( cmp->is_Cmp(), "" );
2162 return (CmpNode*)cmp;
2163 }
2164
2165 //------------------------------sink_use---------------------------------------
2166 // If 'use' was in the loop-exit block, it now needs to be sunk
2167 // below the post-loop merge point.
2168 void PhaseIdealLoop::sink_use( Node *use, Node *post_loop ) {
2169 if (!use->is_CFG() && get_ctrl(use) == post_loop->in(2)) {
2170 set_ctrl(use, post_loop);
2171 for (DUIterator j = use->outs(); use->has_out(j); j++)
2172 sink_use(use->out(j), post_loop);
2173 }
2174 }
2175
2176 void PhaseIdealLoop::clone_loop_handle_data_uses(Node* old, Node_List &old_new,
2177 IdealLoopTree* loop, IdealLoopTree* outer_loop,
2178 Node_List*& split_if_set, Node_List*& split_bool_set,
2179 Node_List*& split_cex_set, Node_List& worklist,
2180 uint new_counter, CloneLoopMode mode) {
2181 Node* nnn = old_new[old->_idx];
2182 // Copy uses to a worklist, so I can munge the def-use info
2183 // with impunity.
2184 for (DUIterator_Fast jmax, j = old->fast_outs(jmax); j < jmax; j++)
2185 worklist.push(old->fast_out(j));
2186
2187 while( worklist.size() ) {
2188 Node *use = worklist.pop();
2189 if (!has_node(use)) continue; // Ignore dead nodes
2190 if (use->in(0) == C->top()) continue;
2191 IdealLoopTree *use_loop = get_loop( has_ctrl(use) ? get_ctrl(use) : use );
2192 // Check for data-use outside of loop - at least one of OLD or USE
2193 // must not be a CFG node.
2194 #ifdef ASSERT
2195 if (loop->_head->as_Loop()->is_strip_mined() && outer_loop->is_member(use_loop) && !loop->is_member(use_loop) && old_new[use->_idx] == NULL) {
2196 Node* sfpt = loop->_head->as_CountedLoop()->outer_safepoint();
2197 assert(mode != IgnoreStripMined, "incorrect cloning mode");
2198 assert((mode == ControlAroundStripMined && use == sfpt) || !use->is_reachable_from_root(), "missed a node");
2199 }
2200 #endif
2201 if (!loop->is_member(use_loop) && !outer_loop->is_member(use_loop) && (!old->is_CFG() || !use->is_CFG())) {
2202
2203 // If the Data use is an IF, that means we have an IF outside of the
2204 // loop that is switching on a condition that is set inside of the
2205 // loop. Happens if people set a loop-exit flag; then test the flag
2206 // in the loop to break the loop, then test is again outside of the
2207 // loop to determine which way the loop exited.
2208 // Loop predicate If node connects to Bool node through Opaque1 node.
2209 if (use->is_If() || use->is_CMove() || C->is_predicate_opaq(use) || use->Opcode() == Op_Opaque4) {
2210 // Since this code is highly unlikely, we lazily build the worklist
2211 // of such Nodes to go split.
2212 if (!split_if_set) {
2213 split_if_set = new Node_List();
2214 }
2215 split_if_set->push(use);
2216 }
2217 if (use->is_Bool()) {
2218 if (!split_bool_set) {
2219 split_bool_set = new Node_List();
2220 }
2221 split_bool_set->push(use);
2222 }
2223 if (use->Opcode() == Op_CreateEx) {
2224 if (!split_cex_set) {
2225 split_cex_set = new Node_List();
2226 }
2227 split_cex_set->push(use);
2228 }
2229
2230
2231 // Get "block" use is in
2232 uint idx = 0;
2233 while( use->in(idx) != old ) idx++;
2234 Node *prev = use->is_CFG() ? use : get_ctrl(use);
2235 assert(!loop->is_member(get_loop(prev)) && !outer_loop->is_member(get_loop(prev)), "" );
2236 Node *cfg = prev->_idx >= new_counter
2237 ? prev->in(2)
2238 : idom(prev);
2239 if( use->is_Phi() ) // Phi use is in prior block
2240 cfg = prev->in(idx); // NOT in block of Phi itself
2241 if (cfg->is_top()) { // Use is dead?
2242 _igvn.replace_input_of(use, idx, C->top());
2243 continue;
2244 }
2245
2246 // If use is referenced through control edge... (idx == 0)
2247 if (mode == IgnoreStripMined && idx == 0) {
2248 LoopNode *head = loop->_head->as_Loop();
2249 if (head->is_strip_mined() && is_dominator(head->outer_loop_exit(), prev)) {
2250 // That node is outside the inner loop, leave it outside the
2251 // outer loop as well to not confuse verification code.
2252 assert(!loop->_parent->is_member(use_loop), "should be out of the outer loop");
2253 _igvn.replace_input_of(use, 0, head->outer_loop_exit());
2254 continue;
2255 }
2256 }
2257
2258 while(!outer_loop->is_member(get_loop(cfg))) {
2259 prev = cfg;
2260 cfg = cfg->_idx >= new_counter ? cfg->in(2) : idom(cfg);
2261 }
2262 // If the use occurs after merging several exits from the loop, then
2263 // old value must have dominated all those exits. Since the same old
2264 // value was used on all those exits we did not need a Phi at this
2265 // merge point. NOW we do need a Phi here. Each loop exit value
2266 // is now merged with the peeled body exit; each exit gets its own
2267 // private Phi and those Phis need to be merged here.
2268 Node *phi;
2269 if( prev->is_Region() ) {
2270 if( idx == 0 ) { // Updating control edge?
2271 phi = prev; // Just use existing control
2272 } else { // Else need a new Phi
2273 phi = PhiNode::make( prev, old );
2274 // Now recursively fix up the new uses of old!
2275 for( uint i = 1; i < prev->req(); i++ ) {
2276 worklist.push(phi); // Onto worklist once for each 'old' input
2277 }
2278 }
2279 } else {
2280 // Get new RegionNode merging old and new loop exits
2281 prev = old_new[prev->_idx];
2282 assert( prev, "just made this in step 7" );
2283 if( idx == 0) { // Updating control edge?
2284 phi = prev; // Just use existing control
2285 } else { // Else need a new Phi
2286 // Make a new Phi merging data values properly
2287 phi = PhiNode::make( prev, old );
2288 phi->set_req( 1, nnn );
2289 }
2290 }
2291 // If inserting a new Phi, check for prior hits
2292 if( idx != 0 ) {
2293 Node *hit = _igvn.hash_find_insert(phi);
2294 if( hit == NULL ) {
2295 _igvn.register_new_node_with_optimizer(phi); // Register new phi
2296 } else { // or
2297 // Remove the new phi from the graph and use the hit
2298 _igvn.remove_dead_node(phi);
2299 phi = hit; // Use existing phi
2300 }
2301 set_ctrl(phi, prev);
2302 }
2303 // Make 'use' use the Phi instead of the old loop body exit value
2304 _igvn.replace_input_of(use, idx, phi);
2305 if( use->_idx >= new_counter ) { // If updating new phis
2306 // Not needed for correctness, but prevents a weak assert
2307 // in AddPNode from tripping (when we end up with different
2308 // base & derived Phis that will become the same after
2309 // IGVN does CSE).
2310 Node *hit = _igvn.hash_find_insert(use);
2311 if( hit ) // Go ahead and re-hash for hits.
2312 _igvn.replace_node( use, hit );
2313 }
2314
2315 // If 'use' was in the loop-exit block, it now needs to be sunk
2316 // below the post-loop merge point.
2317 sink_use( use, prev );
2318 }
2319 }
2320 }
2321
2322 static void clone_outer_loop_helper(Node* n, const IdealLoopTree *loop, const IdealLoopTree* outer_loop,
2323 const Node_List &old_new, Unique_Node_List& wq, PhaseIdealLoop* phase,
2324 bool check_old_new) {
2325 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
2326 Node* u = n->fast_out(j);
2327 assert(check_old_new || old_new[u->_idx] == NULL, "shouldn't have been cloned");
2328 if (!u->is_CFG() && (!check_old_new || old_new[u->_idx] == NULL)) {
2329 Node* c = phase->get_ctrl(u);
2330 IdealLoopTree* u_loop = phase->get_loop(c);
2331 assert(!loop->is_member(u_loop), "can be in outer loop or out of both loops only");
2332 if (outer_loop->is_member(u_loop) ||
2333 // nodes pinned with control in the outer loop but not referenced from the safepoint must be moved out of
2334 // the outer loop too
2335 (u->in(0) != NULL && outer_loop->is_member(phase->get_loop(u->in(0))))) {
2336 wq.push(u);
2337 }
2338 }
2339 }
2340 }
2341
2342 void PhaseIdealLoop::clone_outer_loop(LoopNode* head, CloneLoopMode mode, IdealLoopTree *loop,
2343 IdealLoopTree* outer_loop, int dd, Node_List &old_new,
2344 Node_List& extra_data_nodes) {
2345 if (head->is_strip_mined() && mode != IgnoreStripMined) {
2346 CountedLoopNode* cl = head->as_CountedLoop();
2347 Node* l = cl->outer_loop();
2348 Node* tail = cl->outer_loop_tail();
2349 IfNode* le = cl->outer_loop_end();
2350 Node* sfpt = cl->outer_safepoint();
2351 CountedLoopEndNode* cle = cl->loopexit();
2352 CountedLoopNode* new_cl = old_new[cl->_idx]->as_CountedLoop();
2353 CountedLoopEndNode* new_cle = new_cl->as_CountedLoop()->loopexit_or_null();
2354 Node* cle_out = cle->proj_out(false);
2355
2356 Node* new_sfpt = NULL;
2357 Node* new_cle_out = cle_out->clone();
2358 old_new.map(cle_out->_idx, new_cle_out);
2359 if (mode == CloneIncludesStripMined) {
2360 // clone outer loop body
2361 Node* new_l = l->clone();
2362 Node* new_tail = tail->clone();
2363 IfNode* new_le = le->clone()->as_If();
2364 new_sfpt = sfpt->clone();
2365
2366 set_loop(new_l, outer_loop->_parent);
2367 set_idom(new_l, new_l->in(LoopNode::EntryControl), dd);
2368 set_loop(new_cle_out, outer_loop->_parent);
2369 set_idom(new_cle_out, new_cle, dd);
2370 set_loop(new_sfpt, outer_loop->_parent);
2371 set_idom(new_sfpt, new_cle_out, dd);
2372 set_loop(new_le, outer_loop->_parent);
2373 set_idom(new_le, new_sfpt, dd);
2374 set_loop(new_tail, outer_loop->_parent);
2375 set_idom(new_tail, new_le, dd);
2376 set_idom(new_cl, new_l, dd);
2377
2378 old_new.map(l->_idx, new_l);
2379 old_new.map(tail->_idx, new_tail);
2380 old_new.map(le->_idx, new_le);
2381 old_new.map(sfpt->_idx, new_sfpt);
2382
2383 new_l->set_req(LoopNode::LoopBackControl, new_tail);
2384 new_l->set_req(0, new_l);
2385 new_tail->set_req(0, new_le);
2386 new_le->set_req(0, new_sfpt);
2387 new_sfpt->set_req(0, new_cle_out);
2388 new_cle_out->set_req(0, new_cle);
2389 new_cl->set_req(LoopNode::EntryControl, new_l);
2390
2391 _igvn.register_new_node_with_optimizer(new_l);
2392 _igvn.register_new_node_with_optimizer(new_tail);
2393 _igvn.register_new_node_with_optimizer(new_le);
2394 } else {
2395 Node *newhead = old_new[loop->_head->_idx];
2396 newhead->as_Loop()->clear_strip_mined();
2397 _igvn.replace_input_of(newhead, LoopNode::EntryControl, newhead->in(LoopNode::EntryControl)->in(LoopNode::EntryControl));
2398 set_idom(newhead, newhead->in(LoopNode::EntryControl), dd);
2399 }
2400 // Look at data node that were assigned a control in the outer
2401 // loop: they are kept in the outer loop by the safepoint so start
2402 // from the safepoint node's inputs.
