1 /*
2 * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // Optimization - Graph Style
26
27 #include "incls/_precompiled.incl"
28 #include "incls/_lcm.cpp.incl"
29
30 //------------------------------implicit_null_check----------------------------
31 // Detect implicit-null-check opportunities. Basically, find NULL checks
32 // with suitable memory ops nearby. Use the memory op to do the NULL check.
33 // I can generate a memory op if there is not one nearby.
34 // The proj is the control projection for the not-null case.
35 // The val is the pointer being checked for nullness or
36 // decodeHeapOop_not_null node if it did not fold into address.
37 void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowed_reasons) {
38 // Assume if null check need for 0 offset then always needed
39 // Intel solaris doesn't support any null checks yet and no
40 // mechanism exists (yet) to set the switches at an os_cpu level
41 if( !ImplicitNullChecks || MacroAssembler::needs_explicit_null_check(0)) return;
42
43 // Make sure the ptr-is-null path appears to be uncommon!
44 float f = end()->as_MachIf()->_prob;
45 if( proj->Opcode() == Op_IfTrue ) f = 1.0f - f;
46 if( f > PROB_UNLIKELY_MAG(4) ) return;
47
48 uint bidx = 0; // Capture index of value into memop
49 bool was_store; // Memory op is a store op
50
51 // Get the successor block for if the test ptr is non-null
52 Block* not_null_block; // this one goes with the proj
53 Block* null_block;
54 if (_nodes[_nodes.size()-1] == proj) {
55 null_block = _succs[0];
56 not_null_block = _succs[1];
57 } else {
58 assert(_nodes[_nodes.size()-2] == proj, "proj is one or the other");
59 not_null_block = _succs[0];
60 null_block = _succs[1];
61 }
62 while (null_block->is_Empty() == Block::empty_with_goto) {
63 null_block = null_block->_succs[0];
64 }
65
66 // Search the exception block for an uncommon trap.
67 // (See Parse::do_if and Parse::do_ifnull for the reason
68 // we need an uncommon trap. Briefly, we need a way to
69 // detect failure of this optimization, as in 6366351.)
70 {
71 bool found_trap = false;
72 for (uint i1 = 0; i1 < null_block->_nodes.size(); i1++) {
73 Node* nn = null_block->_nodes[i1];
74 if (nn->is_MachCall() &&
75 nn->as_MachCall()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point()) {
76 const Type* trtype = nn->in(TypeFunc::Parms)->bottom_type();
77 if (trtype->isa_int() && trtype->is_int()->is_con()) {
78 jint tr_con = trtype->is_int()->get_con();
79 Deoptimization::DeoptReason reason = Deoptimization::trap_request_reason(tr_con);
80 Deoptimization::DeoptAction action = Deoptimization::trap_request_action(tr_con);
81 assert((int)reason < (int)BitsPerInt, "recode bit map");
82 if (is_set_nth_bit(allowed_reasons, (int) reason)
83 && action != Deoptimization::Action_none) {
84 // This uncommon trap is sure to recompile, eventually.
85 // When that happens, C->too_many_traps will prevent
86 // this transformation from happening again.
87 found_trap = true;
88 }
89 }
90 break;
91 }
92 }
93 if (!found_trap) {
94 // We did not find an uncommon trap.
95 return;
96 }
97 }
98
99 // Check for decodeHeapOop_not_null node which did not fold into address
100 bool is_decoden = ((intptr_t)val) & 1;
101 val = (Node*)(((intptr_t)val) & ~1);
102
103 assert(!is_decoden || (val->in(0) == NULL) && val->is_Mach() &&
104 (val->as_Mach()->ideal_Opcode() == Op_DecodeN), "sanity");
105
106 // Search the successor block for a load or store who's base value is also
107 // the tested value. There may be several.
108 Node_List *out = new Node_List(Thread::current()->resource_area());
109 MachNode *best = NULL; // Best found so far
110 for (DUIterator i = val->outs(); val->has_out(i); i++) {
111 Node *m = val->out(i);
112 if( !m->is_Mach() ) continue;
113 MachNode *mach = m->as_Mach();
114 was_store = false;
115 int iop = mach->ideal_Opcode();
116 switch( iop ) {
117 case Op_LoadB:
118 case Op_LoadUS:
119 case Op_LoadD:
120 case Op_LoadF:
121 case Op_LoadI:
122 case Op_LoadL:
123 case Op_LoadP:
124 case Op_LoadN:
125 case Op_LoadS:
126 case Op_LoadKlass:
127 case Op_LoadNKlass:
128 case Op_LoadRange:
129 case Op_LoadD_unaligned:
130 case Op_LoadL_unaligned:
131 assert(mach->in(2) == val, "should be address");
132 break;
133 case Op_StoreB:
134 case Op_StoreC:
135 case Op_StoreCM:
136 case Op_StoreD:
137 case Op_StoreF:
138 case Op_StoreI:
139 case Op_StoreL:
140 case Op_StoreP:
141 case Op_StoreN:
142 was_store = true; // Memory op is a store op
143 // Stores will have their address in slot 2 (memory in slot 1).
144 // If the value being nul-checked is in another slot, it means we
145 // are storing the checked value, which does NOT check the value!
