Tue, 17 Oct 2017 12:58:25 +0800
merge
1 /*
2 * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "memory/allocation.inline.hpp"
27 #include "opto/addnode.hpp"
28 #include "opto/connode.hpp"
29 #include "opto/divnode.hpp"
30 #include "opto/loopnode.hpp"
31 #include "opto/matcher.hpp"
32 #include "opto/mulnode.hpp"
33 #include "opto/rootnode.hpp"
34 #include "opto/subnode.hpp"
36 //=============================================================================
37 //------------------------------split_thru_phi---------------------------------
38 // Split Node 'n' through merge point if there is enough win.
39 Node *PhaseIdealLoop::split_thru_phi( Node *n, Node *region, int policy ) {
40 if (n->Opcode() == Op_ConvI2L && n->bottom_type() != TypeLong::LONG) {
41 // ConvI2L may have type information on it which is unsafe to push up
42 // so disable this for now
43 return NULL;
44 }
46 int wins = 0;
47 assert(!n->is_CFG(), "");
48 assert(region->is_Region(), "");
50 const Type* type = n->bottom_type();
51 const TypeOopPtr *t_oop = _igvn.type(n)->isa_oopptr();
52 Node *phi;
53 if (t_oop != NULL && t_oop->is_known_instance_field()) {
54 int iid = t_oop->instance_id();
55 int index = C->get_alias_index(t_oop);
56 int offset = t_oop->offset();
57 phi = new (C) PhiNode(region, type, NULL, iid, index, offset);
58 } else {
59 phi = PhiNode::make_blank(region, n);
60 }
61 uint old_unique = C->unique();
62 for (uint i = 1; i < region->req(); i++) {
63 Node *x;
64 Node* the_clone = NULL;
65 if (region->in(i) == C->top()) {
66 x = C->top(); // Dead path? Use a dead data op
67 } else {
68 x = n->clone(); // Else clone up the data op
69 the_clone = x; // Remember for possible deletion.
70 // Alter data node to use pre-phi inputs
71 if (n->in(0) == region)
72 x->set_req( 0, region->in(i) );
73 for (uint j = 1; j < n->req(); j++) {
74 Node *in = n->in(j);
75 if (in->is_Phi() && in->in(0) == region)
76 x->set_req( j, in->in(i) ); // Use pre-Phi input for the clone
77 }
78 }
79 // Check for a 'win' on some paths
80 const Type *t = x->Value(&_igvn);
82 bool singleton = t->singleton();
84 // A TOP singleton indicates that there are no possible values incoming
85 // along a particular edge. In most cases, this is OK, and the Phi will
86 // be eliminated later in an Ideal call. However, we can't allow this to
87 // happen if the singleton occurs on loop entry, as the elimination of
88 // the PhiNode may cause the resulting node to migrate back to a previous
89 // loop iteration.
90 if (singleton && t == Type::TOP) {
91 // Is_Loop() == false does not confirm the absence of a loop (e.g., an
92 // irreducible loop may not be indicated by an affirmative is_Loop());
93 // therefore, the only top we can split thru a phi is on a backedge of
94 // a loop.
95 singleton &= region->is_Loop() && (i != LoopNode::EntryControl);
96 }
98 if (singleton) {
99 wins++;
100 x = ((PhaseGVN&)_igvn).makecon(t);
101 } else {
102 // We now call Identity to try to simplify the cloned node.
103 // Note that some Identity methods call phase->type(this).
104 // Make sure that the type array is big enough for
105 // our new node, even though we may throw the node away.
106 // (Note: This tweaking with igvn only works because x is a new node.)
107 _igvn.set_type(x, t);
108 // If x is a TypeNode, capture any more-precise type permanently into Node
109 // otherwise it will be not updated during igvn->transform since
110 // igvn->type(x) is set to x->Value() already.
111 x->raise_bottom_type(t);
112 Node *y = x->Identity(&_igvn);
113 if (y != x) {
114 wins++;
115 x = y;
116 } else {
117 y = _igvn.hash_find(x);
118 if (y) {
119 wins++;
120 x = y;
121 } else {
122 // Else x is a new node we are keeping
123 // We do not need register_new_node_with_optimizer
124 // because set_type has already been called.
125 _igvn._worklist.push(x);
126 }
127 }
128 }
129 if (x != the_clone && the_clone != NULL)
130 _igvn.remove_dead_node(the_clone);
131 phi->set_req( i, x );
132 }
133 // Too few wins?
134 if (wins <= policy) {
135 _igvn.remove_dead_node(phi);
136 return NULL;
137 }
139 // Record Phi
140 register_new_node( phi, region );
142 for (uint i2 = 1; i2 < phi->req(); i2++) {
143 Node *x = phi->in(i2);
144 // If we commoned up the cloned 'x' with another existing Node,
145 // the existing Node picks up a new use. We need to make the
146 // existing Node occur higher up so it dominates its uses.
147 Node *old_ctrl;
148 IdealLoopTree *old_loop;
150 if (x->is_Con()) {
151 // Constant's control is always root.
152 set_ctrl(x, C->root());
153 continue;
154 }
155 // The occasional new node
156 if (x->_idx >= old_unique) { // Found a new, unplaced node?
157 old_ctrl = NULL;
158 old_loop = NULL; // Not in any prior loop
159 } else {
160 old_ctrl = get_ctrl(x);
161 old_loop = get_loop(old_ctrl); // Get prior loop
162 }
163 // New late point must dominate new use
164 Node *new_ctrl = dom_lca(old_ctrl, region->in(i2));
165 if (new_ctrl == old_ctrl) // Nothing is changed
166 continue;
168 IdealLoopTree *new_loop = get_loop(new_ctrl);
170 // Don't move x into a loop if its uses are
171 // outside of loop. Otherwise x will be cloned
172 // for each use outside of this loop.
173 IdealLoopTree *use_loop = get_loop(region);
174 if (!new_loop->is_member(use_loop) &&
175 (old_loop == NULL || !new_loop->is_member(old_loop))) {
176 // Take early control, later control will be recalculated
177 // during next iteration of loop optimizations.
178 new_ctrl = get_early_ctrl(x);
179 new_loop = get_loop(new_ctrl);
180 }
181 // Set new location
182 set_ctrl(x, new_ctrl);
183 // If changing loop bodies, see if we need to collect into new body
184 if (old_loop != new_loop) {
185 if (old_loop && !old_loop->_child)
186 old_loop->_body.yank(x);
187 if (!new_loop->_child)
188 new_loop->_body.push(x); // Collect body info
189 }
190 }
192 return phi;
193 }
195 //------------------------------dominated_by------------------------------------
196 // Replace the dominated test with an obvious true or false. Place it on the
197 // IGVN worklist for later cleanup. Move control-dependent data Nodes on the
198 // live path up to the dominating control.
199 void PhaseIdealLoop::dominated_by( Node *prevdom, Node *iff, bool flip, bool exclude_loop_predicate ) {
200 #ifndef PRODUCT
201 if (VerifyLoopOptimizations && PrintOpto) tty->print_cr("dominating test");
202 #endif
205 // prevdom is the dominating projection of the dominating test.
206 assert( iff->is_If(), "" );
207 assert( iff->Opcode() == Op_If || iff->Opcode() == Op_CountedLoopEnd, "Check this code when new subtype is added");
208 int pop = prevdom->Opcode();
209 assert( pop == Op_IfFalse || pop == Op_IfTrue, "" );
210 if (flip) {
211 if (pop == Op_IfTrue)
212 pop = Op_IfFalse;
213 else
214 pop = Op_IfTrue;
215 }
216 // 'con' is set to true or false to kill the dominated test.
217 Node *con = _igvn.makecon(pop == Op_IfTrue ? TypeInt::ONE : TypeInt::ZERO);
218 set_ctrl(con, C->root()); // Constant gets a new use
219 // Hack the dominated test
220 _igvn.replace_input_of(iff, 1, con);
222 // If I dont have a reachable TRUE and FALSE path following the IfNode then
223 // I can assume this path reaches an infinite loop. In this case it's not
224 // important to optimize the data Nodes - either the whole compilation will
225 // be tossed or this path (and all data Nodes) will go dead.
226 if (iff->outcnt() != 2) return;
228 // Make control-dependent data Nodes on the live path (path that will remain
229 // once the dominated IF is removed) become control-dependent on the
230 // dominating projection.
231 Node* dp = iff->as_If()->proj_out(pop == Op_IfTrue);
233 // Loop predicates may have depending checks which should not
234 // be skipped. For example, range check predicate has two checks
235 // for lower and upper bounds.
236 if (dp == NULL)
237 return;
239 ProjNode* dp_proj = dp->as_Proj();
240 ProjNode* unc_proj = iff->as_If()->proj_out(1 - dp_proj->_con)->as_Proj();
241 if (exclude_loop_predicate &&
242 (unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_predicate) ||
243 unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_range_check))) {
244 // If this is a range check (IfNode::is_range_check), do not
245 // reorder because Compile::allow_range_check_smearing might have
246 // changed the check.
247 return; // Let IGVN transformation change control dependence.
248 }
250 IdealLoopTree *old_loop = get_loop(dp);
252 for (DUIterator_Fast imax, i = dp->fast_outs(imax); i < imax; i++) {
253 Node* cd = dp->fast_out(i); // Control-dependent node
254 if (cd->depends_only_on_test()) {
255 assert(cd->in(0) == dp, "");
256 _igvn.replace_input_of(cd, 0, prevdom);
257 set_early_ctrl(cd);
258 IdealLoopTree *new_loop = get_loop(get_ctrl(cd));
259 if (old_loop != new_loop) {
260 if (!old_loop->_child) old_loop->_body.yank(cd);
261 if (!new_loop->_child) new_loop->_body.push(cd);
262 }
263 --i;
264 --imax;
265 }
266 }
267 }
269 //------------------------------has_local_phi_input----------------------------
270 // Return TRUE if 'n' has Phi inputs from its local block and no other
271 // block-local inputs (all non-local-phi inputs come from earlier blocks)
272 Node *PhaseIdealLoop::has_local_phi_input( Node *n ) {
273 Node *n_ctrl = get_ctrl(n);
274 // See if some inputs come from a Phi in this block, or from before
275 // this block.
276 uint i;
277 for( i = 1; i < n->req(); i++ ) {
278 Node *phi = n->in(i);
279 if( phi->is_Phi() && phi->in(0) == n_ctrl )
280 break;
281 }
282 if( i >= n->req() )
283 return NULL; // No Phi inputs; nowhere to clone thru
285 // Check for inputs created between 'n' and the Phi input. These
286 // must split as well; they have already been given the chance
287 // (courtesy of a post-order visit) and since they did not we must
288 // recover the 'cost' of splitting them by being very profitable
289 // when splitting 'n'. Since this is unlikely we simply give up.
290 for( i = 1; i < n->req(); i++ ) {
291 Node *m = n->in(i);
292 if( get_ctrl(m) == n_ctrl && !m->is_Phi() ) {
293 // We allow the special case of AddP's with no local inputs.
294 // This allows us to split-up address expressions.
295 if (m->is_AddP() &&
296 get_ctrl(m->in(2)) != n_ctrl &&
297 get_ctrl(m->in(3)) != n_ctrl) {
298 // Move the AddP up to dominating point
299 set_ctrl_and_loop(m, find_non_split_ctrl(idom(n_ctrl)));
300 continue;
301 }
302 return NULL;
303 }
304 }
306 return n_ctrl;
307 }
309 //------------------------------remix_address_expressions----------------------
310 // Rework addressing expressions to get the most loop-invariant stuff
311 // moved out. We'd like to do all associative operators, but it's especially
312 // important (common) to do address expressions.
313 Node *PhaseIdealLoop::remix_address_expressions( Node *n ) {
314 if (!has_ctrl(n)) return NULL;
315 Node *n_ctrl = get_ctrl(n);
316 IdealLoopTree *n_loop = get_loop(n_ctrl);
318 // See if 'n' mixes loop-varying and loop-invariant inputs and
319 // itself is loop-varying.
321 // Only interested in binary ops (and AddP)
322 if( n->req() < 3 || n->req() > 4 ) return NULL;
324 Node *n1_ctrl = get_ctrl(n->in( 1));
325 Node *n2_ctrl = get_ctrl(n->in( 2));
326 Node *n3_ctrl = get_ctrl(n->in(n->req() == 3 ? 2 : 3));
327 IdealLoopTree *n1_loop = get_loop( n1_ctrl );
328 IdealLoopTree *n2_loop = get_loop( n2_ctrl );
329 IdealLoopTree *n3_loop = get_loop( n3_ctrl );
331 // Does one of my inputs spin in a tighter loop than self?
332 if( (n_loop->is_member( n1_loop ) && n_loop != n1_loop) ||
333 (n_loop->is_member( n2_loop ) && n_loop != n2_loop) ||
334 (n_loop->is_member( n3_loop ) && n_loop != n3_loop) )
335 return NULL; // Leave well enough alone
337 // Is at least one of my inputs loop-invariant?
338 if( n1_loop == n_loop &&
339 n2_loop == n_loop &&
340 n3_loop == n_loop )
341 return NULL; // No loop-invariant inputs
344 int n_op = n->Opcode();
346 // Replace expressions like ((V+I) << 2) with (V<<2 + I<<2).
