1 /*
2 * Copyright (c) 1997, 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 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "code/exceptionHandlerTable.hpp"
29 #include "code/nmethod.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "compiler/oopMap.hpp"
32 #include "opto/addnode.hpp"
33 #include "opto/block.hpp"
34 #include "opto/c2compiler.hpp"
35 #include "opto/callGenerator.hpp"
36 #include "opto/callnode.hpp"
37 #include "opto/cfgnode.hpp"
38 #include "opto/chaitin.hpp"
39 #include "opto/compile.hpp"
40 #include "opto/connode.hpp"
41 #include "opto/divnode.hpp"
42 #include "opto/escape.hpp"
43 #include "opto/idealGraphPrinter.hpp"
44 #include "opto/loopnode.hpp"
45 #include "opto/machnode.hpp"
46 #include "opto/macro.hpp"
47 #include "opto/matcher.hpp"
48 #include "opto/memnode.hpp"
49 #include "opto/mulnode.hpp"
50 #include "opto/node.hpp"
51 #include "opto/opcodes.hpp"
52 #include "opto/output.hpp"
53 #include "opto/parse.hpp"
54 #include "opto/phaseX.hpp"
55 #include "opto/rootnode.hpp"
56 #include "opto/runtime.hpp"
57 #include "opto/stringopts.hpp"
58 #include "opto/type.hpp"
59 #include "opto/vectornode.hpp"
60 #include "runtime/arguments.hpp"
61 #include "runtime/signature.hpp"
62 #include "runtime/stubRoutines.hpp"
63 #include "runtime/timer.hpp"
64 #include "utilities/copy.hpp"
65 #ifdef TARGET_ARCH_MODEL_x86_32
66 # include "adfiles/ad_x86_32.hpp"
67 #endif
68 #ifdef TARGET_ARCH_MODEL_x86_64
69 # include "adfiles/ad_x86_64.hpp"
70 #endif
71 #ifdef TARGET_ARCH_MODEL_sparc
72 # include "adfiles/ad_sparc.hpp"
73 #endif
74 #ifdef TARGET_ARCH_MODEL_zero
75 # include "adfiles/ad_zero.hpp"
76 #endif
77
78 /// Support for intrinsics.
79
80 // Return the index at which m must be inserted (or already exists).
81 // The sort order is by the address of the ciMethod, with is_virtual as minor key.
82 int Compile::intrinsic_insertion_index(ciMethod* m, bool is_virtual) {
83 #ifdef ASSERT
84 for (int i = 1; i < _intrinsics->length(); i++) {
85 CallGenerator* cg1 = _intrinsics->at(i-1);
86 CallGenerator* cg2 = _intrinsics->at(i);
87 assert(cg1->method() != cg2->method()
88 ? cg1->method() < cg2->method()
89 : cg1->is_virtual() < cg2->is_virtual(),
90 "compiler intrinsics list must stay sorted");
91 }
92 #endif
93 // Binary search sorted list, in decreasing intervals [lo, hi].
94 int lo = 0, hi = _intrinsics->length()-1;
95 while (lo <= hi) {
96 int mid = (uint)(hi + lo) / 2;
97 ciMethod* mid_m = _intrinsics->at(mid)->method();
98 if (m < mid_m) {
99 hi = mid-1;
100 } else if (m > mid_m) {
101 lo = mid+1;
102 } else {
103 // look at minor sort key
104 bool mid_virt = _intrinsics->at(mid)->is_virtual();
105 if (is_virtual < mid_virt) {
106 hi = mid-1;
107 } else if (is_virtual > mid_virt) {
108 lo = mid+1;
109 } else {
110 return mid; // exact match
111 }
112 }
113 }
114 return lo; // inexact match
115 }
116
117 void Compile::register_intrinsic(CallGenerator* cg) {
118 if (_intrinsics == NULL) {
119 _intrinsics = new GrowableArray<CallGenerator*>(60);
120 }
121 // This code is stolen from ciObjectFactory::insert.
122 // Really, GrowableArray should have methods for
123 // insert_at, remove_at, and binary_search.
124 int len = _intrinsics->length();
125 int index = intrinsic_insertion_index(cg->method(), cg->is_virtual());
126 if (index == len) {
127 _intrinsics->append(cg);
128 } else {
129 #ifdef ASSERT
130 CallGenerator* oldcg = _intrinsics->at(index);
131 assert(oldcg->method() != cg->method() || oldcg->is_virtual() != cg->is_virtual(), "don't register twice");
132 #endif
133 _intrinsics->append(_intrinsics->at(len-1));
134 int pos;
135 for (pos = len-2; pos >= index; pos--) {
136 _intrinsics->at_put(pos+1,_intrinsics->at(pos));
137 }
138 _intrinsics->at_put(index, cg);
139 }
140 assert(find_intrinsic(cg->method(), cg->is_virtual()) == cg, "registration worked");
141 }
142
143 CallGenerator* Compile::find_intrinsic(ciMethod* m, bool is_virtual) {
144 assert(m->is_loaded(), "don't try this on unloaded methods");
145 if (_intrinsics != NULL) {
146 int index = intrinsic_insertion_index(m, is_virtual);
147 if (index < _intrinsics->length()
148 && _intrinsics->at(index)->method() == m
149 && _intrinsics->at(index)->is_virtual() == is_virtual) {
150 return _intrinsics->at(index);
151 }
152 }
153 // Lazily create intrinsics for intrinsic IDs well-known in the runtime.
154 if (m->intrinsic_id() != vmIntrinsics::_none &&
155 m->intrinsic_id() <= vmIntrinsics::LAST_COMPILER_INLINE) {
156 CallGenerator* cg = make_vm_intrinsic(m, is_virtual);
157 if (cg != NULL) {
158 // Save it for next time:
159 register_intrinsic(cg);
160 return cg;
161 } else {
162 gather_intrinsic_statistics(m->intrinsic_id(), is_virtual, _intrinsic_disabled);
163 }
164 }
165 return NULL;
166 }
167
168 // Compile:: register_library_intrinsics and make_vm_intrinsic are defined
169 // in library_call.cpp.
170
171
172 #ifndef PRODUCT
173 // statistics gathering...
174
175 juint Compile::_intrinsic_hist_count[vmIntrinsics::ID_LIMIT] = {0};
176 jubyte Compile::_intrinsic_hist_flags[vmIntrinsics::ID_LIMIT] = {0};
177
178 bool Compile::gather_intrinsic_statistics(vmIntrinsics::ID id, bool is_virtual, int flags) {
179 assert(id > vmIntrinsics::_none && id < vmIntrinsics::ID_LIMIT, "oob");
180 int oflags = _intrinsic_hist_flags[id];
181 assert(flags != 0, "what happened?");
182 if (is_virtual) {
183 flags |= _intrinsic_virtual;
184 }
185 bool changed = (flags != oflags);
186 if ((flags & _intrinsic_worked) != 0) {
187 juint count = (_intrinsic_hist_count[id] += 1);
188 if (count == 1) {
189 changed = true; // first time
190 }
191 // increment the overall count also:
192 _intrinsic_hist_count[vmIntrinsics::_none] += 1;
193 }
194 if (changed) {
195 if (((oflags ^ flags) & _intrinsic_virtual) != 0) {
196 // Something changed about the intrinsic's virtuality.
197 if ((flags & _intrinsic_virtual) != 0) {
198 // This is the first use of this intrinsic as a virtual call.
199 if (oflags != 0) {
200 // We already saw it as a non-virtual, so note both cases.
201 flags |= _intrinsic_both;
202 }
203 } else if ((oflags & _intrinsic_both) == 0) {
204 // This is the first use of this intrinsic as a non-virtual
205 flags |= _intrinsic_both;
206 }
207 }
208 _intrinsic_hist_flags[id] = (jubyte) (oflags | flags);
209 }
210 // update the overall flags also:
211 _intrinsic_hist_flags[vmIntrinsics::_none] |= (jubyte) flags;
212 return changed;
213 }
214
215 static char* format_flags(int flags, char* buf) {
216 buf[0] = 0;
217 if ((flags & Compile::_intrinsic_worked) != 0) strcat(buf, ",worked");
218 if ((flags & Compile::_intrinsic_failed) != 0) strcat(buf, ",failed");
219 if ((flags & Compile::_intrinsic_disabled) != 0) strcat(buf, ",disabled");
220 if ((flags & Compile::_intrinsic_virtual) != 0) strcat(buf, ",virtual");
221 if ((flags & Compile::_intrinsic_both) != 0) strcat(buf, ",nonvirtual");
222 if (buf[0] == 0) strcat(buf, ",");
223 assert(buf[0] == ',', "must be");
224 return &buf[1];
225 }
226
227 void Compile::print_intrinsic_statistics() {
228 char flagsbuf[100];
229 ttyLocker ttyl;
230 if (xtty != NULL) xtty->head("statistics type='intrinsic'");
231 tty->print_cr("Compiler intrinsic usage:");
232 juint total = _intrinsic_hist_count[vmIntrinsics::_none];
233 if (total == 0) total = 1; // avoid div0 in case of no successes
234 #define PRINT_STAT_LINE(name, c, f) \
235 tty->print_cr(" %4d (%4.1f%%) %s (%s)", (int)(c), ((c) * 100.0) / total, name, f);
236 for (int index = 1 + (int)vmIntrinsics::_none; index < (int)vmIntrinsics::ID_LIMIT; index++) {
237 vmIntrinsics::ID id = (vmIntrinsics::ID) index;
238 int flags = _intrinsic_hist_flags[id];
239 juint count = _intrinsic_hist_count[id];
240 if ((flags | count) != 0) {
241 PRINT_STAT_LINE(vmIntrinsics::name_at(id), count, format_flags(flags, flagsbuf));
242 }
243 }
244 PRINT_STAT_LINE("total", total, format_flags(_intrinsic_hist_flags[vmIntrinsics::_none], flagsbuf));
245 if (xtty != NULL) xtty->tail("statistics");
246 }
247
248 void Compile::print_statistics() {
249 { ttyLocker ttyl;
250 if (xtty != NULL) xtty->head("statistics type='opto'");
251 Parse::print_statistics();
252 PhaseCCP::print_statistics();
253 PhaseRegAlloc::print_statistics();
254 Scheduling::print_statistics();
255 PhasePeephole::print_statistics();
256 PhaseIdealLoop::print_statistics();
257 if (xtty != NULL) xtty->tail("statistics");
258 }
259 if (_intrinsic_hist_flags[vmIntrinsics::_none] != 0) {
260 // put this under its own <statistics> element.
261 print_intrinsic_statistics();
262 }
263 }
264 #endif //PRODUCT
265
266 // Support for bundling info
267 Bundle* Compile::node_bundling(const Node *n) {
268 assert(valid_bundle_info(n), "oob");
269 return &_node_bundling_base[n->_idx];
270 }
271
272 bool Compile::valid_bundle_info(const Node *n) {
273 return (_node_bundling_limit > n->_idx);
274 }
275
276
277 void Compile::gvn_replace_by(Node* n, Node* nn) {
278 for (DUIterator_Last imin, i = n->last_outs(imin); i >= imin; ) {
279 Node* use = n->last_out(i);
280 bool is_in_table = initial_gvn()->hash_delete(use);
281 uint uses_found = 0;
282 for (uint j = 0; j < use->len(); j++) {
283 if (use->in(j) == n) {
284 if (j < use->req())
285 use->set_req(j, nn);
286 else
287 use->set_prec(j, nn);
288 uses_found++;
289 }
290 }
291 if (is_in_table) {
292 // reinsert into table
293 initial_gvn()->hash_find_insert(use);
294 }
295 record_for_igvn(use);
296 i -= uses_found; // we deleted 1 or more copies of this edge
297 }
298 }
299
300
301
302
303 // Identify all nodes that are reachable from below, useful.
304 // Use breadth-first pass that records state in a Unique_Node_List,
305 // recursive traversal is slower.
306 void Compile::identify_useful_nodes(Unique_Node_List &useful) {
307 int estimated_worklist_size = unique();
308 useful.map( estimated_worklist_size, NULL ); // preallocate space
309
310 // Initialize worklist
311 if (root() != NULL) { useful.push(root()); }
312 // If 'top' is cached, declare it useful to preserve cached node
313 if( cached_top_node() ) { useful.push(cached_top_node()); }
314
315 // Push all useful nodes onto the list, breadthfirst
316 for( uint next = 0; next < useful.size(); ++next ) {
317 assert( next < unique(), "Unique useful nodes < total nodes");
318 Node *n = useful.at(next);
319 uint max = n->len();
320 for( uint i = 0; i < max; ++i ) {
321 Node *m = n->in(i);
322 if( m == NULL ) continue;
323 useful.push(m);
324 }
325 }
326 }
327
328 // Disconnect all useless nodes by disconnecting those at the boundary.
329 void Compile::remove_useless_nodes(Unique_Node_List &useful) {
330 uint next = 0;
331 while( next < useful.size() ) {
332 Node *n = useful.at(next++);
333 // Use raw traversal of out edges since this code removes out edges
334 int max = n->outcnt();
335 for (int j = 0; j < max; ++j ) {
336 Node* child = n->raw_out(j);
337 if( ! useful.member(child) ) {
338 assert( !child->is_top() || child != top(),
339 "If top is cached in Compile object it is in useful list");
340 // Only need to remove this out-edge to the useless node
341 n->raw_del_out(j);
342 --j;
343 --max;
344 }
345 }
346 if (n->outcnt() == 1 && n->has_special_unique_user()) {
347 record_for_igvn( n->unique_out() );
348 }
349 }
350 debug_only(verify_graph_edges(true/*check for no_dead_code*/);)
351 }
352
353 //------------------------------frame_size_in_words-----------------------------
354 // frame_slots in units of words
355 int Compile::frame_size_in_words() const {
356 // shift is 0 in LP32 and 1 in LP64
357 const int shift = (LogBytesPerWord - LogBytesPerInt);
358 int words = _frame_slots >> shift;
359 assert( words << shift == _frame_slots, "frame size must be properly aligned in LP64" );
360 return words;
361 }
362
363 // ============================================================================
364 //------------------------------CompileWrapper---------------------------------
365 class CompileWrapper : public StackObj {
366 Compile *const _compile;
367 public:
368 CompileWrapper(Compile* compile);
369
370 ~CompileWrapper();
371 };
372
373 CompileWrapper::CompileWrapper(Compile* compile) : _compile(compile) {
374 // the Compile* pointer is stored in the current ciEnv:
375 ciEnv* env = compile->env();
376 assert(env == ciEnv::current(), "must already be a ciEnv active");
377 assert(env->compiler_data() == NULL, "compile already active?");
378 env->set_compiler_data(compile);
379 assert(compile == Compile::current(), "sanity");
380
381 compile->set_type_dict(NULL);
382 compile->set_type_hwm(NULL);
383 compile->set_type_last_size(0);
384 compile->set_last_tf(NULL, NULL);
385 compile->set_indexSet_arena(NULL);
386 compile->set_indexSet_free_block_list(NULL);
387 compile->init_type_arena();
388 Type::Initialize(compile);
389 _compile->set_scratch_buffer_blob(NULL);
390 _compile->begin_method();
391 }
392 CompileWrapper::~CompileWrapper() {
393 _compile->end_method();
394 if (_compile->scratch_buffer_blob() != NULL)
395 BufferBlob::free(_compile->scratch_buffer_blob());
396 _compile->env()->set_compiler_data(NULL);
397 }
398
399
400 //----------------------------print_compile_messages---------------------------
401 void Compile::print_compile_messages() {
402 #ifndef PRODUCT
403 // Check if recompiling
404 if (_subsume_loads == false && PrintOpto) {
405 // Recompiling without allowing machine instructions to subsume loads
406 tty->print_cr("*********************************************************");
407 tty->print_cr("** Bailout: Recompile without subsuming loads **");
408 tty->print_cr("*********************************************************");
409 }
410 if (_do_escape_analysis != DoEscapeAnalysis && PrintOpto) {
411 // Recompiling without escape analysis
412 tty->print_cr("*********************************************************");
413 tty->print_cr("** Bailout: Recompile without escape analysis **");
414 tty->print_cr("*********************************************************");
415 }
416 if (env()->break_at_compile()) {
417 // Open the debugger when compiling this method.