146 if( mach->in(2) != val ) continue;
147 break; // Found a memory op?
148 case Op_StrComp:
149 case Op_StrEquals:
150 case Op_StrIndexOf:
151 case Op_AryEq:
152 // Not a legit memory op for implicit null check regardless of
153 // embedded loads
154 continue;
155 default: // Also check for embedded loads
156 if( !mach->needs_anti_dependence_check() )
157 continue; // Not an memory op; skip it
158 if( must_clone[iop] ) {
159 // Do not move nodes which produce flags because
160 // RA will try to clone it to place near branch and
161 // it will cause recompilation, see clone_node().
162 continue;
163 }
164 {
165 // Check that value is used in memory address in
166 // instructions with embedded load (CmpP val1,(val2+off)).
167 Node* base;
168 Node* index;
169 const MachOper* oper = mach->memory_inputs(base, index);
170 if (oper == NULL || oper == (MachOper*)-1) {
171 continue; // Not an memory op; skip it
172 }
173 if (val == base ||
174 val == index && val->bottom_type()->isa_narrowoop()) {
175 break; // Found it
176 } else {
177 continue; // Skip it
178 }
179 }
180 break;
181 }
182 // check if the offset is not too high for implicit exception
183 {
184 intptr_t offset = 0;
185 const TypePtr *adr_type = NULL; // Do not need this return value here
186 const Node* base = mach->get_base_and_disp(offset, adr_type);
187 if (base == NULL || base == NodeSentinel) {
188 // Narrow oop address doesn't have base, only index
189 if( val->bottom_type()->isa_narrowoop() &&
190 MacroAssembler::needs_explicit_null_check(offset) )
191 continue; // Give up if offset is beyond page size
192 // cannot reason about it; is probably not implicit null exception
193 } else {
194 const TypePtr* tptr;
195 if (UseCompressedOops && Universe::narrow_oop_shift() == 0) {
196 // 32-bits narrow oop can be the base of address expressions
197 tptr = base->bottom_type()->make_ptr();
198 } else {
199 // only regular oops are expected here
200 tptr = base->bottom_type()->is_ptr();
201 }
202 // Give up if offset is not a compile-time constant
203 if( offset == Type::OffsetBot || tptr->_offset == Type::OffsetBot )
204 continue;
205 offset += tptr->_offset; // correct if base is offseted
206 if( MacroAssembler::needs_explicit_null_check(offset) )
207 continue; // Give up is reference is beyond 4K page size
208 }
209 }
210
211 // Check ctrl input to see if the null-check dominates the memory op
212 Block *cb = cfg->_bbs[mach->_idx];
213 cb = cb->_idom; // Always hoist at least 1 block
214 if( !was_store ) { // Stores can be hoisted only one block
215 while( cb->_dom_depth > (_dom_depth + 1))
216 cb = cb->_idom; // Hoist loads as far as we want
217 // The non-null-block should dominate the memory op, too. Live
218 // range spilling will insert a spill in the non-null-block if it is
219 // needs to spill the memory op for an implicit null check.
220 if (cb->_dom_depth == (_dom_depth + 1)) {
221 if (cb != not_null_block) continue;
222 cb = cb->_idom;
223 }
224 }
225 if( cb != this ) continue;
226
227 // Found a memory user; see if it can be hoisted to check-block
228 uint vidx = 0; // Capture index of value into memop
229 uint j;
230 for( j = mach->req()-1; j > 0; j-- ) {
231 if( mach->in(j) == val ) {
232 vidx = j;
233 // Ignore DecodeN val which could be hoisted to where needed.
234 if( is_decoden ) continue;
235 }
236 // Block of memory-op input
237 Block *inb = cfg->_bbs[mach->in(j)->_idx];
238 Block *b = this; // Start from nul check
239 while( b != inb && b->_dom_depth > inb->_dom_depth )
240 b = b->_idom; // search upwards for input
241 // See if input dominates null check
242 if( b != inb )
243 break;
244 }
245 if( j > 0 )
246 continue;
247 Block *mb = cfg->_bbs[mach->_idx];
248 // Hoisting stores requires more checks for the anti-dependence case.
249 // Give up hoisting if we have to move the store past any load.
250 if( was_store ) {
251 Block *b = mb; // Start searching here for a local load
252 // mach use (faulting) trying to hoist
253 // n might be blocker to hoisting
254 while( b != this ) {
255 uint k;
256 for( k = 1; k < b->_nodes.size(); k++ ) {
257 Node *n = b->_nodes[k];
258 if( n->needs_anti_dependence_check() &&
259 n->in(LoadNode::Memory) == mach->in(StoreNode::Memory) )
260 break; // Found anti-dependent load
261 }
262 if( k < b->_nodes.size() )
263 break; // Found anti-dependent load
264 // Make sure control does not do a merge (would have to check allpaths)
265 if( b->num_preds() != 2 ) break;
266 b = cfg->_bbs[b->pred(1)->_idx]; // Move up to predecessor block
267 }
268 if( b != this ) continue;
269 }
270
271 // Make sure this memory op is not already being used for a NullCheck
272 Node *e = mb->end();
273 if( e->is_MachNullCheck() && e->in(1) == mach )
274 continue; // Already being used as a NULL check
275
276 // Found a candidate! Pick one with least dom depth - the highest
277 // in the dom tree should be closest to the null check.