347 if( n_op == Op_LShiftI ) {
348 // Scale is loop invariant
349 Node *scale = n->in(2);
350 Node *scale_ctrl = get_ctrl(scale);
351 IdealLoopTree *scale_loop = get_loop(scale_ctrl );
352 if( n_loop == scale_loop || !scale_loop->is_member( n_loop ) )
353 return NULL;
354 const TypeInt *scale_t = scale->bottom_type()->isa_int();
355 if( scale_t && scale_t->is_con() && scale_t->get_con() >= 16 )
356 return NULL; // Dont bother with byte/short masking
357 // Add must vary with loop (else shift would be loop-invariant)
358 Node *add = n->in(1);
359 Node *add_ctrl = get_ctrl(add);
360 IdealLoopTree *add_loop = get_loop(add_ctrl);
361 //assert( n_loop == add_loop, "" );
362 if( n_loop != add_loop ) return NULL; // happens w/ evil ZKM loops
364 // Convert I-V into I+ (0-V); same for V-I
365 if( add->Opcode() == Op_SubI &&
366 _igvn.type( add->in(1) ) != TypeInt::ZERO ) {
367 Node *zero = _igvn.intcon(0);
368 set_ctrl(zero, C->root());
369 Node *neg = new (C) SubINode( _igvn.intcon(0), add->in(2) );
370 register_new_node( neg, get_ctrl(add->in(2) ) );
371 add = new (C) AddINode( add->in(1), neg );
372 register_new_node( add, add_ctrl );
373 }
374 if( add->Opcode() != Op_AddI ) return NULL;
375 // See if one add input is loop invariant
376 Node *add_var = add->in(1);
377 Node *add_var_ctrl = get_ctrl(add_var);
378 IdealLoopTree *add_var_loop = get_loop(add_var_ctrl );
379 Node *add_invar = add->in(2);
380 Node *add_invar_ctrl = get_ctrl(add_invar);
381 IdealLoopTree *add_invar_loop = get_loop(add_invar_ctrl );
382 if( add_var_loop == n_loop ) {
383 } else if( add_invar_loop == n_loop ) {
384 // Swap to find the invariant part
385 add_invar = add_var;
386 add_invar_ctrl = add_var_ctrl;
387 add_invar_loop = add_var_loop;
388 add_var = add->in(2);
389 Node *add_var_ctrl = get_ctrl(add_var);
390 IdealLoopTree *add_var_loop = get_loop(add_var_ctrl );
391 } else // Else neither input is loop invariant
392 return NULL;
393 if( n_loop == add_invar_loop || !add_invar_loop->is_member( n_loop ) )
394 return NULL; // No invariant part of the add?
396 // Yes! Reshape address expression!
397 Node *inv_scale = new (C) LShiftINode( add_invar, scale );
398 Node *inv_scale_ctrl =
399 dom_depth(add_invar_ctrl) > dom_depth(scale_ctrl) ?
400 add_invar_ctrl : scale_ctrl;
401 register_new_node( inv_scale, inv_scale_ctrl );
402 Node *var_scale = new (C) LShiftINode( add_var, scale );
403 register_new_node( var_scale, n_ctrl );
404 Node *var_add = new (C) AddINode( var_scale, inv_scale );
405 register_new_node( var_add, n_ctrl );
406 _igvn.replace_node( n, var_add );
407 return var_add;
408 }
410 // Replace (I+V) with (V+I)
411 if( n_op == Op_AddI ||
412 n_op == Op_AddL ||
413 n_op == Op_AddF ||
414 n_op == Op_AddD ||
415 n_op == Op_MulI ||
416 n_op == Op_MulL ||
417 n_op == Op_MulF ||
418 n_op == Op_MulD ) {
419 if( n2_loop == n_loop ) {
420 assert( n1_loop != n_loop, "" );
421 n->swap_edges(1, 2);
422 }
423 }
425 // Replace ((I1 +p V) +p I2) with ((I1 +p I2) +p V),
426 // but not if I2 is a constant.
427 if( n_op == Op_AddP ) {
428 if( n2_loop == n_loop && n3_loop != n_loop ) {
429 if( n->in(2)->Opcode() == Op_AddP && !n->in(3)->is_Con() ) {
430 Node *n22_ctrl = get_ctrl(n->in(2)->in(2));
431 Node *n23_ctrl = get_ctrl(n->in(2)->in(3));
432 IdealLoopTree *n22loop = get_loop( n22_ctrl );
433 IdealLoopTree *n23_loop = get_loop( n23_ctrl );
434 if( n22loop != n_loop && n22loop->is_member(n_loop) &&
435 n23_loop == n_loop ) {
436 Node *add1 = new (C) AddPNode( n->in(1), n->in(2)->in(2), n->in(3) );
437 // Stuff new AddP in the loop preheader
438 register_new_node( add1, n_loop->_head->in(LoopNode::EntryControl) );
439 Node *add2 = new (C) AddPNode( n->in(1), add1, n->in(2)->in(3) );
440 register_new_node( add2, n_ctrl );
441 _igvn.replace_node( n, add2 );
442 return add2;
443 }
444 }
445 }
447 // Replace (I1 +p (I2 + V)) with ((I1 +p I2) +p V)
448 if( n2_loop != n_loop && n3_loop == n_loop ) {
449 if( n->in(3)->Opcode() == Op_AddI ) {
450 Node *V = n->in(3)->in(1);
451 Node *I = n->in(3)->in(2);
452 if( is_member(n_loop,get_ctrl(V)) ) {
453 } else {
454 Node *tmp = V; V = I; I = tmp;
455 }
456 if( !is_member(n_loop,get_ctrl(I)) ) {
457 Node *add1 = new (C) AddPNode( n->in(1), n->in(2), I );
458 // Stuff new AddP in the loop preheader
459 register_new_node( add1, n_loop->_head->in(LoopNode::EntryControl) );
460 Node *add2 = new (C) AddPNode( n->in(1), add1, V );
461 register_new_node( add2, n_ctrl );
462 _igvn.replace_node( n, add2 );
463 return add2;
464 }
465 }
466 }
467 }
469 return NULL;
470 }
472 //------------------------------conditional_move-------------------------------
473 // Attempt to replace a Phi with a conditional move. We have some pretty
474 // strict profitability requirements. All Phis at the merge point must
475 // be converted, so we can remove the control flow. We need to limit the
476 // number of c-moves to a small handful. All code that was in the side-arms
477 // of the CFG diamond is now speculatively executed. This code has to be
478 // "cheap enough". We are pretty much limited to CFG diamonds that merge
479 // 1 or 2 items with a total of 1 or 2 ops executed speculatively.
480 Node *PhaseIdealLoop::conditional_move( Node *region ) {
482 assert(region->is_Region(), "sanity check");
483 if (region->req() != 3) return NULL;
485 // Check for CFG diamond
486 Node *lp = region->in(1);
487 Node *rp = region->in(2);
488 if (!lp || !rp) return NULL;
489 Node *lp_c = lp->in(0);
490 if (lp_c == NULL || lp_c != rp->in(0) || !lp_c->is_If()) return NULL;
491 IfNode *iff = lp_c->as_If();
493 // Check for ops pinned in an arm of the diamond.
494 // Can't remove the control flow in this case
495 if (lp->outcnt() > 1) return NULL;
496 if (rp->outcnt() > 1) return NULL;
498 IdealLoopTree* r_loop = get_loop(region);
499 assert(r_loop == get_loop(iff), "sanity");
500 // Always convert to CMOVE if all results are used only outside this loop.
501 bool used_inside_loop = (r_loop == _ltree_root);
503 // Check profitability
504 int cost = 0;
505 int phis = 0;
506 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
507 Node *out = region->fast_out(i);
508 if (!out->is_Phi()) continue; // Ignore other control edges, etc
509 phis++;
510 PhiNode* phi = out->as_Phi();
511 BasicType bt = phi->type()->basic_type();
512 switch (bt) {
513 case T_FLOAT:
514 case T_DOUBLE: {
515 cost += Matcher::float_cmove_cost(); // Could be very expensive
516 break;
517 }
518 case T_LONG: {
519 cost += Matcher::long_cmove_cost(); // May encodes as 2 CMOV's
520 }
521 case T_INT: // These all CMOV fine
522 case T_ADDRESS: { // (RawPtr)
523 cost++;
524 break;
525 }
526 case T_NARROWOOP: // Fall through
527 case T_OBJECT: { // Base oops are OK, but not derived oops
528 const TypeOopPtr *tp = phi->type()->make_ptr()->isa_oopptr();
529 // Derived pointers are Bad (tm): what's the Base (for GC purposes) of a
530 // CMOVE'd derived pointer? It's a CMOVE'd derived base. Thus
531 // CMOVE'ing a derived pointer requires we also CMOVE the base. If we
532 // have a Phi for the base here that we convert to a CMOVE all is well
533 // and good. But if the base is dead, we'll not make a CMOVE. Later
534 // the allocator will have to produce a base by creating a CMOVE of the
535 // relevant bases. This puts the allocator in the business of
536 // manufacturing expensive instructions, generally a bad plan.
537 // Just Say No to Conditionally-Moved Derived Pointers.
538 if (tp && tp->offset() != 0)
539 return NULL;
540 cost++;
541 break;
542 }
543 default:
544 return NULL; // In particular, can't do memory or I/O
545 }
546 // Add in cost any speculative ops
547 for (uint j = 1; j < region->req(); j++) {
548 Node *proj = region->in(j);
549 Node *inp = phi->in(j);
550 if (get_ctrl(inp) == proj) { // Found local op
551 cost++;
552 // Check for a chain of dependent ops; these will all become
553 // speculative in a CMOV.
554 for (uint k = 1; k < inp->req(); k++)
555 if (get_ctrl(inp->in(k)) == proj)
556 cost += ConditionalMoveLimit; // Too much speculative goo
557 }
558 }
559 // See if the Phi is used by a Cmp or Narrow oop Decode/Encode.
560 // This will likely Split-If, a higher-payoff operation.
561 for (DUIterator_Fast kmax, k = phi->fast_outs(kmax); k < kmax; k++) {
562 Node* use = phi->fast_out(k);
563 if (use->is_Cmp() || use->is_DecodeNarrowPtr() || use->is_EncodeNarrowPtr())
564 cost += ConditionalMoveLimit;
565 // Is there a use inside the loop?
566 // Note: check only basic types since CMoveP is pinned.
567 if (!used_inside_loop && is_java_primitive(bt)) {
568 IdealLoopTree* u_loop = get_loop(has_ctrl(use) ? get_ctrl(use) : use);
569 if (r_loop == u_loop || r_loop->is_member(u_loop)) {
570 used_inside_loop = true;
571 }
572 }
573 }
574 }
575 Node* bol = iff->in(1);
576 assert(bol->Opcode() == Op_Bool, "");
577 int cmp_op = bol->in(1)->Opcode();
578 // It is expensive to generate flags from a float compare.
579 // Avoid duplicated float compare.
580 if (phis > 1 && (cmp_op == Op_CmpF || cmp_op == Op_CmpD)) return NULL;
582 float infrequent_prob = PROB_UNLIKELY_MAG(3);
583 // Ignore cost and blocks frequency if CMOVE can be moved outside the loop.
584 if (used_inside_loop) {
585 if (cost >= ConditionalMoveLimit) return NULL; // Too much goo
587 // BlockLayoutByFrequency optimization moves infrequent branch
588 // from hot path. No point in CMOV'ing in such case (110 is used
589 // instead of 100 to take into account not exactness of float value).
590 if (BlockLayoutByFrequency) {
591 infrequent_prob = MAX2(infrequent_prob, (float)BlockLayoutMinDiamondPercentage/110.0f);
592 }
593 }
594 // Check for highly predictable branch. No point in CMOV'ing if
595 // we are going to predict accurately all the time.
596 if (iff->_prob < infrequent_prob ||
597 iff->_prob > (1.0f - infrequent_prob))
598 return NULL;
600 // --------------
601 // Now replace all Phis with CMOV's
602 Node *cmov_ctrl = iff->in(0);
603 uint flip = (lp->Opcode() == Op_IfTrue);
604 while (1) {
605 PhiNode* phi = NULL;
606 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
607 Node *out = region->fast_out(i);
608 if (out->is_Phi()) {
609 phi = out->as_Phi();
610 break;
611 }
612 }
613 if (phi == NULL) break;
614 #ifndef PRODUCT
615 if (PrintOpto && VerifyLoopOptimizations) tty->print_cr("CMOV");
616 #endif
617 // Move speculative ops
618 for (uint j = 1; j < region->req(); j++) {
619 Node *proj = region->in(j);
620 Node *inp = phi->in(j);
621 if (get_ctrl(inp) == proj) { // Found local op
622 #ifndef PRODUCT
623 if (PrintOpto && VerifyLoopOptimizations) {
624 tty->print(" speculate: ");
625 inp->dump();
626 }
627 #endif
628 set_ctrl(inp, cmov_ctrl);
629 }
630 }
631 Node *cmov = CMoveNode::make( C, cmov_ctrl, iff->in(1), phi->in(1+flip), phi->in(2-flip), _igvn.type(phi) );
632 register_new_node( cmov, cmov_ctrl );
633 _igvn.replace_node( phi, cmov );
634 #ifndef PRODUCT
635 if (TraceLoopOpts) {
636 tty->print("CMOV ");
637 r_loop->dump_head();
638 if (Verbose) {
639 bol->in(1)->dump(1);
640 cmov->dump(1);
641 }
642 }
643 if (VerifyLoopOptimizations) verify();
644 #endif
645 }
647 // The useless CFG diamond will fold up later; see the optimization in
648 // RegionNode::Ideal.
649 _igvn._worklist.push(region);
651 return iff->in(1);
652 }
654 //------------------------------split_if_with_blocks_pre-----------------------
655 // Do the real work in a non-recursive function. Data nodes want to be
656 // cloned in the pre-order so they can feed each other nicely.
657 Node *PhaseIdealLoop::split_if_with_blocks_pre( Node *n ) {
658 // Cloning these guys is unlikely to win
659 int n_op = n->Opcode();
660 if( n_op == Op_MergeMem ) return n;
661 if( n->is_Proj() ) return n;
662 // Do not clone-up CmpFXXX variations, as these are always
663 // followed by a CmpI
664 if( n->is_Cmp() ) return n;
665 // Attempt to use a conditional move instead of a phi/branch
666 if( ConditionalMoveLimit > 0 && n_op == Op_Region ) {
667 Node *cmov = conditional_move( n );
668 if( cmov ) return cmov;
669 }
670 if( n->is_CFG() || n->is_LoadStore() )
671 return n;
672 if( n_op == Op_Opaque1 || // Opaque nodes cannot be mod'd
673 n_op == Op_Opaque2 ) {
674 if( !C->major_progress() ) // If chance of no more loop opts...
675 _igvn._worklist.push(n); // maybe we'll remove them
676 return n;
677 }
679 if( n->is_Con() ) return n; // No cloning for Con nodes
681 Node *n_ctrl = get_ctrl(n);
682 if( !n_ctrl ) return n; // Dead node
684 // Attempt to remix address expressions for loop invariants
685 Node *m = remix_address_expressions( n );
686 if( m ) return m;
688 // Determine if the Node has inputs from some local Phi.