2403 IdealLoopTree* outer_loop = get_loop(l);
2404 Node_Stack stack(2);
2405 stack.push(sfpt, 1);
2406 uint new_counter = C->unique();
2407 while (stack.size() > 0) {
2408 Node* n = stack.node();
2409 uint i = stack.index();
2410 while (i < n->req() &&
2411 (n->in(i) == NULL ||
2412 !has_ctrl(n->in(i)) ||
2413 get_loop(get_ctrl(n->in(i))) != outer_loop ||
2414 (old_new[n->in(i)->_idx] != NULL && old_new[n->in(i)->_idx]->_idx >= new_counter))) {
2415 i++;
2416 }
2417 if (i < n->req()) {
2418 stack.set_index(i+1);
2419 stack.push(n->in(i), 0);
2420 } else {
2421 assert(old_new[n->_idx] == NULL || n == sfpt || old_new[n->_idx]->_idx < new_counter, "no clone yet");
2422 Node* m = n == sfpt ? new_sfpt : n->clone();
2423 if (m != NULL) {
2424 for (uint i = 0; i < n->req(); i++) {
2425 if (m->in(i) != NULL && old_new[m->in(i)->_idx] != NULL) {
2426 m->set_req(i, old_new[m->in(i)->_idx]);
2427 }
2428 }
2429 } else {
2430 assert(n == sfpt && mode != CloneIncludesStripMined, "where's the safepoint clone?");
2431 }
2432 if (n != sfpt) {
2433 extra_data_nodes.push(n);
2434 _igvn.register_new_node_with_optimizer(m);
2435 assert(get_ctrl(n) == cle_out, "what other control?");
2436 set_ctrl(m, new_cle_out);
2437 old_new.map(n->_idx, m);
2438 }
2439 stack.pop();
2440 }
2441 }
2442 if (mode == CloneIncludesStripMined) {
2443 _igvn.register_new_node_with_optimizer(new_sfpt);
2444 _igvn.register_new_node_with_optimizer(new_cle_out);
2445 }
2446 // Some other transformation may have pessimistically assign some
2447 // data nodes to the outer loop. Set their control so they are out
2448 // of the outer loop.
2449 ResourceMark rm;
2450 Unique_Node_List wq;
2451 for (uint i = 0; i < extra_data_nodes.size(); i++) {
2452 Node* old = extra_data_nodes.at(i);
2453 clone_outer_loop_helper(old, loop, outer_loop, old_new, wq, this, true);
2454 }
2455
2456 Node* inner_out = sfpt->in(0);
2457 if (inner_out->outcnt() > 1) {
2458 clone_outer_loop_helper(inner_out, loop, outer_loop, old_new, wq, this, true);
2459 }
2460
2461 Node* new_ctrl = cl->outer_loop_exit();
2462 assert(get_loop(new_ctrl) != outer_loop, "must be out of the loop nest");
2463 for (uint i = 0; i < wq.size(); i++) {
2464 Node* n = wq.at(i);
2465 set_ctrl(n, new_ctrl);
2466 if (n->in(0) != NULL) {
2467 _igvn.replace_input_of(n, 0, new_ctrl);
2468 }
2469 clone_outer_loop_helper(n, loop, outer_loop, old_new, wq, this, false);
2470 }
2471 } else {
2472 Node *newhead = old_new[loop->_head->_idx];
2473 set_idom(newhead, newhead->in(LoopNode::EntryControl), dd);
2474 }
2475 }
2476
2477 //------------------------------clone_loop-------------------------------------
2478 //
2479 // C L O N E A L O O P B O D Y
2480 //
2481 // This is the basic building block of the loop optimizations. It clones an
2482 // entire loop body. It makes an old_new loop body mapping; with this mapping
2483 // you can find the new-loop equivalent to an old-loop node. All new-loop
2484 // nodes are exactly equal to their old-loop counterparts, all edges are the
2485 // same. All exits from the old-loop now have a RegionNode that merges the
2486 // equivalent new-loop path. This is true even for the normal "loop-exit"
2487 // condition. All uses of loop-invariant old-loop values now come from (one
2488 // or more) Phis that merge their new-loop equivalents.
2489 //
2490 // This operation leaves the graph in an illegal state: there are two valid
2491 // control edges coming from the loop pre-header to both loop bodies. I'll
2492 // definitely have to hack the graph after running this transform.
2493 //
2494 // From this building block I will further edit edges to perform loop peeling
2495 // or loop unrolling or iteration splitting (Range-Check-Elimination), etc.
2496 //
2497 // Parameter side_by_size_idom:
2498 // When side_by_size_idom is NULL, the dominator tree is constructed for
2499 // the clone loop to dominate the original. Used in construction of
2500 // pre-main-post loop sequence.
2501 // When nonnull, the clone and original are side-by-side, both are
2502 // dominated by the side_by_side_idom node. Used in construction of
2503 // unswitched loops.
2504 void PhaseIdealLoop::clone_loop( IdealLoopTree *loop, Node_List &old_new, int dd,
2505 CloneLoopMode mode, Node* side_by_side_idom) {
2506
2507 LoopNode* head = loop->_head->as_Loop();
2508 head->verify_strip_mined(1);
2509
2510 if (C->do_vector_loop() && PrintOpto) {
2511 const char* mname = C->method()->name()->as_quoted_ascii();
2512 if (mname != NULL) {
2513 tty->print("PhaseIdealLoop::clone_loop: for vectorize method %s\n", mname);
2514 }
2515 }
2516
2517 CloneMap& cm = C->clone_map();
2518 Dict* dict = cm.dict();
2519 if (C->do_vector_loop()) {
2520 cm.set_clone_idx(cm.max_gen()+1);
2521 #ifndef PRODUCT
2522 if (PrintOpto) {
2523 tty->print_cr("PhaseIdealLoop::clone_loop: _clone_idx %d", cm.clone_idx());
2524 loop->dump_head();
2525 }
2526 #endif
2527 }
2528
2529 // Step 1: Clone the loop body. Make the old->new mapping.
2530 uint i;
2531 for (i = 0; i < loop->_body.size(); i++) {
2532 Node* old = loop->_body.at(i);
2533 Node* nnn = old->clone();
2534 old_new.map(old->_idx, nnn);
2535 if (old->is_reduction()) {
2536 // Reduction flag is not copied by default. Copy it here when cloning the entire loop body.
2537 nnn->add_flag(Node::Flag_is_reduction);
2538 }
2539 if (C->do_vector_loop()) {
2540 cm.verify_insert_and_clone(old, nnn, cm.clone_idx());
2541 }
2542 _igvn.register_new_node_with_optimizer(nnn);
2543 }
2544
2545 IdealLoopTree* outer_loop = (head->is_strip_mined() && mode != IgnoreStripMined) ? get_loop(head->as_CountedLoop()->outer_loop()) : loop;
2546
2547 // Step 2: Fix the edges in the new body. If the old input is outside the
2548 // loop use it. If the old input is INside the loop, use the corresponding
2549 // new node instead.