278 if( !best ||
279 cfg->_bbs[mach->_idx]->_dom_depth < cfg->_bbs[best->_idx]->_dom_depth ) {
280 best = mach;
281 bidx = vidx;
282
283 }
284 }
285 // No candidate!
286 if( !best ) return;
287
288 // ---- Found an implicit null check
289 extern int implicit_null_checks;
290 implicit_null_checks++;
291
292 if( is_decoden ) {
293 // Check if we need to hoist decodeHeapOop_not_null first.
294 Block *valb = cfg->_bbs[val->_idx];
295 if( this != valb && this->_dom_depth < valb->_dom_depth ) {
296 // Hoist it up to the end of the test block.
297 valb->find_remove(val);
298 this->add_inst(val);
299 cfg->_bbs.map(val->_idx,this);
300 // DecodeN on x86 may kill flags. Check for flag-killing projections
301 // that also need to be hoisted.
302 for (DUIterator_Fast jmax, j = val->fast_outs(jmax); j < jmax; j++) {
303 Node* n = val->fast_out(j);
304 if( n->Opcode() == Op_MachProj ) {
305 cfg->_bbs[n->_idx]->find_remove(n);
306 this->add_inst(n);
307 cfg->_bbs.map(n->_idx,this);
308 }
309 }
310 }
311 }
312 // Hoist the memory candidate up to the end of the test block.
313 Block *old_block = cfg->_bbs[best->_idx];
314 old_block->find_remove(best);
315 add_inst(best);
316 cfg->_bbs.map(best->_idx,this);
317
318 // Move the control dependence
319 if (best->in(0) && best->in(0) == old_block->_nodes[0])
320 best->set_req(0, _nodes[0]);
321
322 // Check for flag-killing projections that also need to be hoisted
323 // Should be DU safe because no edge updates.
324 for (DUIterator_Fast jmax, j = best->fast_outs(jmax); j < jmax; j++) {
325 Node* n = best->fast_out(j);
326 if( n->Opcode() == Op_MachProj ) {
327 cfg->_bbs[n->_idx]->find_remove(n);
328 add_inst(n);
329 cfg->_bbs.map(n->_idx,this);
330 }
331 }
332
333 Compile *C = cfg->C;
334 // proj==Op_True --> ne test; proj==Op_False --> eq test.
335 // One of two graph shapes got matched:
336 // (IfTrue (If (Bool NE (CmpP ptr NULL))))
337 // (IfFalse (If (Bool EQ (CmpP ptr NULL))))
338 // NULL checks are always branch-if-eq. If we see a IfTrue projection
339 // then we are replacing a 'ne' test with a 'eq' NULL check test.
340 // We need to flip the projections to keep the same semantics.
341 if( proj->Opcode() == Op_IfTrue ) {
342 // Swap order of projections in basic block to swap branch targets
343 Node *tmp1 = _nodes[end_idx()+1];
344 Node *tmp2 = _nodes[end_idx()+2];
345 _nodes.map(end_idx()+1, tmp2);
346 _nodes.map(end_idx()+2, tmp1);
347 Node *tmp = new (C, 1) Node(C->top()); // Use not NULL input
348 tmp1->replace_by(tmp);
349 tmp2->replace_by(tmp1);
350 tmp->replace_by(tmp2);
351 tmp->destruct();
352 }
353
354 // Remove the existing null check; use a new implicit null check instead.
355 // Since schedule-local needs precise def-use info, we need to correct
356 // it as well.
357 Node *old_tst = proj->in(0);
358 MachNode *nul_chk = new (C) MachNullCheckNode(old_tst->in(0),best,bidx);
359 _nodes.map(end_idx(),nul_chk);
360 cfg->_bbs.map(nul_chk->_idx,this);
361 // Redirect users of old_test to nul_chk
362 for (DUIterator_Last i2min, i2 = old_tst->last_outs(i2min); i2 >= i2min; --i2)
363 old_tst->last_out(i2)->set_req(0, nul_chk);
364 // Clean-up any dead code
365 for (uint i3 = 0; i3 < old_tst->req(); i3++)
366 old_tst->set_req(i3, NULL);
367
368 cfg->latency_from_uses(nul_chk);
369 cfg->latency_from_uses(best);
370 }
371
372
373 //------------------------------select-----------------------------------------
374 // Select a nice fellow from the worklist to schedule next. If there is only
375 // one choice, then use it. Projections take top priority for correctness
376 // reasons - if I see a projection, then it is next. There are a number of
377 // other special cases, for instructions that consume condition codes, et al.
378 // These are chosen immediately. Some instructions are required to immediately
379 // precede the last instruction in the block, and these are taken last. Of the
380 // remaining cases (most), choose the instruction with the greatest latency
381 // (that is, the most number of pseudo-cycles required to the end of the
382 // routine). If there is a tie, choose the instruction with the most inputs.