689 // Returns the block to clone thru.
690 Node *n_blk = has_local_phi_input( n );
691 if( !n_blk ) return n;
692 // Do not clone the trip counter through on a CountedLoop
693 // (messes up the canonical shape).
694 if( n_blk->is_CountedLoop() && n->Opcode() == Op_AddI ) return n;
696 // Check for having no control input; not pinned. Allow
697 // dominating control.
698 if( n->in(0) ) {
699 Node *dom = idom(n_blk);
700 if( dom_lca( n->in(0), dom ) != n->in(0) )
701 return n;
702 }
703 // Policy: when is it profitable. You must get more wins than
704 // policy before it is considered profitable. Policy is usually 0,
705 // so 1 win is considered profitable. Big merges will require big
706 // cloning, so get a larger policy.
707 int policy = n_blk->req() >> 2;
709 // If the loop is a candidate for range check elimination,
710 // delay splitting through it's phi until a later loop optimization
711 if (n_blk->is_CountedLoop()) {
712 IdealLoopTree *lp = get_loop(n_blk);
713 if (lp && lp->_rce_candidate) {
714 return n;
715 }
716 }
718 // Use same limit as split_if_with_blocks_post
719 if( C->unique() > 35000 ) return n; // Method too big
721 // Split 'n' through the merge point if it is profitable
722 Node *phi = split_thru_phi( n, n_blk, policy );
723 if (!phi) return n;
725 // Found a Phi to split thru!
726 // Replace 'n' with the new phi
727 _igvn.replace_node( n, phi );
728 // Moved a load around the loop, 'en-registering' something.
729 if (n_blk->is_Loop() && n->is_Load() &&
730 !phi->in(LoopNode::LoopBackControl)->is_Load())
731 C->set_major_progress();
733 return phi;
734 }
736 static bool merge_point_too_heavy(Compile* C, Node* region) {
737 // Bail out if the region and its phis have too many users.
738 int weight = 0;
739 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
740 weight += region->fast_out(i)->outcnt();
741 }
742 int nodes_left = C->max_node_limit() - C->live_nodes();
743 if (weight * 8 > nodes_left) {
744 #ifndef PRODUCT
745 if (PrintOpto)
746 tty->print_cr("*** Split-if bails out: %d nodes, region weight %d", C->unique(), weight);
747 #endif
748 return true;
749 } else {
750 return false;
751 }
752 }
754 static bool merge_point_safe(Node* region) {
755 // 4799512: Stop split_if_with_blocks from splitting a block with a ConvI2LNode
756 // having a PhiNode input. This sidesteps the dangerous case where the split
757 // ConvI2LNode may become TOP if the input Value() does not
758 // overlap the ConvI2L range, leaving a node which may not dominate its
759 // uses.
760 // A better fix for this problem can be found in the BugTraq entry, but
761 // expediency for Mantis demands this hack.
762 // 6855164: If the merge point has a FastLockNode with a PhiNode input, we stop
763 // split_if_with_blocks from splitting a block because we could not move around
764 // the FastLockNode.
765 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
766 Node* n = region->fast_out(i);
767 if (n->is_Phi()) {
768 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
769 Node* m = n->fast_out(j);
770 if (m->is_FastLock())
771 return false;
772 #ifdef _LP64
773 if (m->Opcode() == Op_ConvI2L)
774 return false;
775 #endif
776 }
777 }
778 }
779 return true;
780 }
783 //------------------------------place_near_use---------------------------------
784 // Place some computation next to use but not inside inner loops.
785 // For inner loop uses move it to the preheader area.
786 Node *PhaseIdealLoop::place_near_use( Node *useblock ) const {
787 IdealLoopTree *u_loop = get_loop( useblock );
788 return (u_loop->_irreducible || u_loop->_child)
789 ? useblock
790 : u_loop->_head->in(LoopNode::EntryControl);
791 }
794 //------------------------------split_if_with_blocks_post----------------------
795 // Do the real work in a non-recursive function. CFG hackery wants to be
796 // in the post-order, so it can dirty the I-DOM info and not use the dirtied
797 // info.
798 void PhaseIdealLoop::split_if_with_blocks_post( Node *n ) {
800 // Cloning Cmp through Phi's involves the split-if transform.
801 // FastLock is not used by an If
802 if( n->is_Cmp() && !n->is_FastLock() ) {
803 if( C->unique() > 35000 ) return; // Method too big
805 // Do not do 'split-if' if irreducible loops are present.
806 if( _has_irreducible_loops )
807 return;
809 Node *n_ctrl = get_ctrl(n);
810 // Determine if the Node has inputs from some local Phi.
811 // Returns the block to clone thru.
812 Node *n_blk = has_local_phi_input( n );
813 if( n_blk != n_ctrl ) return;
815 if( merge_point_too_heavy(C, n_ctrl) )
816 return;
818 if( n->outcnt() != 1 ) return; // Multiple bool's from 1 compare?
819 Node *bol = n->unique_out();
820 assert( bol->is_Bool(), "expect a bool here" );
821 if( bol->outcnt() != 1 ) return;// Multiple branches from 1 compare?
822 Node *iff = bol->unique_out();
824 // Check some safety conditions
825 if( iff->is_If() ) { // Classic split-if?
826 if( iff->in(0) != n_ctrl ) return; // Compare must be in same blk as if
827 } else if (iff->is_CMove()) { // Trying to split-up a CMOVE
828 // Can't split CMove with different control edge.
829 if (iff->in(0) != NULL && iff->in(0) != n_ctrl ) return;
830 if( get_ctrl(iff->in(2)) == n_ctrl ||
831 get_ctrl(iff->in(3)) == n_ctrl )
832 return; // Inputs not yet split-up
833 if ( get_loop(n_ctrl) != get_loop(get_ctrl(iff)) ) {
834 return; // Loop-invar test gates loop-varying CMOVE
835 }
836 } else {
837 return; // some other kind of node, such as an Allocate
838 }
840 // Do not do 'split-if' if some paths are dead. First do dead code
841 // elimination and then see if its still profitable.
842 for( uint i = 1; i < n_ctrl->req(); i++ )
843 if( n_ctrl->in(i) == C->top() )
844 return;
846 // When is split-if profitable? Every 'win' on means some control flow
847 // goes dead, so it's almost always a win.
848 int policy = 0;
849 // If trying to do a 'Split-If' at the loop head, it is only
850 // profitable if the cmp folds up on BOTH paths. Otherwise we
851 // risk peeling a loop forever.
853 // CNC - Disabled for now. Requires careful handling of loop
854 // body selection for the cloned code. Also, make sure we check
855 // for any input path not being in the same loop as n_ctrl. For
856 // irreducible loops we cannot check for 'n_ctrl->is_Loop()'
857 // because the alternative loop entry points won't be converted
858 // into LoopNodes.
859 IdealLoopTree *n_loop = get_loop(n_ctrl);
860 for( uint j = 1; j < n_ctrl->req(); j++ )
861 if( get_loop(n_ctrl->in(j)) != n_loop )
862 return;
864 // Check for safety of the merge point.
865 if( !merge_point_safe(n_ctrl) ) {
866 return;
867 }
869 // Split compare 'n' through the merge point if it is profitable
870 Node *phi = split_thru_phi( n, n_ctrl, policy );
871 if( !phi ) return;
873 // Found a Phi to split thru!
874 // Replace 'n' with the new phi
875 _igvn.replace_node( n, phi );
877 // Now split the bool up thru the phi
878 Node *bolphi = split_thru_phi( bol, n_ctrl, -1 );
879 guarantee(bolphi != NULL, "null boolean phi node");
881 _igvn.replace_node( bol, bolphi );
882 assert( iff->in(1) == bolphi, "" );
884 if( bolphi->Value(&_igvn)->singleton() )
885 return;
887 // Conditional-move? Must split up now
888 if( !iff->is_If() ) {
889 Node *cmovphi = split_thru_phi( iff, n_ctrl, -1 );
890 _igvn.replace_node( iff, cmovphi );
891 return;
892 }
894 // Now split the IF
895 do_split_if( iff );
896 return;
897 }
899 // Check for an IF ready to split; one that has its
900 // condition codes input coming from a Phi at the block start.
901 int n_op = n->Opcode();
903 // Check for an IF being dominated by another IF same test
904 if (n_op == Op_If) {
905 Node *bol = n->in(1);
906 uint max = bol->outcnt();
907 // Check for same test used more than once?
908 if (max > 1 && bol->is_Bool()) {
909 // Search up IDOMs to see if this IF is dominated.
910 Node *cutoff = get_ctrl(bol);
912 // Now search up IDOMs till cutoff, looking for a dominating test
913 Node *prevdom = n;
914 Node *dom = idom(prevdom);
915 while (dom != cutoff) {
916 if (dom->req() > 1 && dom->in(1) == bol && prevdom->in(0) == dom) {
917 // Replace the dominated test with an obvious true or false.
918 // Place it on the IGVN worklist for later cleanup.
919 C->set_major_progress();
920 dominated_by(prevdom, n, false, true);
921 #ifndef PRODUCT
922 if( VerifyLoopOptimizations ) verify();
923 #endif
924 return;
925 }
926 prevdom = dom;
927 dom = idom(prevdom);
928 }
929 }
930 }
932 // See if a shared loop-varying computation has no loop-varying uses.
933 // Happens if something is only used for JVM state in uncommon trap exits,
934 // like various versions of induction variable+offset. Clone the
935 // computation per usage to allow it to sink out of the loop.
936 if (has_ctrl(n) && !n->in(0)) {// n not dead and has no control edge (can float about)
937 Node *n_ctrl = get_ctrl(n);
938 IdealLoopTree *n_loop = get_loop(n_ctrl);
939 if( n_loop != _ltree_root ) {
940 DUIterator_Fast imax, i = n->fast_outs(imax);
941 for (; i < imax; i++) {
942 Node* u = n->fast_out(i);
943 if( !has_ctrl(u) ) break; // Found control user
944 IdealLoopTree *u_loop = get_loop(get_ctrl(u));
945 if( u_loop == n_loop ) break; // Found loop-varying use
946 if( n_loop->is_member( u_loop ) ) break; // Found use in inner loop
947 if( u->Opcode() == Op_Opaque1 ) break; // Found loop limit, bugfix for 4677003
948 }
949 bool did_break = (i < imax); // Did we break out of the previous loop?
950 if (!did_break && n->outcnt() > 1) { // All uses in outer loops!
951 Node *late_load_ctrl = NULL;
952 if (n->is_Load()) {
953 // If n is a load, get and save the result from get_late_ctrl(),
954 // to be later used in calculating the control for n's clones.
955 clear_dom_lca_tags();
956 late_load_ctrl = get_late_ctrl(n, n_ctrl);
957 }
958 // If n is a load, and the late control is the same as the current
959 // control, then the cloning of n is a pointless exercise, because
960 // GVN will ensure that we end up where we started.
961 if (!n->is_Load() || late_load_ctrl != n_ctrl) {
962 for (DUIterator_Last jmin, j = n->last_outs(jmin); j >= jmin; ) {
963 Node *u = n->last_out(j); // Clone private computation per use
964 _igvn.rehash_node_delayed(u);
965 Node *x = n->clone(); // Clone computation
966 Node *x_ctrl = NULL;
967 if( u->is_Phi() ) {
968 // Replace all uses of normal nodes. Replace Phi uses
969 // individually, so the separate Nodes can sink down
970 // different paths.
971 uint k = 1;
972 while( u->in(k) != n ) k++;
973 u->set_req( k, x );
974 // x goes next to Phi input path
975 x_ctrl = u->in(0)->in(k);
976 --j;
977 } else { // Normal use
978 // Replace all uses
979 for( uint k = 0; k < u->req(); k++ ) {
980 if( u->in(k) == n ) {
981 u->set_req( k, x );
982 --j;
983 }
984 }
985 x_ctrl = get_ctrl(u);
986 }
988 // Find control for 'x' next to use but not inside inner loops.
989 // For inner loop uses get the preheader area.
990 x_ctrl = place_near_use(x_ctrl);
992 if (n->is_Load()) {
993 // For loads, add a control edge to a CFG node outside of the loop
994 // to force them to not combine and return back inside the loop
995 // during GVN optimization (4641526).
996 //
997 // Because we are setting the actual control input, factor in
998 // the result from get_late_ctrl() so we respect any
999 // anti-dependences. (6233005).
1000 x_ctrl = dom_lca(late_load_ctrl, x_ctrl);
1002 // Don't allow the control input to be a CFG splitting node.
1003 // Such nodes should only have ProjNodes as outs, e.g. IfNode
1004 // should only have IfTrueNode and IfFalseNode (4985384).
1005 x_ctrl = find_non_split_ctrl(x_ctrl);
1006 assert(dom_depth(n_ctrl) <= dom_depth(x_ctrl), "n is later than its clone");
1008 x->set_req(0, x_ctrl);
1009 }
1010 register_new_node(x, x_ctrl);
1012 // Some institutional knowledge is needed here: 'x' is
1013 // yanked because if the optimizer runs GVN on it all the
1014 // cloned x's will common up and undo this optimization and
1015 // be forced back in the loop. This is annoying because it
1016 // makes +VerifyOpto report false-positives on progress. I
1017 // tried setting control edges on the x's to force them to
1018 // not combine, but the matching gets worried when it tries
1019 // to fold a StoreP and an AddP together (as part of an
1020 // address expression) and the AddP and StoreP have
1021 // different controls.
1022 if (!x->is_Load() && !x->is_DecodeNarrowPtr()) _igvn._worklist.yank(x);
1023 }
1024 _igvn.remove_dead_node(n);
1025 }
1026 }
1027 }
1028 }
1030 // Check for Opaque2's who's loop has disappeared - who's input is in the
1031 // same loop nest as their output. Remove 'em, they are no longer useful.
1032 if( n_op == Op_Opaque2 &&
1033 n->in(1) != NULL &&
1034 get_loop(get_ctrl(n)) == get_loop(get_ctrl(n->in(1))) ) {
1035 _igvn.replace_node( n, n->in(1) );
1036 }
1037 }
1039 //------------------------------split_if_with_blocks---------------------------
1040 // Check for aggressive application of 'split-if' optimization,
1041 // using basic block level info.