2550 for( i = 0; i < loop->_body.size(); i++ ) {
2551 Node *old = loop->_body.at(i);
2552 Node *nnn = old_new[old->_idx];
2553 // Fix CFG/Loop controlling the new node
2554 if (has_ctrl(old)) {
2555 set_ctrl(nnn, old_new[get_ctrl(old)->_idx]);
2556 } else {
2557 set_loop(nnn, outer_loop->_parent);
2558 if (old->outcnt() > 0) {
2559 set_idom( nnn, old_new[idom(old)->_idx], dd );
2560 }
2561 }
2562 // Correct edges to the new node
2563 for( uint j = 0; j < nnn->req(); j++ ) {
2564 Node *n = nnn->in(j);
2565 if( n ) {
2566 IdealLoopTree *old_in_loop = get_loop( has_ctrl(n) ? get_ctrl(n) : n );
2567 if( loop->is_member( old_in_loop ) )
2568 nnn->set_req(j, old_new[n->_idx]);
2569 }
2570 }
2571 _igvn.hash_find_insert(nnn);
2572 }
2573
2574 Node_List extra_data_nodes; // data nodes in the outer strip mined loop
2575 clone_outer_loop(head, mode, loop, outer_loop, dd, old_new, extra_data_nodes);
2576
2577 // Step 3: Now fix control uses. Loop varying control uses have already
2578 // been fixed up (as part of all input edges in Step 2). Loop invariant
2579 // control uses must be either an IfFalse or an IfTrue. Make a merge
2580 // point to merge the old and new IfFalse/IfTrue nodes; make the use
2581 // refer to this.
2582 Node_List worklist;
2583 uint new_counter = C->unique();
2584 for( i = 0; i < loop->_body.size(); i++ ) {
2585 Node* old = loop->_body.at(i);
2586 if( !old->is_CFG() ) continue;
2587
2588 // Copy uses to a worklist, so I can munge the def-use info
2589 // with impunity.
2590 for (DUIterator_Fast jmax, j = old->fast_outs(jmax); j < jmax; j++)
2591 worklist.push(old->fast_out(j));
2592
2593 while( worklist.size() ) { // Visit all uses
2594 Node *use = worklist.pop();
2595 if (!has_node(use)) continue; // Ignore dead nodes
2596 IdealLoopTree *use_loop = get_loop( has_ctrl(use) ? get_ctrl(use) : use );
2597 if( !loop->is_member( use_loop ) && use->is_CFG() ) {
2598 // Both OLD and USE are CFG nodes here.
2599 assert( use->is_Proj(), "" );
2600 Node* nnn = old_new[old->_idx];
2601
2602 Node* newuse = NULL;
2603 if (head->is_strip_mined() && mode != IgnoreStripMined) {
2604 CountedLoopNode* cl = head->as_CountedLoop();
2605 CountedLoopEndNode* cle = cl->loopexit();
2606 Node* cle_out = cle->proj_out_or_null(false);
2607 if (use == cle_out) {
2608 IfNode* le = cl->outer_loop_end();
2609 use = le->proj_out(false);
2610 use_loop = get_loop(use);
2611 if (mode == CloneIncludesStripMined) {
2612 nnn = old_new[le->_idx];
2613 } else {
2614 newuse = old_new[cle_out->_idx];
2615 }
2616 }
2617 }
2618 if (newuse == NULL) {
2619 newuse = use->clone();
2620 }
2621
2622 // Clone the loop exit control projection
2623 if (C->do_vector_loop()) {
2624 cm.verify_insert_and_clone(use, newuse, cm.clone_idx());
2625 }
2626 newuse->set_req(0,nnn);
2627 _igvn.register_new_node_with_optimizer(newuse);
2628 set_loop(newuse, use_loop);
2629 set_idom(newuse, nnn, dom_depth(nnn) + 1 );
2630
2631 // We need a Region to merge the exit from the peeled body and the
2632 // exit from the old loop body.
2633 RegionNode *r = new RegionNode(3);
2634 // Map the old use to the new merge point
2635 old_new.map( use->_idx, r );
2636 uint dd_r = MIN2(dom_depth(newuse),dom_depth(use));
2637 assert( dd_r >= dom_depth(dom_lca(newuse,use)), "" );
2638
2639 // The original user of 'use' uses 'r' instead.
2640 for (DUIterator_Last lmin, l = use->last_outs(lmin); l >= lmin;) {
2641 Node* useuse = use->last_out(l);
2642 _igvn.rehash_node_delayed(useuse);
2643 uint uses_found = 0;
2644 if( useuse->in(0) == use ) {
2645 useuse->set_req(0, r);
2646 uses_found++;
2647 if( useuse->is_CFG() ) {
2648 // This is not a dom_depth > dd_r because when new
2649 // control flow is constructed by a loop opt, a node and
2650 // its dominator can end up at the same dom_depth
2651 assert(dom_depth(useuse) >= dd_r, "");
2652 set_idom(useuse, r, dom_depth(useuse));
2653 }
2654 }
2655 for( uint k = 1; k < useuse->req(); k++ ) {
2656 if( useuse->in(k) == use ) {
2657 useuse->set_req(k, r);
2658 uses_found++;
2659 if (useuse->is_Loop() && k == LoopNode::EntryControl) {
2660 // This is not a dom_depth > dd_r because when new
2661 // control flow is constructed by a loop opt, a node
2662 // and its dominator can end up at the same dom_depth
2663 assert(dom_depth(useuse) >= dd_r , "");
2664 set_idom(useuse, r, dom_depth(useuse));
2665 }
2666 }
2667 }
2668 l -= uses_found; // we deleted 1 or more copies of this edge
2669 }
2670
2671 // Now finish up 'r'
2672 r->set_req( 1, newuse );
2673 r->set_req( 2, use );
2674 _igvn.register_new_node_with_optimizer(r);
2675 set_loop(r, use_loop);
2676 set_idom(r, !side_by_side_idom ? newuse->in(0) : side_by_side_idom, dd_r);
2677 } // End of if a loop-exit test
2678 }
2679 }
2680
2681 // Step 4: If loop-invariant use is not control, it must be dominated by a
2682 // loop exit IfFalse/IfTrue. Find "proper" loop exit. Make a Region
2683 // there if needed. Make a Phi there merging old and new used values.
2684 Node_List *split_if_set = NULL;
2685 Node_List *split_bool_set = NULL;
2686 Node_List *split_cex_set = NULL;
2687 for( i = 0; i < loop->_body.size(); i++ ) {
2688 Node* old = loop->_body.at(i);
2689 clone_loop_handle_data_uses(old, old_new, loop, outer_loop, split_if_set,
2690 split_bool_set, split_cex_set, worklist, new_counter,
2691 mode);
2692 }
2693
2694 for (i = 0; i < extra_data_nodes.size(); i++) {
2695 Node* old = extra_data_nodes.at(i);
2696 clone_loop_handle_data_uses(old, old_new, loop, outer_loop, split_if_set,
2697 split_bool_set, split_cex_set, worklist, new_counter,
2698 mode);
2699 }
2700
2701 // Check for IFs that need splitting/cloning. Happens if an IF outside of
2702 // the loop uses a condition set in the loop. The original IF probably
2703 // takes control from one or more OLD Regions (which in turn get from NEW
2704 // Regions). In any case, there will be a set of Phis for each merge point
2705 // from the IF up to where the original BOOL def exists the loop.
2706 if (split_if_set) {
2707 while (split_if_set->size()) {
2708 Node *iff = split_if_set->pop();
2709 if (iff->in(1)->is_Phi()) {
2710 Node *b = clone_iff(iff->in(1)->as_Phi(), loop);
2711 _igvn.replace_input_of(iff, 1, b);
2712 }
2713 }
2714 }
2715 if (split_bool_set) {
2716 while (split_bool_set->size()) {
2717 Node *b = split_bool_set->pop();
2718 Node *phi = b->in(1);
2719 assert(phi->is_Phi(), "");
2720 CmpNode *cmp = clone_bool((PhiNode*)phi, loop);
2721 _igvn.replace_input_of(b, 1, cmp);
2722 }
2723 }
2724 if (split_cex_set) {
2725 while (split_cex_set->size()) {
2726 Node *b = split_cex_set->pop();
2727 assert(b->in(0)->is_Region(), "");
2728 assert(b->in(1)->is_Phi(), "");
2729 assert(b->in(0)->in(0) == b->in(1)->in(0), "");
2730 split_up(b, b->in(0), NULL);
2731 }
2732 }
2733
2734 }
2735
2736
2737 //---------------------- stride_of_possible_iv -------------------------------------
2738 // Looks for an iff/bool/comp with one operand of the compare
2739 // being a cycle involving an add and a phi,
2740 // with an optional truncation (left-shift followed by a right-shift)
2741 // of the add. Returns zero if not an iv.
2742 int PhaseIdealLoop::stride_of_possible_iv(Node* iff) {
2743 Node* trunc1 = NULL;
2744 Node* trunc2 = NULL;
2745 const TypeInteger* ttype = NULL;
2746 if (!iff->is_If() || iff->in(1) == NULL || !iff->in(1)->is_Bool()) {
2747 return 0;
2748 }
2749 BoolNode* bl = iff->in(1)->as_Bool();
2750 Node* cmp = bl->in(1);
2751 if (!cmp || (cmp->Opcode() != Op_CmpI && cmp->Opcode() != Op_CmpU)) {
2752 return 0;
2753 }
2754 // Must have an invariant operand
2755 if (is_member(get_loop(iff), get_ctrl(cmp->in(2)))) {
2756 return 0;
2757 }
2758 Node* add2 = NULL;
2759 Node* cmp1 = cmp->in(1);
2760 if (cmp1->is_Phi()) {
2761 // (If (Bool (CmpX phi:(Phi ...(Optional-trunc(AddI phi add2))) )))
2762 Node* phi = cmp1;
2763 for (uint i = 1; i < phi->req(); i++) {
2764 Node* in = phi->in(i);
2765 Node* add = CountedLoopNode::match_incr_with_optional_truncation(in,
2766 &trunc1, &trunc2, &ttype, T_INT);
2767 if (add && add->in(1) == phi) {
2768 add2 = add->in(2);
2769 break;
2770 }
2771 }
2772 } else {
2773 // (If (Bool (CmpX addtrunc:(Optional-trunc((AddI (Phi ...addtrunc...) add2)) )))
2774 Node* addtrunc = cmp1;
2775 Node* add = CountedLoopNode::match_incr_with_optional_truncation(addtrunc,
2776 &trunc1, &trunc2, &ttype, T_INT);
2777 if (add && add->in(1)->is_Phi()) {
2778 Node* phi = add->in(1);
2779 for (uint i = 1; i < phi->req(); i++) {
2780 if (phi->in(i) == addtrunc) {
2781 add2 = add->in(2);
2782 break;
2783 }
2784 }
2785 }
2786 }
2787 if (add2 != NULL) {
2788 const TypeInt* add2t = _igvn.type(add2)->is_int();
2789 if (add2t->is_con()) {
2790 return add2t->get_con();
2791 }
2792 }
2793 return 0;
2794 }
2795
2796
2797 //---------------------- stay_in_loop -------------------------------------
2798 // Return the (unique) control output node that's in the loop (if it exists.)