383 Node *Block::select(PhaseCFG *cfg, Node_List &worklist, int *ready_cnt, VectorSet &next_call, uint sched_slot) {
384
385 // If only a single entry on the stack, use it
386 uint cnt = worklist.size();
387 if (cnt == 1) {
388 Node *n = worklist[0];
389 worklist.map(0,worklist.pop());
390 return n;
391 }
392
393 uint choice = 0; // Bigger is most important
394 uint latency = 0; // Bigger is scheduled first
395 uint score = 0; // Bigger is better
396 int idx = -1; // Index in worklist
397
398 for( uint i=0; i<cnt; i++ ) { // Inspect entire worklist
399 // Order in worklist is used to break ties.
400 // See caller for how this is used to delay scheduling
401 // of induction variable increments to after the other
402 // uses of the phi are scheduled.
403 Node *n = worklist[i]; // Get Node on worklist
404
405 int iop = n->is_Mach() ? n->as_Mach()->ideal_Opcode() : 0;
406 if( n->is_Proj() || // Projections always win
407 n->Opcode()== Op_Con || // So does constant 'Top'
408 iop == Op_CreateEx || // Create-exception must start block
409 iop == Op_CheckCastPP
410 ) {
411 worklist.map(i,worklist.pop());
412 return n;
413 }
414
415 // Final call in a block must be adjacent to 'catch'
416 Node *e = end();
417 if( e->is_Catch() && e->in(0)->in(0) == n )
418 continue;
419
420 // Memory op for an implicit null check has to be at the end of the block
421 if( e->is_MachNullCheck() && e->in(1) == n )
422 continue;
423
424 uint n_choice = 2;
425
426 // See if this instruction is consumed by a branch. If so, then (as the
427 // branch is the last instruction in the basic block) force it to the
428 // end of the basic block
429 if ( must_clone[iop] ) {
430 // See if any use is a branch
431 bool found_machif = false;
432
433 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
434 Node* use = n->fast_out(j);
435
436 // The use is a conditional branch, make them adjacent
437 if (use->is_MachIf() && cfg->_bbs[use->_idx]==this ) {
438 found_machif = true;
439 break;
440 }
441
442 // More than this instruction pending for successor to be ready,
443 // don't choose this if other opportunities are ready
444 if (ready_cnt[use->_idx] > 1)
445 n_choice = 1;
446 }
447
448 // loop terminated, prefer not to use this instruction
449 if (found_machif)
450 continue;
451 }
452
453 // See if this has a predecessor that is "must_clone", i.e. sets the
454 // condition code. If so, choose this first
455 for (uint j = 0; j < n->req() ; j++) {
456 Node *inn = n->in(j);
457 if (inn) {
458 if (inn->is_Mach() && must_clone[inn->as_Mach()->ideal_Opcode()] ) {
459 n_choice = 3;
460 break;
461 }
462 }
463 }
464
465 // MachTemps should be scheduled last so they are near their uses
466 if (n->is_MachTemp()) {
467 n_choice = 1;
468 }
469
470 uint n_latency = cfg->_node_latency->at_grow(n->_idx);
471 uint n_score = n->req(); // Many inputs get high score to break ties
472
473 // Keep best latency found
474 if( choice < n_choice ||
475 ( choice == n_choice &&
476 ( latency < n_latency ||
477 ( latency == n_latency &&
478 ( score < n_score ))))) {
479 choice = n_choice;
480 latency = n_latency;
481 score = n_score;
482 idx = i; // Also keep index in worklist
483 }
484 } // End of for all ready nodes in worklist
485
486 assert(idx >= 0, "index should be set");
487 Node *n = worklist[(uint)idx]; // Get the winner
488
489 worklist.map((uint)idx, worklist.pop()); // Compress worklist
490 return n;
491 }
492
493
494 //------------------------------set_next_call----------------------------------
495 void Block::set_next_call( Node *n, VectorSet &next_call, Block_Array &bbs ) {
496 if( next_call.test_set(n->_idx) ) return;
497 for( uint i=0; i<n->len(); i++ ) {
498 Node *m = n->in(i);
499 if( !m ) continue; // must see all nodes in block that precede call
500 if( bbs[m->_idx] == this )
501 set_next_call( m, next_call, bbs );
502 }
503 }
504
505 //------------------------------needed_for_next_call---------------------------
506 // Set the flag 'next_call' for each Node that is needed for the next call to
507 // be scheduled. This flag lets me bias scheduling so Nodes needed for the
508 // next subroutine call get priority - basically it moves things NOT needed
509 // for the next call till after the call. This prevents me from trying to
510 // carry lots of stuff live across a call.
511 void Block::needed_for_next_call(Node *this_call, VectorSet &next_call, Block_Array &bbs) {
512 // Find the next control-defining Node in this block
513 Node* call = NULL;
514 for (DUIterator_Fast imax, i = this_call->fast_outs(imax); i < imax; i++) {
515 Node* m = this_call->fast_out(i);
516 if( bbs[m->_idx] == this && // Local-block user
517 m != this_call && // Not self-start node
518 m->is_Call() )
519 call = m;
520 break;
521 }
522 if (call == NULL) return; // No next call (e.g., block end is near)
523 // Set next-call for all inputs to this call
524 set_next_call(call, next_call, bbs);
525 }
526
527 //------------------------------sched_call-------------------------------------
528 uint Block::sched_call( Matcher &matcher, Block_Array &bbs, uint node_cnt, Node_List &worklist, int *ready_cnt, MachCallNode *mcall, VectorSet &next_call ) {
529 RegMask regs;
530
531 // Schedule all the users of the call right now. All the users are
532 // projection Nodes, so they must be scheduled next to the call.