1042 void PhaseIdealLoop::split_if_with_blocks( VectorSet &visited, Node_Stack &nstack ) {
1043 Node *n = C->root();
1044 visited.set(n->_idx); // first, mark node as visited
1045 // Do pre-visit work for root
1046 n = split_if_with_blocks_pre( n );
1047 uint cnt = n->outcnt();
1048 uint i = 0;
1049 while (true) {
1050 // Visit all children
1051 if (i < cnt) {
1052 Node* use = n->raw_out(i);
1053 ++i;
1054 if (use->outcnt() != 0 && !visited.test_set(use->_idx)) {
1055 // Now do pre-visit work for this use
1056 use = split_if_with_blocks_pre( use );
1057 nstack.push(n, i); // Save parent and next use's index.
1058 n = use; // Process all children of current use.
1059 cnt = use->outcnt();
1060 i = 0;
1061 }
1062 }
1063 else {
1064 // All of n's children have been processed, complete post-processing.
1065 if (cnt != 0 && !n->is_Con()) {
1066 assert(has_node(n), "no dead nodes");
1067 split_if_with_blocks_post( n );
1068 }
1069 if (nstack.is_empty()) {
1070 // Finished all nodes on stack.
1071 break;
1072 }
1073 // Get saved parent node and next use's index. Visit the rest of uses.
1074 n = nstack.node();
1075 cnt = n->outcnt();
1076 i = nstack.index();
1077 nstack.pop();
1078 }
1079 }
1080 }
1083 //=============================================================================
1084 //
1085 // C L O N E A L O O P B O D Y
1086 //
1088 //------------------------------clone_iff--------------------------------------
1089 // Passed in a Phi merging (recursively) some nearly equivalent Bool/Cmps.
1090 // "Nearly" because all Nodes have been cloned from the original in the loop,
1091 // but the fall-in edges to the Cmp are different. Clone bool/Cmp pairs
1092 // through the Phi recursively, and return a Bool.
1093 BoolNode *PhaseIdealLoop::clone_iff( PhiNode *phi, IdealLoopTree *loop ) {
1095 // Convert this Phi into a Phi merging Bools
1096 uint i;
1097 for( i = 1; i < phi->req(); i++ ) {
1098 Node *b = phi->in(i);
1099 if( b->is_Phi() ) {
1100 _igvn.replace_input_of(phi, i, clone_iff( b->as_Phi(), loop ));
1101 } else {
1102 assert( b->is_Bool(), "" );
1103 }
1104 }
1106 Node *sample_bool = phi->in(1);
1107 Node *sample_cmp = sample_bool->in(1);
1109 // Make Phis to merge the Cmp's inputs.
1110 PhiNode *phi1 = new (C) PhiNode( phi->in(0), Type::TOP );
1111 PhiNode *phi2 = new (C) PhiNode( phi->in(0), Type::TOP );
1112 for( i = 1; i < phi->req(); i++ ) {
1113 Node *n1 = phi->in(i)->in(1)->in(1);
1114 Node *n2 = phi->in(i)->in(1)->in(2);
1115 phi1->set_req( i, n1 );
1116 phi2->set_req( i, n2 );
1117 phi1->set_type( phi1->type()->meet_speculative(n1->bottom_type()));
1118 phi2->set_type( phi2->type()->meet_speculative(n2->bottom_type()));
1119 }
1120 // See if these Phis have been made before.
1121 // Register with optimizer
1122 Node *hit1 = _igvn.hash_find_insert(phi1);
1123 if( hit1 ) { // Hit, toss just made Phi
1124 _igvn.remove_dead_node(phi1); // Remove new phi
1125 assert( hit1->is_Phi(), "" );
1126 phi1 = (PhiNode*)hit1; // Use existing phi
1127 } else { // Miss
1128 _igvn.register_new_node_with_optimizer(phi1);
1129 }
1130 Node *hit2 = _igvn.hash_find_insert(phi2);
1131 if( hit2 ) { // Hit, toss just made Phi
1132 _igvn.remove_dead_node(phi2); // Remove new phi
1133 assert( hit2->is_Phi(), "" );
1134 phi2 = (PhiNode*)hit2; // Use existing phi
1135 } else { // Miss
1136 _igvn.register_new_node_with_optimizer(phi2);
1137 }
1138 // Register Phis with loop/block info
1139 set_ctrl(phi1, phi->in(0));
1140 set_ctrl(phi2, phi->in(0));
1141 // Make a new Cmp
1142 Node *cmp = sample_cmp->clone();
1143 cmp->set_req( 1, phi1 );
1144 cmp->set_req( 2, phi2 );
1145 _igvn.register_new_node_with_optimizer(cmp);
1146 set_ctrl(cmp, phi->in(0));
1148 // Make a new Bool
1149 Node *b = sample_bool->clone();
1150 b->set_req(1,cmp);
1151 _igvn.register_new_node_with_optimizer(b);
1152 set_ctrl(b, phi->in(0));
1154 assert( b->is_Bool(), "" );
1155 return (BoolNode*)b;
1156 }
1158 //------------------------------clone_bool-------------------------------------
1159 // Passed in a Phi merging (recursively) some nearly equivalent Bool/Cmps.
1160 // "Nearly" because all Nodes have been cloned from the original in the loop,
1161 // but the fall-in edges to the Cmp are different. Clone bool/Cmp pairs
1162 // through the Phi recursively, and return a Bool.
1163 CmpNode *PhaseIdealLoop::clone_bool( PhiNode *phi, IdealLoopTree *loop ) {
1164 uint i;
1165 // Convert this Phi into a Phi merging Bools
1166 for( i = 1; i < phi->req(); i++ ) {
1167 Node *b = phi->in(i);
1168 if( b->is_Phi() ) {
1169 _igvn.replace_input_of(phi, i, clone_bool( b->as_Phi(), loop ));
1170 } else {
1171 assert( b->is_Cmp() || b->is_top(), "inputs are all Cmp or TOP" );
1172 }
1173 }
1175 Node *sample_cmp = phi->in(1);
1177 // Make Phis to merge the Cmp's inputs.
1178 PhiNode *phi1 = new (C) PhiNode( phi->in(0), Type::TOP );
1179 PhiNode *phi2 = new (C) PhiNode( phi->in(0), Type::TOP );
1180 for( uint j = 1; j < phi->req(); j++ ) {
1181 Node *cmp_top = phi->in(j); // Inputs are all Cmp or TOP
1182 Node *n1, *n2;
1183 if( cmp_top->is_Cmp() ) {
1184 n1 = cmp_top->in(1);
1185 n2 = cmp_top->in(2);
1186 } else {
1187 n1 = n2 = cmp_top;
1188 }
1189 phi1->set_req( j, n1 );
1190 phi2->set_req( j, n2 );
1191 phi1->set_type(phi1->type()->meet_speculative(n1->bottom_type()));
1192 phi2->set_type(phi2->type()->meet_speculative(n2->bottom_type()));
1193 }
1195 // See if these Phis have been made before.
1196 // Register with optimizer
1197 Node *hit1 = _igvn.hash_find_insert(phi1);
1198 if( hit1 ) { // Hit, toss just made Phi
1199 _igvn.remove_dead_node(phi1); // Remove new phi
1200 assert( hit1->is_Phi(), "" );
1201 phi1 = (PhiNode*)hit1; // Use existing phi
1202 } else { // Miss
1203 _igvn.register_new_node_with_optimizer(phi1);
1204 }
1205 Node *hit2 = _igvn.hash_find_insert(phi2);
1206 if( hit2 ) { // Hit, toss just made Phi
1207 _igvn.remove_dead_node(phi2); // Remove new phi
1208 assert( hit2->is_Phi(), "" );
1209 phi2 = (PhiNode*)hit2; // Use existing phi
1210 } else { // Miss
1211 _igvn.register_new_node_with_optimizer(phi2);
1212 }
1213 // Register Phis with loop/block info
1214 set_ctrl(phi1, phi->in(0));
1215 set_ctrl(phi2, phi->in(0));
1216 // Make a new Cmp
1217 Node *cmp = sample_cmp->clone();
1218 cmp->set_req( 1, phi1 );
1219 cmp->set_req( 2, phi2 );
1220 _igvn.register_new_node_with_optimizer(cmp);
1221 set_ctrl(cmp, phi->in(0));
1223 assert( cmp->is_Cmp(), "" );
1224 return (CmpNode*)cmp;
1225 }
1227 //------------------------------sink_use---------------------------------------
1228 // If 'use' was in the loop-exit block, it now needs to be sunk
1229 // below the post-loop merge point.
1230 void PhaseIdealLoop::sink_use( Node *use, Node *post_loop ) {
1231 if (!use->is_CFG() && get_ctrl(use) == post_loop->in(2)) {
1232 set_ctrl(use, post_loop);
1233 for (DUIterator j = use->outs(); use->has_out(j); j++)
1234 sink_use(use->out(j), post_loop);
1235 }
1236 }
1238 //------------------------------clone_loop-------------------------------------
1239 //
1240 // C L O N E A L O O P B O D Y
1241 //
1242 // This is the basic building block of the loop optimizations. It clones an
1243 // entire loop body. It makes an old_new loop body mapping; with this mapping
1244 // you can find the new-loop equivalent to an old-loop node. All new-loop
1245 // nodes are exactly equal to their old-loop counterparts, all edges are the
1246 // same. All exits from the old-loop now have a RegionNode that merges the
1247 // equivalent new-loop path. This is true even for the normal "loop-exit"
1248 // condition. All uses of loop-invariant old-loop values now come from (one
1249 // or more) Phis that merge their new-loop equivalents.
1250 //
1251 // This operation leaves the graph in an illegal state: there are two valid
1252 // control edges coming from the loop pre-header to both loop bodies. I'll
1253 // definitely have to hack the graph after running this transform.
1254 //
1255 // From this building block I will further edit edges to perform loop peeling
1256 // or loop unrolling or iteration splitting (Range-Check-Elimination), etc.
1257 //
1258 // Parameter side_by_size_idom:
1259 // When side_by_size_idom is NULL, the dominator tree is constructed for
1260 // the clone loop to dominate the original. Used in construction of
1261 // pre-main-post loop sequence.
1262 // When nonnull, the clone and original are side-by-side, both are
1263 // dominated by the side_by_side_idom node. Used in construction of
1264 // unswitched loops.
1265 void PhaseIdealLoop::clone_loop( IdealLoopTree *loop, Node_List &old_new, int dd,
1266 Node* side_by_side_idom) {
1268 // Step 1: Clone the loop body. Make the old->new mapping.
1269 uint i;
1270 for( i = 0; i < loop->_body.size(); i++ ) {
1271 Node *old = loop->_body.at(i);
1272 Node *nnn = old->clone();
1273 old_new.map( old->_idx, nnn );
1274 _igvn.register_new_node_with_optimizer(nnn);
1275 }
1278 // Step 2: Fix the edges in the new body. If the old input is outside the
1279 // loop use it. If the old input is INside the loop, use the corresponding
1280 // new node instead.
1281 for( i = 0; i < loop->_body.size(); i++ ) {
1282 Node *old = loop->_body.at(i);
1283 Node *nnn = old_new[old->_idx];
1284 // Fix CFG/Loop controlling the new node
1285 if (has_ctrl(old)) {
1286 set_ctrl(nnn, old_new[get_ctrl(old)->_idx]);
1287 } else {
1288 set_loop(nnn, loop->_parent);
1289 if (old->outcnt() > 0) {
1290 set_idom( nnn, old_new[idom(old)->_idx], dd );
1291 }
1292 }
1293 // Correct edges to the new node
1294 for( uint j = 0; j < nnn->req(); j++ ) {
1295 Node *n = nnn->in(j);
1296 if( n ) {
1297 IdealLoopTree *old_in_loop = get_loop( has_ctrl(n) ? get_ctrl(n) : n );
1298 if( loop->is_member( old_in_loop ) )
1299 nnn->set_req(j, old_new[n->_idx]);
1300 }
1301 }
1302 _igvn.hash_find_insert(nnn);
1303 }
1304 Node *newhead = old_new[loop->_head->_idx];
1305 set_idom(newhead, newhead->in(LoopNode::EntryControl), dd);
1308 // Step 3: Now fix control uses. Loop varying control uses have already
1309 // been fixed up (as part of all input edges in Step 2). Loop invariant
1310 // control uses must be either an IfFalse or an IfTrue. Make a merge
1311 // point to merge the old and new IfFalse/IfTrue nodes; make the use
1312 // refer to this.
1313 ResourceArea *area = Thread::current()->resource_area();
1314 Node_List worklist(area);
1315 uint new_counter = C->unique();
1316 for( i = 0; i < loop->_body.size(); i++ ) {
1317 Node* old = loop->_body.at(i);
1318 if( !old->is_CFG() ) continue;
1319 Node* nnn = old_new[old->_idx];
1321 // Copy uses to a worklist, so I can munge the def-use info
1322 // with impunity.
1323 for (DUIterator_Fast jmax, j = old->fast_outs(jmax); j < jmax; j++)
1324 worklist.push(old->fast_out(j));
1326 while( worklist.size() ) { // Visit all uses
1327 Node *use = worklist.pop();
1328 if (!has_node(use)) continue; // Ignore dead nodes
1329 IdealLoopTree *use_loop = get_loop( has_ctrl(use) ? get_ctrl(use) : use );
1330 if( !loop->is_member( use_loop ) && use->is_CFG() ) {
1331 // Both OLD and USE are CFG nodes here.
1332 assert( use->is_Proj(), "" );
1334 // Clone the loop exit control projection
1335 Node *newuse = use->clone();
1336 newuse->set_req(0,nnn);
1337 _igvn.register_new_node_with_optimizer(newuse);
1338 set_loop(newuse, use_loop);
1339 set_idom(newuse, nnn, dom_depth(nnn) + 1 );
1341 // We need a Region to merge the exit from the peeled body and the
1342 // exit from the old loop body.
1343 RegionNode *r = new (C) RegionNode(3);
1344 // Map the old use to the new merge point
1345 old_new.map( use->_idx, r );
1346 uint dd_r = MIN2(dom_depth(newuse),dom_depth(use));
1347 assert( dd_r >= dom_depth(dom_lca(newuse,use)), "" );
1349 // The original user of 'use' uses 'r' instead.