418 tty->print("### Breaking when compiling: ");
419 method()->print_short_name();
420 tty->cr();
421 BREAKPOINT;
422 }
423
424 if( PrintOpto ) {
425 if (is_osr_compilation()) {
426 tty->print("[OSR]%3d", _compile_id);
427 } else {
428 tty->print("%3d", _compile_id);
429 }
430 }
431 #endif
432 }
433
434
435 void Compile::init_scratch_buffer_blob() {
436 if( scratch_buffer_blob() != NULL ) return;
437
438 // Construct a temporary CodeBuffer to have it construct a BufferBlob
439 // Cache this BufferBlob for this compile.
440 ResourceMark rm;
441 int size = (MAX_inst_size + MAX_stubs_size + MAX_const_size);
442 BufferBlob* blob = BufferBlob::create("Compile::scratch_buffer", size);
443 // Record the buffer blob for next time.
444 set_scratch_buffer_blob(blob);
445 // Have we run out of code space?
446 if (scratch_buffer_blob() == NULL) {
447 // Let CompilerBroker disable further compilations.
448 record_failure("Not enough space for scratch buffer in CodeCache");
449 return;
450 }
451
452 // Initialize the relocation buffers
453 relocInfo* locs_buf = (relocInfo*) blob->content_end() - MAX_locs_size;
454 set_scratch_locs_memory(locs_buf);
455 }
456
457
458 //-----------------------scratch_emit_size-------------------------------------
459 // Helper function that computes size by emitting code
460 uint Compile::scratch_emit_size(const Node* n) {
461 // Emit into a trash buffer and count bytes emitted.
462 // This is a pretty expensive way to compute a size,
463 // but it works well enough if seldom used.
464 // All common fixed-size instructions are given a size
465 // method by the AD file.
466 // Note that the scratch buffer blob and locs memory are
467 // allocated at the beginning of the compile task, and
468 // may be shared by several calls to scratch_emit_size.
469 // The allocation of the scratch buffer blob is particularly
470 // expensive, since it has to grab the code cache lock.
471 BufferBlob* blob = this->scratch_buffer_blob();
472 assert(blob != NULL, "Initialize BufferBlob at start");
473 assert(blob->size() > MAX_inst_size, "sanity");
474 relocInfo* locs_buf = scratch_locs_memory();
475 address blob_begin = blob->content_begin();
476 address blob_end = (address)locs_buf;
477 assert(blob->content_contains(blob_end), "sanity");
478 CodeBuffer buf(blob_begin, blob_end - blob_begin);
479 buf.initialize_consts_size(MAX_const_size);
480 buf.initialize_stubs_size(MAX_stubs_size);
481 assert(locs_buf != NULL, "sanity");
482 int lsize = MAX_locs_size / 2;
483 buf.insts()->initialize_shared_locs(&locs_buf[0], lsize);
484 buf.stubs()->initialize_shared_locs(&locs_buf[lsize], lsize);
485 n->emit(buf, this->regalloc());
486 return buf.insts_size();
487 }
488
489
490 // ============================================================================
491 //------------------------------Compile standard-------------------------------
492 debug_only( int Compile::_debug_idx = 100000; )
493
494 // Compile a method. entry_bci is -1 for normal compilations and indicates
495 // the continuation bci for on stack replacement.
496
497
498 Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr_bci, bool subsume_loads, bool do_escape_analysis )
499 : Phase(Compiler),
500 _env(ci_env),
501 _log(ci_env->log()),
502 _compile_id(ci_env->compile_id()),
503 _save_argument_registers(false),
504 _stub_name(NULL),
505 _stub_function(NULL),
506 _stub_entry_point(NULL),
507 _method(target),
508 _entry_bci(osr_bci),
509 _initial_gvn(NULL),
510 _for_igvn(NULL),
511 _warm_calls(NULL),
512 _subsume_loads(subsume_loads),
513 _do_escape_analysis(do_escape_analysis),
514 _failure_reason(NULL),
515 _code_buffer("Compile::Fill_buffer"),
516 _orig_pc_slot(0),
517 _orig_pc_slot_offset_in_bytes(0),
518 _has_method_handle_invokes(false),
519 _node_bundling_limit(0),
520 _node_bundling_base(NULL),
521 _java_calls(0),
522 _inner_loops(0),
523 #ifndef PRODUCT
524 _trace_opto_output(TraceOptoOutput || method()->has_option("TraceOptoOutput")),
525 _printer(IdealGraphPrinter::printer()),
526 #endif
527 _congraph(NULL) {
528 C = this;
529
530 CompileWrapper cw(this);
531 #ifndef PRODUCT
532 if (TimeCompiler2) {
533 tty->print(" ");
534 target->holder()->name()->print();
535 tty->print(".");
536 target->print_short_name();
537 tty->print(" ");
538 }
539 TraceTime t1("Total compilation time", &_t_totalCompilation, TimeCompiler, TimeCompiler2);
540 TraceTime t2(NULL, &_t_methodCompilation, TimeCompiler, false);
541 bool print_opto_assembly = PrintOptoAssembly || _method->has_option("PrintOptoAssembly");
542 if (!print_opto_assembly) {
543 bool print_assembly = (PrintAssembly || _method->should_print_assembly());
544 if (print_assembly && !Disassembler::can_decode()) {
545 tty->print_cr("PrintAssembly request changed to PrintOptoAssembly");
546 print_opto_assembly = true;
547 }
548 }
549 set_print_assembly(print_opto_assembly);
550 set_parsed_irreducible_loop(false);
551 #endif
552
553 if (ProfileTraps) {
554 // Make sure the method being compiled gets its own MDO,
555 // so we can at least track the decompile_count().
556 method()->build_method_data();
557 }
558
559 Init(::AliasLevel);
560
561
562 print_compile_messages();
563
564 if (UseOldInlining || PrintCompilation NOT_PRODUCT( || PrintOpto) )
565 _ilt = InlineTree::build_inline_tree_root();
566 else
567 _ilt = NULL;
568
569 // Even if NO memory addresses are used, MergeMem nodes must have at least 1 slice
570 assert(num_alias_types() >= AliasIdxRaw, "");
571
572 #define MINIMUM_NODE_HASH 1023
573 // Node list that Iterative GVN will start with
574 Unique_Node_List for_igvn(comp_arena());
575 set_for_igvn(&for_igvn);
576
577 // GVN that will be run immediately on new nodes
578 uint estimated_size = method()->code_size()*4+64;
579 estimated_size = (estimated_size < MINIMUM_NODE_HASH ? MINIMUM_NODE_HASH : estimated_size);
580 PhaseGVN gvn(node_arena(), estimated_size);
581 set_initial_gvn(&gvn);
582
583 { // Scope for timing the parser
584 TracePhase t3("parse", &_t_parser, true);
585
586 // Put top into the hash table ASAP.
587 initial_gvn()->transform_no_reclaim(top());
588
589 // Set up tf(), start(), and find a CallGenerator.
590 CallGenerator* cg;
591 if (is_osr_compilation()) {
592 const TypeTuple *domain = StartOSRNode::osr_domain();
593 const TypeTuple *range = TypeTuple::make_range(method()->signature());
594 init_tf(TypeFunc::make(domain, range));
595 StartNode* s = new (this, 2) StartOSRNode(root(), domain);
596 initial_gvn()->set_type_bottom(s);
597 init_start(s);
598 cg = CallGenerator::for_osr(method(), entry_bci());
599 } else {
600 // Normal case.
601 init_tf(TypeFunc::make(method()));
602 StartNode* s = new (this, 2) StartNode(root(), tf()->domain());
603 initial_gvn()->set_type_bottom(s);
604 init_start(s);
605 float past_uses = method()->interpreter_invocation_count();
606 float expected_uses = past_uses;
607 cg = CallGenerator::for_inline(method(), expected_uses);
608 }
609 if (failing()) return;
610 if (cg == NULL) {
611 record_method_not_compilable_all_tiers("cannot parse method");
612 return;
613 }
614 JVMState* jvms = build_start_state(start(), tf());
615 if ((jvms = cg->generate(jvms)) == NULL) {
616 record_method_not_compilable("method parse failed");
617 return;
618 }
619 GraphKit kit(jvms);
620
621 if (!kit.stopped()) {
622 // Accept return values, and transfer control we know not where.
623 // This is done by a special, unique ReturnNode bound to root.
624 return_values(kit.jvms());
625 }
626
627 if (kit.has_exceptions()) {
628 // Any exceptions that escape from this call must be rethrown
629 // to whatever caller is dynamically above us on the stack.
630 // This is done by a special, unique RethrowNode bound to root.
631 rethrow_exceptions(kit.transfer_exceptions_into_jvms());
632 }
633
634 if (!failing() && has_stringbuilder()) {
635 {
636 // remove useless nodes to make the usage analysis simpler
637 ResourceMark rm;
638 PhaseRemoveUseless pru(initial_gvn(), &for_igvn);
639 }
640
641 {
642 ResourceMark rm;
643 print_method("Before StringOpts", 3);
644 PhaseStringOpts pso(initial_gvn(), &for_igvn);
645 print_method("After StringOpts", 3);
646 }
647
648 // now inline anything that we skipped the first time around
649 while (_late_inlines.length() > 0) {
650 CallGenerator* cg = _late_inlines.pop();
651 cg->do_late_inline();
652 }
653 }
654 assert(_late_inlines.length() == 0, "should have been processed");
655
656 print_method("Before RemoveUseless", 3);
657
658 // Remove clutter produced by parsing.
659 if (!failing()) {
660 ResourceMark rm;
661 PhaseRemoveUseless pru(initial_gvn(), &for_igvn);
662 }
663 }
664
665 // Note: Large methods are capped off in do_one_bytecode().
666 if (failing()) return;
667
668 // After parsing, node notes are no longer automagic.
669 // They must be propagated by register_new_node_with_optimizer(),
670 // clone(), or the like.
671 set_default_node_notes(NULL);
672
673 for (;;) {
674 int successes = Inline_Warm();
675 if (failing()) return;
676 if (successes == 0) break;
677 }
678
679 // Drain the list.
680 Finish_Warm();
681 #ifndef PRODUCT
682 if (_printer) {
683 _printer->print_inlining(this);
684 }
685 #endif
686
687 if (failing()) return;
688 NOT_PRODUCT( verify_graph_edges(); )
689
690 // Now optimize
691 Optimize();
692 if (failing()) return;
693 NOT_PRODUCT( verify_graph_edges(); )
694
695 #ifndef PRODUCT
696 if (PrintIdeal) {
697 ttyLocker ttyl; // keep the following output all in one block
698 // This output goes directly to the tty, not the compiler log.
699 // To enable tools to match it up with the compilation activity,
700 // be sure to tag this tty output with the compile ID.
701 if (xtty != NULL) {
702 xtty->head("ideal compile_id='%d'%s", compile_id(),
703 is_osr_compilation() ? " compile_kind='osr'" :
704 "");
705 }
706 root()->dump(9999);
707 if (xtty != NULL) {
708 xtty->tail("ideal");
709 }
710 }
711 #endif
712
713 // Now that we know the size of all the monitors we can add a fixed slot
714 // for the original deopt pc.
715
716 _orig_pc_slot = fixed_slots();
717 int next_slot = _orig_pc_slot + (sizeof(address) / VMRegImpl::stack_slot_size);
718 set_fixed_slots(next_slot);
719
720 // Now generate code
721 Code_Gen();
722 if (failing()) return;
723
724 // Check if we want to skip execution of all compiled code.