2799 Node* PhaseIdealLoop::stay_in_loop( Node* n, IdealLoopTree *loop) {
2800 Node* unique = NULL;
2801 if (!n) return NULL;
2802 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2803 Node* use = n->fast_out(i);
2804 if (!has_ctrl(use) && loop->is_member(get_loop(use))) {
2805 if (unique != NULL) {
2806 return NULL;
2807 }
2808 unique = use;
2809 }
2810 }
2811 return unique;
2812 }
2813
2814 //------------------------------ register_node -------------------------------------
2815 // Utility to register node "n" with PhaseIdealLoop
2816 void PhaseIdealLoop::register_node(Node* n, IdealLoopTree *loop, Node* pred, int ddepth) {
2817 _igvn.register_new_node_with_optimizer(n);
2818 loop->_body.push(n);
2819 if (n->is_CFG()) {
2820 set_loop(n, loop);
2821 set_idom(n, pred, ddepth);
2822 } else {
2823 set_ctrl(n, pred);
2824 }
2825 }
2826
2827 //------------------------------ proj_clone -------------------------------------
2828 // Utility to create an if-projection
2829 ProjNode* PhaseIdealLoop::proj_clone(ProjNode* p, IfNode* iff) {
2830 ProjNode* c = p->clone()->as_Proj();
2831 c->set_req(0, iff);
2832 return c;
2833 }
2834
2835 //------------------------------ short_circuit_if -------------------------------------
2836 // Force the iff control output to be the live_proj
2837 Node* PhaseIdealLoop::short_circuit_if(IfNode* iff, ProjNode* live_proj) {
2838 guarantee(live_proj != NULL, "null projection");
2839 int proj_con = live_proj->_con;
2840 assert(proj_con == 0 || proj_con == 1, "false or true projection");
2841 Node *con = _igvn.intcon(proj_con);
2842 set_ctrl(con, C->root());
2843 if (iff) {
2844 iff->set_req(1, con);
2845 }
2846 return con;
2847 }
2848
2849 //------------------------------ insert_if_before_proj -------------------------------------
2850 // Insert a new if before an if projection (* - new node)
2851 //
2852 // before
2853 // if(test)
2854 // / \
2855 // v v
2856 // other-proj proj (arg)
2857 //
2858 // after
2859 // if(test)
2860 // / \
2861 // / v
2862 // | * proj-clone
2863 // v |
2864 // other-proj v
2865 // * new_if(relop(cmp[IU](left,right)))
2866 // / \
2867 // v v
2868 // * new-proj proj
2869 // (returned)
2870 //
2871 ProjNode* PhaseIdealLoop::insert_if_before_proj(Node* left, bool Signed, BoolTest::mask relop, Node* right, ProjNode* proj) {
2872 IfNode* iff = proj->in(0)->as_If();
2873 IdealLoopTree *loop = get_loop(proj);
2874 ProjNode *other_proj = iff->proj_out(!proj->is_IfTrue())->as_Proj();
2875 int ddepth = dom_depth(proj);
2876
2877 _igvn.rehash_node_delayed(iff);
2878 _igvn.rehash_node_delayed(proj);
2879
2880 proj->set_req(0, NULL); // temporary disconnect
2881 ProjNode* proj2 = proj_clone(proj, iff);
2882 register_node(proj2, loop, iff, ddepth);
2883
2884 Node* cmp = Signed ? (Node*) new CmpINode(left, right) : (Node*) new CmpUNode(left, right);
2885 register_node(cmp, loop, proj2, ddepth);
2886
2887 BoolNode* bol = new BoolNode(cmp, relop);
2888 register_node(bol, loop, proj2, ddepth);
2889
2890 int opcode = iff->Opcode();
2891 assert(opcode == Op_If || opcode == Op_RangeCheck, "unexpected opcode");
2892 IfNode* new_if = (opcode == Op_If) ? new IfNode(proj2, bol, iff->_prob, iff->_fcnt):
2893 new RangeCheckNode(proj2, bol, iff->_prob, iff->_fcnt);
2894 register_node(new_if, loop, proj2, ddepth);
2895
2896 proj->set_req(0, new_if); // reattach
2897 set_idom(proj, new_if, ddepth);
2898
2899 ProjNode* new_exit = proj_clone(other_proj, new_if)->as_Proj();
2900 guarantee(new_exit != NULL, "null exit node");
2901 register_node(new_exit, get_loop(other_proj), new_if, ddepth);
2902
2903 return new_exit;
2904 }
2905
2906 //------------------------------ insert_region_before_proj -------------------------------------
2907 // Insert a region before an if projection (* - new node)
2908 //
2909 // before
2910 // if(test)
2911 // / |
2912 // v |
2913 // proj v
2914 // other-proj
2915 //
2916 // after
2917 // if(test)
2918 // / |
2919 // v |
2920 // * proj-clone v
2921 // | other-proj
2922 // v
2923 // * new-region
2924 // |
2925 // v
2926 // * dum_if
2927 // / \
2928 // v \
2929 // * dum-proj v
2930 // proj
2931 //
2932 RegionNode* PhaseIdealLoop::insert_region_before_proj(ProjNode* proj) {
2933 IfNode* iff = proj->in(0)->as_If();
2934 IdealLoopTree *loop = get_loop(proj);
2935 ProjNode *other_proj = iff->proj_out(!proj->is_IfTrue())->as_Proj();
2936 int ddepth = dom_depth(proj);
2937
2938 _igvn.rehash_node_delayed(iff);
2939 _igvn.rehash_node_delayed(proj);
2940
2941 proj->set_req(0, NULL); // temporary disconnect
2942 ProjNode* proj2 = proj_clone(proj, iff);
2943 register_node(proj2, loop, iff, ddepth);
2944
2945 RegionNode* reg = new RegionNode(2);
2946 reg->set_req(1, proj2);
2947 register_node(reg, loop, iff, ddepth);
2948
2949 IfNode* dum_if = new IfNode(reg, short_circuit_if(NULL, proj), iff->_prob, iff->_fcnt);
2950 register_node(dum_if, loop, reg, ddepth);
2951
2952 proj->set_req(0, dum_if); // reattach
2953 set_idom(proj, dum_if, ddepth);
2954
2955 ProjNode* dum_proj = proj_clone(other_proj, dum_if);
2956 register_node(dum_proj, loop, dum_if, ddepth);
2957
2958 return reg;
2959 }
2960
2961 //------------------------------ insert_cmpi_loop_exit -------------------------------------
2962 // Clone a signed compare loop exit from an unsigned compare and
2963 // insert it before the unsigned cmp on the stay-in-loop path.
2964 // All new nodes inserted in the dominator tree between the original
2965 // if and it's projections. The original if test is replaced with
2966 // a constant to force the stay-in-loop path.
2967 //
2968 // This is done to make sure that the original if and it's projections
2969 // still dominate the same set of control nodes, that the ctrl() relation
2970 // from data nodes to them is preserved, and that their loop nesting is
2971 // preserved.
2972 //
2973 // before
2974 // if(i <u limit) unsigned compare loop exit
2975 // / |
2976 // v v
2977 // exit-proj stay-in-loop-proj
2978 //
2979 // after
2980 // if(stay-in-loop-const) original if
2981 // / |
2982 // / v
2983 // / if(i < limit) new signed test
2984 // / / |
2985 // / / v
2986 // / / if(i <u limit) new cloned unsigned test
2987 // / / / |
2988 // v v v |
2989 // region |
2990 // | |
2991 // dum-if |
2992 // / | |
2993 // ether | |
2994 // v v
2995 // exit-proj stay-in-loop-proj
2996 //
2997 IfNode* PhaseIdealLoop::insert_cmpi_loop_exit(IfNode* if_cmpu, IdealLoopTree *loop) {
2998 const bool Signed = true;
2999 const bool Unsigned = false;
3000
3001 BoolNode* bol = if_cmpu->in(1)->as_Bool();
3002 if (bol->_test._test != BoolTest::lt) return NULL;
3003 CmpNode* cmpu = bol->in(1)->as_Cmp();
3004 if (cmpu->Opcode() != Op_CmpU) return NULL;
3005 int stride = stride_of_possible_iv(if_cmpu);
3006 if (stride == 0) return NULL;
3007
3008 Node* lp_proj = stay_in_loop(if_cmpu, loop);
3009 guarantee(lp_proj != NULL, "null loop node");
3010
3011 ProjNode* lp_continue = lp_proj->as_Proj();
3012 ProjNode* lp_exit = if_cmpu->proj_out(!lp_continue->is_IfTrue())->as_Proj();
3013 if (!lp_exit->is_IfFalse()) {
3014 // The loop exit condition is (i <u limit) ==> (i >= 0 && i < limit).
3015 // We therefore can't add a single exit condition.
3016 return NULL;
3017 }
3018 // The loop exit condition is !(i <u limit) ==> (i < 0 || i >= limit).
3019 // Split out the exit condition (i < 0) for stride < 0 or (i >= limit) for stride > 0.