1350 for (DUIterator_Last lmin, l = use->last_outs(lmin); l >= lmin;) {
1351 Node* useuse = use->last_out(l);
1352 _igvn.rehash_node_delayed(useuse);
1353 uint uses_found = 0;
1354 if( useuse->in(0) == use ) {
1355 useuse->set_req(0, r);
1356 uses_found++;
1357 if( useuse->is_CFG() ) {
1358 assert( dom_depth(useuse) > dd_r, "" );
1359 set_idom(useuse, r, dom_depth(useuse));
1360 }
1361 }
1362 for( uint k = 1; k < useuse->req(); k++ ) {
1363 if( useuse->in(k) == use ) {
1364 useuse->set_req(k, r);
1365 uses_found++;
1366 }
1367 }
1368 l -= uses_found; // we deleted 1 or more copies of this edge
1369 }
1371 // Now finish up 'r'
1372 r->set_req( 1, newuse );
1373 r->set_req( 2, use );
1374 _igvn.register_new_node_with_optimizer(r);
1375 set_loop(r, use_loop);
1376 set_idom(r, !side_by_side_idom ? newuse->in(0) : side_by_side_idom, dd_r);
1377 } // End of if a loop-exit test
1378 }
1379 }
1381 // Step 4: If loop-invariant use is not control, it must be dominated by a
1382 // loop exit IfFalse/IfTrue. Find "proper" loop exit. Make a Region
1383 // there if needed. Make a Phi there merging old and new used values.
1384 Node_List *split_if_set = NULL;
1385 Node_List *split_bool_set = NULL;
1386 Node_List *split_cex_set = NULL;
1387 for( i = 0; i < loop->_body.size(); i++ ) {
1388 Node* old = loop->_body.at(i);
1389 Node* nnn = old_new[old->_idx];
1390 // Copy uses to a worklist, so I can munge the def-use info
1391 // with impunity.
1392 for (DUIterator_Fast jmax, j = old->fast_outs(jmax); j < jmax; j++)
1393 worklist.push(old->fast_out(j));
1395 while( worklist.size() ) {
1396 Node *use = worklist.pop();
1397 if (!has_node(use)) continue; // Ignore dead nodes
1398 if (use->in(0) == C->top()) continue;
1399 IdealLoopTree *use_loop = get_loop( has_ctrl(use) ? get_ctrl(use) : use );
1400 // Check for data-use outside of loop - at least one of OLD or USE
1401 // must not be a CFG node.
1402 if( !loop->is_member( use_loop ) && (!old->is_CFG() || !use->is_CFG())) {
1404 // If the Data use is an IF, that means we have an IF outside of the
1405 // loop that is switching on a condition that is set inside of the
1406 // loop. Happens if people set a loop-exit flag; then test the flag
1407 // in the loop to break the loop, then test is again outside of the
1408 // loop to determine which way the loop exited.
1409 // Loop predicate If node connects to Bool node through Opaque1 node.
1410 if (use->is_If() || use->is_CMove() || C->is_predicate_opaq(use)) {
1411 // Since this code is highly unlikely, we lazily build the worklist
1412 // of such Nodes to go split.
1413 if( !split_if_set )
1414 split_if_set = new Node_List(area);
1415 split_if_set->push(use);
1416 }
1417 if( use->is_Bool() ) {
1418 if( !split_bool_set )
1419 split_bool_set = new Node_List(area);
1420 split_bool_set->push(use);
1421 }
1422 if( use->Opcode() == Op_CreateEx ) {
1423 if( !split_cex_set )
1424 split_cex_set = new Node_List(area);
1425 split_cex_set->push(use);
1426 }
1429 // Get "block" use is in
1430 uint idx = 0;
1431 while( use->in(idx) != old ) idx++;
1432 Node *prev = use->is_CFG() ? use : get_ctrl(use);
1433 assert( !loop->is_member( get_loop( prev ) ), "" );
1434 Node *cfg = prev->_idx >= new_counter
1435 ? prev->in(2)
1436 : idom(prev);
1437 if( use->is_Phi() ) // Phi use is in prior block
1438 cfg = prev->in(idx); // NOT in block of Phi itself
1439 if (cfg->is_top()) { // Use is dead?
1440 _igvn.replace_input_of(use, idx, C->top());
1441 continue;
1442 }
1444 while( !loop->is_member( get_loop( cfg ) ) ) {
1445 prev = cfg;
1446 cfg = cfg->_idx >= new_counter ? cfg->in(2) : idom(cfg);
1447 }
1448 // If the use occurs after merging several exits from the loop, then
1449 // old value must have dominated all those exits. Since the same old
1450 // value was used on all those exits we did not need a Phi at this
1451 // merge point. NOW we do need a Phi here. Each loop exit value
1452 // is now merged with the peeled body exit; each exit gets its own
1453 // private Phi and those Phis need to be merged here.
1454 Node *phi;
1455 if( prev->is_Region() ) {
1456 if( idx == 0 ) { // Updating control edge?
1457 phi = prev; // Just use existing control
1458 } else { // Else need a new Phi
1459 phi = PhiNode::make( prev, old );
1460 // Now recursively fix up the new uses of old!
1461 for( uint i = 1; i < prev->req(); i++ ) {
1462 worklist.push(phi); // Onto worklist once for each 'old' input
1463 }
1464 }
1465 } else {
1466 // Get new RegionNode merging old and new loop exits
1467 prev = old_new[prev->_idx];
1468 assert( prev, "just made this in step 7" );
1469 if( idx == 0 ) { // Updating control edge?
1470 phi = prev; // Just use existing control
1471 } else { // Else need a new Phi
1472 // Make a new Phi merging data values properly
1473 phi = PhiNode::make( prev, old );
1474 phi->set_req( 1, nnn );
1475 }
1476 }
1477 // If inserting a new Phi, check for prior hits
1478 if( idx != 0 ) {
1479 Node *hit = _igvn.hash_find_insert(phi);
1480 if( hit == NULL ) {
1481 _igvn.register_new_node_with_optimizer(phi); // Register new phi
1482 } else { // or
1483 // Remove the new phi from the graph and use the hit
1484 _igvn.remove_dead_node(phi);
1485 phi = hit; // Use existing phi
1486 }
1487 set_ctrl(phi, prev);
1488 }
1489 // Make 'use' use the Phi instead of the old loop body exit value
1490 _igvn.replace_input_of(use, idx, phi);
1491 if( use->_idx >= new_counter ) { // If updating new phis
1492 // Not needed for correctness, but prevents a weak assert
1493 // in AddPNode from tripping (when we end up with different
1494 // base & derived Phis that will become the same after
1495 // IGVN does CSE).
1496 Node *hit = _igvn.hash_find_insert(use);
1497 if( hit ) // Go ahead and re-hash for hits.
1498 _igvn.replace_node( use, hit );
1499 }
1501 // If 'use' was in the loop-exit block, it now needs to be sunk
1502 // below the post-loop merge point.
1503 sink_use( use, prev );
1504 }
1505 }
1506 }
1508 // Check for IFs that need splitting/cloning. Happens if an IF outside of
1509 // the loop uses a condition set in the loop. The original IF probably
1510 // takes control from one or more OLD Regions (which in turn get from NEW
1511 // Regions). In any case, there will be a set of Phis for each merge point
1512 // from the IF up to where the original BOOL def exists the loop.
1513 if( split_if_set ) {
1514 while( split_if_set->size() ) {
1515 Node *iff = split_if_set->pop();
1516 if( iff->in(1)->is_Phi() ) {
1517 BoolNode *b = clone_iff( iff->in(1)->as_Phi(), loop );
1518 _igvn.replace_input_of(iff, 1, b);
1519 }
1520 }
1521 }
1522 if( split_bool_set ) {
1523 while( split_bool_set->size() ) {
1524 Node *b = split_bool_set->pop();
1525 Node *phi = b->in(1);
1526 assert( phi->is_Phi(), "" );
1527 CmpNode *cmp = clone_bool( (PhiNode*)phi, loop );
1528 _igvn.replace_input_of(b, 1, cmp);
1529 }
1530 }
1531 if( split_cex_set ) {
1532 while( split_cex_set->size() ) {
1533 Node *b = split_cex_set->pop();
1534 assert( b->in(0)->is_Region(), "" );
1535 assert( b->in(1)->is_Phi(), "" );
1536 assert( b->in(0)->in(0) == b->in(1)->in(0), "" );
1537 split_up( b, b->in(0), NULL );
1538 }
1539 }
1541 }
1544 //---------------------- stride_of_possible_iv -------------------------------------
1545 // Looks for an iff/bool/comp with one operand of the compare
1546 // being a cycle involving an add and a phi,
1547 // with an optional truncation (left-shift followed by a right-shift)
1548 // of the add. Returns zero if not an iv.
1549 int PhaseIdealLoop::stride_of_possible_iv(Node* iff) {
1550 Node* trunc1 = NULL;
1551 Node* trunc2 = NULL;
1552 const TypeInt* ttype = NULL;
1553 if (!iff->is_If() || iff->in(1) == NULL || !iff->in(1)->is_Bool()) {
1554 return 0;
1555 }
1556 BoolNode* bl = iff->in(1)->as_Bool();
1557 Node* cmp = bl->in(1);
1558 if (!cmp || cmp->Opcode() != Op_CmpI && cmp->Opcode() != Op_CmpU) {
1559 return 0;
1560 }
1561 // Must have an invariant operand
1562 if (is_member(get_loop(iff), get_ctrl(cmp->in(2)))) {
1563 return 0;
1564 }
1565 Node* add2 = NULL;
1566 Node* cmp1 = cmp->in(1);
1567 if (cmp1->is_Phi()) {
1568 // (If (Bool (CmpX phi:(Phi ...(Optional-trunc(AddI phi add2))) )))
1569 Node* phi = cmp1;
1570 for (uint i = 1; i < phi->req(); i++) {
1571 Node* in = phi->in(i);
1572 Node* add = CountedLoopNode::match_incr_with_optional_truncation(in,
1573 &trunc1, &trunc2, &ttype);
1574 if (add && add->in(1) == phi) {
1575 add2 = add->in(2);
1576 break;
1577 }
1578 }
1579 } else {
1580 // (If (Bool (CmpX addtrunc:(Optional-trunc((AddI (Phi ...addtrunc...) add2)) )))
1581 Node* addtrunc = cmp1;
1582 Node* add = CountedLoopNode::match_incr_with_optional_truncation(addtrunc,
1583 &trunc1, &trunc2, &ttype);
1584 if (add && add->in(1)->is_Phi()) {
1585 Node* phi = add->in(1);
1586 for (uint i = 1; i < phi->req(); i++) {
1587 if (phi->in(i) == addtrunc) {
1588 add2 = add->in(2);
1589 break;
1590 }
1591 }
1592 }
1593 }
1594 if (add2 != NULL) {
1595 const TypeInt* add2t = _igvn.type(add2)->is_int();
1596 if (add2t->is_con()) {
1597 return add2t->get_con();
1598 }
1599 }
1600 return 0;
1601 }
1604 //---------------------- stay_in_loop -------------------------------------
1605 // Return the (unique) control output node that's in the loop (if it exists.)
1606 Node* PhaseIdealLoop::stay_in_loop( Node* n, IdealLoopTree *loop) {
1607 Node* unique = NULL;
1608 if (!n) return NULL;
1609 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1610 Node* use = n->fast_out(i);
1611 if (!has_ctrl(use) && loop->is_member(get_loop(use))) {
1612 if (unique != NULL) {
1613 return NULL;
1614 }
1615 unique = use;
1616 }
1617 }
1618 return unique;
1619 }
1621 //------------------------------ register_node -------------------------------------
1622 // Utility to register node "n" with PhaseIdealLoop
1623 void PhaseIdealLoop::register_node(Node* n, IdealLoopTree *loop, Node* pred, int ddepth) {
1624 _igvn.register_new_node_with_optimizer(n);
1625 loop->_body.push(n);
1626 if (n->is_CFG()) {
1627 set_loop(n, loop);
1628 set_idom(n, pred, ddepth);
1629 } else {
1630 set_ctrl(n, pred);
1631 }
1632 }
1634 //------------------------------ proj_clone -------------------------------------
1635 // Utility to create an if-projection
1636 ProjNode* PhaseIdealLoop::proj_clone(ProjNode* p, IfNode* iff) {
1637 ProjNode* c = p->clone()->as_Proj();
1638 c->set_req(0, iff);
1639 return c;
1640 }
1642 //------------------------------ short_circuit_if -------------------------------------
1643 // Force the iff control output to be the live_proj
1644 Node* PhaseIdealLoop::short_circuit_if(IfNode* iff, ProjNode* live_proj) {
1645 guarantee(live_proj != NULL, "null projection");
1646 int proj_con = live_proj->_con;
1647 assert(proj_con == 0 || proj_con == 1, "false or true projection");
1648 Node *con = _igvn.intcon(proj_con);
1649 set_ctrl(con, C->root());
1650 if (iff) {
1651 iff->set_req(1, con);
1652 }
1653 return con;
1654 }
1656 //------------------------------ insert_if_before_proj -------------------------------------
1657 // Insert a new if before an if projection (* - new node)
1658 //
1659 // before
1660 // if(test)
1661 // / \
1662 // v v
1663 // other-proj proj (arg)
1664 //
1665 // after
1666 // if(test)
1667 // / \
1668 // / v
1669 // | * proj-clone
1670 // v |
1671 // other-proj v
1672 // * new_if(relop(cmp[IU](left,right)))
1673 // / \
1674 // v v
1675 // * new-proj proj
1676 // (returned)
1677 //
1678 ProjNode* PhaseIdealLoop::insert_if_before_proj(Node* left, bool Signed, BoolTest::mask relop, Node* right, ProjNode* proj) {
1679 IfNode* iff = proj->in(0)->as_If();
1680 IdealLoopTree *loop = get_loop(proj);
1681 ProjNode *other_proj = iff->proj_out(!proj->is_IfTrue())->as_Proj();
1682 int ddepth = dom_depth(proj);
1684 _igvn.rehash_node_delayed(iff);
1685 _igvn.rehash_node_delayed(proj);
1687 proj->set_req(0, NULL); // temporary disconnect
1688 ProjNode* proj2 = proj_clone(proj, iff);
1689 register_node(proj2, loop, iff, ddepth);
1691 Node* cmp = Signed ? (Node*) new (C)CmpINode(left, right) : (Node*) new (C)CmpUNode(left, right);
1692 register_node(cmp, loop, proj2, ddepth);
1694 BoolNode* bol = new (C)BoolNode(cmp, relop);
1695 register_node(bol, loop, proj2, ddepth);
1697 IfNode* new_if = new (C)IfNode(proj2, bol, iff->_prob, iff->_fcnt);
1698 register_node(new_if, loop, proj2, ddepth);
1700 proj->set_req(0, new_if); // reattach
1701 set_idom(proj, new_if, ddepth);
1703 ProjNode* new_exit = proj_clone(other_proj, new_if)->as_Proj();
1704 guarantee(new_exit != NULL, "null exit node");
1705 register_node(new_exit, get_loop(other_proj), new_if, ddepth);
1707 return new_exit;
1708 }
1710 //------------------------------ insert_region_before_proj -------------------------------------
1711 // Insert a region before an if projection (* - new node)
1712 //
1713 // before
1714 // if(test)
1715 // / |
1716 // v |
1717 // proj v
1718 // other-proj
1719 //
1720 // after
1721 // if(test)
1722 // / |
1723 // v |
1724 // * proj-clone v
1725 // | other-proj
1726 // v
1727 // * new-region
1728 // |
1729 // v
1730 // * dum_if
1731 // / \
1732 // v \
1733 // * dum-proj v
1734 // proj
1735 //
1736 RegionNode* PhaseIdealLoop::insert_region_before_proj(ProjNode* proj) {
1737 IfNode* iff = proj->in(0)->as_If();
1738 IdealLoopTree *loop = get_loop(proj);
1739 ProjNode *other_proj = iff->proj_out(!proj->is_IfTrue())->as_Proj();
1740 int ddepth = dom_depth(proj);
1742 _igvn.rehash_node_delayed(iff);
1743 _igvn.rehash_node_delayed(proj);
1745 proj->set_req(0, NULL); // temporary disconnect
1746 ProjNode* proj2 = proj_clone(proj, iff);
1747 register_node(proj2, loop, iff, ddepth);
1749 RegionNode* reg = new (C)RegionNode(2);
1750 reg->set_req(1, proj2);
1751 register_node(reg, loop, iff, ddepth);
1753 IfNode* dum_if = new (C)IfNode(reg, short_circuit_if(NULL, proj), iff->_prob, iff->_fcnt);
1754 register_node(dum_if, loop, reg, ddepth);
1756 proj->set_req(0, dum_if); // reattach
1757 set_idom(proj, dum_if, ddepth);
1759 ProjNode* dum_proj = proj_clone(other_proj, dum_if);
1760 register_node(dum_proj, loop, dum_if, ddepth);
1762 return reg;
1763 }
1765 //------------------------------ insert_cmpi_loop_exit -------------------------------------
1766 // Clone a signed compare loop exit from an unsigned compare and
1767 // insert it before the unsigned cmp on the stay-in-loop path.