725 {
726 #ifndef PRODUCT
727 if (OptoNoExecute) {
728 record_method_not_compilable("+OptoNoExecute"); // Flag as failed
729 return;
730 }
731 TracePhase t2("install_code", &_t_registerMethod, TimeCompiler);
732 #endif
733
734 if (is_osr_compilation()) {
735 _code_offsets.set_value(CodeOffsets::Verified_Entry, 0);
736 _code_offsets.set_value(CodeOffsets::OSR_Entry, _first_block_size);
737 } else {
738 _code_offsets.set_value(CodeOffsets::Verified_Entry, _first_block_size);
739 _code_offsets.set_value(CodeOffsets::OSR_Entry, 0);
740 }
741
742 env()->register_method(_method, _entry_bci,
743 &_code_offsets,
744 _orig_pc_slot_offset_in_bytes,
745 code_buffer(),
746 frame_size_in_words(), _oop_map_set,
747 &_handler_table, &_inc_table,
748 compiler,
749 env()->comp_level(),
750 true, /*has_debug_info*/
751 has_unsafe_access()
752 );
753 }
754 }
755
756 //------------------------------Compile----------------------------------------
757 // Compile a runtime stub
758 Compile::Compile( ciEnv* ci_env,
759 TypeFunc_generator generator,
760 address stub_function,
761 const char *stub_name,
762 int is_fancy_jump,
763 bool pass_tls,
764 bool save_arg_registers,
765 bool return_pc )
766 : Phase(Compiler),
767 _env(ci_env),
768 _log(ci_env->log()),
769 _compile_id(-1),
770 _save_argument_registers(save_arg_registers),
771 _method(NULL),
772 _stub_name(stub_name),
773 _stub_function(stub_function),
774 _stub_entry_point(NULL),
775 _entry_bci(InvocationEntryBci),
776 _initial_gvn(NULL),
777 _for_igvn(NULL),
778 _warm_calls(NULL),
779 _orig_pc_slot(0),
780 _orig_pc_slot_offset_in_bytes(0),
781 _subsume_loads(true),
782 _do_escape_analysis(false),
783 _failure_reason(NULL),
784 _code_buffer("Compile::Fill_buffer"),
785 _has_method_handle_invokes(false),
786 _node_bundling_limit(0),
787 _node_bundling_base(NULL),
788 _java_calls(0),
789 _inner_loops(0),
790 #ifndef PRODUCT
791 _trace_opto_output(TraceOptoOutput),
792 _printer(NULL),
793 #endif
794 _congraph(NULL) {
795 C = this;
796
797 #ifndef PRODUCT
798 TraceTime t1(NULL, &_t_totalCompilation, TimeCompiler, false);
799 TraceTime t2(NULL, &_t_stubCompilation, TimeCompiler, false);
800 set_print_assembly(PrintFrameConverterAssembly);
801 set_parsed_irreducible_loop(false);
802 #endif
803 CompileWrapper cw(this);
804 Init(/*AliasLevel=*/ 0);
805 init_tf((*generator)());
806
807 {
808 // The following is a dummy for the sake of GraphKit::gen_stub
809 Unique_Node_List for_igvn(comp_arena());
810 set_for_igvn(&for_igvn); // not used, but some GraphKit guys push on this
811 PhaseGVN gvn(Thread::current()->resource_area(),255);
812 set_initial_gvn(&gvn); // not significant, but GraphKit guys use it pervasively
813 gvn.transform_no_reclaim(top());
814
815 GraphKit kit;
816 kit.gen_stub(stub_function, stub_name, is_fancy_jump, pass_tls, return_pc);
817 }
818
819 NOT_PRODUCT( verify_graph_edges(); )
820 Code_Gen();
821 if (failing()) return;
822
823
824 // Entry point will be accessed using compile->stub_entry_point();
825 if (code_buffer() == NULL) {
826 Matcher::soft_match_failure();
827 } else {
828 if (PrintAssembly && (WizardMode || Verbose))
829 tty->print_cr("### Stub::%s", stub_name);
830
831 if (!failing()) {
832 assert(_fixed_slots == 0, "no fixed slots used for runtime stubs");
833
834 // Make the NMethod
835 // For now we mark the frame as never safe for profile stackwalking
836 RuntimeStub *rs = RuntimeStub::new_runtime_stub(stub_name,
837 code_buffer(),
838 CodeOffsets::frame_never_safe,
839 // _code_offsets.value(CodeOffsets::Frame_Complete),
840 frame_size_in_words(),
841 _oop_map_set,
842 save_arg_registers);
843 assert(rs != NULL && rs->is_runtime_stub(), "sanity check");
844
845 _stub_entry_point = rs->entry_point();
846 }
847 }
848 }
849
850 #ifndef PRODUCT
851 void print_opto_verbose_signature( const TypeFunc *j_sig, const char *stub_name ) {
852 if(PrintOpto && Verbose) {
853 tty->print("%s ", stub_name); j_sig->print_flattened(); tty->cr();
854 }
855 }
856 #endif
857
858 void Compile::print_codes() {
859 }
860
861 //------------------------------Init-------------------------------------------
862 // Prepare for a single compilation
863 void Compile::Init(int aliaslevel) {
864 _unique = 0;
865 _regalloc = NULL;
866
867 _tf = NULL; // filled in later
868 _top = NULL; // cached later
869 _matcher = NULL; // filled in later
870 _cfg = NULL; // filled in later
871
872 set_24_bit_selection_and_mode(Use24BitFP, false);
873
874 _node_note_array = NULL;
875 _default_node_notes = NULL;
876
877 _immutable_memory = NULL; // filled in at first inquiry
878
879 // Globally visible Nodes
880 // First set TOP to NULL to give safe behavior during creation of RootNode
881 set_cached_top_node(NULL);
882 set_root(new (this, 3) RootNode());
883 // Now that you have a Root to point to, create the real TOP
884 set_cached_top_node( new (this, 1) ConNode(Type::TOP) );
885 set_recent_alloc(NULL, NULL);
886
887 // Create Debug Information Recorder to record scopes, oopmaps, etc.
888 env()->set_oop_recorder(new OopRecorder(comp_arena()));
889 env()->set_debug_info(new DebugInformationRecorder(env()->oop_recorder()));
890 env()->set_dependencies(new Dependencies(env()));
891
892 _fixed_slots = 0;
893 set_has_split_ifs(false);
894 set_has_loops(has_method() && method()->has_loops()); // first approximation
895 set_has_stringbuilder(false);
896 _trap_can_recompile = false; // no traps emitted yet
897 _major_progress = true; // start out assuming good things will happen
898 set_has_unsafe_access(false);
899 Copy::zero_to_bytes(_trap_hist, sizeof(_trap_hist));
900 set_decompile_count(0);
901
902 set_do_freq_based_layout(BlockLayoutByFrequency || method_has_option("BlockLayoutByFrequency"));
903 set_num_loop_opts(LoopOptsCount);
904 set_do_inlining(Inline);
905 set_max_inline_size(MaxInlineSize);
906 set_freq_inline_size(FreqInlineSize);
907 set_do_scheduling(OptoScheduling);
908 set_do_count_invocations(false);
909 set_do_method_data_update(false);
910
911 if (debug_info()->recording_non_safepoints()) {
912 set_node_note_array(new(comp_arena()) GrowableArray<Node_Notes*>
913 (comp_arena(), 8, 0, NULL));
914 set_default_node_notes(Node_Notes::make(this));
915 }
916
917 // // -- Initialize types before each compile --
918 // // Update cached type information
919 // if( _method && _method->constants() )
920 // Type::update_loaded_types(_method, _method->constants());
921
922 // Init alias_type map.
923 if (!_do_escape_analysis && aliaslevel == 3)
924 aliaslevel = 2; // No unique types without escape analysis
925 _AliasLevel = aliaslevel;
926 const int grow_ats = 16;
927 _max_alias_types = grow_ats;
928 _alias_types = NEW_ARENA_ARRAY(comp_arena(), AliasType*, grow_ats);
929 AliasType* ats = NEW_ARENA_ARRAY(comp_arena(), AliasType, grow_ats);
930 Copy::zero_to_bytes(ats, sizeof(AliasType)*grow_ats);
931 {
932 for (int i = 0; i < grow_ats; i++) _alias_types[i] = &ats[i];
933 }
934 // Initialize the first few types.
935 _alias_types[AliasIdxTop]->Init(AliasIdxTop, NULL);
936 _alias_types[AliasIdxBot]->Init(AliasIdxBot, TypePtr::BOTTOM);
937 _alias_types[AliasIdxRaw]->Init(AliasIdxRaw, TypeRawPtr::BOTTOM);
938 _num_alias_types = AliasIdxRaw+1;
939 // Zero out the alias type cache.
940 Copy::zero_to_bytes(_alias_cache, sizeof(_alias_cache));
941 // A NULL adr_type hits in the cache right away. Preload the right answer.
942 probe_alias_cache(NULL)->_index = AliasIdxTop;
943
944 _intrinsics = NULL;
945 _macro_nodes = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL);
946 _predicate_opaqs = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL);
947 register_library_intrinsics();
948 }
949
950 //---------------------------init_start----------------------------------------
951 // Install the StartNode on this compile object.
952 void Compile::init_start(StartNode* s) {
953 if (failing())
954 return; // already failing
955 assert(s == start(), "");
956 }
957
958 StartNode* Compile::start() const {
959 assert(!failing(), "");
960 for (DUIterator_Fast imax, i = root()->fast_outs(imax); i < imax; i++) {
961 Node* start = root()->fast_out(i);
962 if( start->is_Start() )
963 return start->as_Start();
964 }
965 ShouldNotReachHere();
966 return NULL;
967 }
968
969 //-------------------------------immutable_memory-------------------------------------
970 // Access immutable memory
971 Node* Compile::immutable_memory() {
972 if (_immutable_memory != NULL) {
973 return _immutable_memory;
974 }
975 StartNode* s = start();
976 for (DUIterator_Fast imax, i = s->fast_outs(imax); true; i++) {
977 Node *p = s->fast_out(i);
978 if (p != s && p->as_Proj()->_con == TypeFunc::Memory) {
979 _immutable_memory = p;
980 return _immutable_memory;
981 }
982 }
983 ShouldNotReachHere();
984 return NULL;
985 }
986
987 //----------------------set_cached_top_node------------------------------------
988 // Install the cached top node, and make sure Node::is_top works correctly.
989 void Compile::set_cached_top_node(Node* tn) {
990 if (tn != NULL) verify_top(tn);
991 Node* old_top = _top;
992 _top = tn;
993 // Calling Node::setup_is_top allows the nodes the chance to adjust
994 // their _out arrays.
995 if (_top != NULL) _top->setup_is_top();
996 if (old_top != NULL) old_top->setup_is_top();
997 assert(_top == NULL || top()->is_top(), "");
998 }
999
1000 #ifndef PRODUCT
1001 void Compile::verify_top(Node* tn) const {
1002 if (tn != NULL) {
1003 assert(tn->is_Con(), "top node must be a constant");
1004 assert(((ConNode*)tn)->type() == Type::TOP, "top node must have correct type");
1005 assert(tn->in(0) != NULL, "must have live top node");
1006 }
1007 }
1008 #endif
1009
1010
1011 ///-------------------Managing Per-Node Debug & Profile Info-------------------
1012
1013 void Compile::grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by) {
1014 guarantee(arr != NULL, "");
1015 int num_blocks = arr->length();
1016 if (grow_by < num_blocks) grow_by = num_blocks;
1017 int num_notes = grow_by * _node_notes_block_size;
1018 Node_Notes* notes = NEW_ARENA_ARRAY(node_arena(), Node_Notes, num_notes);
1019 Copy::zero_to_bytes(notes, num_notes * sizeof(Node_Notes));
1020 while (num_notes > 0) {
1021 arr->append(notes);
1022 notes += _node_notes_block_size;
1023 num_notes -= _node_notes_block_size;
1024 }
1025 assert(num_notes == 0, "exact multiple, please");
1026 }
1027
1028 bool Compile::copy_node_notes_to(Node* dest, Node* source) {
1029 if (source == NULL || dest == NULL) return false;
1030
1031 if (dest->is_Con())
1032 return false; // Do not push debug info onto constants.
1033
1034 #ifdef ASSERT
1035 // Leave a bread crumb trail pointing to the original node:
1036 if (dest != NULL && dest != source && dest->debug_orig() == NULL) {
1037 dest->set_debug_orig(source);
1038 }
1039 #endif
1040
1041 if (node_note_array() == NULL)
1042 return false; // Not collecting any notes now.
1043
1044 // This is a copy onto a pre-existing node, which may already have notes.
1045 // If both nodes have notes, do not overwrite any pre-existing notes.
1046 Node_Notes* source_notes = node_notes_at(source->_idx);
1047 if (source_notes == NULL || source_notes->is_clear()) return false;
1048 Node_Notes* dest_notes = node_notes_at(dest->_idx);
1049 if (dest_notes == NULL || dest_notes->is_clear()) {
1050 return set_node_notes_at(dest->_idx, source_notes);
1051 }
1052
1053 Node_Notes merged_notes = (*source_notes);
1054 // The order of operations here ensures that dest notes will win...
1055 merged_notes.update_from(dest_notes);
1056 return set_node_notes_at(dest->_idx, &merged_notes);
1057 }
1058
1059
1060 //--------------------------allow_range_check_smearing-------------------------
1061 // Gating condition for coalescing similar range checks.
1062 // Sometimes we try 'speculatively' replacing a series of a range checks by a
1063 // single covering check that is at least as strong as any of them.
1064 // If the optimization succeeds, the simplified (strengthened) range check
1065 // will always succeed. If it fails, we will deopt, and then give up
1066 // on the optimization.
1067 bool Compile::allow_range_check_smearing() const {
1068 // If this method has already thrown a range-check,
1069 // assume it was because we already tried range smearing
1070 // and it failed.
1071 uint already_trapped = trap_count(Deoptimization::Reason_range_check);
1072 return !already_trapped;
1073 }
1074
1075
1076 //------------------------------flatten_alias_type-----------------------------
1077 const TypePtr *Compile::flatten_alias_type( const TypePtr *tj ) const {
1078 int offset = tj->offset();
1079 TypePtr::PTR ptr = tj->ptr();
1080
1081 // Known instance (scalarizable allocation) alias only with itself.
1082 bool is_known_inst = tj->isa_oopptr() != NULL &&
1083 tj->is_oopptr()->is_known_instance();
1084
1085 // Process weird unsafe references.
1086 if (offset == Type::OffsetBot && (tj->isa_instptr() /*|| tj->isa_klassptr()*/)) {
1087 assert(InlineUnsafeOps, "indeterminate pointers come only from unsafe ops");
1088 assert(!is_known_inst, "scalarizable allocation should not have unsafe references");
1089 tj = TypeOopPtr::BOTTOM;
1090 ptr = tj->ptr();
1091 offset = tj->offset();
1092 }
1093
1094 // Array pointers need some flattening
1095 const TypeAryPtr *ta = tj->isa_aryptr();
1096 if( ta && is_known_inst ) {
1097 if ( offset != Type::OffsetBot &&
1098 offset > arrayOopDesc::length_offset_in_bytes() ) {
1099 offset = Type::OffsetBot; // Flatten constant access into array body only
1100 tj = ta = TypeAryPtr::make(ptr, ta->ary(), ta->klass(), true, offset, ta->instance_id());
1101 }
1102 } else if( ta && _AliasLevel >= 2 ) {
1103 // For arrays indexed by constant indices, we flatten the alias
1104 // space to include all of the array body. Only the header, klass
1105 // and array length can be accessed un-aliased.
1106 if( offset != Type::OffsetBot ) {
1107 if( ta->const_oop() ) { // methodDataOop or methodOop
1108 offset = Type::OffsetBot; // Flatten constant access into array body
1109 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),ta->ary(),ta->klass(),false,offset);
1110 } else if( offset == arrayOopDesc::length_offset_in_bytes() ) {
1111 // range is OK as-is.
1112 tj = ta = TypeAryPtr::RANGE;
1113 } else if( offset == oopDesc::klass_offset_in_bytes() ) {
1114 tj = TypeInstPtr::KLASS; // all klass loads look alike
1115 ta = TypeAryPtr::RANGE; // generic ignored junk
1116 ptr = TypePtr::BotPTR;
1117 } else if( offset == oopDesc::mark_offset_in_bytes() ) {
1118 tj = TypeInstPtr::MARK;
1119 ta = TypeAryPtr::RANGE; // generic ignored junk
1120 ptr = TypePtr::BotPTR;
1121 } else { // Random constant offset into array body
1122 offset = Type::OffsetBot; // Flatten constant access into array body
1123 tj = ta = TypeAryPtr::make(ptr,ta->ary(),ta->klass(),false,offset);
1124 }
1125 }
1126 // Arrays of fixed size alias with arrays of unknown size.
1127 if (ta->size() != TypeInt::POS) {
1128 const TypeAry *tary = TypeAry::make(ta->elem(), TypeInt::POS);
1129 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,ta->klass(),false,offset);
1130 }
1131 // Arrays of known objects become arrays of unknown objects.