3020 Node* limit = NULL;
3021 if (stride > 0) {
3022 limit = cmpu->in(2);
3023 } else {
3024 limit = _igvn.makecon(TypeInt::ZERO);
3025 set_ctrl(limit, C->root());
3026 }
3027 // Create a new region on the exit path
3028 RegionNode* reg = insert_region_before_proj(lp_exit);
3029 guarantee(reg != NULL, "null region node");
3030
3031 // Clone the if-cmpu-true-false using a signed compare
3032 BoolTest::mask rel_i = stride > 0 ? bol->_test._test : BoolTest::ge;
3033 ProjNode* cmpi_exit = insert_if_before_proj(cmpu->in(1), Signed, rel_i, limit, lp_continue);
3034 reg->add_req(cmpi_exit);
3035
3036 // Clone the if-cmpu-true-false
3037 BoolTest::mask rel_u = bol->_test._test;
3038 ProjNode* cmpu_exit = insert_if_before_proj(cmpu->in(1), Unsigned, rel_u, cmpu->in(2), lp_continue);
3039 reg->add_req(cmpu_exit);
3040
3041 // Force original if to stay in loop.
3042 short_circuit_if(if_cmpu, lp_continue);
3043
3044 return cmpi_exit->in(0)->as_If();
3045 }
3046
3047 //------------------------------ remove_cmpi_loop_exit -------------------------------------
3048 // Remove a previously inserted signed compare loop exit.
3049 void PhaseIdealLoop::remove_cmpi_loop_exit(IfNode* if_cmp, IdealLoopTree *loop) {
3050 Node* lp_proj = stay_in_loop(if_cmp, loop);
3051 assert(if_cmp->in(1)->in(1)->Opcode() == Op_CmpI &&
3052 stay_in_loop(lp_proj, loop)->is_If() &&
3053 stay_in_loop(lp_proj, loop)->in(1)->in(1)->Opcode() == Op_CmpU, "inserted cmpi before cmpu");
3054 Node *con = _igvn.makecon(lp_proj->is_IfTrue() ? TypeInt::ONE : TypeInt::ZERO);
3055 set_ctrl(con, C->root());
3056 if_cmp->set_req(1, con);
3057 }
3058
3059 //------------------------------ scheduled_nodelist -------------------------------------
3060 // Create a post order schedule of nodes that are in the
3061 // "member" set. The list is returned in "sched".
3062 // The first node in "sched" is the loop head, followed by
3063 // nodes which have no inputs in the "member" set, and then
3064 // followed by the nodes that have an immediate input dependence
3065 // on a node in "sched".
3066 void PhaseIdealLoop::scheduled_nodelist( IdealLoopTree *loop, VectorSet& member, Node_List &sched ) {
3067
3068 assert(member.test(loop->_head->_idx), "loop head must be in member set");
3069 VectorSet visited;
3070 Node_Stack nstack(loop->_body.size());
3071
3072 Node* n = loop->_head; // top of stack is cached in "n"
3073 uint idx = 0;
3074 visited.set(n->_idx);
3075
3076 // Initially push all with no inputs from within member set
3077 for(uint i = 0; i < loop->_body.size(); i++ ) {
3078 Node *elt = loop->_body.at(i);
3079 if (member.test(elt->_idx)) {
3080 bool found = false;
3081 for (uint j = 0; j < elt->req(); j++) {
3082 Node* def = elt->in(j);
3083 if (def && member.test(def->_idx) && def != elt) {
3084 found = true;
3085 break;
3086 }
3087 }
3088 if (!found && elt != loop->_head) {
3089 nstack.push(n, idx);
3090 n = elt;
3091 assert(!visited.test(n->_idx), "not seen yet");
3092 visited.set(n->_idx);
3093 }
3094 }
3095 }
3096
3097 // traverse out's that are in the member set
3098 while (true) {
3099 if (idx < n->outcnt()) {
3100 Node* use = n->raw_out(idx);
3101 idx++;
3102 if (!visited.test_set(use->_idx)) {
3103 if (member.test(use->_idx)) {
3104 nstack.push(n, idx);
3105 n = use;
3106 idx = 0;
3107 }
3108 }
3109 } else {
3110 // All outputs processed
3111 sched.push(n);
3112 if (nstack.is_empty()) break;
3113 n = nstack.node();
3114 idx = nstack.index();
3115 nstack.pop();
3116 }
3117 }
3118 }
3119
3120
3121 //------------------------------ has_use_in_set -------------------------------------
3122 // Has a use in the vector set
3123 bool PhaseIdealLoop::has_use_in_set( Node* n, VectorSet& vset ) {
3124 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3125 Node* use = n->fast_out(j);
3126 if (vset.test(use->_idx)) {
3127 return true;
3128 }
3129 }
3130 return false;
3131 }
3132
3133
3134 //------------------------------ has_use_internal_to_set -------------------------------------
3135 // Has use internal to the vector set (ie. not in a phi at the loop head)
3136 bool PhaseIdealLoop::has_use_internal_to_set( Node* n, VectorSet& vset, IdealLoopTree *loop ) {
3137 Node* head = loop->_head;
3138 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3139 Node* use = n->fast_out(j);
3140 if (vset.test(use->_idx) && !(use->is_Phi() && use->in(0) == head)) {
3141 return true;
3142 }
3143 }
3144 return false;
3145 }
3146
3147
3148 //------------------------------ clone_for_use_outside_loop -------------------------------------
3149 // clone "n" for uses that are outside of loop
3150 int PhaseIdealLoop::clone_for_use_outside_loop( IdealLoopTree *loop, Node* n, Node_List& worklist ) {
3151 int cloned = 0;
3152 assert(worklist.size() == 0, "should be empty");
3153 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3154 Node* use = n->fast_out(j);
3155 if( !loop->is_member(get_loop(has_ctrl(use) ? get_ctrl(use) : use)) ) {
3156 worklist.push(use);
3157 }
3158 }
3159
3160 if (C->check_node_count(worklist.size() + NodeLimitFudgeFactor,
3161 "Too many clones required in clone_for_use_outside_loop in partial peeling")) {
3162 return -1;
3163 }
3164
3165 while( worklist.size() ) {
3166 Node *use = worklist.pop();
3167 if (!has_node(use) || use->in(0) == C->top()) continue;
3168 uint j;
3169 for (j = 0; j < use->req(); j++) {
3170 if (use->in(j) == n) break;
3171 }
3172 assert(j < use->req(), "must be there");
3173
3174 // clone "n" and insert it between the inputs of "n" and the use outside the loop
3175 Node* n_clone = n->clone();
3176 _igvn.replace_input_of(use, j, n_clone);
3177 cloned++;
3178 Node* use_c;
3179 if (!use->is_Phi()) {
3180 use_c = has_ctrl(use) ? get_ctrl(use) : use->in(0);
3181 } else {
3182 // Use in a phi is considered a use in the associated predecessor block
3183 use_c = use->in(0)->in(j);
3184 }
3185 set_ctrl(n_clone, use_c);
3186 assert(!loop->is_member(get_loop(use_c)), "should be outside loop");
3187 get_loop(use_c)->_body.push(n_clone);
3188 _igvn.register_new_node_with_optimizer(n_clone);
3189 #ifndef PRODUCT
3190 if (TracePartialPeeling) {
3191 tty->print_cr("loop exit cloning old: %d new: %d newbb: %d", n->_idx, n_clone->_idx, get_ctrl(n_clone)->_idx);
3192 }
3193 #endif
3194 }
3195 return cloned;
3196 }
3197
3198
3199 //------------------------------ clone_for_special_use_inside_loop -------------------------------------
3200 // clone "n" for special uses that are in the not_peeled region.
3201 // If these def-uses occur in separate blocks, the code generator
3202 // marks the method as not compilable. For example, if a "BoolNode"
3203 // is in a different basic block than the "IfNode" that uses it, then
3204 // the compilation is aborted in the code generator.