533 // Collect all the defined registers.
534 for (DUIterator_Fast imax, i = mcall->fast_outs(imax); i < imax; i++) {
535 Node* n = mcall->fast_out(i);
536 assert( n->Opcode()==Op_MachProj, "" );
537 --ready_cnt[n->_idx];
538 assert( !ready_cnt[n->_idx], "" );
539 // Schedule next to call
540 _nodes.map(node_cnt++, n);
541 // Collect defined registers
542 regs.OR(n->out_RegMask());
543 // Check for scheduling the next control-definer
544 if( n->bottom_type() == Type::CONTROL )
545 // Warm up next pile of heuristic bits
546 needed_for_next_call(n, next_call, bbs);
547
548 // Children of projections are now all ready
549 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
550 Node* m = n->fast_out(j); // Get user
551 if( bbs[m->_idx] != this ) continue;
552 if( m->is_Phi() ) continue;
553 if( !--ready_cnt[m->_idx] )
554 worklist.push(m);
555 }
556
557 }
558
559 // Act as if the call defines the Frame Pointer.
560 // Certainly the FP is alive and well after the call.
561 regs.Insert(matcher.c_frame_pointer());
562
563 // Set all registers killed and not already defined by the call.
564 uint r_cnt = mcall->tf()->range()->cnt();
565 int op = mcall->ideal_Opcode();
566 MachProjNode *proj = new (matcher.C, 1) MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj );
567 bbs.map(proj->_idx,this);
568 _nodes.insert(node_cnt++, proj);
569
570 // Select the right register save policy.
571 const char * save_policy;
572 switch (op) {
573 case Op_CallRuntime:
574 case Op_CallLeaf:
575 case Op_CallLeafNoFP:
576 // Calling C code so use C calling convention
577 save_policy = matcher._c_reg_save_policy;
578 break;
579
580 case Op_CallStaticJava:
581 case Op_CallDynamicJava:
582 // Calling Java code so use Java calling convention
583 save_policy = matcher._register_save_policy;
584 break;
585
586 default:
587 ShouldNotReachHere();
588 }
589
590 // When using CallRuntime mark SOE registers as killed by the call
591 // so values that could show up in the RegisterMap aren't live in a
592 // callee saved register since the register wouldn't know where to
593 // find them. CallLeaf and CallLeafNoFP are ok because they can't
594 // have debug info on them. Strictly speaking this only needs to be
595 // done for oops since idealreg2debugmask takes care of debug info
596 // references but there no way to handle oops differently than other
597 // pointers as far as the kill mask goes.
598 bool exclude_soe = op == Op_CallRuntime;
599
600 // If the call is a MethodHandle invoke, we need to exclude the
601 // register which is used to save the SP value over MH invokes from
602 // the mask. Otherwise this register could be used for
603 // deoptimization information.
604 if (op == Op_CallStaticJava) {
605 MachCallStaticJavaNode* mcallstaticjava = (MachCallStaticJavaNode*) mcall;
606 if (mcallstaticjava->_method_handle_invoke)
607 proj->_rout.OR(Matcher::method_handle_invoke_SP_save_mask());
608 }
609
610 // Fill in the kill mask for the call
611 for( OptoReg::Name r = OptoReg::Name(0); r < _last_Mach_Reg; r=OptoReg::add(r,1) ) {
612 if( !regs.Member(r) ) { // Not already defined by the call
613 // Save-on-call register?
614 if ((save_policy[r] == 'C') ||
615 (save_policy[r] == 'A') ||
616 ((save_policy[r] == 'E') && exclude_soe)) {
617 proj->_rout.Insert(r);
618 }
619 }
620 }
621
622 return node_cnt;
623 }
624
625
626 //------------------------------schedule_local---------------------------------
627 // Topological sort within a block. Someday become a real scheduler.
628 bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, int *ready_cnt, VectorSet &next_call) {
629 // Already "sorted" are the block start Node (as the first entry), and
630 // the block-ending Node and any trailing control projections. We leave
631 // these alone. PhiNodes and ParmNodes are made to follow the block start
632 // Node. Everything else gets topo-sorted.
633
634 #ifndef PRODUCT
635 if (cfg->trace_opto_pipelining()) {
636 tty->print_cr("# --- schedule_local B%d, before: ---", _pre_order);
637 for (uint i = 0;i < _nodes.size();i++) {
638 tty->print("# ");
639 _nodes[i]->fast_dump();
640 }
641 tty->print_cr("#");
642 }
643 #endif
644
645 // RootNode is already sorted
646 if( _nodes.size() == 1 ) return true;
647
648 // Move PhiNodes and ParmNodes from 1 to cnt up to the start
649 uint node_cnt = end_idx();
650 uint phi_cnt = 1;
651 uint i;
652 for( i = 1; i<node_cnt; i++ ) { // Scan for Phi
653 Node *n = _nodes[i];
654 if( n->is_Phi() || // Found a PhiNode or ParmNode
655 (n->is_Proj() && n->in(0) == head()) ) {
656 // Move guy at 'phi_cnt' to the end; makes a hole at phi_cnt
657 _nodes.map(i,_nodes[phi_cnt]);
658 _nodes.map(phi_cnt++,n); // swap Phi/Parm up front
659 } else { // All others
660 // Count block-local inputs to 'n'
661 uint cnt = n->len(); // Input count
662 uint local = 0;
663 for( uint j=0; j<cnt; j++ ) {
664 Node *m = n->in(j);
665 if( m && cfg->_bbs[m->_idx] == this && !m->is_top() )
666 local++; // One more block-local input
667 }
668 ready_cnt[n->_idx] = local; // Count em up
669
670 // A few node types require changing a required edge to a precedence edge
671 // before allocation.