1768 // All new nodes inserted in the dominator tree between the original
1769 // if and it's projections. The original if test is replaced with
1770 // a constant to force the stay-in-loop path.
1771 //
1772 // This is done to make sure that the original if and it's projections
1773 // still dominate the same set of control nodes, that the ctrl() relation
1774 // from data nodes to them is preserved, and that their loop nesting is
1775 // preserved.
1776 //
1777 // before
1778 // if(i <u limit) unsigned compare loop exit
1779 // / |
1780 // v v
1781 // exit-proj stay-in-loop-proj
1782 //
1783 // after
1784 // if(stay-in-loop-const) original if
1785 // / |
1786 // / v
1787 // / if(i < limit) new signed test
1788 // / / |
1789 // / / v
1790 // / / if(i <u limit) new cloned unsigned test
1791 // / / / |
1792 // v v v |
1793 // region |
1794 // | |
1795 // dum-if |
1796 // / | |
1797 // ether | |
1798 // v v
1799 // exit-proj stay-in-loop-proj
1800 //
1801 IfNode* PhaseIdealLoop::insert_cmpi_loop_exit(IfNode* if_cmpu, IdealLoopTree *loop) {
1802 const bool Signed = true;
1803 const bool Unsigned = false;
1805 BoolNode* bol = if_cmpu->in(1)->as_Bool();
1806 if (bol->_test._test != BoolTest::lt) return NULL;
1807 CmpNode* cmpu = bol->in(1)->as_Cmp();
1808 if (cmpu->Opcode() != Op_CmpU) return NULL;
1809 int stride = stride_of_possible_iv(if_cmpu);
1810 if (stride == 0) return NULL;
1812 Node* lp_proj = stay_in_loop(if_cmpu, loop);
1813 guarantee(lp_proj != NULL, "null loop node");
1815 ProjNode* lp_continue = lp_proj->as_Proj();
1816 ProjNode* lp_exit = if_cmpu->proj_out(!lp_continue->is_IfTrue())->as_Proj();
1818 Node* limit = NULL;
1819 if (stride > 0) {
1820 limit = cmpu->in(2);
1821 } else {
1822 limit = _igvn.makecon(TypeInt::ZERO);
1823 set_ctrl(limit, C->root());
1824 }
1825 // Create a new region on the exit path
1826 RegionNode* reg = insert_region_before_proj(lp_exit);
1827 guarantee(reg != NULL, "null region node");
1829 // Clone the if-cmpu-true-false using a signed compare
1830 BoolTest::mask rel_i = stride > 0 ? bol->_test._test : BoolTest::ge;
1831 ProjNode* cmpi_exit = insert_if_before_proj(cmpu->in(1), Signed, rel_i, limit, lp_continue);
1832 reg->add_req(cmpi_exit);
1834 // Clone the if-cmpu-true-false
1835 BoolTest::mask rel_u = bol->_test._test;
1836 ProjNode* cmpu_exit = insert_if_before_proj(cmpu->in(1), Unsigned, rel_u, cmpu->in(2), lp_continue);
1837 reg->add_req(cmpu_exit);
1839 // Force original if to stay in loop.
1840 short_circuit_if(if_cmpu, lp_continue);
1842 return cmpi_exit->in(0)->as_If();
1843 }
1845 //------------------------------ remove_cmpi_loop_exit -------------------------------------
1846 // Remove a previously inserted signed compare loop exit.
1847 void PhaseIdealLoop::remove_cmpi_loop_exit(IfNode* if_cmp, IdealLoopTree *loop) {
1848 Node* lp_proj = stay_in_loop(if_cmp, loop);
1849 assert(if_cmp->in(1)->in(1)->Opcode() == Op_CmpI &&
1850 stay_in_loop(lp_proj, loop)->is_If() &&
1851 stay_in_loop(lp_proj, loop)->in(1)->in(1)->Opcode() == Op_CmpU, "inserted cmpi before cmpu");
1852 Node *con = _igvn.makecon(lp_proj->is_IfTrue() ? TypeInt::ONE : TypeInt::ZERO);
1853 set_ctrl(con, C->root());
1854 if_cmp->set_req(1, con);
1855 }
1857 //------------------------------ scheduled_nodelist -------------------------------------
1858 // Create a post order schedule of nodes that are in the
1859 // "member" set. The list is returned in "sched".
1860 // The first node in "sched" is the loop head, followed by
1861 // nodes which have no inputs in the "member" set, and then
1862 // followed by the nodes that have an immediate input dependence
1863 // on a node in "sched".
1864 void PhaseIdealLoop::scheduled_nodelist( IdealLoopTree *loop, VectorSet& member, Node_List &sched ) {
1866 assert(member.test(loop->_head->_idx), "loop head must be in member set");
1867 Arena *a = Thread::current()->resource_area();
1868 VectorSet visited(a);
1869 Node_Stack nstack(a, loop->_body.size());
1871 Node* n = loop->_head; // top of stack is cached in "n"
1872 uint idx = 0;
1873 visited.set(n->_idx);
1875 // Initially push all with no inputs from within member set
1876 for(uint i = 0; i < loop->_body.size(); i++ ) {
1877 Node *elt = loop->_body.at(i);
1878 if (member.test(elt->_idx)) {
1879 bool found = false;
1880 for (uint j = 0; j < elt->req(); j++) {
1881 Node* def = elt->in(j);
1882 if (def && member.test(def->_idx) && def != elt) {
1883 found = true;
1884 break;
1885 }
1886 }
1887 if (!found && elt != loop->_head) {
1888 nstack.push(n, idx);
1889 n = elt;
1890 assert(!visited.test(n->_idx), "not seen yet");
1891 visited.set(n->_idx);
1892 }
1893 }
1894 }
1896 // traverse out's that are in the member set
1897 while (true) {
1898 if (idx < n->outcnt()) {
1899 Node* use = n->raw_out(idx);
1900 idx++;
1901 if (!visited.test_set(use->_idx)) {
1902 if (member.test(use->_idx)) {
1903 nstack.push(n, idx);
1904 n = use;
1905 idx = 0;
1906 }
1907 }
1908 } else {
1909 // All outputs processed
1910 sched.push(n);
1911 if (nstack.is_empty()) break;
1912 n = nstack.node();
1913 idx = nstack.index();
1914 nstack.pop();
1915 }
1916 }
1917 }
1920 //------------------------------ has_use_in_set -------------------------------------
1921 // Has a use in the vector set
1922 bool PhaseIdealLoop::has_use_in_set( Node* n, VectorSet& vset ) {
1923 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
1924 Node* use = n->fast_out(j);
1925 if (vset.test(use->_idx)) {
1926 return true;
1927 }
1928 }
1929 return false;
1930 }
1933 //------------------------------ has_use_internal_to_set -------------------------------------
1934 // Has use internal to the vector set (ie. not in a phi at the loop head)
1935 bool PhaseIdealLoop::has_use_internal_to_set( Node* n, VectorSet& vset, IdealLoopTree *loop ) {
1936 Node* head = loop->_head;
1937 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
1938 Node* use = n->fast_out(j);
1939 if (vset.test(use->_idx) && !(use->is_Phi() && use->in(0) == head)) {
1940 return true;
1941 }
1942 }
1943 return false;
1944 }
1947 //------------------------------ clone_for_use_outside_loop -------------------------------------
1948 // clone "n" for uses that are outside of loop
1949 int PhaseIdealLoop::clone_for_use_outside_loop( IdealLoopTree *loop, Node* n, Node_List& worklist ) {
1950 int cloned = 0;
1951 assert(worklist.size() == 0, "should be empty");
1952 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
1953 Node* use = n->fast_out(j);
1954 if( !loop->is_member(get_loop(has_ctrl(use) ? get_ctrl(use) : use)) ) {
1955 worklist.push(use);
1956 }
1957 }
1958 while( worklist.size() ) {
1959 Node *use = worklist.pop();
1960 if (!has_node(use) || use->in(0) == C->top()) continue;
1961 uint j;
1962 for (j = 0; j < use->req(); j++) {
1963 if (use->in(j) == n) break;
1964 }
1965 assert(j < use->req(), "must be there");
1967 // clone "n" and insert it between the inputs of "n" and the use outside the loop
1968 Node* n_clone = n->clone();
1969 _igvn.replace_input_of(use, j, n_clone);
1970 cloned++;
1971 Node* use_c;
1972 if (!use->is_Phi()) {
1973 use_c = has_ctrl(use) ? get_ctrl(use) : use->in(0);
1974 } else {
1975 // Use in a phi is considered a use in the associated predecessor block
1976 use_c = use->in(0)->in(j);
1977 }
1978 set_ctrl(n_clone, use_c);
1979 assert(!loop->is_member(get_loop(use_c)), "should be outside loop");
1980 get_loop(use_c)->_body.push(n_clone);
1981 _igvn.register_new_node_with_optimizer(n_clone);
1982 #if !defined(PRODUCT)
1983 if (TracePartialPeeling) {
1984 tty->print_cr("loop exit cloning old: %d new: %d newbb: %d", n->_idx, n_clone->_idx, get_ctrl(n_clone)->_idx);
1985 }
1986 #endif
1987 }
1988 return cloned;
1989 }
1992 //------------------------------ clone_for_special_use_inside_loop -------------------------------------
1993 // clone "n" for special uses that are in the not_peeled region.
1994 // If these def-uses occur in separate blocks, the code generator
1995 // marks the method as not compilable. For example, if a "BoolNode"
1996 // is in a different basic block than the "IfNode" that uses it, then
1997 // the compilation is aborted in the code generator.
1998 void PhaseIdealLoop::clone_for_special_use_inside_loop( IdealLoopTree *loop, Node* n,
1999 VectorSet& not_peel, Node_List& sink_list, Node_List& worklist ) {
2000 if (n->is_Phi() || n->is_Load()) {
2001 return;
2002 }
2003 assert(worklist.size() == 0, "should be empty");
2004 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
2005 Node* use = n->fast_out(j);
2006 if ( not_peel.test(use->_idx) &&
2007 (use->is_If() || use->is_CMove() || use->is_Bool()) &&
2008 use->in(1) == n) {
2009 worklist.push(use);
2010 }
2011 }
2012 if (worklist.size() > 0) {
2013 // clone "n" and insert it between inputs of "n" and the use
2014 Node* n_clone = n->clone();
2015 loop->_body.push(n_clone);
2016 _igvn.register_new_node_with_optimizer(n_clone);
2017 set_ctrl(n_clone, get_ctrl(n));
2018 sink_list.push(n_clone);
2019 not_peel <<= n_clone->_idx; // add n_clone to not_peel set.