3205 void PhaseIdealLoop::clone_for_special_use_inside_loop( IdealLoopTree *loop, Node* n,
3206 VectorSet& not_peel, Node_List& sink_list, Node_List& worklist ) {
3207 if (n->is_Phi() || n->is_Load()) {
3208 return;
3209 }
3210 assert(worklist.size() == 0, "should be empty");
3211 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3212 Node* use = n->fast_out(j);
3213 if ( not_peel.test(use->_idx) &&
3214 (use->is_If() || use->is_CMove() || use->is_Bool()) &&
3215 use->in(1) == n) {
3216 worklist.push(use);
3217 }
3218 }
3219 if (worklist.size() > 0) {
3220 // clone "n" and insert it between inputs of "n" and the use
3221 Node* n_clone = n->clone();
3222 loop->_body.push(n_clone);
3223 _igvn.register_new_node_with_optimizer(n_clone);
3224 set_ctrl(n_clone, get_ctrl(n));
3225 sink_list.push(n_clone);
3226 not_peel.set(n_clone->_idx);
3227 #ifndef PRODUCT
3228 if (TracePartialPeeling) {
3229 tty->print_cr("special not_peeled cloning old: %d new: %d", n->_idx, n_clone->_idx);
3230 }
3231 #endif
3232 while( worklist.size() ) {
3233 Node *use = worklist.pop();
3234 _igvn.rehash_node_delayed(use);
3235 for (uint j = 1; j < use->req(); j++) {
3236 if (use->in(j) == n) {
3237 use->set_req(j, n_clone);
3238 }
3239 }
3240 }
3241 }
3242 }
3243
3244
3245 //------------------------------ insert_phi_for_loop -------------------------------------
3246 // Insert phi(lp_entry_val, back_edge_val) at use->in(idx) for loop lp if phi does not already exist
3247 void PhaseIdealLoop::insert_phi_for_loop( Node* use, uint idx, Node* lp_entry_val, Node* back_edge_val, LoopNode* lp ) {
3248 Node *phi = PhiNode::make(lp, back_edge_val);
3249 phi->set_req(LoopNode::EntryControl, lp_entry_val);
3250 // Use existing phi if it already exists
3251 Node *hit = _igvn.hash_find_insert(phi);
3252 if( hit == NULL ) {
3253 _igvn.register_new_node_with_optimizer(phi);
3254 set_ctrl(phi, lp);
3255 } else {
3256 // Remove the new phi from the graph and use the hit
3257 _igvn.remove_dead_node(phi);
3258 phi = hit;
3259 }
3260 _igvn.replace_input_of(use, idx, phi);
3261 }
3262
3263 #ifdef ASSERT
3264 //------------------------------ is_valid_loop_partition -------------------------------------
3265 // Validate the loop partition sets: peel and not_peel
3266 bool PhaseIdealLoop::is_valid_loop_partition( IdealLoopTree *loop, VectorSet& peel, Node_List& peel_list,
3267 VectorSet& not_peel ) {
3268 uint i;
3269 // Check that peel_list entries are in the peel set
3270 for (i = 0; i < peel_list.size(); i++) {
3271 if (!peel.test(peel_list.at(i)->_idx)) {
3272 return false;
3273 }
3274 }
3275 // Check at loop members are in one of peel set or not_peel set
3276 for (i = 0; i < loop->_body.size(); i++ ) {
3277 Node *def = loop->_body.at(i);
3278 uint di = def->_idx;
3279 // Check that peel set elements are in peel_list
3280 if (peel.test(di)) {
3281 if (not_peel.test(di)) {
3282 return false;
3283 }
3284 // Must be in peel_list also
3285 bool found = false;
3286 for (uint j = 0; j < peel_list.size(); j++) {
3287 if (peel_list.at(j)->_idx == di) {
3288 found = true;
3289 break;
3290 }
3291 }
3292 if (!found) {
3293 return false;
3294 }
3295 } else if (not_peel.test(di)) {
3296 if (peel.test(di)) {
3297 return false;
3298 }
3299 } else {
3300 return false;
3301 }
3302 }
3303 return true;
3304 }
3305
3306 //------------------------------ is_valid_clone_loop_exit_use -------------------------------------
3307 // Ensure a use outside of loop is of the right form
3308 bool PhaseIdealLoop::is_valid_clone_loop_exit_use( IdealLoopTree *loop, Node* use, uint exit_idx) {
3309 Node *use_c = has_ctrl(use) ? get_ctrl(use) : use;
3310 return (use->is_Phi() &&
3311 use_c->is_Region() && use_c->req() == 3 &&
3312 (use_c->in(exit_idx)->Opcode() == Op_IfTrue ||
3313 use_c->in(exit_idx)->Opcode() == Op_IfFalse ||
3314 use_c->in(exit_idx)->Opcode() == Op_JumpProj) &&
3315 loop->is_member( get_loop( use_c->in(exit_idx)->in(0) ) ) );
3316 }
3317
3318 //------------------------------ is_valid_clone_loop_form -------------------------------------
3319 // Ensure that all uses outside of loop are of the right form
3320 bool PhaseIdealLoop::is_valid_clone_loop_form( IdealLoopTree *loop, Node_List& peel_list,
3321 uint orig_exit_idx, uint clone_exit_idx) {
3322 uint len = peel_list.size();
3323 for (uint i = 0; i < len; i++) {
3324 Node *def = peel_list.at(i);
3325
3326 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
3327 Node *use = def->fast_out(j);
3328 Node *use_c = has_ctrl(use) ? get_ctrl(use) : use;
3329 if (!loop->is_member(get_loop(use_c))) {
3330 // use is not in the loop, check for correct structure
3331 if (use->in(0) == def) {
3332 // Okay
3333 } else if (!is_valid_clone_loop_exit_use(loop, use, orig_exit_idx)) {
3334 return false;
3335 }
3336 }
3337 }
3338 }
3339 return true;
3340 }
3341 #endif
3342
3343 //------------------------------ partial_peel -------------------------------------
3344 // Partially peel (aka loop rotation) the top portion of a loop (called
3345 // the peel section below) by cloning it and placing one copy just before
3346 // the new loop head and the other copy at the bottom of the new loop.
3347 //
3348 // before after where it came from
3349 //
3350 // stmt1 stmt1
3351 // loop: stmt2 clone
3352 // stmt2 if condA goto exitA clone
3353 // if condA goto exitA new_loop: new
3354 // stmt3 stmt3 clone
3355 // if !condB goto loop if condB goto exitB clone
3356 // exitB: stmt2 orig
3357 // stmt4 if !condA goto new_loop orig
3358 // exitA: goto exitA
3359 // exitB:
3360 // stmt4
3361 // exitA:
3362 //
3363 // Step 1: find the cut point: an exit test on probable
3364 // induction variable.
3365 // Step 2: schedule (with cloning) operations in the peel
3366 // section that can be executed after the cut into
3367 // the section that is not peeled. This may need
3368 // to clone operations into exit blocks. For
3369 // instance, a reference to A[i] in the not-peel
3370 // section and a reference to B[i] in an exit block
3371 // may cause a left-shift of i by 2 to be placed
3372 // in the peel block. This step will clone the left
3373 // shift into the exit block and sink the left shift
3374 // from the peel to the not-peel section.
3375 // Step 3: clone the loop, retarget the control, and insert
3376 // phis for values that are live across the new loop
3377 // head. This is very dependent on the graph structure
3378 // from clone_loop. It creates region nodes for
3379 // exit control and associated phi nodes for values
3380 // flow out of the loop through that exit. The region
3381 // node is dominated by the clone's control projection.
3382 // So the clone's peel section is placed before the
3383 // new loop head, and the clone's not-peel section is
3384 // forms the top part of the new loop. The original
3385 // peel section forms the tail of the new loop.
3386 // Step 4: update the dominator tree and recompute the
3387 // dominator depth.
3388 //
3389 // orig
3390 //
3391 // stmt1
3392 // |
3393 // v
3394 // loop predicate
3395 // |
3396 // v
3397 // loop<----+
3398 // | |
3399 // stmt2 |
3400 // | |
3401 // v |
3402 // ifA |
3403 // / | |
3404 // v v |
3405 // false true ^ <-- last_peel
3406 // / | |
3407 // / ===|==cut |
3408 // / stmt3 | <-- first_not_peel
3409 // / | |
3410 // | v |
3411 // v ifB |
3412 // exitA: / \ |
3413 // / \ |
3414 // v v |
3415 // false true |
3416 // / \ |
3417 // / ----+
3418 // |
3419 // v
3420 // exitB:
3421 // stmt4
3422 //
3423 //
3424 // after clone loop
3425 //
3426 // stmt1
3427 // |
3428 // v
3429 // loop predicate
3430 // / \
3431 // clone / \ orig
3432 // / \
3433 // / \
3434 // v v
3435 // +---->loop loop<----+
3436 // | | | |
3437 // | stmt2 stmt2 |
3438 // | | | |
3439 // | v v |
3440 // | ifA ifA |
3441 // | | \ / | |
3442 // | v v v v |
3443 // ^ true false false true ^ <-- last_peel
3444 // | | ^ \ / | |
3445 // | cut==|== \ \ / ===|==cut |
3446 // | stmt3 \ \ / stmt3 | <-- first_not_peel
3447 // | | dom | | | |
3448 // | v \ 1v v2 v |
3449 // | ifB regionA ifB |
3450 // | / \ | / \ |
3451 // | / \ v / \ |
3452 // | v v exitA: v v |
3453 // | true false false true |
3454 // | / ^ \ / \ |
3455 // +---- \ \ / ----+
3456 // dom \ /
3457 // \ 1v v2
3458 // regionB
3459 // |
3460 // v
3461 // exitB:
3462 // stmt4
3463 //
3464 //
3465 // after partial peel
3466 //
3467 // stmt1
3468 // |
3469 // v
3470 // loop predicate
3471 // /
3472 // clone / orig
3473 // / TOP
3474 // / \
3475 // v v
3476 // TOP->loop loop----+
3477 // | | |
3478 // stmt2 stmt2 |
3479 // | | |
3480 // v v |
3481 // ifA ifA |
3482 // | \ / | |
3483 // v v v v |
3484 // true false false true | <-- last_peel
3485 // | ^ \ / +------|---+
3486 // +->newloop \ \ / === ==cut | |
3487 // | stmt3 \ \ / TOP | |
3488 // | | dom | | stmt3 | | <-- first_not_peel
3489 // | v \ 1v v2 v | |
3490 // | ifB regionA ifB ^ v
3491 // | / \ | / \ | |
3492 // | / \ v / \ | |
3493 // | v v exitA: v v | |
3494 // | true false false true | |
3495 // | / ^ \ / \ | |
3496 // | | \ \ / v | |
3497 // | | dom \ / TOP | |
3498 // | | \ 1v v2 | |
3499 // ^ v regionB | |
3500 // | | | | |
3501 // | | v ^ v
3502 // | | exitB: | |
3503 // | | stmt4 | |
3504 // | +------------>-----------------+ |
3505 // | |
3506 // +-----------------<---------------------+
3507 //
3508 //
3509 // final graph
3510 //
3511 // stmt1
3512 // |
3513 // v
3514 // loop predicate
3515 // |
3516 // v
3517 // stmt2 clone
3518 // |
3519 // v
3520 // ........> ifA clone
3521 // : / |
3522 // dom / |
3523 // : v v
3524 // : false true
3525 // : | |
3526 // : | v
3527 // : | newloop<-----+
3528 // : | | |
3529 // : | stmt3 clone |
3530 // : | | |
3531 // : | v |
3532 // : | ifB |
3533 // : | / \ |
3534 // : | v v |
3535 // : | false true |
3536 // : | | | |
3537 // : | v stmt2 |
3538 // : | exitB: | |
3539 // : | stmt4 v |
3540 // : | ifA orig |
3541 // : | / \ |
3542 // : | / \ |
3543 // : | v v |
3544 // : | false true |
3545 // : | / \ |
3546 // : v v -----+
3547 // RegionA
3548 // |
3549 // v
3550 // exitA
3551 //
3552 bool PhaseIdealLoop::partial_peel( IdealLoopTree *loop, Node_List &old_new ) {
3553
3554 assert(!loop->_head->is_CountedLoop(), "Non-counted loop only");
3555 if (!loop->_head->is_Loop()) {
3556 return false;
3557 }
3558 LoopNode *head = loop->_head->as_Loop();
3559
3560 if (head->is_partial_peel_loop() || head->partial_peel_has_failed()) {
3561 return false;
3562 }
3563
3564 // Check for complex exit control
3565 for (uint ii = 0; ii < loop->_body.size(); ii++) {
3566 Node *n = loop->_body.at(ii);
3567 int opc = n->Opcode();
3568 if (n->is_Call() ||
3569 opc == Op_Catch ||
3570 opc == Op_CatchProj ||
3571 opc == Op_Jump ||
3572 opc == Op_JumpProj) {
3573 #ifndef PRODUCT
3574 if (TracePartialPeeling) {
3575 tty->print_cr("\nExit control too complex: lp: %d", head->_idx);
3576 }
3577 #endif
3578 return false;
3579 }
3580 }
3581
3582 int dd = dom_depth(head);
3583
3584 // Step 1: find cut point
3585
3586 // Walk up dominators to loop head looking for first loop exit
3587 // which is executed on every path thru loop.