672 if( UseConcMarkSweepGC || UseG1GC ) {
673 if( n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_StoreCM ) {
674 // Note: Required edges with an index greater than oper_input_base
675 // are not supported by the allocator.
676 // Note2: Can only depend on unmatched edge being last,
677 // can not depend on its absolute position.
678 Node *oop_store = n->in(n->req() - 1);
679 n->del_req(n->req() - 1);
680 n->add_prec(oop_store);
681 assert(cfg->_bbs[oop_store->_idx]->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
682 }
683 }
684 if( n->is_Mach() && n->req() > TypeFunc::Parms &&
685 (n->as_Mach()->ideal_Opcode() == Op_MemBarAcquire ||
686 n->as_Mach()->ideal_Opcode() == Op_MemBarVolatile) ) {
687 // MemBarAcquire could be created without Precedent edge.
688 // del_req() replaces the specified edge with the last input edge
689 // and then removes the last edge. If the specified edge > number of
690 // edges the last edge will be moved outside of the input edges array
691 // and the edge will be lost. This is why this code should be
692 // executed only when Precedent (== TypeFunc::Parms) edge is present.
693 Node *x = n->in(TypeFunc::Parms);
694 n->del_req(TypeFunc::Parms);
695 n->add_prec(x);
696 }
697 }
698 }
699 for(uint i2=i; i2<_nodes.size(); i2++ ) // Trailing guys get zapped count
700 ready_cnt[_nodes[i2]->_idx] = 0;
701
702 // All the prescheduled guys do not hold back internal nodes
703 uint i3;
704 for(i3 = 0; i3<phi_cnt; i3++ ) { // For all pre-scheduled
705 Node *n = _nodes[i3]; // Get pre-scheduled
706 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
707 Node* m = n->fast_out(j);
708 if( cfg->_bbs[m->_idx] ==this ) // Local-block user
709 ready_cnt[m->_idx]--; // Fix ready count
710 }
711 }
712
713 Node_List delay;
714 // Make a worklist
715 Node_List worklist;
716 for(uint i4=i3; i4<node_cnt; i4++ ) { // Put ready guys on worklist
717 Node *m = _nodes[i4];
718 if( !ready_cnt[m->_idx] ) { // Zero ready count?
719 if (m->is_iteratively_computed()) {
720 // Push induction variable increments last to allow other uses
721 // of the phi to be scheduled first. The select() method breaks
722 // ties in scheduling by worklist order.
723 delay.push(m);
724 } else if (m->is_Mach() && m->as_Mach()->ideal_Opcode() == Op_CreateEx) {
725 // Force the CreateEx to the top of the list so it's processed
726 // first and ends up at the start of the block.
727 worklist.insert(0, m);
728 } else {
729 worklist.push(m); // Then on to worklist!
730 }
731 }
732 }
733 while (delay.size()) {
734 Node* d = delay.pop();
735 worklist.push(d);
736 }
737
738 // Warm up the 'next_call' heuristic bits
739 needed_for_next_call(_nodes[0], next_call, cfg->_bbs);
740
741 #ifndef PRODUCT
742 if (cfg->trace_opto_pipelining()) {
743 for (uint j=0; j<_nodes.size(); j++) {
744 Node *n = _nodes[j];
745 int idx = n->_idx;
746 tty->print("# ready cnt:%3d ", ready_cnt[idx]);
747 tty->print("latency:%3d ", cfg->_node_latency->at_grow(idx));
748 tty->print("%4d: %s\n", idx, n->Name());
749 }
750 }
751 #endif
752
753 // Pull from worklist and schedule
754 while( worklist.size() ) { // Worklist is not ready
755
756 #ifndef PRODUCT
757 if (cfg->trace_opto_pipelining()) {
758 tty->print("# ready list:");
759 for( uint i=0; i<worklist.size(); i++ ) { // Inspect entire worklist
760 Node *n = worklist[i]; // Get Node on worklist
761 tty->print(" %d", n->_idx);
762 }
763 tty->cr();
764 }
765 #endif
766
767 // Select and pop a ready guy from worklist
768 Node* n = select(cfg, worklist, ready_cnt, next_call, phi_cnt);
769 _nodes.map(phi_cnt++,n); // Schedule him next
770
771 #ifndef PRODUCT
772 if (cfg->trace_opto_pipelining()) {
773 tty->print("# select %d: %s", n->_idx, n->Name());
774 tty->print(", latency:%d", cfg->_node_latency->at_grow(n->_idx));
775 n->dump();
776 if (Verbose) {
777 tty->print("# ready list:");
778 for( uint i=0; i<worklist.size(); i++ ) { // Inspect entire worklist
779 Node *n = worklist[i]; // Get Node on worklist
780 tty->print(" %d", n->_idx);
781 }
782 tty->cr();
783 }
784 }
785
786 #endif
787 if( n->is_MachCall() ) {
788 MachCallNode *mcall = n->as_MachCall();
789 phi_cnt = sched_call(matcher, cfg->_bbs, phi_cnt, worklist, ready_cnt, mcall, next_call);
790 continue;
791 }
792 // Children are now all ready
793 for (DUIterator_Fast i5max, i5 = n->fast_outs(i5max); i5 < i5max; i5++) {
794 Node* m = n->fast_out(i5); // Get user
795 if( cfg->_bbs[m->_idx] != this ) continue;
796 if( m->is_Phi() ) continue;
797 if( !--ready_cnt[m->_idx] )
798 worklist.push(m);
799 }
800 }
801
802 if( phi_cnt != end_idx() ) {
803 // did not schedule all. Retry, Bailout, or Die
804 Compile* C = matcher.C;
805 if (C->subsume_loads() == true && !C->failing()) {
806 // Retry with subsume_loads == false
807 // If this is the first failure, the sentinel string will "stick"
808 // to the Compile object, and the C2Compiler will see it and retry.