2020 #if !defined(PRODUCT)
2021 if (TracePartialPeeling) {
2022 tty->print_cr("special not_peeled cloning old: %d new: %d", n->_idx, n_clone->_idx);
2023 }
2024 #endif
2025 while( worklist.size() ) {
2026 Node *use = worklist.pop();
2027 _igvn.rehash_node_delayed(use);
2028 for (uint j = 1; j < use->req(); j++) {
2029 if (use->in(j) == n) {
2030 use->set_req(j, n_clone);
2031 }
2032 }
2033 }
2034 }
2035 }
2038 //------------------------------ insert_phi_for_loop -------------------------------------
2039 // Insert phi(lp_entry_val, back_edge_val) at use->in(idx) for loop lp if phi does not already exist
2040 void PhaseIdealLoop::insert_phi_for_loop( Node* use, uint idx, Node* lp_entry_val, Node* back_edge_val, LoopNode* lp ) {
2041 Node *phi = PhiNode::make(lp, back_edge_val);
2042 phi->set_req(LoopNode::EntryControl, lp_entry_val);
2043 // Use existing phi if it already exists
2044 Node *hit = _igvn.hash_find_insert(phi);
2045 if( hit == NULL ) {
2046 _igvn.register_new_node_with_optimizer(phi);
2047 set_ctrl(phi, lp);
2048 } else {
2049 // Remove the new phi from the graph and use the hit
2050 _igvn.remove_dead_node(phi);
2051 phi = hit;
2052 }
2053 _igvn.replace_input_of(use, idx, phi);
2054 }
2056 #ifdef ASSERT
2057 //------------------------------ is_valid_loop_partition -------------------------------------
2058 // Validate the loop partition sets: peel and not_peel
2059 bool PhaseIdealLoop::is_valid_loop_partition( IdealLoopTree *loop, VectorSet& peel, Node_List& peel_list,
2060 VectorSet& not_peel ) {
2061 uint i;
2062 // Check that peel_list entries are in the peel set
2063 for (i = 0; i < peel_list.size(); i++) {
2064 if (!peel.test(peel_list.at(i)->_idx)) {
2065 return false;
2066 }
2067 }
2068 // Check at loop members are in one of peel set or not_peel set
2069 for (i = 0; i < loop->_body.size(); i++ ) {
2070 Node *def = loop->_body.at(i);
2071 uint di = def->_idx;
2072 // Check that peel set elements are in peel_list
2073 if (peel.test(di)) {
2074 if (not_peel.test(di)) {
2075 return false;
2076 }
2077 // Must be in peel_list also
2078 bool found = false;
2079 for (uint j = 0; j < peel_list.size(); j++) {
2080 if (peel_list.at(j)->_idx == di) {
2081 found = true;
2082 break;
2083 }
2084 }
2085 if (!found) {
2086 return false;
2087 }
2088 } else if (not_peel.test(di)) {
2089 if (peel.test(di)) {
2090 return false;
2091 }
2092 } else {
2093 return false;
2094 }
2095 }
2096 return true;
2097 }
2099 //------------------------------ is_valid_clone_loop_exit_use -------------------------------------
2100 // Ensure a use outside of loop is of the right form
2101 bool PhaseIdealLoop::is_valid_clone_loop_exit_use( IdealLoopTree *loop, Node* use, uint exit_idx) {
2102 Node *use_c = has_ctrl(use) ? get_ctrl(use) : use;
2103 return (use->is_Phi() &&
2104 use_c->is_Region() && use_c->req() == 3 &&
2105 (use_c->in(exit_idx)->Opcode() == Op_IfTrue ||
2106 use_c->in(exit_idx)->Opcode() == Op_IfFalse ||
2107 use_c->in(exit_idx)->Opcode() == Op_JumpProj) &&
2108 loop->is_member( get_loop( use_c->in(exit_idx)->in(0) ) ) );
2109 }
2111 //------------------------------ is_valid_clone_loop_form -------------------------------------
2112 // Ensure that all uses outside of loop are of the right form
2113 bool PhaseIdealLoop::is_valid_clone_loop_form( IdealLoopTree *loop, Node_List& peel_list,
2114 uint orig_exit_idx, uint clone_exit_idx) {
2115 uint len = peel_list.size();
2116 for (uint i = 0; i < len; i++) {
2117 Node *def = peel_list.at(i);
2119 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
2120 Node *use = def->fast_out(j);
2121 Node *use_c = has_ctrl(use) ? get_ctrl(use) : use;
2122 if (!loop->is_member(get_loop(use_c))) {
2123 // use is not in the loop, check for correct structure
2124 if (use->in(0) == def) {
2125 // Okay
2126 } else if (!is_valid_clone_loop_exit_use(loop, use, orig_exit_idx)) {
2127 return false;
2128 }
2129 }
2130 }
2131 }
2132 return true;
2133 }
2134 #endif
2136 //------------------------------ partial_peel -------------------------------------
2137 // Partially peel (aka loop rotation) the top portion of a loop (called
2138 // the peel section below) by cloning it and placing one copy just before
2139 // the new loop head and the other copy at the bottom of the new loop.
2140 //
2141 // before after where it came from
2142 //
2143 // stmt1 stmt1
2144 // loop: stmt2 clone
2145 // stmt2 if condA goto exitA clone
2146 // if condA goto exitA new_loop: new
2147 // stmt3 stmt3 clone
2148 // if !condB goto loop if condB goto exitB clone
2149 // exitB: stmt2 orig
2150 // stmt4 if !condA goto new_loop orig
2151 // exitA: goto exitA
2152 // exitB:
2153 // stmt4
2154 // exitA:
2155 //
2156 // Step 1: find the cut point: an exit test on probable
2157 // induction variable.
2158 // Step 2: schedule (with cloning) operations in the peel
2159 // section that can be executed after the cut into
2160 // the section that is not peeled. This may need
2161 // to clone operations into exit blocks. For
2162 // instance, a reference to A[i] in the not-peel
2163 // section and a reference to B[i] in an exit block
2164 // may cause a left-shift of i by 2 to be placed
2165 // in the peel block. This step will clone the left
2166 // shift into the exit block and sink the left shift
2167 // from the peel to the not-peel section.
2168 // Step 3: clone the loop, retarget the control, and insert
2169 // phis for values that are live across the new loop
2170 // head. This is very dependent on the graph structure
2171 // from clone_loop. It creates region nodes for
2172 // exit control and associated phi nodes for values
2173 // flow out of the loop through that exit. The region
2174 // node is dominated by the clone's control projection.
2175 // So the clone's peel section is placed before the
2176 // new loop head, and the clone's not-peel section is
2177 // forms the top part of the new loop. The original
2178 // peel section forms the tail of the new loop.
2179 // Step 4: update the dominator tree and recompute the
2180 // dominator depth.
2181 //
2182 // orig
2183 //
2184 // stmt1
2185 // |
2186 // v
2187 // loop predicate
2188 // |
2189 // v
2190 // loop<----+
2191 // | |
2192 // stmt2 |
2193 // | |
2194 // v |
2195 // ifA |
2196 // / | |
2197 // v v |
2198 // false true ^ <-- last_peel
2199 // / | |
2200 // / ===|==cut |
2201 // / stmt3 | <-- first_not_peel
2202 // / | |
2203 // | v |
2204 // v ifB |
2205 // exitA: / \ |
2206 // / \ |
2207 // v v |
2208 // false true |
2209 // / \ |
2210 // / ----+
2211 // |
2212 // v
2213 // exitB:
2214 // stmt4
2215 //
2216 //
2217 // after clone loop
2218 //
2219 // stmt1
2220 // |
2221 // v
2222 // loop predicate
2223 // / \
2224 // clone / \ orig
2225 // / \
2226 // / \
2227 // v v
2228 // +---->loop loop<----+
2229 // | | | |
2230 // | stmt2 stmt2 |
2231 // | | | |
2232 // | v v |
2233 // | ifA ifA |
2234 // | | \ / | |
2235 // | v v v v |
2236 // ^ true false false true ^ <-- last_peel
2237 // | | ^ \ / | |
2238 // | cut==|== \ \ / ===|==cut |
2239 // | stmt3 \ \ / stmt3 | <-- first_not_peel
2240 // | | dom | | | |
2241 // | v \ 1v v2 v |
2242 // | ifB regionA ifB |
2243 // | / \ | / \ |
2244 // | / \ v / \ |
2245 // | v v exitA: v v |
2246 // | true false false true |
2247 // | / ^ \ / \ |
2248 // +---- \ \ / ----+
2249 // dom \ /
2250 // \ 1v v2
2251 // regionB
2252 // |
2253 // v
2254 // exitB:
2255 // stmt4
2256 //
2257 //
2258 // after partial peel
2259 //
2260 // stmt1
2261 // |
2262 // v
2263 // loop predicate
2264 // /
2265 // clone / orig
2266 // / TOP
2267 // / \
2268 // v v
2269 // TOP->loop loop----+
2270 // | | |
2271 // stmt2 stmt2 |
2272 // | | |
2273 // v v |
2274 // ifA ifA |
2275 // | \ / | |
2276 // v v v v |
2277 // true false false true | <-- last_peel
2278 // | ^ \ / +------|---+
2279 // +->newloop \ \ / === ==cut | |
2280 // | stmt3 \ \ / TOP | |
2281 // | | dom | | stmt3 | | <-- first_not_peel
2282 // | v \ 1v v2 v | |
2283 // | ifB regionA ifB ^ v
2284 // | / \ | / \ | |
2285 // | / \ v / \ | |
2286 // | v v exitA: v v | |
2287 // | true false false true | |
2288 // | / ^ \ / \ | |
2289 // | | \ \ / v | |
2290 // | | dom \ / TOP | |
2291 // | | \ 1v v2 | |
2292 // ^ v regionB | |
2293 // | | | | |
2294 // | | v ^ v
2295 // | | exitB: | |
2296 // | | stmt4 | |
2297 // | +------------>-----------------+ |
2298 // | |
2299 // +-----------------<---------------------+
2300 //
2301 //
2302 // final graph
2303 //
2304 // stmt1
2305 // |
2306 // v
2307 // loop predicate
2308 // |
2309 // v
2310 // stmt2 clone
2311 // |
2312 // v
2313 // ........> ifA clone
2314 // : / |
2315 // dom / |
2316 // : v v
2317 // : false true
2318 // : | |
2319 // : | v
2320 // : | newloop<-----+
2321 // : | | |
2322 // : | stmt3 clone |
2323 // : | | |
2324 // : | v |
2325 // : | ifB |
2326 // : | / \ |
2327 // : | v v |
2328 // : | false true |
2329 // : | | | |
2330 // : | v stmt2 |
2331 // : | exitB: | |
2332 // : | stmt4 v |
2333 // : | ifA orig |
2334 // : | / \ |
2335 // : | / \ |
2336 // : | v v |
2337 // : | false true |
2338 // : | / \ |
2339 // : v v -----+
2340 // RegionA
2341 // |
2342 // v
2343 // exitA
2344 //
2345 bool PhaseIdealLoop::partial_peel( IdealLoopTree *loop, Node_List &old_new ) {
2347 assert(!loop->_head->is_CountedLoop(), "Non-counted loop only");
2348 if (!loop->_head->is_Loop()) {
2349 return false; }
2351 LoopNode *head = loop->_head->as_Loop();
2353 if (head->is_partial_peel_loop() || head->partial_peel_has_failed()) {
2354 return false;
2355 }
2357 // Check for complex exit control
2358 for(uint ii = 0; ii < loop->_body.size(); ii++ ) {
2359 Node *n = loop->_body.at(ii);
2360 int opc = n->Opcode();
2361 if (n->is_Call() ||
2362 opc == Op_Catch ||
2363 opc == Op_CatchProj ||
2364 opc == Op_Jump ||
2365 opc == Op_JumpProj) {
2366 #if !defined(PRODUCT)
2367 if (TracePartialPeeling) {
2368 tty->print_cr("\nExit control too complex: lp: %d", head->_idx);
2369 }
2370 #endif
2371 return false;
2372 }
2373 }
2375 int dd = dom_depth(head);
2377 // Step 1: find cut point
2379 // Walk up dominators to loop head looking for first loop exit
2380 // which is executed on every path thru loop.
2381 IfNode *peel_if = NULL;
2382 IfNode *peel_if_cmpu = NULL;
2384 Node *iff = loop->tail();
2385 while( iff != head ) {
2386 if( iff->is_If() ) {
2387 Node *ctrl = get_ctrl(iff->in(1));
2388 if (ctrl->is_top()) return false; // Dead test on live IF.
2389 // If loop-varying exit-test, check for induction variable
2390 if( loop->is_member(get_loop(ctrl)) &&
2391 loop->is_loop_exit(iff) &&
2392 is_possible_iv_test(iff)) {
2393 Node* cmp = iff->in(1)->in(1);
2394 if (cmp->Opcode() == Op_CmpI) {
2395 peel_if = iff->as_If();
2396 } else {
2397 assert(cmp->Opcode() == Op_CmpU, "must be CmpI or CmpU");
2398 peel_if_cmpu = iff->as_If();
2399 }
2400 }
2401 }
2402 iff = idom(iff);
2403 }
2404 // Prefer signed compare over unsigned compare.
2405 IfNode* new_peel_if = NULL;
2406 if (peel_if == NULL) {
2407 if (!PartialPeelAtUnsignedTests || peel_if_cmpu == NULL) {
2408 return false; // No peel point found
2409 }
2410 new_peel_if = insert_cmpi_loop_exit(peel_if_cmpu, loop);
2411 if (new_peel_if == NULL) {
2412 return false; // No peel point found
2413 }
2414 peel_if = new_peel_if;
2415 }
2416 Node* last_peel = stay_in_loop(peel_if, loop);
2417 Node* first_not_peeled = stay_in_loop(last_peel, loop);
2418 if (first_not_peeled == NULL || first_not_peeled == head) {
2419 return false;
2420 }
2422 #if !defined(PRODUCT)
2423 if (TraceLoopOpts) {
2424 tty->print("PartialPeel ");
2425 loop->dump_head();
2426 }
2428 if (TracePartialPeeling) {
2429 tty->print_cr("before partial peel one iteration");
2430 Node_List wl;
2431 Node* t = head->in(2);
2432 while (true) {
2433 wl.push(t);
2434 if (t == head) break;
2435 t = idom(t);
2436 }
2437 while (wl.size() > 0) {
2438 Node* tt = wl.pop();
2439 tt->dump();
2440 if (tt == last_peel) tty->print_cr("-- cut --");
2441 }
2442 }
2443 #endif
2444 ResourceArea *area = Thread::current()->resource_area();
2445 VectorSet peel(area);
2446 VectorSet not_peel(area);
2447 Node_List peel_list(area);
2448 Node_List worklist(area);
2449 Node_List sink_list(area);
2451 // Set of cfg nodes to peel are those that are executable from
2452 // the head through last_peel.