1132 if (ta->elem()->isa_narrowoop() && ta->elem() != TypeNarrowOop::BOTTOM) {
1133 const TypeAry *tary = TypeAry::make(TypeNarrowOop::BOTTOM, ta->size());
1134 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,NULL,false,offset);
1135 }
1136 if (ta->elem()->isa_oopptr() && ta->elem() != TypeInstPtr::BOTTOM) {
1137 const TypeAry *tary = TypeAry::make(TypeInstPtr::BOTTOM, ta->size());
1138 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,NULL,false,offset);
1139 }
1140 // Arrays of bytes and of booleans both use 'bastore' and 'baload' so
1141 // cannot be distinguished by bytecode alone.
1142 if (ta->elem() == TypeInt::BOOL) {
1143 const TypeAry *tary = TypeAry::make(TypeInt::BYTE, ta->size());
1144 ciKlass* aklass = ciTypeArrayKlass::make(T_BYTE);
1145 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,aklass,false,offset);
1146 }
1147 // During the 2nd round of IterGVN, NotNull castings are removed.
1148 // Make sure the Bottom and NotNull variants alias the same.
1149 // Also, make sure exact and non-exact variants alias the same.
1150 if( ptr == TypePtr::NotNull || ta->klass_is_exact() ) {
1151 if (ta->const_oop()) {
1152 tj = ta = TypeAryPtr::make(TypePtr::Constant,ta->const_oop(),ta->ary(),ta->klass(),false,offset);
1153 } else {
1154 tj = ta = TypeAryPtr::make(TypePtr::BotPTR,ta->ary(),ta->klass(),false,offset);
1155 }
1156 }
1157 }
1158
1159 // Oop pointers need some flattening
1160 const TypeInstPtr *to = tj->isa_instptr();
1161 if( to && _AliasLevel >= 2 && to != TypeOopPtr::BOTTOM ) {
1162 if( ptr == TypePtr::Constant ) {
1163 // No constant oop pointers (such as Strings); they alias with
1164 // unknown strings.
1165 assert(!is_known_inst, "not scalarizable allocation");
1166 tj = to = TypeInstPtr::make(TypePtr::BotPTR,to->klass(),false,0,offset);
1167 } else if( is_known_inst ) {
1168 tj = to; // Keep NotNull and klass_is_exact for instance type
1169 } else if( ptr == TypePtr::NotNull || to->klass_is_exact() ) {
1170 // During the 2nd round of IterGVN, NotNull castings are removed.
1171 // Make sure the Bottom and NotNull variants alias the same.
1172 // Also, make sure exact and non-exact variants alias the same.
1173 tj = to = TypeInstPtr::make(TypePtr::BotPTR,to->klass(),false,0,offset);
1174 }
1175 // Canonicalize the holder of this field
1176 ciInstanceKlass *k = to->klass()->as_instance_klass();
1177 if (offset >= 0 && offset < instanceOopDesc::base_offset_in_bytes()) {
1178 // First handle header references such as a LoadKlassNode, even if the
1179 // object's klass is unloaded at compile time (4965979).
1180 if (!is_known_inst) { // Do it only for non-instance types
1181 tj = to = TypeInstPtr::make(TypePtr::BotPTR, env()->Object_klass(), false, NULL, offset);
1182 }
1183 } else if (offset < 0 || offset >= k->size_helper() * wordSize) {
1184 to = NULL;
1185 tj = TypeOopPtr::BOTTOM;
1186 offset = tj->offset();
1187 } else {
1188 ciInstanceKlass *canonical_holder = k->get_canonical_holder(offset);
1189 if (!k->equals(canonical_holder) || tj->offset() != offset) {
1190 if( is_known_inst ) {
1191 tj = to = TypeInstPtr::make(to->ptr(), canonical_holder, true, NULL, offset, to->instance_id());
1192 } else {
1193 tj = to = TypeInstPtr::make(to->ptr(), canonical_holder, false, NULL, offset);
1194 }
1195 }
1196 }
1197 }
1198
1199 // Klass pointers to object array klasses need some flattening
1200 const TypeKlassPtr *tk = tj->isa_klassptr();
1201 if( tk ) {
1202 // If we are referencing a field within a Klass, we need
1203 // to assume the worst case of an Object. Both exact and
1204 // inexact types must flatten to the same alias class.
1205 // Since the flattened result for a klass is defined to be
1206 // precisely java.lang.Object, use a constant ptr.
1207 if ( offset == Type::OffsetBot || (offset >= 0 && (size_t)offset < sizeof(Klass)) ) {
1208
1209 tj = tk = TypeKlassPtr::make(TypePtr::Constant,
1210 TypeKlassPtr::OBJECT->klass(),
1211 offset);
1212 }
1213
1214 ciKlass* klass = tk->klass();
1215 if( klass->is_obj_array_klass() ) {
1216 ciKlass* k = TypeAryPtr::OOPS->klass();
1217 if( !k || !k->is_loaded() ) // Only fails for some -Xcomp runs
1218 k = TypeInstPtr::BOTTOM->klass();
1219 tj = tk = TypeKlassPtr::make( TypePtr::NotNull, k, offset );
1220 }
1221
1222 // Check for precise loads from the primary supertype array and force them
1223 // to the supertype cache alias index. Check for generic array loads from
1224 // the primary supertype array and also force them to the supertype cache
1225 // alias index. Since the same load can reach both, we need to merge
1226 // these 2 disparate memories into the same alias class. Since the
1227 // primary supertype array is read-only, there's no chance of confusion
1228 // where we bypass an array load and an array store.
1229 uint off2 = offset - Klass::primary_supers_offset_in_bytes();
1230 if( offset == Type::OffsetBot ||
1231 off2 < Klass::primary_super_limit()*wordSize ) {
1232 offset = sizeof(oopDesc) +Klass::secondary_super_cache_offset_in_bytes();
1233 tj = tk = TypeKlassPtr::make( TypePtr::NotNull, tk->klass(), offset );
1234 }
1235 }
1236
1237 // Flatten all Raw pointers together.
1238 if (tj->base() == Type::RawPtr)
1239 tj = TypeRawPtr::BOTTOM;
1240
1241 if (tj->base() == Type::AnyPtr)
1242 tj = TypePtr::BOTTOM; // An error, which the caller must check for.
1243
1244 // Flatten all to bottom for now
1245 switch( _AliasLevel ) {
1246 case 0:
1247 tj = TypePtr::BOTTOM;
1248 break;
1249 case 1: // Flatten to: oop, static, field or array
1250 switch (tj->base()) {
1251 //case Type::AryPtr: tj = TypeAryPtr::RANGE; break;
1252 case Type::RawPtr: tj = TypeRawPtr::BOTTOM; break;
1253 case Type::AryPtr: // do not distinguish arrays at all
1254 case Type::InstPtr: tj = TypeInstPtr::BOTTOM; break;
1255 case Type::KlassPtr: tj = TypeKlassPtr::OBJECT; break;
1256 case Type::AnyPtr: tj = TypePtr::BOTTOM; break; // caller checks it
1257 default: ShouldNotReachHere();
1258 }
1259 break;
1260 case 2: // No collapsing at level 2; keep all splits
1261 case 3: // No collapsing at level 3; keep all splits
1262 break;
1263 default:
1264 Unimplemented();
1265 }
1266
1267 offset = tj->offset();
1268 assert( offset != Type::OffsetTop, "Offset has fallen from constant" );
1269
1270 assert( (offset != Type::OffsetBot && tj->base() != Type::AryPtr) ||
1271 (offset == Type::OffsetBot && tj->base() == Type::AryPtr) ||
1272 (offset == Type::OffsetBot && tj == TypeOopPtr::BOTTOM) ||
1273 (offset == Type::OffsetBot && tj == TypePtr::BOTTOM) ||
1274 (offset == oopDesc::mark_offset_in_bytes() && tj->base() == Type::AryPtr) ||
1275 (offset == oopDesc::klass_offset_in_bytes() && tj->base() == Type::AryPtr) ||
1276 (offset == arrayOopDesc::length_offset_in_bytes() && tj->base() == Type::AryPtr) ,
1277 "For oops, klasses, raw offset must be constant; for arrays the offset is never known" );
1278 assert( tj->ptr() != TypePtr::TopPTR &&
1279 tj->ptr() != TypePtr::AnyNull &&
1280 tj->ptr() != TypePtr::Null, "No imprecise addresses" );
1281 // assert( tj->ptr() != TypePtr::Constant ||
1282 // tj->base() == Type::RawPtr ||
1283 // tj->base() == Type::KlassPtr, "No constant oop addresses" );
1284
1285 return tj;
1286 }
1287
1288 void Compile::AliasType::Init(int i, const TypePtr* at) {
1289 _index = i;
1290 _adr_type = at;
1291 _field = NULL;
1292 _is_rewritable = true; // default
1293 const TypeOopPtr *atoop = (at != NULL) ? at->isa_oopptr() : NULL;
1294 if (atoop != NULL && atoop->is_known_instance()) {
1295 const TypeOopPtr *gt = atoop->cast_to_instance_id(TypeOopPtr::InstanceBot);
1296 _general_index = Compile::current()->get_alias_index(gt);
1297 } else {
1298 _general_index = 0;
1299 }
1300 }
1301
1302 //---------------------------------print_on------------------------------------
1303 #ifndef PRODUCT
1304 void Compile::AliasType::print_on(outputStream* st) {
1305 if (index() < 10)
1306 st->print("@ <%d> ", index());
1307 else st->print("@ <%d>", index());
1308 st->print(is_rewritable() ? " " : " RO");
1309 int offset = adr_type()->offset();
1310 if (offset == Type::OffsetBot)
1311 st->print(" +any");
1312 else st->print(" +%-3d", offset);
1313 st->print(" in ");
1314 adr_type()->dump_on(st);
1315 const TypeOopPtr* tjp = adr_type()->isa_oopptr();
1316 if (field() != NULL && tjp) {
1317 if (tjp->klass() != field()->holder() ||
1318 tjp->offset() != field()->offset_in_bytes()) {
1319 st->print(" != ");
1320 field()->print();
1321 st->print(" ***");
1322 }
1323 }
1324 }
1325
1326 void print_alias_types() {
1327 Compile* C = Compile::current();
1328 tty->print_cr("--- Alias types, AliasIdxBot .. %d", C->num_alias_types()-1);
1329 for (int idx = Compile::AliasIdxBot; idx < C->num_alias_types(); idx++) {
1330 C->alias_type(idx)->print_on(tty);
1331 tty->cr();
1332 }
1333 }
1334 #endif
1335
1336
1337 //----------------------------probe_alias_cache--------------------------------
1338 Compile::AliasCacheEntry* Compile::probe_alias_cache(const TypePtr* adr_type) {
1339 intptr_t key = (intptr_t) adr_type;
1340 key ^= key >> logAliasCacheSize;
1341 return &_alias_cache[key & right_n_bits(logAliasCacheSize)];
1342 }
1343
1344
1345 //-----------------------------grow_alias_types--------------------------------
1346 void Compile::grow_alias_types() {
1347 const int old_ats = _max_alias_types; // how many before?
1348 const int new_ats = old_ats; // how many more?
1349 const int grow_ats = old_ats+new_ats; // how many now?
1350 _max_alias_types = grow_ats;
1351 _alias_types = REALLOC_ARENA_ARRAY(comp_arena(), AliasType*, _alias_types, old_ats, grow_ats);
1352 AliasType* ats = NEW_ARENA_ARRAY(comp_arena(), AliasType, new_ats);
1353 Copy::zero_to_bytes(ats, sizeof(AliasType)*new_ats);
1354 for (int i = 0; i < new_ats; i++) _alias_types[old_ats+i] = &ats[i];
1355 }
1356
1357
1358 //--------------------------------find_alias_type------------------------------
1359 Compile::AliasType* Compile::find_alias_type(const TypePtr* adr_type, bool no_create) {
1360 if (_AliasLevel == 0)
1361 return alias_type(AliasIdxBot);
1362
1363 AliasCacheEntry* ace = probe_alias_cache(adr_type);
1364 if (ace->_adr_type == adr_type) {
1365 return alias_type(ace->_index);
1366 }
1367
1368 // Handle special cases.
1369 if (adr_type == NULL) return alias_type(AliasIdxTop);
1370 if (adr_type == TypePtr::BOTTOM) return alias_type(AliasIdxBot);
1371
1372 // Do it the slow way.
1373 const TypePtr* flat = flatten_alias_type(adr_type);
1374
1375 #ifdef ASSERT
1376 assert(flat == flatten_alias_type(flat), "idempotent");
1377 assert(flat != TypePtr::BOTTOM, "cannot alias-analyze an untyped ptr");
1378 if (flat->isa_oopptr() && !flat->isa_klassptr()) {
1379 const TypeOopPtr* foop = flat->is_oopptr();
1380 // Scalarizable allocations have exact klass always.
1381 bool exact = !foop->klass_is_exact() || foop->is_known_instance();
1382 const TypePtr* xoop = foop->cast_to_exactness(exact)->is_ptr();
1383 assert(foop == flatten_alias_type(xoop), "exactness must not affect alias type");
1384 }
1385 assert(flat == flatten_alias_type(flat), "exact bit doesn't matter");
1386 #endif
1387
1388 int idx = AliasIdxTop;
1389 for (int i = 0; i < num_alias_types(); i++) {
1390 if (alias_type(i)->adr_type() == flat) {
1391 idx = i;
1392 break;
1393 }
1394 }
1395
1396 if (idx == AliasIdxTop) {
1397 if (no_create) return NULL;
1398 // Grow the array if necessary.
1399 if (_num_alias_types == _max_alias_types) grow_alias_types();
1400 // Add a new alias type.
1401 idx = _num_alias_types++;
1402 _alias_types[idx]->Init(idx, flat);
1403 if (flat == TypeInstPtr::KLASS) alias_type(idx)->set_rewritable(false);
1404 if (flat == TypeAryPtr::RANGE) alias_type(idx)->set_rewritable(false);
1405 if (flat->isa_instptr()) {
1406 if (flat->offset() == java_lang_Class::klass_offset_in_bytes()
1407 && flat->is_instptr()->klass() == env()->Class_klass())
1408 alias_type(idx)->set_rewritable(false);
1409 }
1410 if (flat->isa_klassptr()) {
1411 if (flat->offset() == Klass::super_check_offset_offset_in_bytes() + (int)sizeof(oopDesc))
1412 alias_type(idx)->set_rewritable(false);
1413 if (flat->offset() == Klass::modifier_flags_offset_in_bytes() + (int)sizeof(oopDesc))
1414 alias_type(idx)->set_rewritable(false);
1415 if (flat->offset() == Klass::access_flags_offset_in_bytes() + (int)sizeof(oopDesc))
1416 alias_type(idx)->set_rewritable(false);
1417 if (flat->offset() == Klass::java_mirror_offset_in_bytes() + (int)sizeof(oopDesc))
1418 alias_type(idx)->set_rewritable(false);
1419 }
1420 // %%% (We would like to finalize JavaThread::threadObj_offset(),
1421 // but the base pointer type is not distinctive enough to identify
1422 // references into JavaThread.)