809 C->record_failure(C2Compiler::retry_no_subsuming_loads());
810 }
811 // assert( phi_cnt == end_idx(), "did not schedule all" );
812 return false;
813 }
814
815 #ifndef PRODUCT
816 if (cfg->trace_opto_pipelining()) {
817 tty->print_cr("#");
818 tty->print_cr("# after schedule_local");
819 for (uint i = 0;i < _nodes.size();i++) {
820 tty->print("# ");
821 _nodes[i]->fast_dump();
822 }
823 tty->cr();
824 }
825 #endif
826
827
828 return true;
829 }
830
831 //--------------------------catch_cleanup_fix_all_inputs-----------------------
832 static void catch_cleanup_fix_all_inputs(Node *use, Node *old_def, Node *new_def) {
833 for (uint l = 0; l < use->len(); l++) {
834 if (use->in(l) == old_def) {
835 if (l < use->req()) {
836 use->set_req(l, new_def);
837 } else {
838 use->rm_prec(l);
839 use->add_prec(new_def);
840 l--;
841 }
842 }
843 }
844 }
845
846 //------------------------------catch_cleanup_find_cloned_def------------------
847 static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, Block_Array &bbs, int n_clone_idx) {
848 assert( use_blk != def_blk, "Inter-block cleanup only");
849
850 // The use is some block below the Catch. Find and return the clone of the def
851 // that dominates the use. If there is no clone in a dominating block, then
852 // create a phi for the def in a dominating block.
853
854 // Find which successor block dominates this use. The successor
855 // blocks must all be single-entry (from the Catch only; I will have
856 // split blocks to make this so), hence they all dominate.
857 while( use_blk->_dom_depth > def_blk->_dom_depth+1 )
858 use_blk = use_blk->_idom;
859
860 // Find the successor
861 Node *fixup = NULL;
862
863 uint j;
864 for( j = 0; j < def_blk->_num_succs; j++ )
865 if( use_blk == def_blk->_succs[j] )
866 break;
867
868 if( j == def_blk->_num_succs ) {
869 // Block at same level in dom-tree is not a successor. It needs a
870 // PhiNode, the PhiNode uses from the def and IT's uses need fixup.
871 Node_Array inputs = new Node_List(Thread::current()->resource_area());
872 for(uint k = 1; k < use_blk->num_preds(); k++) {
873 inputs.map(k, catch_cleanup_find_cloned_def(bbs[use_blk->pred(k)->_idx], def, def_blk, bbs, n_clone_idx));
874 }
875
876 // Check to see if the use_blk already has an identical phi inserted.
877 // If it exists, it will be at the first position since all uses of a
878 // def are processed together.
879 Node *phi = use_blk->_nodes[1];
880 if( phi->is_Phi() ) {
881 fixup = phi;
882 for (uint k = 1; k < use_blk->num_preds(); k++) {
883 if (phi->in(k) != inputs[k]) {
884 // Not a match
885 fixup = NULL;
886 break;
887 }
888 }
889 }
890
891 // If an existing PhiNode was not found, make a new one.
892 if (fixup == NULL) {
893 Node *new_phi = PhiNode::make(use_blk->head(), def);
894 use_blk->_nodes.insert(1, new_phi);
895 bbs.map(new_phi->_idx, use_blk);
896 for (uint k = 1; k < use_blk->num_preds(); k++) {
897 new_phi->set_req(k, inputs[k]);
898 }
899 fixup = new_phi;
900 }
901
902 } else {
903 // Found the use just below the Catch. Make it use the clone.
904 fixup = use_blk->_nodes[n_clone_idx];
905 }
906
907 return fixup;
908 }
909
910 //--------------------------catch_cleanup_intra_block--------------------------
911 // Fix all input edges in use that reference "def". The use is in the same
912 // block as the def and both have been cloned in each successor block.
913 static void catch_cleanup_intra_block(Node *use, Node *def, Block *blk, int beg, int n_clone_idx) {
914
915 // Both the use and def have been cloned. For each successor block,
916 // get the clone of the use, and make its input the clone of the def
917 // found in that block.