3588 IfNode *peel_if = NULL;
3589 IfNode *peel_if_cmpu = NULL;
3590
3591 Node *iff = loop->tail();
3592 while (iff != head) {
3593 if (iff->is_If()) {
3594 Node *ctrl = get_ctrl(iff->in(1));
3595 if (ctrl->is_top()) return false; // Dead test on live IF.
3596 // If loop-varying exit-test, check for induction variable
3597 if (loop->is_member(get_loop(ctrl)) &&
3598 loop->is_loop_exit(iff) &&
3599 is_possible_iv_test(iff)) {
3600 Node* cmp = iff->in(1)->in(1);
3601 if (cmp->Opcode() == Op_CmpI) {
3602 peel_if = iff->as_If();
3603 } else {
3604 assert(cmp->Opcode() == Op_CmpU, "must be CmpI or CmpU");
3605 peel_if_cmpu = iff->as_If();
3606 }
3607 }
3608 }
3609 iff = idom(iff);
3610 }
3611
3612 // Prefer signed compare over unsigned compare.
3613 IfNode* new_peel_if = NULL;
3614 if (peel_if == NULL) {
3615 if (!PartialPeelAtUnsignedTests || peel_if_cmpu == NULL) {
3616 return false; // No peel point found
3617 }
3618 new_peel_if = insert_cmpi_loop_exit(peel_if_cmpu, loop);
3619 if (new_peel_if == NULL) {
3620 return false; // No peel point found
3621 }
3622 peel_if = new_peel_if;
3623 }
3624 Node* last_peel = stay_in_loop(peel_if, loop);
3625 Node* first_not_peeled = stay_in_loop(last_peel, loop);
3626 if (first_not_peeled == NULL || first_not_peeled == head) {
3627 return false;
3628 }
3629
3630 #ifndef PRODUCT
3631 if (TraceLoopOpts) {
3632 tty->print("PartialPeel ");
3633 loop->dump_head();
3634 }
3635
3636 if (TracePartialPeeling) {
3637 tty->print_cr("before partial peel one iteration");
3638 Node_List wl;
3639 Node* t = head->in(2);
3640 while (true) {
3641 wl.push(t);
3642 if (t == head) break;
3643 t = idom(t);
3644 }
3645 while (wl.size() > 0) {
3646 Node* tt = wl.pop();
3647 tt->dump();
3648 if (tt == last_peel) tty->print_cr("-- cut --");
3649 }
3650 }
3651 #endif
3652 VectorSet peel;
3653 VectorSet not_peel;
3654 Node_List peel_list;
3655 Node_List worklist;
3656 Node_List sink_list;
3657
3658 uint estimate = loop->est_loop_clone_sz(1);
3659 if (exceeding_node_budget(estimate)) {
3660 return false;
3661 }
3662
3663 // Set of cfg nodes to peel are those that are executable from
3664 // the head through last_peel.
3665 assert(worklist.size() == 0, "should be empty");
3666 worklist.push(head);
3667 peel.set(head->_idx);
3668 while (worklist.size() > 0) {
3669 Node *n = worklist.pop();
3670 if (n != last_peel) {
3671 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3672 Node* use = n->fast_out(j);
3673 if (use->is_CFG() &&
3674 loop->is_member(get_loop(use)) &&
3675 !peel.test_set(use->_idx)) {
3676 worklist.push(use);
3677 }
3678 }
3679 }
3680 }
3681
3682 // Set of non-cfg nodes to peel are those that are control
3683 // dependent on the cfg nodes.
3684 for (uint i = 0; i < loop->_body.size(); i++) {
3685 Node *n = loop->_body.at(i);
3686 Node *n_c = has_ctrl(n) ? get_ctrl(n) : n;
3687 if (peel.test(n_c->_idx)) {
3688 peel.set(n->_idx);
3689 } else {
3690 not_peel.set(n->_idx);
3691 }
3692 }
3693
3694 // Step 2: move operations from the peeled section down into the
3695 // not-peeled section
3696
3697 // Get a post order schedule of nodes in the peel region
3698 // Result in right-most operand.
3699 scheduled_nodelist(loop, peel, peel_list);
3700
3701 assert(is_valid_loop_partition(loop, peel, peel_list, not_peel), "bad partition");
3702
3703 // For future check for too many new phis
3704 uint old_phi_cnt = 0;
3705 for (DUIterator_Fast jmax, j = head->fast_outs(jmax); j < jmax; j++) {
3706 Node* use = head->fast_out(j);
3707 if (use->is_Phi()) old_phi_cnt++;
3708 }
3709
3710 #ifndef PRODUCT
3711 if (TracePartialPeeling) {
3712 tty->print_cr("\npeeled list");
3713 }
3714 #endif
3715
3716 // Evacuate nodes in peel region into the not_peeled region if possible
3717 bool too_many_clones = false;
3718 uint new_phi_cnt = 0;
3719 uint cloned_for_outside_use = 0;
3720 for (uint i = 0; i < peel_list.size();) {
3721 Node* n = peel_list.at(i);
3722 #ifndef PRODUCT
3723 if (TracePartialPeeling) n->dump();
3724 #endif
3725 bool incr = true;
3726 if (!n->is_CFG()) {
3727 if (has_use_in_set(n, not_peel)) {
3728 // If not used internal to the peeled region,
3729 // move "n" from peeled to not_peeled region.
3730 if (!has_use_internal_to_set(n, peel, loop)) {
3731 // if not pinned and not a load (which maybe anti-dependent on a store)
3732 // and not a CMove (Matcher expects only bool->cmove).
3733 if (n->in(0) == NULL && !n->is_Load() && !n->is_CMove()) {
3734 int new_clones = clone_for_use_outside_loop(loop, n, worklist);
3735 if (new_clones == -1) {
3736 too_many_clones = true;
3737 break;
3738 }
3739 cloned_for_outside_use += new_clones;
3740 sink_list.push(n);
3741 peel.remove(n->_idx);
3742 not_peel.set(n->_idx);
3743 peel_list.remove(i);
3744 incr = false;
3745 #ifndef PRODUCT
3746 if (TracePartialPeeling) {
3747 tty->print_cr("sink to not_peeled region: %d newbb: %d",
3748 n->_idx, get_ctrl(n)->_idx);
3749 }
3750 #endif
3751 }
3752 } else {
3753 // Otherwise check for special def-use cases that span
3754 // the peel/not_peel boundary such as bool->if
3755 clone_for_special_use_inside_loop(loop, n, not_peel, sink_list, worklist);
3756 new_phi_cnt++;
3757 }
3758 }
3759 }
3760 if (incr) i++;
3761 }
3762
3763 estimate += cloned_for_outside_use + new_phi_cnt;
3764 bool exceed_node_budget = !may_require_nodes(estimate);
3765 bool exceed_phi_limit = new_phi_cnt > old_phi_cnt + PartialPeelNewPhiDelta;
3766
3767 if (too_many_clones || exceed_node_budget || exceed_phi_limit) {
3768 #ifndef PRODUCT
3769 if (TracePartialPeeling && exceed_phi_limit) {
3770 tty->print_cr("\nToo many new phis: %d old %d new cmpi: %c",
3771 new_phi_cnt, old_phi_cnt, new_peel_if != NULL?'T':'F');
3772 }
3773 #endif
3774 if (new_peel_if != NULL) {
3775 remove_cmpi_loop_exit(new_peel_if, loop);
3776 }
3777 // Inhibit more partial peeling on this loop
3778 assert(!head->is_partial_peel_loop(), "not partial peeled");
3779 head->mark_partial_peel_failed();
3780 if (cloned_for_outside_use > 0) {
3781 // Terminate this round of loop opts because
3782 // the graph outside this loop was changed.
3783 C->set_major_progress();
3784 return true;
3785 }
3786 return false;
3787 }
3788
3789 // Step 3: clone loop, retarget control, and insert new phis
3790
3791 // Create new loop head for new phis and to hang
3792 // the nodes being moved (sinked) from the peel region.