1423
1424 // Check for final instance fields.
1425 const TypeInstPtr* tinst = flat->isa_instptr();
1426 if (tinst && tinst->offset() >= instanceOopDesc::base_offset_in_bytes()) {
1427 ciInstanceKlass *k = tinst->klass()->as_instance_klass();
1428 ciField* field = k->get_field_by_offset(tinst->offset(), false);
1429 // Set field() and is_rewritable() attributes.
1430 if (field != NULL) alias_type(idx)->set_field(field);
1431 }
1432 const TypeKlassPtr* tklass = flat->isa_klassptr();
1433 // Check for final static fields.
1434 if (tklass && tklass->klass()->is_instance_klass()) {
1435 ciInstanceKlass *k = tklass->klass()->as_instance_klass();
1436 ciField* field = k->get_field_by_offset(tklass->offset(), true);
1437 // Set field() and is_rewritable() attributes.
1438 if (field != NULL) alias_type(idx)->set_field(field);
1439 }
1440 }
1441
1442 // Fill the cache for next time.
1443 ace->_adr_type = adr_type;
1444 ace->_index = idx;
1445 assert(alias_type(adr_type) == alias_type(idx), "type must be installed");
1446
1447 // Might as well try to fill the cache for the flattened version, too.
1448 AliasCacheEntry* face = probe_alias_cache(flat);
1449 if (face->_adr_type == NULL) {
1450 face->_adr_type = flat;
1451 face->_index = idx;
1452 assert(alias_type(flat) == alias_type(idx), "flat type must work too");
1453 }
1454
1455 return alias_type(idx);
1456 }
1457
1458
1459 Compile::AliasType* Compile::alias_type(ciField* field) {
1460 const TypeOopPtr* t;
1461 if (field->is_static())
1462 t = TypeKlassPtr::make(field->holder());
1463 else
1464 t = TypeOopPtr::make_from_klass_raw(field->holder());
1465 AliasType* atp = alias_type(t->add_offset(field->offset_in_bytes()));
1466 assert(field->is_final() == !atp->is_rewritable(), "must get the rewritable bits correct");
1467 return atp;
1468 }
1469
1470
1471 //------------------------------have_alias_type--------------------------------
1472 bool Compile::have_alias_type(const TypePtr* adr_type) {
1473 AliasCacheEntry* ace = probe_alias_cache(adr_type);
1474 if (ace->_adr_type == adr_type) {
1475 return true;
1476 }
1477
1478 // Handle special cases.
1479 if (adr_type == NULL) return true;
1480 if (adr_type == TypePtr::BOTTOM) return true;
1481
1482 return find_alias_type(adr_type, true) != NULL;
1483 }
1484
1485 //-----------------------------must_alias--------------------------------------
1486 // True if all values of the given address type are in the given alias category.
1487 bool Compile::must_alias(const TypePtr* adr_type, int alias_idx) {
1488 if (alias_idx == AliasIdxBot) return true; // the universal category
1489 if (adr_type == NULL) return true; // NULL serves as TypePtr::TOP
1490 if (alias_idx == AliasIdxTop) return false; // the empty category
1491 if (adr_type->base() == Type::AnyPtr) return false; // TypePtr::BOTTOM or its twins
1492
1493 // the only remaining possible overlap is identity
1494 int adr_idx = get_alias_index(adr_type);
1495 assert(adr_idx != AliasIdxBot && adr_idx != AliasIdxTop, "");
1496 assert(adr_idx == alias_idx ||
1497 (alias_type(alias_idx)->adr_type() != TypeOopPtr::BOTTOM
1498 && adr_type != TypeOopPtr::BOTTOM),
1499 "should not be testing for overlap with an unsafe pointer");
1500 return adr_idx == alias_idx;
1501 }
1502
1503 //------------------------------can_alias--------------------------------------
1504 // True if any values of the given address type are in the given alias category.
1505 bool Compile::can_alias(const TypePtr* adr_type, int alias_idx) {
1506 if (alias_idx == AliasIdxTop) return false; // the empty category
1507 if (adr_type == NULL) return false; // NULL serves as TypePtr::TOP
1508 if (alias_idx == AliasIdxBot) return true; // the universal category
1509 if (adr_type->base() == Type::AnyPtr) return true; // TypePtr::BOTTOM or its twins
1510
1511 // the only remaining possible overlap is identity
1512 int adr_idx = get_alias_index(adr_type);
1513 assert(adr_idx != AliasIdxBot && adr_idx != AliasIdxTop, "");
1514 return adr_idx == alias_idx;
1515 }
1516
1517
1518
1519 //---------------------------pop_warm_call-------------------------------------
1520 WarmCallInfo* Compile::pop_warm_call() {
1521 WarmCallInfo* wci = _warm_calls;
1522 if (wci != NULL) _warm_calls = wci->remove_from(wci);
1523 return wci;
1524 }
1525
1526 //----------------------------Inline_Warm--------------------------------------
1527 int Compile::Inline_Warm() {
1528 // If there is room, try to inline some more warm call sites.
1529 // %%% Do a graph index compaction pass when we think we're out of space?
1530 if (!InlineWarmCalls) return 0;
1531
1532 int calls_made_hot = 0;
1533 int room_to_grow = NodeCountInliningCutoff - unique();
1534 int amount_to_grow = MIN2(room_to_grow, (int)NodeCountInliningStep);
1535 int amount_grown = 0;
1536 WarmCallInfo* call;
1537 while (amount_to_grow > 0 && (call = pop_warm_call()) != NULL) {
1538 int est_size = (int)call->size();
1539 if (est_size > (room_to_grow - amount_grown)) {
1540 // This one won't fit anyway. Get rid of it.
1541 call->make_cold();
1542 continue;
1543 }
1544 call->make_hot();
1545 calls_made_hot++;
1546 amount_grown += est_size;
1547 amount_to_grow -= est_size;
1548 }
1549
1550 if (calls_made_hot > 0) set_major_progress();
1551 return calls_made_hot;
1552 }
1553
1554
1555 //----------------------------Finish_Warm--------------------------------------
1556 void Compile::Finish_Warm() {
1557 if (!InlineWarmCalls) return;
1558 if (failing()) return;
1559 if (warm_calls() == NULL) return;
1560
1561 // Clean up loose ends, if we are out of space for inlining.
1562 WarmCallInfo* call;
1563 while ((call = pop_warm_call()) != NULL) {
1564 call->make_cold();
1565 }
1566 }
1567
1568 //---------------------cleanup_loop_predicates-----------------------
1569 // Remove the opaque nodes that protect the predicates so that all unused
1570 // checks and uncommon_traps will be eliminated from the ideal graph
1571 void Compile::cleanup_loop_predicates(PhaseIterGVN &igvn) {
1572 if (predicate_count()==0) return;
1573 for (int i = predicate_count(); i > 0; i--) {
1574 Node * n = predicate_opaque1_node(i-1);
1575 assert(n->Opcode() == Op_Opaque1, "must be");
1576 igvn.replace_node(n, n->in(1));
1577 }
1578 assert(predicate_count()==0, "should be clean!");
1579 igvn.optimize();
1580 }
1581
1582 //------------------------------Optimize---------------------------------------
1583 // Given a graph, optimize it.
1584 void Compile::Optimize() {
1585 TracePhase t1("optimizer", &_t_optimizer, true);
1586
1587 #ifndef PRODUCT
1588 if (env()->break_at_compile()) {
1589 BREAKPOINT;
1590 }
1591
1592 #endif
1593
1594 ResourceMark rm;
1595 int loop_opts_cnt;
1596
1597 NOT_PRODUCT( verify_graph_edges(); )
1598
1599 print_method("After Parsing");
1600
1601 {
1602 // Iterative Global Value Numbering, including ideal transforms
1603 // Initialize IterGVN with types and values from parse-time GVN
1604 PhaseIterGVN igvn(initial_gvn());
1605 {
1606 NOT_PRODUCT( TracePhase t2("iterGVN", &_t_iterGVN, TimeCompiler); )
1607 igvn.optimize();
1608 }
1609
1610 print_method("Iter GVN 1", 2);
1611
1612 if (failing()) return;
1613
1614 // Perform escape analysis
1615 if (_do_escape_analysis && ConnectionGraph::has_candidates(this)) {
1616 TracePhase t2("escapeAnalysis", &_t_escapeAnalysis, true);
1617 ConnectionGraph::do_analysis(this, &igvn);
1618
1619 if (failing()) return;
1620
1621 igvn.optimize();
1622 print_method("Iter GVN 3", 2);
1623
1624 if (failing()) return;
1625
1626 }
1627
1628 // Loop transforms on the ideal graph. Range Check Elimination,
1629 // peeling, unrolling, etc.
1630
1631 // Set loop opts counter
1632 loop_opts_cnt = num_loop_opts();
1633 if((loop_opts_cnt > 0) && (has_loops() || has_split_ifs())) {
1634 {
1635 TracePhase t2("idealLoop", &_t_idealLoop, true);
1636 PhaseIdealLoop ideal_loop( igvn, true, UseLoopPredicate);
1637 loop_opts_cnt--;
1638 if (major_progress()) print_method("PhaseIdealLoop 1", 2);
1639 if (failing()) return;
1640 }
1641 // Loop opts pass if partial peeling occurred in previous pass
1642 if(PartialPeelLoop && major_progress() && (loop_opts_cnt > 0)) {
1643 TracePhase t3("idealLoop", &_t_idealLoop, true);
1644 PhaseIdealLoop ideal_loop( igvn, false, UseLoopPredicate);
1645 loop_opts_cnt--;
1646 if (major_progress()) print_method("PhaseIdealLoop 2", 2);
1647 if (failing()) return;
1648 }
1649 // Loop opts pass for loop-unrolling before CCP
1650 if(major_progress() && (loop_opts_cnt > 0)) {
1651 TracePhase t4("idealLoop", &_t_idealLoop, true);
1652 PhaseIdealLoop ideal_loop( igvn, false, UseLoopPredicate);
1653 loop_opts_cnt--;
1654 if (major_progress()) print_method("PhaseIdealLoop 3", 2);
1655 }
1656 if (!failing()) {
1657 // Verify that last round of loop opts produced a valid graph
1658 NOT_PRODUCT( TracePhase t2("idealLoopVerify", &_t_idealLoopVerify, TimeCompiler); )
1659 PhaseIdealLoop::verify(igvn);
1660 }
1661 }
1662 if (failing()) return;
1663
1664 // Conditional Constant Propagation;
1665 PhaseCCP ccp( &igvn );
1666 assert( true, "Break here to ccp.dump_nodes_and_types(_root,999,1)");
1667 {
1668 TracePhase t2("ccp", &_t_ccp, true);
1669 ccp.do_transform();
1670 }
1671 print_method("PhaseCPP 1", 2);
1672
1673 assert( true, "Break here to ccp.dump_old2new_map()");
1674
1675 // Iterative Global Value Numbering, including ideal transforms
1676 {
1677 NOT_PRODUCT( TracePhase t2("iterGVN2", &_t_iterGVN2, TimeCompiler); )
1678 igvn = ccp;
1679 igvn.optimize();
1680 }
1681
1682 print_method("Iter GVN 2", 2);
1683
1684 if (failing()) return;
1685
1686 // Loop transforms on the ideal graph. Range Check Elimination,
1687 // peeling, unrolling, etc.
1688 if(loop_opts_cnt > 0) {
1689 debug_only( int cnt = 0; );
1690 bool loop_predication = UseLoopPredicate;
1691 while(major_progress() && (loop_opts_cnt > 0)) {
1692 TracePhase t2("idealLoop", &_t_idealLoop, true);
1693 assert( cnt++ < 40, "infinite cycle in loop optimization" );
1694 PhaseIdealLoop ideal_loop( igvn, true, loop_predication);
1695 loop_opts_cnt--;
1696 if (major_progress()) print_method("PhaseIdealLoop iterations", 2);
1697 if (failing()) return;
1698 // Perform loop predication optimization during first iteration after CCP.
1699 // After that switch it off and cleanup unused loop predicates.
1700 if (loop_predication) {
1701 loop_predication = false;
1702 cleanup_loop_predicates(igvn);
1703 if (failing()) return;
1704 }
1705 }
1706 }
1707
1708 {
1709 // Verify that all previous optimizations produced a valid graph
1710 // at least to this point, even if no loop optimizations were done.
1711 NOT_PRODUCT( TracePhase t2("idealLoopVerify", &_t_idealLoopVerify, TimeCompiler); )
1712 PhaseIdealLoop::verify(igvn);
1713 }
1714
1715 {
1716 NOT_PRODUCT( TracePhase t2("macroExpand", &_t_macroExpand, TimeCompiler); )
1717 PhaseMacroExpand mex(igvn);
1718 if (mex.expand_macro_nodes()) {
1719 assert(failing(), "must bail out w/ explicit message");
1720 return;
1721 }
1722 }
1723
1724 } // (End scope of igvn; run destructor if necessary for asserts.)
1725
1726 // A method with only infinite loops has no edges entering loops from root
1727 {
1728 NOT_PRODUCT( TracePhase t2("graphReshape", &_t_graphReshaping, TimeCompiler); )
1729 if (final_graph_reshaping()) {
1730 assert(failing(), "must bail out w/ explicit message");
1731 return;
1732 }
1733 }
1734
1735 print_method("Optimize finished", 2);
1736 }
1737
1738
1739 //------------------------------Code_Gen---------------------------------------
1740 // Given a graph, generate code for it
1741 void Compile::Code_Gen() {
1742 if (failing()) return;
1743
1744 // Perform instruction selection. You might think we could reclaim Matcher
1745 // memory PDQ, but actually the Matcher is used in generating spill code.
1746 // Internals of the Matcher (including some VectorSets) must remain live
1747 // for awhile - thus I cannot reclaim Matcher memory lest a VectorSet usage
1748 // set a bit in reclaimed memory.
1749
1750 // In debug mode can dump m._nodes.dump() for mapping of ideal to machine
1751 // nodes. Mapping is only valid at the root of each matched subtree.
1752 NOT_PRODUCT( verify_graph_edges(); )
1753
1754 Node_List proj_list;
1755 Matcher m(proj_list);
1756 _matcher = &m;
1757 {
1758 TracePhase t2("matcher", &_t_matcher, true);
1759 m.match();
1760 }
1761 // In debug mode can dump m._nodes.dump() for mapping of ideal to machine
1762 // nodes. Mapping is only valid at the root of each matched subtree.