918
919 uint use_idx = blk->find_node(use);
920 uint offset_idx = use_idx - beg;
921 for( uint k = 0; k < blk->_num_succs; k++ ) {
922 // Get clone in each successor block
923 Block *sb = blk->_succs[k];
924 Node *clone = sb->_nodes[offset_idx+1];
925 assert( clone->Opcode() == use->Opcode(), "" );
926
927 // Make use-clone reference the def-clone
928 catch_cleanup_fix_all_inputs(clone, def, sb->_nodes[n_clone_idx]);
929 }
930 }
931
932 //------------------------------catch_cleanup_inter_block---------------------
933 // Fix all input edges in use that reference "def". The use is in a different
934 // block than the def.
935 static void catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, Block_Array &bbs, int n_clone_idx) {
936 if( !use_blk ) return; // Can happen if the use is a precedence edge
937
938 Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, bbs, n_clone_idx);
939 catch_cleanup_fix_all_inputs(use, def, new_def);
940 }
941
942 //------------------------------call_catch_cleanup-----------------------------
943 // If we inserted any instructions between a Call and his CatchNode,
944 // clone the instructions on all paths below the Catch.
945 void Block::call_catch_cleanup(Block_Array &bbs) {
946
947 // End of region to clone
948 uint end = end_idx();
949 if( !_nodes[end]->is_Catch() ) return;
950 // Start of region to clone
951 uint beg = end;
952 while( _nodes[beg-1]->Opcode() != Op_MachProj ||
953 !_nodes[beg-1]->in(0)->is_Call() ) {
954 beg--;
955 assert(beg > 0,"Catch cleanup walking beyond block boundary");
956 }
957 // Range of inserted instructions is [beg, end)
958 if( beg == end ) return;
959
960 // Clone along all Catch output paths. Clone area between the 'beg' and
961 // 'end' indices.
962 for( uint i = 0; i < _num_succs; i++ ) {
963 Block *sb = _succs[i];
964 // Clone the entire area; ignoring the edge fixup for now.
965 for( uint j = end; j > beg; j-- ) {
966 // It is safe here to clone a node with anti_dependence
967 // since clones dominate on each path.
968 Node *clone = _nodes[j-1]->clone();
969 sb->_nodes.insert( 1, clone );
970 bbs.map(clone->_idx,sb);
971 }
972 }
973
974
975 // Fixup edges. Check the def-use info per cloned Node
976 for(uint i2 = beg; i2 < end; i2++ ) {
977 uint n_clone_idx = i2-beg+1; // Index of clone of n in each successor block
978 Node *n = _nodes[i2]; // Node that got cloned
979 // Need DU safe iterator because of edge manipulation in calls.
980 Unique_Node_List *out = new Unique_Node_List(Thread::current()->resource_area());
981 for (DUIterator_Fast j1max, j1 = n->fast_outs(j1max); j1 < j1max; j1++) {
982 out->push(n->fast_out(j1));
983 }
984 uint max = out->size();
985 for (uint j = 0; j < max; j++) {// For all users
986 Node *use = out->pop();
987 Block *buse = bbs[use->_idx];
988 if( use->is_Phi() ) {
989 for( uint k = 1; k < use->req(); k++ )
990 if( use->in(k) == n ) {
991 Node *fixup = catch_cleanup_find_cloned_def(bbs[buse->pred(k)->_idx], n, this, bbs, n_clone_idx);
992 use->set_req(k, fixup);
993 }
994 } else {
995 if (this == buse) {
996 catch_cleanup_intra_block(use, n, this, beg, n_clone_idx);
997 } else {
998 catch_cleanup_inter_block(use, buse, n, this, bbs, n_clone_idx);
999 }
1000 }
1001 } // End for all users
1002
1003 } // End of for all Nodes in cloned area
1004
1005 // Remove the now-dead cloned ops
1006 for(uint i3 = beg; i3 < end; i3++ ) {
1007 _nodes[beg]->disconnect_inputs(NULL);
1008 _nodes.remove(beg);
1009 }
1010
1011 // If the successor blocks have a CreateEx node, move it back to the top
1012 for(uint i4 = 0; i4 < _num_succs; i4++ ) {
1013 Block *sb = _succs[i4];
1014 uint new_cnt = end - beg;
1015 // Remove any newly created, but dead, nodes.
1016 for( uint j = new_cnt; j > 0; j-- ) {
1017 Node *n = sb->_nodes[j];
1018 if (n->outcnt() == 0 &&
1019 (!n->is_Proj() || n->as_Proj()->in(0)->outcnt() == 1) ){
1020 n->disconnect_inputs(NULL);
1021 sb->_nodes.remove(j);
1022 new_cnt--;
1023 }
1024 }
1025 // If any newly created nodes remain, move the CreateEx node to the top
1026 if (new_cnt > 0) {
1027 Node *cex = sb->_nodes[1+new_cnt];
1028 if( cex->is_Mach() && cex->as_Mach()->ideal_Opcode() == Op_CreateEx ) {
1029 sb->_nodes.remove(1+new_cnt);
1030 sb->_nodes.insert(1,cex);
1031 }
1032 }
1033 }
1034 }