2453 assert(worklist.size() == 0, "should be empty");
2454 worklist.push(head);
2455 peel.set(head->_idx);
2456 while (worklist.size() > 0) {
2457 Node *n = worklist.pop();
2458 if (n != last_peel) {
2459 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
2460 Node* use = n->fast_out(j);
2461 if (use->is_CFG() &&
2462 loop->is_member(get_loop(use)) &&
2463 !peel.test_set(use->_idx)) {
2464 worklist.push(use);
2465 }
2466 }
2467 }
2468 }
2470 // Set of non-cfg nodes to peel are those that are control
2471 // dependent on the cfg nodes.
2472 uint i;
2473 for(i = 0; i < loop->_body.size(); i++ ) {
2474 Node *n = loop->_body.at(i);
2475 Node *n_c = has_ctrl(n) ? get_ctrl(n) : n;
2476 if (peel.test(n_c->_idx)) {
2477 peel.set(n->_idx);
2478 } else {
2479 not_peel.set(n->_idx);
2480 }
2481 }
2483 // Step 2: move operations from the peeled section down into the
2484 // not-peeled section
2486 // Get a post order schedule of nodes in the peel region
2487 // Result in right-most operand.
2488 scheduled_nodelist(loop, peel, peel_list );
2490 assert(is_valid_loop_partition(loop, peel, peel_list, not_peel), "bad partition");
2492 // For future check for too many new phis
2493 uint old_phi_cnt = 0;
2494 for (DUIterator_Fast jmax, j = head->fast_outs(jmax); j < jmax; j++) {
2495 Node* use = head->fast_out(j);
2496 if (use->is_Phi()) old_phi_cnt++;
2497 }
2499 #if !defined(PRODUCT)
2500 if (TracePartialPeeling) {
2501 tty->print_cr("\npeeled list");
2502 }
2503 #endif
2505 // Evacuate nodes in peel region into the not_peeled region if possible
2506 uint new_phi_cnt = 0;
2507 uint cloned_for_outside_use = 0;
2508 for (i = 0; i < peel_list.size();) {
2509 Node* n = peel_list.at(i);
2510 #if !defined(PRODUCT)
2511 if (TracePartialPeeling) n->dump();
2512 #endif
2513 bool incr = true;
2514 if ( !n->is_CFG() ) {
2516 if ( has_use_in_set(n, not_peel) ) {
2518 // If not used internal to the peeled region,
2519 // move "n" from peeled to not_peeled region.
2521 if ( !has_use_internal_to_set(n, peel, loop) ) {
2523 // if not pinned and not a load (which maybe anti-dependent on a store)
2524 // and not a CMove (Matcher expects only bool->cmove).
2525 if ( n->in(0) == NULL && !n->is_Load() && !n->is_CMove() ) {
2526 cloned_for_outside_use += clone_for_use_outside_loop( loop, n, worklist );
2527 sink_list.push(n);
2528 peel >>= n->_idx; // delete n from peel set.
2529 not_peel <<= n->_idx; // add n to not_peel set.
2530 peel_list.remove(i);
2531 incr = false;
2532 #if !defined(PRODUCT)
2533 if (TracePartialPeeling) {
2534 tty->print_cr("sink to not_peeled region: %d newbb: %d",
2535 n->_idx, get_ctrl(n)->_idx);
2536 }
2537 #endif
2538 }
2539 } else {
2540 // Otherwise check for special def-use cases that span
2541 // the peel/not_peel boundary such as bool->if
2542 clone_for_special_use_inside_loop( loop, n, not_peel, sink_list, worklist );
2543 new_phi_cnt++;
2544 }
2545 }
2546 }
2547 if (incr) i++;
2548 }
2550 if (new_phi_cnt > old_phi_cnt + PartialPeelNewPhiDelta) {
2551 #if !defined(PRODUCT)
2552 if (TracePartialPeeling) {
2553 tty->print_cr("\nToo many new phis: %d old %d new cmpi: %c",
2554 new_phi_cnt, old_phi_cnt, new_peel_if != NULL?'T':'F');
2555 }
2556 #endif
2557 if (new_peel_if != NULL) {
2558 remove_cmpi_loop_exit(new_peel_if, loop);
2559 }
2560 // Inhibit more partial peeling on this loop
2561 assert(!head->is_partial_peel_loop(), "not partial peeled");
2562 head->mark_partial_peel_failed();
2563 if (cloned_for_outside_use > 0) {
2564 // Terminate this round of loop opts because
2565 // the graph outside this loop was changed.
2566 C->set_major_progress();
2567 return true;
2568 }
2569 return false;
2570 }
2572 // Step 3: clone loop, retarget control, and insert new phis
2574 // Create new loop head for new phis and to hang
2575 // the nodes being moved (sinked) from the peel region.
2576 LoopNode* new_head = new (C) LoopNode(last_peel, last_peel);
2577 new_head->set_unswitch_count(head->unswitch_count()); // Preserve
2578 _igvn.register_new_node_with_optimizer(new_head);
2579 assert(first_not_peeled->in(0) == last_peel, "last_peel <- first_not_peeled");
2580 first_not_peeled->set_req(0, new_head);
2581 set_loop(new_head, loop);
2582 loop->_body.push(new_head);
2583 not_peel.set(new_head->_idx);
2584 set_idom(new_head, last_peel, dom_depth(first_not_peeled));
2585 set_idom(first_not_peeled, new_head, dom_depth(first_not_peeled));
2587 while (sink_list.size() > 0) {
2588 Node* n = sink_list.pop();
2589 set_ctrl(n, new_head);
2590 }
2592 assert(is_valid_loop_partition(loop, peel, peel_list, not_peel), "bad partition");
2594 clone_loop( loop, old_new, dd );
2596 const uint clone_exit_idx = 1;
2597 const uint orig_exit_idx = 2;
2598 assert(is_valid_clone_loop_form( loop, peel_list, orig_exit_idx, clone_exit_idx ), "bad clone loop");
2600 Node* head_clone = old_new[head->_idx];
2601 LoopNode* new_head_clone = old_new[new_head->_idx]->as_Loop();
2602 Node* orig_tail_clone = head_clone->in(2);
2604 // Add phi if "def" node is in peel set and "use" is not
2606 for(i = 0; i < peel_list.size(); i++ ) {
2607 Node *def = peel_list.at(i);
2608 if (!def->is_CFG()) {
2609 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
2610 Node *use = def->fast_out(j);
2611 if (has_node(use) && use->in(0) != C->top() &&
2612 (!peel.test(use->_idx) ||
2613 (use->is_Phi() && use->in(0) == head)) ) {
2614 worklist.push(use);
2615 }
2616 }
2617 while( worklist.size() ) {
2618 Node *use = worklist.pop();
2619 for (uint j = 1; j < use->req(); j++) {
2620 Node* n = use->in(j);
2621 if (n == def) {
2623 // "def" is in peel set, "use" is not in peel set
2624 // or "use" is in the entry boundary (a phi) of the peel set
2626 Node* use_c = has_ctrl(use) ? get_ctrl(use) : use;
2628 if ( loop->is_member(get_loop( use_c )) ) {
2629 // use is in loop
2630 if (old_new[use->_idx] != NULL) { // null for dead code
2631 Node* use_clone = old_new[use->_idx];
2632 _igvn.replace_input_of(use, j, C->top());
2633 insert_phi_for_loop( use_clone, j, old_new[def->_idx], def, new_head_clone );
2634 }
2635 } else {
2636 assert(is_valid_clone_loop_exit_use(loop, use, orig_exit_idx), "clone loop format");
2637 // use is not in the loop, check if the live range includes the cut
2638 Node* lp_if = use_c->in(orig_exit_idx)->in(0);
2639 if (not_peel.test(lp_if->_idx)) {
2640 assert(j == orig_exit_idx, "use from original loop");
2641 insert_phi_for_loop( use, clone_exit_idx, old_new[def->_idx], def, new_head_clone );
2642 }
2643 }
2644 }
2645 }
2646 }
2647 }
2648 }
2650 // Step 3b: retarget control
2652 // Redirect control to the new loop head if a cloned node in
2653 // the not_peeled region has control that points into the peeled region.
2654 // This necessary because the cloned peeled region will be outside
2655 // the loop.
2656 // from to
2657 // cloned-peeled <---+
2658 // new_head_clone: | <--+
2659 // cloned-not_peeled in(0) in(0)
2660 // orig-peeled
2662 for(i = 0; i < loop->_body.size(); i++ ) {
2663 Node *n = loop->_body.at(i);
2664 if (!n->is_CFG() && n->in(0) != NULL &&
2665 not_peel.test(n->_idx) && peel.test(n->in(0)->_idx)) {
2666 Node* n_clone = old_new[n->_idx];
2667 _igvn.replace_input_of(n_clone, 0, new_head_clone);
2668 }
2669 }
2671 // Backedge of the surviving new_head (the clone) is original last_peel
2672 _igvn.replace_input_of(new_head_clone, LoopNode::LoopBackControl, last_peel);
2674 // Cut first node in original not_peel set
2675 _igvn.rehash_node_delayed(new_head); // Multiple edge updates:
2676 new_head->set_req(LoopNode::EntryControl, C->top()); // use rehash_node_delayed / set_req instead of
2677 new_head->set_req(LoopNode::LoopBackControl, C->top()); // multiple replace_input_of calls
2679 // Copy head_clone back-branch info to original head
2680 // and remove original head's loop entry and
2681 // clone head's back-branch
2682 _igvn.rehash_node_delayed(head); // Multiple edge updates
2683 head->set_req(LoopNode::EntryControl, head_clone->in(LoopNode::LoopBackControl));
2684 head->set_req(LoopNode::LoopBackControl, C->top());
2685 _igvn.replace_input_of(head_clone, LoopNode::LoopBackControl, C->top());
2687 // Similarly modify the phis
2688 for (DUIterator_Fast kmax, k = head->fast_outs(kmax); k < kmax; k++) {
2689 Node* use = head->fast_out(k);
2690 if (use->is_Phi() && use->outcnt() > 0) {
2691 Node* use_clone = old_new[use->_idx];
2692 _igvn.rehash_node_delayed(use); // Multiple edge updates
2693 use->set_req(LoopNode::EntryControl, use_clone->in(LoopNode::LoopBackControl));
2694 use->set_req(LoopNode::LoopBackControl, C->top());
2695 _igvn.replace_input_of(use_clone, LoopNode::LoopBackControl, C->top());
2696 }
2697 }
2699 // Step 4: update dominator tree and dominator depth
2701 set_idom(head, orig_tail_clone, dd);
2702 recompute_dom_depth();
2704 // Inhibit more partial peeling on this loop
2705 new_head_clone->set_partial_peel_loop();
2706 C->set_major_progress();
2707 loop->record_for_igvn();
2709 #if !defined(PRODUCT)
2710 if (TracePartialPeeling) {
2711 tty->print_cr("\nafter partial peel one iteration");
2712 Node_List wl(area);
2713 Node* t = last_peel;
2714 while (true) {
2715 wl.push(t);
2716 if (t == head_clone) break;
2717 t = idom(t);
2718 }
2719 while (wl.size() > 0) {
2720 Node* tt = wl.pop();
2721 if (tt == head) tty->print_cr("orig head");
2722 else if (tt == new_head_clone) tty->print_cr("new head");
2723 else if (tt == head_clone) tty->print_cr("clone head");
2724 tt->dump();
2725 }
2726 }
2727 #endif
2728 return true;
2729 }
2731 //------------------------------reorg_offsets----------------------------------
2732 // Reorganize offset computations to lower register pressure. Mostly
2733 // prevent loop-fallout uses of the pre-incremented trip counter (which are
2734 // then alive with the post-incremented trip counter forcing an extra
2735 // register move)
2736 void PhaseIdealLoop::reorg_offsets(IdealLoopTree *loop) {
2737 // Perform it only for canonical counted loops.
2738 // Loop's shape could be messed up by iteration_split_impl.
2739 if (!loop->_head->is_CountedLoop())
2740 return;
2741 if (!loop->_head->as_Loop()->is_valid_counted_loop())
2742 return;
2744 CountedLoopNode *cl = loop->_head->as_CountedLoop();
2745 CountedLoopEndNode *cle = cl->loopexit();
2746 Node *exit = cle->proj_out(false);
2747 Node *phi = cl->phi();
2749 // Check for the special case of folks using the pre-incremented
2750 // trip-counter on the fall-out path (forces the pre-incremented
2751 // and post-incremented trip counter to be live at the same time).
2752 // Fix this by adjusting to use the post-increment trip counter.
2754 bool progress = true;
2755 while (progress) {
2756 progress = false;
2757 for (DUIterator_Fast imax, i = phi->fast_outs(imax); i < imax; i++) {
2758 Node* use = phi->fast_out(i); // User of trip-counter
2759 if (!has_ctrl(use)) continue;
2760 Node *u_ctrl = get_ctrl(use);
2761 if (use->is_Phi()) {
2762 u_ctrl = NULL;
2763 for (uint j = 1; j < use->req(); j++)
2764 if (use->in(j) == phi)
2765 u_ctrl = dom_lca(u_ctrl, use->in(0)->in(j));
2766 }
2767 IdealLoopTree *u_loop = get_loop(u_ctrl);
2768 // Look for loop-invariant use
2769 if (u_loop == loop) continue;
2770 if (loop->is_member(u_loop)) continue;
2771 // Check that use is live out the bottom. Assuming the trip-counter
2772 // update is right at the bottom, uses of of the loop middle are ok.
2773 if (dom_lca(exit, u_ctrl) != exit) continue;
2774 // Hit! Refactor use to use the post-incremented tripcounter.
2775 // Compute a post-increment tripcounter.
2776 Node *opaq = new (C) Opaque2Node( C, cle->incr() );
2777 register_new_node(opaq, exit);
2778 Node *neg_stride = _igvn.intcon(-cle->stride_con());
2779 set_ctrl(neg_stride, C->root());
2780 Node *post = new (C) AddINode( opaq, neg_stride);
2781 register_new_node(post, exit);
2782 _igvn.rehash_node_delayed(use);
2783 for (uint j = 1; j < use->req(); j++) {
2784 if (use->in(j) == phi)
2785 use->set_req(j, post);
2786 }
2787 // Since DU info changed, rerun loop
2788 progress = true;
2789 break;
2790 }
2791 }
2793 }