1763 NOT_PRODUCT( verify_graph_edges(); )
1764
1765 // If you have too many nodes, or if matching has failed, bail out
1766 check_node_count(0, "out of nodes matching instructions");
1767 if (failing()) return;
1768
1769 // Build a proper-looking CFG
1770 PhaseCFG cfg(node_arena(), root(), m);
1771 _cfg = &cfg;
1772 {
1773 NOT_PRODUCT( TracePhase t2("scheduler", &_t_scheduler, TimeCompiler); )
1774 cfg.Dominators();
1775 if (failing()) return;
1776
1777 NOT_PRODUCT( verify_graph_edges(); )
1778
1779 cfg.Estimate_Block_Frequency();
1780 cfg.GlobalCodeMotion(m,unique(),proj_list);
1781
1782 print_method("Global code motion", 2);
1783
1784 if (failing()) return;
1785 NOT_PRODUCT( verify_graph_edges(); )
1786
1787 debug_only( cfg.verify(); )
1788 }
1789 NOT_PRODUCT( verify_graph_edges(); )
1790
1791 PhaseChaitin regalloc(unique(),cfg,m);
1792 _regalloc = ®alloc;
1793 {
1794 TracePhase t2("regalloc", &_t_registerAllocation, true);
1795 // Perform any platform dependent preallocation actions. This is used,
1796 // for example, to avoid taking an implicit null pointer exception
1797 // using the frame pointer on win95.
1798 _regalloc->pd_preallocate_hook();
1799
1800 // Perform register allocation. After Chaitin, use-def chains are
1801 // no longer accurate (at spill code) and so must be ignored.
1802 // Node->LRG->reg mappings are still accurate.
1803 _regalloc->Register_Allocate();
1804
1805 // Bail out if the allocator builds too many nodes
1806 if (failing()) return;
1807 }
1808
1809 // Prior to register allocation we kept empty basic blocks in case the
1810 // the allocator needed a place to spill. After register allocation we
1811 // are not adding any new instructions. If any basic block is empty, we
1812 // can now safely remove it.
1813 {
1814 NOT_PRODUCT( TracePhase t2("blockOrdering", &_t_blockOrdering, TimeCompiler); )
1815 cfg.remove_empty();
1816 if (do_freq_based_layout()) {
1817 PhaseBlockLayout layout(cfg);
1818 } else {
1819 cfg.set_loop_alignment();
1820 }
1821 cfg.fixup_flow();
1822 }
1823
1824 // Perform any platform dependent postallocation verifications.
1825 debug_only( _regalloc->pd_postallocate_verify_hook(); )
1826
1827 // Apply peephole optimizations
1828 if( OptoPeephole ) {
1829 NOT_PRODUCT( TracePhase t2("peephole", &_t_peephole, TimeCompiler); )
1830 PhasePeephole peep( _regalloc, cfg);
1831 peep.do_transform();
1832 }
1833
1834 // Convert Nodes to instruction bits in a buffer
1835 {
1836 // %%%% workspace merge brought two timers together for one job
1837 TracePhase t2a("output", &_t_output, true);
1838 NOT_PRODUCT( TraceTime t2b(NULL, &_t_codeGeneration, TimeCompiler, false); )
1839 Output();
1840 }
1841
1842 print_method("Final Code");
1843
1844 // He's dead, Jim.
1845 _cfg = (PhaseCFG*)0xdeadbeef;
1846 _regalloc = (PhaseChaitin*)0xdeadbeef;
1847 }
1848
1849
1850 //------------------------------dump_asm---------------------------------------
1851 // Dump formatted assembly
1852 #ifndef PRODUCT
1853 void Compile::dump_asm(int *pcs, uint pc_limit) {
1854 bool cut_short = false;
1855 tty->print_cr("#");
1856 tty->print("# "); _tf->dump(); tty->cr();
1857 tty->print_cr("#");
1858
1859 // For all blocks
1860 int pc = 0x0; // Program counter
1861 char starts_bundle = ' ';
1862 _regalloc->dump_frame();
1863
1864 Node *n = NULL;
1865 for( uint i=0; i<_cfg->_num_blocks; i++ ) {
1866 if (VMThread::should_terminate()) { cut_short = true; break; }
1867 Block *b = _cfg->_blocks[i];
1868 if (b->is_connector() && !Verbose) continue;
1869 n = b->_nodes[0];
1870 if (pcs && n->_idx < pc_limit)
1871 tty->print("%3.3x ", pcs[n->_idx]);
1872 else
1873 tty->print(" ");
1874 b->dump_head( &_cfg->_bbs );
1875 if (b->is_connector()) {
1876 tty->print_cr(" # Empty connector block");
1877 } else if (b->num_preds() == 2 && b->pred(1)->is_CatchProj() && b->pred(1)->as_CatchProj()->_con == CatchProjNode::fall_through_index) {
1878 tty->print_cr(" # Block is sole successor of call");
1879 }
1880
1881 // For all instructions
1882 Node *delay = NULL;
1883 for( uint j = 0; j<b->_nodes.size(); j++ ) {
1884 if (VMThread::should_terminate()) { cut_short = true; break; }
1885 n = b->_nodes[j];
1886 if (valid_bundle_info(n)) {
1887 Bundle *bundle = node_bundling(n);
1888 if (bundle->used_in_unconditional_delay()) {
1889 delay = n;
1890 continue;
1891 }
1892 if (bundle->starts_bundle())
1893 starts_bundle = '+';
1894 }
1895
1896 if (WizardMode) n->dump();
1897
1898 if( !n->is_Region() && // Dont print in the Assembly
1899 !n->is_Phi() && // a few noisely useless nodes
1900 !n->is_Proj() &&
1901 !n->is_MachTemp() &&
1902 !n->is_SafePointScalarObject() &&
1903 !n->is_Catch() && // Would be nice to print exception table targets
1904 !n->is_MergeMem() && // Not very interesting
1905 !n->is_top() && // Debug info table constants
1906 !(n->is_Con() && !n->is_Mach())// Debug info table constants
1907 ) {
1908 if (pcs && n->_idx < pc_limit)
1909 tty->print("%3.3x", pcs[n->_idx]);
1910 else
1911 tty->print(" ");
1912 tty->print(" %c ", starts_bundle);
1913 starts_bundle = ' ';
1914 tty->print("\t");
1915 n->format(_regalloc, tty);
1916 tty->cr();
1917 }
1918
1919 // If we have an instruction with a delay slot, and have seen a delay,
1920 // then back up and print it
1921 if (valid_bundle_info(n) && node_bundling(n)->use_unconditional_delay()) {
1922 assert(delay != NULL, "no unconditional delay instruction");
1923 if (WizardMode) delay->dump();
1924
1925 if (node_bundling(delay)->starts_bundle())
1926 starts_bundle = '+';
1927 if (pcs && n->_idx < pc_limit)
1928 tty->print("%3.3x", pcs[n->_idx]);
1929 else
1930 tty->print(" ");
1931 tty->print(" %c ", starts_bundle);
1932 starts_bundle = ' ';
1933 tty->print("\t");
1934 delay->format(_regalloc, tty);
1935 tty->print_cr("");
1936 delay = NULL;
1937 }
1938
1939 // Dump the exception table as well
1940 if( n->is_Catch() && (Verbose || WizardMode) ) {
1941 // Print the exception table for this offset
1942 _handler_table.print_subtable_for(pc);
1943 }
1944 }
1945
1946 if (pcs && n->_idx < pc_limit)
1947 tty->print_cr("%3.3x", pcs[n->_idx]);
1948 else
1949 tty->print_cr("");
1950
1951 assert(cut_short || delay == NULL, "no unconditional delay branch");
1952
1953 } // End of per-block dump
1954 tty->print_cr("");
1955
1956 if (cut_short) tty->print_cr("*** disassembly is cut short ***");
1957 }
1958 #endif
1959
1960 //------------------------------Final_Reshape_Counts---------------------------
1961 // This class defines counters to help identify when a method
1962 // may/must be executed using hardware with only 24-bit precision.
1963 struct Final_Reshape_Counts : public StackObj {
1964 int _call_count; // count non-inlined 'common' calls
1965 int _float_count; // count float ops requiring 24-bit precision
1966 int _double_count; // count double ops requiring more precision
1967 int _java_call_count; // count non-inlined 'java' calls
1968 int _inner_loop_count; // count loops which need alignment
1969 VectorSet _visited; // Visitation flags
1970 Node_List _tests; // Set of IfNodes & PCTableNodes
1971
1972 Final_Reshape_Counts() :
1973 _call_count(0), _float_count(0), _double_count(0),
1974 _java_call_count(0), _inner_loop_count(0),
1975 _visited( Thread::current()->resource_area() ) { }
1976
1977 void inc_call_count () { _call_count ++; }
1978 void inc_float_count () { _float_count ++; }
1979 void inc_double_count() { _double_count++; }
1980 void inc_java_call_count() { _java_call_count++; }
1981 void inc_inner_loop_count() { _inner_loop_count++; }
1982
1983 int get_call_count () const { return _call_count ; }
1984 int get_float_count () const { return _float_count ; }
1985 int get_double_count() const { return _double_count; }
1986 int get_java_call_count() const { return _java_call_count; }
1987 int get_inner_loop_count() const { return _inner_loop_count; }
1988 };
1989
1990 static bool oop_offset_is_sane(const TypeInstPtr* tp) {
1991 ciInstanceKlass *k = tp->klass()->as_instance_klass();
1992 // Make sure the offset goes inside the instance layout.
1993 return k->contains_field_offset(tp->offset());
1994 // Note that OffsetBot and OffsetTop are very negative.
1995 }
1996
1997 //------------------------------final_graph_reshaping_impl----------------------
1998 // Implement items 1-5 from final_graph_reshaping below.
1999 static void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc ) {
2000
2001 if ( n->outcnt() == 0 ) return; // dead node
2002 uint nop = n->Opcode();
2003
2004 // Check for 2-input instruction with "last use" on right input.
2005 // Swap to left input. Implements item (2).
2006 if( n->req() == 3 && // two-input instruction
2007 n->in(1)->outcnt() > 1 && // left use is NOT a last use
2008 (!n->in(1)->is_Phi() || n->in(1)->in(2) != n) && // it is not data loop
2009 n->in(2)->outcnt() == 1 &&// right use IS a last use
2010 !n->in(2)->is_Con() ) { // right use is not a constant
2011 // Check for commutative opcode
2012 switch( nop ) {
2013 case Op_AddI: case Op_AddF: case Op_AddD: case Op_AddL:
2014 case Op_MaxI: case Op_MinI:
2015 case Op_MulI: case Op_MulF: case Op_MulD: case Op_MulL:
2016 case Op_AndL: case Op_XorL: case Op_OrL:
2017 case Op_AndI: case Op_XorI: case Op_OrI: {
2018 // Move "last use" input to left by swapping inputs
2019 n->swap_edges(1, 2);
2020 break;
2021 }
2022 default:
2023 break;
2024 }
2025 }
2026
2027 #ifdef ASSERT
2028 if( n->is_Mem() ) {
2029 Compile* C = Compile::current();
2030 int alias_idx = C->get_alias_index(n->as_Mem()->adr_type());
2031 assert( n->in(0) != NULL || alias_idx != Compile::AliasIdxRaw ||
2032 // oop will be recorded in oop map if load crosses safepoint
2033 n->is_Load() && (n->as_Load()->bottom_type()->isa_oopptr() ||
2034 LoadNode::is_immutable_value(n->in(MemNode::Address))),
2035 "raw memory operations should have control edge");
2036 }
2037 #endif
2038 // Count FPU ops and common calls, implements item (3)
2039 switch( nop ) {
2040 // Count all float operations that may use FPU
2041 case Op_AddF:
2042 case Op_SubF:
2043 case Op_MulF:
2044 case Op_DivF:
2045 case Op_NegF:
2046 case Op_ModF:
2047 case Op_ConvI2F:
2048 case Op_ConF:
2049 case Op_CmpF:
2050 case Op_CmpF3:
2051 // case Op_ConvL2F: // longs are split into 32-bit halves
2052 frc.inc_float_count();
2053 break;
2054
2055 case Op_ConvF2D:
2056 case Op_ConvD2F:
2057 frc.inc_float_count();
2058 frc.inc_double_count();
2059 break;
2060
2061 // Count all double operations that may use FPU
2062 case Op_AddD:
2063 case Op_SubD:
2064 case Op_MulD:
2065 case Op_DivD:
2066 case Op_NegD:
2067 case Op_ModD:
2068 case Op_ConvI2D:
2069 case Op_ConvD2I:
2070 // case Op_ConvL2D: // handled by leaf call
2071 // case Op_ConvD2L: // handled by leaf call
2072 case Op_ConD:
2073 case Op_CmpD:
2074 case Op_CmpD3:
2075 frc.inc_double_count();
2076 break;
2077 case Op_Opaque1: // Remove Opaque Nodes before matching
2078 case Op_Opaque2: // Remove Opaque Nodes before matching
2079 n->subsume_by(n->in(1));
2080 break;
2081 case Op_CallStaticJava:
2082 case Op_CallJava:
2083 case Op_CallDynamicJava:
2084 frc.inc_java_call_count(); // Count java call site;
2085 case Op_CallRuntime:
2086 case Op_CallLeaf:
2087 case Op_CallLeafNoFP: {
2088 assert( n->is_Call(), "" );
2089 CallNode *call = n->as_Call();
2090 // Count call sites where the FP mode bit would have to be flipped.
2091 // Do not count uncommon runtime calls:
2092 // uncommon_trap, _complete_monitor_locking, _complete_monitor_unlocking,
2093 // _new_Java, _new_typeArray, _new_objArray, _rethrow_Java, ...
2094 if( !call->is_CallStaticJava() || !call->as_CallStaticJava()->_name ) {
2095 frc.inc_call_count(); // Count the call site
2096 } else { // See if uncommon argument is shared
2097 Node *n = call->in(TypeFunc::Parms);
2098 int nop = n->Opcode();
2099 // Clone shared simple arguments to uncommon calls, item (1).
2100 if( n->outcnt() > 1 &&
2101 !n->is_Proj() &&
2102 nop != Op_CreateEx &&
2103 nop != Op_CheckCastPP &&
2104 nop != Op_DecodeN &&
2105 !n->is_Mem() ) {
2106 Node *x = n->clone();
2107 call->set_req( TypeFunc::Parms, x );
2108 }
2109 }
2110 break;
2111 }
2112
2113 case Op_StoreD:
2114 case Op_LoadD:
2115 case Op_LoadD_unaligned:
2116 frc.inc_double_count();
2117 goto handle_mem;
2118 case Op_StoreF:
2119 case Op_LoadF:
2120 frc.inc_float_count();
2121 goto handle_mem;
2122
2123 case Op_StoreB:
2124 case Op_StoreC:
2125 case Op_StoreCM:
2126 case Op_StorePConditional:
2127 case Op_StoreI:
2128 case Op_StoreL:
2129 case Op_StoreIConditional:
2130 case Op_StoreLConditional:
2131 case Op_CompareAndSwapI:
2132 case Op_CompareAndSwapL:
2133 case Op_CompareAndSwapP:
2134 case Op_CompareAndSwapN:
2135 case Op_StoreP:
2136 case Op_StoreN:
2137 case Op_LoadB:
2138 case Op_LoadUB:
2139 case Op_LoadUS:
2140 case Op_LoadI:
2141 case Op_LoadUI2L:
2142 case Op_LoadKlass:
2143 case Op_LoadNKlass:
2144 case Op_LoadL:
2145 case Op_LoadL_unaligned:
2146 case Op_LoadPLocked:
2147 case Op_LoadLLocked:
2148 case Op_LoadP:
2149 case Op_LoadN:
2150 case Op_LoadRange:
2151 case Op_LoadS: {
2152 handle_mem:
2153 #ifdef ASSERT
2154 if( VerifyOptoOopOffsets ) {
2155 assert( n->is_Mem(), "" );
2156 MemNode *mem = (MemNode*)n;
2157 // Check to see if address types have grounded out somehow.