3793 LoopNode* new_head = new LoopNode(last_peel, last_peel);
3794 new_head->set_unswitch_count(head->unswitch_count()); // Preserve
3795 _igvn.register_new_node_with_optimizer(new_head);
3796 assert(first_not_peeled->in(0) == last_peel, "last_peel <- first_not_peeled");
3797 _igvn.replace_input_of(first_not_peeled, 0, new_head);
3798 set_loop(new_head, loop);
3799 loop->_body.push(new_head);
3800 not_peel.set(new_head->_idx);
3801 set_idom(new_head, last_peel, dom_depth(first_not_peeled));
3802 set_idom(first_not_peeled, new_head, dom_depth(first_not_peeled));
3803
3804 while (sink_list.size() > 0) {
3805 Node* n = sink_list.pop();
3806 set_ctrl(n, new_head);
3807 }
3808
3809 assert(is_valid_loop_partition(loop, peel, peel_list, not_peel), "bad partition");
3810
3811 clone_loop(loop, old_new, dd, IgnoreStripMined);
3812
3813 const uint clone_exit_idx = 1;
3814 const uint orig_exit_idx = 2;
3815 assert(is_valid_clone_loop_form(loop, peel_list, orig_exit_idx, clone_exit_idx), "bad clone loop");
3816
3817 Node* head_clone = old_new[head->_idx];
3818 LoopNode* new_head_clone = old_new[new_head->_idx]->as_Loop();
3819 Node* orig_tail_clone = head_clone->in(2);
3820
3821 // Add phi if "def" node is in peel set and "use" is not
3822
3823 for (uint i = 0; i < peel_list.size(); i++) {
3824 Node *def = peel_list.at(i);
3825 if (!def->is_CFG()) {
3826 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
3827 Node *use = def->fast_out(j);
3828 if (has_node(use) && use->in(0) != C->top() &&
3829 (!peel.test(use->_idx) ||
3830 (use->is_Phi() && use->in(0) == head)) ) {
3831 worklist.push(use);
3832 }
3833 }
3834 while( worklist.size() ) {
3835 Node *use = worklist.pop();
3836 for (uint j = 1; j < use->req(); j++) {
3837 Node* n = use->in(j);
3838 if (n == def) {
3839
3840 // "def" is in peel set, "use" is not in peel set
3841 // or "use" is in the entry boundary (a phi) of the peel set
3842
3843 Node* use_c = has_ctrl(use) ? get_ctrl(use) : use;
3844
3845 if ( loop->is_member(get_loop( use_c )) ) {
3846 // use is in loop
3847 if (old_new[use->_idx] != NULL) { // null for dead code
3848 Node* use_clone = old_new[use->_idx];
3849 _igvn.replace_input_of(use, j, C->top());
3850 insert_phi_for_loop( use_clone, j, old_new[def->_idx], def, new_head_clone );
3851 }
3852 } else {
3853 assert(is_valid_clone_loop_exit_use(loop, use, orig_exit_idx), "clone loop format");
3854 // use is not in the loop, check if the live range includes the cut
3855 Node* lp_if = use_c->in(orig_exit_idx)->in(0);
3856 if (not_peel.test(lp_if->_idx)) {
3857 assert(j == orig_exit_idx, "use from original loop");
3858 insert_phi_for_loop( use, clone_exit_idx, old_new[def->_idx], def, new_head_clone );
3859 }
3860 }
3861 }
3862 }
3863 }
3864 }
3865 }
3866
3867 // Step 3b: retarget control
3868
3869 // Redirect control to the new loop head if a cloned node in
3870 // the not_peeled region has control that points into the peeled region.
3871 // This necessary because the cloned peeled region will be outside
3872 // the loop.
3873 // from to
3874 // cloned-peeled <---+
3875 // new_head_clone: | <--+
3876 // cloned-not_peeled in(0) in(0)
3877 // orig-peeled
3878
3879 for (uint i = 0; i < loop->_body.size(); i++) {
3880 Node *n = loop->_body.at(i);
3881 if (!n->is_CFG() && n->in(0) != NULL &&
3882 not_peel.test(n->_idx) && peel.test(n->in(0)->_idx)) {
3883 Node* n_clone = old_new[n->_idx];
3884 _igvn.replace_input_of(n_clone, 0, new_head_clone);
3885 }
3886 }
3887
3888 // Backedge of the surviving new_head (the clone) is original last_peel
3889 _igvn.replace_input_of(new_head_clone, LoopNode::LoopBackControl, last_peel);
3890
3891 // Cut first node in original not_peel set
3892 _igvn.rehash_node_delayed(new_head); // Multiple edge updates:
3893 new_head->set_req(LoopNode::EntryControl, C->top()); // use rehash_node_delayed / set_req instead of
3894 new_head->set_req(LoopNode::LoopBackControl, C->top()); // multiple replace_input_of calls
3895
3896 // Copy head_clone back-branch info to original head
3897 // and remove original head's loop entry and
3898 // clone head's back-branch
3899 _igvn.rehash_node_delayed(head); // Multiple edge updates
3900 head->set_req(LoopNode::EntryControl, head_clone->in(LoopNode::LoopBackControl));
3901 head->set_req(LoopNode::LoopBackControl, C->top());
3902 _igvn.replace_input_of(head_clone, LoopNode::LoopBackControl, C->top());
3903
3904 // Similarly modify the phis
3905 for (DUIterator_Fast kmax, k = head->fast_outs(kmax); k < kmax; k++) {
3906 Node* use = head->fast_out(k);
3907 if (use->is_Phi() && use->outcnt() > 0) {
3908 Node* use_clone = old_new[use->_idx];
3909 _igvn.rehash_node_delayed(use); // Multiple edge updates
3910 use->set_req(LoopNode::EntryControl, use_clone->in(LoopNode::LoopBackControl));
3911 use->set_req(LoopNode::LoopBackControl, C->top());
3912 _igvn.replace_input_of(use_clone, LoopNode::LoopBackControl, C->top());
3913 }
3914 }
3915
3916 // Step 4: update dominator tree and dominator depth
3917
3918 set_idom(head, orig_tail_clone, dd);
3919 recompute_dom_depth();
3920
3921 // Inhibit more partial peeling on this loop
3922 new_head_clone->set_partial_peel_loop();
3923 C->set_major_progress();
3924 loop->record_for_igvn();
3925
3926 #ifndef PRODUCT
3927 if (TracePartialPeeling) {
3928 tty->print_cr("\nafter partial peel one iteration");
3929 Node_List wl;
3930 Node* t = last_peel;
3931 while (true) {
3932 wl.push(t);
3933 if (t == head_clone) break;
3934 t = idom(t);
3935 }
3936 while (wl.size() > 0) {
3937 Node* tt = wl.pop();
3938 if (tt == head) tty->print_cr("orig head");
3939 else if (tt == new_head_clone) tty->print_cr("new head");
3940 else if (tt == head_clone) tty->print_cr("clone head");
3941 tt->dump();
3942 }
3943 }
3944 #endif
3945 return true;
3946 }
3947
3948 //------------------------------reorg_offsets----------------------------------
3949 // Reorganize offset computations to lower register pressure. Mostly
3950 // prevent loop-fallout uses of the pre-incremented trip counter (which are
3951 // then alive with the post-incremented trip counter forcing an extra
3952 // register move)
3953 void PhaseIdealLoop::reorg_offsets(IdealLoopTree *loop) {
3954 // Perform it only for canonical counted loops.
3955 // Loop's shape could be messed up by iteration_split_impl.
3956 if (!loop->_head->is_CountedLoop())
3957 return;
3958 if (!loop->_head->as_Loop()->is_valid_counted_loop(T_INT))
3959 return;
3960
3961 CountedLoopNode *cl = loop->_head->as_CountedLoop();
3962 CountedLoopEndNode *cle = cl->loopexit();
3963 Node *exit = cle->proj_out(false);
3964 Node *phi = cl->phi();
3965
3966 // Check for the special case when using the pre-incremented trip-counter on
3967 // the fall-out path (forces the pre-incremented and post-incremented trip
3968 // counter to be live at the same time). Fix this by adjusting to use the
3969 // post-increment trip counter.
3970
3971 bool progress = true;
3972 while (progress) {
3973 progress = false;
3974 for (DUIterator_Fast imax, i = phi->fast_outs(imax); i < imax; i++) {
3975 Node* use = phi->fast_out(i); // User of trip-counter
3976 if (!has_ctrl(use)) continue;
3977 Node *u_ctrl = get_ctrl(use);
3978 if (use->is_Phi()) {
3979 u_ctrl = NULL;
3980 for (uint j = 1; j < use->req(); j++)
3981 if (use->in(j) == phi)
3982 u_ctrl = dom_lca(u_ctrl, use->in(0)->in(j));
3983 }
3984 IdealLoopTree *u_loop = get_loop(u_ctrl);
3985 // Look for loop-invariant use
3986 if (u_loop == loop) continue;
3987 if (loop->is_member(u_loop)) continue;
3988 // Check that use is live out the bottom. Assuming the trip-counter
3989 // update is right at the bottom, uses of of the loop middle are ok.
3990 if (dom_lca(exit, u_ctrl) != exit) continue;
3991 // Hit! Refactor use to use the post-incremented tripcounter.
3992 // Compute a post-increment tripcounter.
3993 Node* c = exit;
3994 if (cl->is_strip_mined()) {
3995 IdealLoopTree* outer_loop = get_loop(cl->outer_loop());
3996 if (!outer_loop->is_member(u_loop)) {
3997 c = cl->outer_loop_exit();
3998 }
3999 }
4000 Node *opaq = new Opaque2Node(C, cle->incr());
4001 register_new_node(opaq, c);
4002 Node *neg_stride = _igvn.intcon(-cle->stride_con());
4003 set_ctrl(neg_stride, C->root());
4004 Node *post = new AddINode(opaq, neg_stride);
4005 register_new_node(post, c);
4006 post = new CastIINode(post, phi->bottom_type()); // preserve the iv phi's type
4007 register_new_node(post, c);
4008 _igvn.rehash_node_delayed(use);
4009 for (uint j = 1; j < use->req(); j++) {
4010 if (use->in(j) == phi)
4011 use->set_req(j, post);
4012 }
4013 // Since DU info changed, rerun loop
4014 progress = true;
4015 break;
4016 }
4017 }
4018
4019 }