2158 const TypeInstPtr *tp = mem->in(MemNode::Address)->bottom_type()->isa_instptr();
2159 assert( !tp || oop_offset_is_sane(tp), "" );
2160 }
2161 #endif
2162 break;
2163 }
2164
2165 case Op_AddP: { // Assert sane base pointers
2166 Node *addp = n->in(AddPNode::Address);
2167 assert( !addp->is_AddP() ||
2168 addp->in(AddPNode::Base)->is_top() || // Top OK for allocation
2169 addp->in(AddPNode::Base) == n->in(AddPNode::Base),
2170 "Base pointers must match" );
2171 #ifdef _LP64
2172 if (UseCompressedOops &&
2173 addp->Opcode() == Op_ConP &&
2174 addp == n->in(AddPNode::Base) &&
2175 n->in(AddPNode::Offset)->is_Con()) {
2176 // Use addressing with narrow klass to load with offset on x86.
2177 // On sparc loading 32-bits constant and decoding it have less
2178 // instructions (4) then load 64-bits constant (7).
2179 // Do this transformation here since IGVN will convert ConN back to ConP.
2180 const Type* t = addp->bottom_type();
2181 if (t->isa_oopptr()) {
2182 Node* nn = NULL;
2183
2184 // Look for existing ConN node of the same exact type.
2185 Compile* C = Compile::current();
2186 Node* r = C->root();
2187 uint cnt = r->outcnt();
2188 for (uint i = 0; i < cnt; i++) {
2189 Node* m = r->raw_out(i);
2190 if (m!= NULL && m->Opcode() == Op_ConN &&
2191 m->bottom_type()->make_ptr() == t) {
2192 nn = m;
2193 break;
2194 }
2195 }
2196 if (nn != NULL) {
2197 // Decode a narrow oop to match address
2198 // [R12 + narrow_oop_reg<<3 + offset]
2199 nn = new (C, 2) DecodeNNode(nn, t);
2200 n->set_req(AddPNode::Base, nn);
2201 n->set_req(AddPNode::Address, nn);
2202 if (addp->outcnt() == 0) {
2203 addp->disconnect_inputs(NULL);
2204 }
2205 }
2206 }
2207 }
2208 #endif
2209 break;
2210 }
2211
2212 #ifdef _LP64
2213 case Op_CastPP:
2214 if (n->in(1)->is_DecodeN() && Matcher::gen_narrow_oop_implicit_null_checks()) {
2215 Compile* C = Compile::current();
2216 Node* in1 = n->in(1);
2217 const Type* t = n->bottom_type();
2218 Node* new_in1 = in1->clone();
2219 new_in1->as_DecodeN()->set_type(t);
2220
2221 if (!Matcher::narrow_oop_use_complex_address()) {
2222 //
2223 // x86, ARM and friends can handle 2 adds in addressing mode
2224 // and Matcher can fold a DecodeN node into address by using
2225 // a narrow oop directly and do implicit NULL check in address:
2226 //
2227 // [R12 + narrow_oop_reg<<3 + offset]
2228 // NullCheck narrow_oop_reg
2229 //
2230 // On other platforms (Sparc) we have to keep new DecodeN node and
2231 // use it to do implicit NULL check in address:
2232 //
2233 // decode_not_null narrow_oop_reg, base_reg
2234 // [base_reg + offset]
2235 // NullCheck base_reg
2236 //
2237 // Pin the new DecodeN node to non-null path on these platform (Sparc)
2238 // to keep the information to which NULL check the new DecodeN node
2239 // corresponds to use it as value in implicit_null_check().
2240 //
2241 new_in1->set_req(0, n->in(0));
2242 }
2243
2244 n->subsume_by(new_in1);
2245 if (in1->outcnt() == 0) {
2246 in1->disconnect_inputs(NULL);
2247 }
2248 }
2249 break;
2250
2251 case Op_CmpP:
2252 // Do this transformation here to preserve CmpPNode::sub() and
2253 // other TypePtr related Ideal optimizations (for example, ptr nullness).
2254 if (n->in(1)->is_DecodeN() || n->in(2)->is_DecodeN()) {
2255 Node* in1 = n->in(1);
2256 Node* in2 = n->in(2);
2257 if (!in1->is_DecodeN()) {
2258 in2 = in1;
2259 in1 = n->in(2);
2260 }
2261 assert(in1->is_DecodeN(), "sanity");
2262
2263 Compile* C = Compile::current();
2264 Node* new_in2 = NULL;
2265 if (in2->is_DecodeN()) {
2266 new_in2 = in2->in(1);
2267 } else if (in2->Opcode() == Op_ConP) {
2268 const Type* t = in2->bottom_type();
2269 if (t == TypePtr::NULL_PTR) {
2270 // Don't convert CmpP null check into CmpN if compressed
2271 // oops implicit null check is not generated.
2272 // This will allow to generate normal oop implicit null check.
2273 if (Matcher::gen_narrow_oop_implicit_null_checks())
2274 new_in2 = ConNode::make(C, TypeNarrowOop::NULL_PTR);
2275 //
2276 // This transformation together with CastPP transformation above
2277 // will generated code for implicit NULL checks for compressed oops.
2278 //
2279 // The original code after Optimize()
2280 //
2281 // LoadN memory, narrow_oop_reg
2282 // decode narrow_oop_reg, base_reg
2283 // CmpP base_reg, NULL
2284 // CastPP base_reg // NotNull
2285 // Load [base_reg + offset], val_reg
2286 //
2287 // after these transformations will be
2288 //
2289 // LoadN memory, narrow_oop_reg
2290 // CmpN narrow_oop_reg, NULL
2291 // decode_not_null narrow_oop_reg, base_reg
2292 // Load [base_reg + offset], val_reg
2293 //
2294 // and the uncommon path (== NULL) will use narrow_oop_reg directly
2295 // since narrow oops can be used in debug info now (see the code in
2296 // final_graph_reshaping_walk()).
2297 //
2298 // At the end the code will be matched to
2299 // on x86:
2300 //
2301 // Load_narrow_oop memory, narrow_oop_reg
2302 // Load [R12 + narrow_oop_reg<<3 + offset], val_reg
2303 // NullCheck narrow_oop_reg
2304 //
2305 // and on sparc:
2306 //
2307 // Load_narrow_oop memory, narrow_oop_reg
2308 // decode_not_null narrow_oop_reg, base_reg
2309 // Load [base_reg + offset], val_reg
2310 // NullCheck base_reg
2311 //
2312 } else if (t->isa_oopptr()) {
2313 new_in2 = ConNode::make(C, t->make_narrowoop());
2314 }
2315 }
2316 if (new_in2 != NULL) {
2317 Node* cmpN = new (C, 3) CmpNNode(in1->in(1), new_in2);
2318 n->subsume_by( cmpN );
2319 if (in1->outcnt() == 0) {
2320 in1->disconnect_inputs(NULL);
2321 }
2322 if (in2->outcnt() == 0) {
2323 in2->disconnect_inputs(NULL);
2324 }
2325 }
2326 }
2327 break;
2328
2329 case Op_DecodeN:
2330 assert(!n->in(1)->is_EncodeP(), "should be optimized out");
2331 // DecodeN could be pinned when it can't be fold into
2332 // an address expression, see the code for Op_CastPP above.
2333 assert(n->in(0) == NULL || !Matcher::narrow_oop_use_complex_address(), "no control");
2334 break;
2335
2336 case Op_EncodeP: {
2337 Node* in1 = n->in(1);
2338 if (in1->is_DecodeN()) {
2339 n->subsume_by(in1->in(1));
2340 } else if (in1->Opcode() == Op_ConP) {
2341 Compile* C = Compile::current();
2342 const Type* t = in1->bottom_type();
2343 if (t == TypePtr::NULL_PTR) {
2344 n->subsume_by(ConNode::make(C, TypeNarrowOop::NULL_PTR));
2345 } else if (t->isa_oopptr()) {
2346 n->subsume_by(ConNode::make(C, t->make_narrowoop()));
2347 }
2348 }
2349 if (in1->outcnt() == 0) {
2350 in1->disconnect_inputs(NULL);
2351 }
2352 break;
2353 }
2354
2355 case Op_Proj: {
2356 if (OptimizeStringConcat) {
2357 ProjNode* p = n->as_Proj();
2358 if (p->_is_io_use) {
2359 // Separate projections were used for the exception path which
2360 // are normally removed by a late inline. If it wasn't inlined
2361 // then they will hang around and should just be replaced with
2362 // the original one.
2363 Node* proj = NULL;
2364 // Replace with just one
2365 for (SimpleDUIterator i(p->in(0)); i.has_next(); i.next()) {
2366 Node *use = i.get();
2367 if (use->is_Proj() && p != use && use->as_Proj()->_con == p->_con) {
2368 proj = use;
2369 break;
2370 }
2371 }
2372 assert(p != NULL, "must be found");
2373 p->subsume_by(proj);
2374 }
2375 }
2376 break;
2377 }
2378
2379 case Op_Phi:
2380 if (n->as_Phi()->bottom_type()->isa_narrowoop()) {
2381 // The EncodeP optimization may create Phi with the same edges
2382 // for all paths. It is not handled well by Register Allocator.
2383 Node* unique_in = n->in(1);
2384 assert(unique_in != NULL, "");
2385 uint cnt = n->req();
2386 for (uint i = 2; i < cnt; i++) {
2387 Node* m = n->in(i);
2388 assert(m != NULL, "");
2389 if (unique_in != m)
2390 unique_in = NULL;
2391 }
2392 if (unique_in != NULL) {
2393 n->subsume_by(unique_in);
2394 }
2395 }
2396 break;
2397
2398 #endif
2399
2400 case Op_ModI:
2401 if (UseDivMod) {
2402 // Check if a%b and a/b both exist
2403 Node* d = n->find_similar(Op_DivI);
2404 if (d) {
2405 // Replace them with a fused divmod if supported
2406 Compile* C = Compile::current();
2407 if (Matcher::has_match_rule(Op_DivModI)) {
2408 DivModINode* divmod = DivModINode::make(C, n);
2409 d->subsume_by(divmod->div_proj());
2410 n->subsume_by(divmod->mod_proj());
2411 } else {
2412 // replace a%b with a-((a/b)*b)
2413 Node* mult = new (C, 3) MulINode(d, d->in(2));
2414 Node* sub = new (C, 3) SubINode(d->in(1), mult);
2415 n->subsume_by( sub );
2416 }
2417 }
2418 }
2419 break;
2420
2421 case Op_ModL:
2422 if (UseDivMod) {
2423 // Check if a%b and a/b both exist
2424 Node* d = n->find_similar(Op_DivL);
2425 if (d) {
2426 // Replace them with a fused divmod if supported
2427 Compile* C = Compile::current();
2428 if (Matcher::has_match_rule(Op_DivModL)) {
2429 DivModLNode* divmod = DivModLNode::make(C, n);
2430 d->subsume_by(divmod->div_proj());
2431 n->subsume_by(divmod->mod_proj());
2432 } else {
2433 // replace a%b with a-((a/b)*b)
2434 Node* mult = new (C, 3) MulLNode(d, d->in(2));
2435 Node* sub = new (C, 3) SubLNode(d->in(1), mult);
2436 n->subsume_by( sub );
2437 }
2438 }
2439 }
2440 break;
2441
2442 case Op_Load16B:
2443 case Op_Load8B:
2444 case Op_Load4B:
2445 case Op_Load8S:
2446 case Op_Load4S:
2447 case Op_Load2S:
2448 case Op_Load8C:
2449 case Op_Load4C:
2450 case Op_Load2C:
2451 case Op_Load4I:
2452 case Op_Load2I:
2453 case Op_Load2L:
2454 case Op_Load4F:
2455 case Op_Load2F:
2456 case Op_Load2D:
2457 case Op_Store16B:
2458 case Op_Store8B:
2459 case Op_Store4B:
2460 case Op_Store8C:
2461 case Op_Store4C:
2462 case Op_Store2C:
2463 case Op_Store4I:
2464 case Op_Store2I:
2465 case Op_Store2L:
2466 case Op_Store4F:
2467 case Op_Store2F:
2468 case Op_Store2D:
2469 break;
2470
2471 case Op_PackB:
2472 case Op_PackS:
2473 case Op_PackC:
2474 case Op_PackI:
2475 case Op_PackF:
2476 case Op_PackL:
2477 case Op_PackD:
2478 if (n->req()-1 > 2) {
2479 // Replace many operand PackNodes with a binary tree for matching
2480 PackNode* p = (PackNode*) n;
2481 Node* btp = p->binaryTreePack(Compile::current(), 1, n->req());
2482 n->subsume_by(btp);
2483 }
2484 break;
2485 case Op_Loop:
2486 case Op_CountedLoop:
2487 if (n->as_Loop()->is_inner_loop()) {
2488 frc.inc_inner_loop_count();
2489 }
2490 break;
2491 default:
2492 assert( !n->is_Call(), "" );
2493 assert( !n->is_Mem(), "" );
2494 break;
2495 }
2496
2497 // Collect CFG split points
2498 if (n->is_MultiBranch())
2499 frc._tests.push(n);
2500 }
2501
2502 //------------------------------final_graph_reshaping_walk---------------------
2503 // Replacing Opaque nodes with their input in final_graph_reshaping_impl(),
2504 // requires that the walk visits a node's inputs before visiting the node.
2505 static void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc ) {
2506 ResourceArea *area = Thread::current()->resource_area();
2507 Unique_Node_List sfpt(area);
2508
2509 frc._visited.set(root->_idx); // first, mark node as visited
2510 uint cnt = root->req();
2511 Node *n = root;
2512 uint i = 0;
2513 while (true) {
2514 if (i < cnt) {
2515 // Place all non-visited non-null inputs onto stack
2516 Node* m = n->in(i);
2517 ++i;
2518 if (m != NULL && !frc._visited.test_set(m->_idx)) {
2519 if (m->is_SafePoint() && m->as_SafePoint()->jvms() != NULL)
2520 sfpt.push(m);
2521 cnt = m->req();
2522 nstack.push(n, i); // put on stack parent and next input's index
2523 n = m;
2524 i = 0;
2525 }
2526 } else {
2527 // Now do post-visit work
2528 final_graph_reshaping_impl( n, frc );
2529 if (nstack.is_empty())
2530 break; // finished
2531 n = nstack.node(); // Get node from stack
2532 cnt = n->req();
2533 i = nstack.index();
2534 nstack.pop(); // Shift to the next node on stack
2535 }
2536 }
2537
2538 // Skip next transformation if compressed oops are not used.
2539 if (!UseCompressedOops || !Matcher::gen_narrow_oop_implicit_null_checks())
2540 return;
2541
2542 // Go over safepoints nodes to skip DecodeN nodes for debug edges.
2543 // It could be done for an uncommon traps or any safepoints/calls
2544 // if the DecodeN node is referenced only in a debug info.
2545 while (sfpt.size() > 0) {
2546 n = sfpt.pop();
2547 JVMState *jvms = n->as_SafePoint()->jvms();
2548 assert(jvms != NULL, "sanity");
2549 int start = jvms->debug_start();
2550 int end = n->req();
2551 bool is_uncommon = (n->is_CallStaticJava() &&
2552 n->as_CallStaticJava()->uncommon_trap_request() != 0);
2553 for (int j = start; j < end; j++) {
2554 Node* in = n->in(j);
2555 if (in->is_DecodeN()) {
2556 bool safe_to_skip = true;
2557 if (!is_uncommon ) {
2558 // Is it safe to skip?
2559 for (uint i = 0; i < in->outcnt(); i++) {
2560 Node* u = in->raw_out(i);
2561 if (!u->is_SafePoint() ||
2562 u->is_Call() && u->as_Call()->has_non_debug_use(n)) {
2563 safe_to_skip = false;
2564 }
2565 }
2566 }
2567 if (safe_to_skip) {
2568 n->set_req(j, in->in(1));
2569 }
2570 if (in->outcnt() == 0) {
2571 in->disconnect_inputs(NULL);
2572 }
2573 }
2574 }
2575 }
2576 }
2577
2578 //------------------------------final_graph_reshaping--------------------------
2579 // Final Graph Reshaping.
2580 //
2581 // (1) Clone simple inputs to uncommon calls, so they can be scheduled late
2582 // and not commoned up and forced early. Must come after regular
2583 // optimizations to avoid GVN undoing the cloning. Clone constant
2584 // inputs to Loop Phis; these will be split by the allocator anyways.
2585 // Remove Opaque nodes.
2586 // (2) Move last-uses by commutative operations to the left input to encourage
2587 // Intel update-in-place two-address operations and better register usage
2588 // on RISCs. Must come after regular optimizations to avoid GVN Ideal
2589 // calls canonicalizing them back.
2590 // (3) Count the number of double-precision FP ops, single-precision FP ops
2591 // and call sites. On Intel, we can get correct rounding either by
2592 // forcing singles to memory (requires extra stores and loads after each
2593 // FP bytecode) or we can set a rounding mode bit (requires setting and
2594 // clearing the mode bit around call sites). The mode bit is only used
2595 // if the relative frequency of single FP ops to calls is low enough.
2596 // This is a key transform for SPEC mpeg_audio.
2597 // (4) Detect infinite loops; blobs of code reachable from above but not
2598 // below. Several of the Code_Gen algorithms fail on such code shapes,
2599 // so we simply bail out. Happens a lot in ZKM.jar, but also happens
2600 // from time to time in other codes (such as -Xcomp finalizer loops, etc).
2601 // Detection is by looking for IfNodes where only 1 projection is
2602 // reachable from below or CatchNodes missing some targets.
2603 // (5) Assert for insane oop offsets in debug mode.
2604
2605 bool Compile::final_graph_reshaping() {
2606 // an infinite loop may have been eliminated by the optimizer,
2607 // in which case the graph will be empty.
2608 if (root()->req() == 1) {
2609 record_method_not_compilable("trivial infinite loop");
2610 return true;
2611 }
2612
2613 Final_Reshape_Counts frc;
2614
2615 // Visit everybody reachable!
2616 // Allocate stack of size C->unique()/2 to avoid frequent realloc
2617 Node_Stack nstack(unique() >> 1);
2618 final_graph_reshaping_walk(nstack, root(), frc);
2619
2620 // Check for unreachable (from below) code (i.e., infinite loops).
2621 for( uint i = 0; i < frc._tests.size(); i++ ) {
2622 MultiBranchNode *n = frc._tests[i]->as_MultiBranch();
2623 // Get number of CFG targets.
2624 // Note that PCTables include exception targets after calls.
2625 uint required_outcnt = n->required_outcnt();
2626 if (n->outcnt() != required_outcnt) {
2627 // Check for a few special cases. Rethrow Nodes never take the
2628 // 'fall-thru' path, so expected kids is 1 less.
2629 if (n->is_PCTable() && n->in(0) && n->in(0)->in(0)) {
2630 if (n->in(0)->in(0)->is_Call()) {
2631 CallNode *call = n->in(0)->in(0)->as_Call();
2632 if (call->entry_point() == OptoRuntime::rethrow_stub()) {
2633 required_outcnt--; // Rethrow always has 1 less kid
2634 } else if (call->req() > TypeFunc::Parms &&
2635 call->is_CallDynamicJava()) {
2636 // Check for null receiver. In such case, the optimizer has
2637 // detected that the virtual call will always result in a null
2638 // pointer exception. The fall-through projection of this CatchNode
2639 // will not be populated.
2640 Node *arg0 = call->in(TypeFunc::Parms);
2641 if (arg0->is_Type() &&
2642 arg0->as_Type()->type()->higher_equal(TypePtr::NULL_PTR)) {
2643 required_outcnt--;
2644 }
2645 } else if (call->entry_point() == OptoRuntime::new_array_Java() &&
2646 call->req() > TypeFunc::Parms+1 &&
2647 call->is_CallStaticJava()) {
2648 // Check for negative array length. In such case, the optimizer has
2649 // detected that the allocation attempt will always result in an
2650 // exception. There is no fall-through projection of this CatchNode .
2651 Node *arg1 = call->in(TypeFunc::Parms+1);
2652 if (arg1->is_Type() &&
2653 arg1->as_Type()->type()->join(TypeInt::POS)->empty()) {
2654 required_outcnt--;
2655 }
2656 }
2657 }
2658 }
2659 // Recheck with a better notion of 'required_outcnt'
2660 if (n->outcnt() != required_outcnt) {
2661 record_method_not_compilable("malformed control flow");
2662 return true; // Not all targets reachable!
2663 }
2664 }
2665 // Check that I actually visited all kids. Unreached kids
2666 // must be infinite loops.
2667 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++)
2668 if (!frc._visited.test(n->fast_out(j)->_idx)) {
2669 record_method_not_compilable("infinite loop");
2670 return true; // Found unvisited kid; must be unreach
2671 }
2672 }
2673
2674 // If original bytecodes contained a mixture of floats and doubles
2675 // check if the optimizer has made it homogenous, item (3).
2676 if( Use24BitFPMode && Use24BitFP && UseSSE == 0 &&
2677 frc.get_float_count() > 32 &&
2678 frc.get_double_count() == 0 &&
2679 (10 * frc.get_call_count() < frc.get_float_count()) ) {
2680 set_24_bit_selection_and_mode( false, true );
2681 }
2682
2683 set_java_calls(frc.get_java_call_count());
2684 set_inner_loops(frc.get_inner_loop_count());
2685
2686 // No infinite loops, no reason to bail out.
2687 return false;
2688 }
2689
2690 //-----------------------------too_many_traps----------------------------------
2691 // Report if there are too many traps at the current method and bci.
2692 // Return true if there was a trap, and/or PerMethodTrapLimit is exceeded.
2693 bool Compile::too_many_traps(ciMethod* method,
2694 int bci,
2695 Deoptimization::DeoptReason reason) {
2696 ciMethodData* md = method->method_data();
2697 if (md->is_empty()) {
2698 // Assume the trap has not occurred, or that it occurred only
2699 // because of a transient condition during start-up in the interpreter.
2700 return false;
2701 }
2702 if (md->has_trap_at(bci, reason) != 0) {
2703 // Assume PerBytecodeTrapLimit==0, for a more conservative heuristic.
2704 // Also, if there are multiple reasons, or if there is no per-BCI record,
2705 // assume the worst.
2706 if (log())
2707 log()->elem("observe trap='%s' count='%d'",
2708 Deoptimization::trap_reason_name(reason),
2709 md->trap_count(reason));
2710 return true;
2711 } else {
2712 // Ignore method/bci and see if there have been too many globally.
2713 return too_many_traps(reason, md);
2714 }
2715 }
2716
2717 // Less-accurate variant which does not require a method and bci.
2718 bool Compile::too_many_traps(Deoptimization::DeoptReason reason,
2719 ciMethodData* logmd) {
2720 if (trap_count(reason) >= (uint)PerMethodTrapLimit) {
2721 // Too many traps globally.
2722 // Note that we use cumulative trap_count, not just md->trap_count.
2723 if (log()) {
2724 int mcount = (logmd == NULL)? -1: (int)logmd->trap_count(reason);
2725 log()->elem("observe trap='%s' count='0' mcount='%d' ccount='%d'",
2726 Deoptimization::trap_reason_name(reason),
2727 mcount, trap_count(reason));
2728 }
2729 return true;
2730 } else {
2731 // The coast is clear.
2732 return false;
2733 }
2734 }
2735
2736 //--------------------------too_many_recompiles--------------------------------
2737 // Report if there are too many recompiles at the current method and bci.
2738 // Consults PerBytecodeRecompilationCutoff and PerMethodRecompilationCutoff.
2739 // Is not eager to return true, since this will cause the compiler to use
2740 // Action_none for a trap point, to avoid too many recompilations.
2741 bool Compile::too_many_recompiles(ciMethod* method,
2742 int bci,
2743 Deoptimization::DeoptReason reason) {
2744 ciMethodData* md = method->method_data();
2745 if (md->is_empty()) {
2746 // Assume the trap has not occurred, or that it occurred only
2747 // because of a transient condition during start-up in the interpreter.
2748 return false;
2749 }
2750 // Pick a cutoff point well within PerBytecodeRecompilationCutoff.
2751 uint bc_cutoff = (uint) PerBytecodeRecompilationCutoff / 8;
2752 uint m_cutoff = (uint) PerMethodRecompilationCutoff / 2 + 1; // not zero
2753 Deoptimization::DeoptReason per_bc_reason
2754 = Deoptimization::reason_recorded_per_bytecode_if_any(reason);
2755 if ((per_bc_reason == Deoptimization::Reason_none
2756 || md->has_trap_at(bci, reason) != 0)
2757 // The trap frequency measure we care about is the recompile count:
2758 && md->trap_recompiled_at(bci)
2759 && md->overflow_recompile_count() >= bc_cutoff) {
2760 // Do not emit a trap here if it has already caused recompilations.
2761 // Also, if there are multiple reasons, or if there is no per-BCI record,
2762 // assume the worst.
2763 if (log())
2764 log()->elem("observe trap='%s recompiled' count='%d' recompiles2='%d'",
2765 Deoptimization::trap_reason_name(reason),
2766 md->trap_count(reason),
2767 md->overflow_recompile_count());
2768 return true;
2769 } else if (trap_count(reason) != 0
2770 && decompile_count() >= m_cutoff) {
2771 // Too many recompiles globally, and we have seen this sort of trap.
2772 // Use cumulative decompile_count, not just md->decompile_count.
2773 if (log())
2774 log()->elem("observe trap='%s' count='%d' mcount='%d' decompiles='%d' mdecompiles='%d'",
2775 Deoptimization::trap_reason_name(reason),
2776 md->trap_count(reason), trap_count(reason),
2777 md->decompile_count(), decompile_count());
2778 return true;
2779 } else {
2780 // The coast is clear.
2781 return false;
2782 }
2783 }
2784
2785
2786 #ifndef PRODUCT
2787 //------------------------------verify_graph_edges---------------------------
2788 // Walk the Graph and verify that there is a one-to-one correspondence
2789 // between Use-Def edges and Def-Use edges in the graph.
2790 void Compile::verify_graph_edges(bool no_dead_code) {
2791 if (VerifyGraphEdges) {
2792 ResourceArea *area = Thread::current()->resource_area();
2793 Unique_Node_List visited(area);
2794 // Call recursive graph walk to check edges
2795 _root->verify_edges(visited);
2796 if (no_dead_code) {
2797 // Now make sure that no visited node is used by an unvisited node.
2798 bool dead_nodes = 0;
2799 Unique_Node_List checked(area);
2800 while (visited.size() > 0) {
2801 Node* n = visited.pop();
2802 checked.push(n);
2803 for (uint i = 0; i < n->outcnt(); i++) {
2804 Node* use = n->raw_out(i);
2805 if (checked.member(use)) continue; // already checked
2806 if (visited.member(use)) continue; // already in the graph
2807 if (use->is_Con()) continue; // a dead ConNode is OK
2808 // At this point, we have found a dead node which is DU-reachable.
2809 if (dead_nodes++ == 0)
2810 tty->print_cr("*** Dead nodes reachable via DU edges:");
2811 use->dump(2);
2812 tty->print_cr("---");
2813 checked.push(use); // No repeats; pretend it is now checked.
2814 }
2815 }
2816 assert(dead_nodes == 0, "using nodes must be reachable from root");
2817 }
2818 }
2819 }
2820 #endif
2821
2822 // The Compile object keeps track of failure reasons separately from the ciEnv.
2823 // This is required because there is not quite a 1-1 relation between the
2824 // ciEnv and its compilation task and the Compile object. Note that one
2825 // ciEnv might use two Compile objects, if C2Compiler::compile_method decides
2826 // to backtrack and retry without subsuming loads. Other than this backtracking
2827 // behavior, the Compile's failure reason is quietly copied up to the ciEnv
2828 // by the logic in C2Compiler.
2829 void Compile::record_failure(const char* reason) {
2830 if (log() != NULL) {
2831 log()->elem("failure reason='%s' phase='compile'", reason);
2832 }
2833 if (_failure_reason == NULL) {
2834 // Record the first failure reason.
2835 _failure_reason = reason;
2836 }
2837 if (!C->failure_reason_is(C2Compiler::retry_no_subsuming_loads())) {
2838 C->print_method(_failure_reason);
2839 }
2840 _root = NULL; // flush the graph, too
2841 }
2842
2843 Compile::TracePhase::TracePhase(const char* name, elapsedTimer* accumulator, bool dolog)
2844 : TraceTime(NULL, accumulator, false NOT_PRODUCT( || TimeCompiler ), false)
2845 {
2846 if (dolog) {
2847 C = Compile::current();
2848 _log = C->log();
2849 } else {
2850 C = NULL;
2851 _log = NULL;
2852 }
2853 if (_log != NULL) {
2854 _log->begin_head("phase name='%s' nodes='%d'", name, C->unique());
2855 _log->stamp();
2856 _log->end_head();
2857 }
2858 }
2859
2860 Compile::TracePhase::~TracePhase() {
2861 if (_log != NULL) {
2862 _log->done("phase nodes='%d'", C->unique());
2863 }
2864 }