Mercurial > hg > graal-jvmci-8
annotate src/share/vm/opto/compile.cpp @ 2192:b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
Reviewed-by: dholmes, bdelsart
author | bobv |
---|---|
date | Wed, 02 Feb 2011 11:35:26 -0500 |
parents | 51bd2d261853 |
children | c7f3d0b4570f |
rev | line source |
---|---|
0 | 1 /* |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1397
diff
changeset
|
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved. |
0 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1397
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1397
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1397
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 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 | |
2192
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
2091
diff
changeset
|
77 #ifdef TARGET_ARCH_MODEL_arm |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
2091
diff
changeset
|
78 # include "adfiles/ad_arm.hpp" |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
2091
diff
changeset
|
79 #endif |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
2091
diff
changeset
|
80 #ifdef TARGET_ARCH_MODEL_ppc |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
2091
diff
changeset
|
81 # include "adfiles/ad_ppc.hpp" |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
2091
diff
changeset
|
82 #endif |
0 | 83 |
2008 | 84 |
85 // -------------------- Compile::mach_constant_base_node ----------------------- | |
86 // Constant table base node singleton. | |
87 MachConstantBaseNode* Compile::mach_constant_base_node() { | |
88 if (_mach_constant_base_node == NULL) { | |
89 _mach_constant_base_node = new (C) MachConstantBaseNode(); | |
90 _mach_constant_base_node->add_req(C->root()); | |
91 } | |
92 return _mach_constant_base_node; | |
93 } | |
94 | |
95 | |
0 | 96 /// Support for intrinsics. |
97 | |
98 // Return the index at which m must be inserted (or already exists). | |
99 // The sort order is by the address of the ciMethod, with is_virtual as minor key. | |
100 int Compile::intrinsic_insertion_index(ciMethod* m, bool is_virtual) { | |
101 #ifdef ASSERT | |
102 for (int i = 1; i < _intrinsics->length(); i++) { | |
103 CallGenerator* cg1 = _intrinsics->at(i-1); | |
104 CallGenerator* cg2 = _intrinsics->at(i); | |
105 assert(cg1->method() != cg2->method() | |
106 ? cg1->method() < cg2->method() | |
107 : cg1->is_virtual() < cg2->is_virtual(), | |
108 "compiler intrinsics list must stay sorted"); | |
109 } | |
110 #endif | |
111 // Binary search sorted list, in decreasing intervals [lo, hi]. | |
112 int lo = 0, hi = _intrinsics->length()-1; | |
113 while (lo <= hi) { | |
114 int mid = (uint)(hi + lo) / 2; | |
115 ciMethod* mid_m = _intrinsics->at(mid)->method(); | |
116 if (m < mid_m) { | |
117 hi = mid-1; | |
118 } else if (m > mid_m) { | |
119 lo = mid+1; | |
120 } else { | |
121 // look at minor sort key | |
122 bool mid_virt = _intrinsics->at(mid)->is_virtual(); | |
123 if (is_virtual < mid_virt) { | |
124 hi = mid-1; | |
125 } else if (is_virtual > mid_virt) { | |
126 lo = mid+1; | |
127 } else { | |
128 return mid; // exact match | |
129 } | |
130 } | |
131 } | |
132 return lo; // inexact match | |
133 } | |
134 | |
135 void Compile::register_intrinsic(CallGenerator* cg) { | |
136 if (_intrinsics == NULL) { | |
137 _intrinsics = new GrowableArray<CallGenerator*>(60); | |
138 } | |
139 // This code is stolen from ciObjectFactory::insert. | |
140 // Really, GrowableArray should have methods for | |
141 // insert_at, remove_at, and binary_search. | |
142 int len = _intrinsics->length(); | |
143 int index = intrinsic_insertion_index(cg->method(), cg->is_virtual()); | |
144 if (index == len) { | |
145 _intrinsics->append(cg); | |
146 } else { | |
147 #ifdef ASSERT | |
148 CallGenerator* oldcg = _intrinsics->at(index); | |
149 assert(oldcg->method() != cg->method() || oldcg->is_virtual() != cg->is_virtual(), "don't register twice"); | |
150 #endif | |
151 _intrinsics->append(_intrinsics->at(len-1)); | |
152 int pos; | |
153 for (pos = len-2; pos >= index; pos--) { | |
154 _intrinsics->at_put(pos+1,_intrinsics->at(pos)); | |
155 } | |
156 _intrinsics->at_put(index, cg); | |
157 } | |
158 assert(find_intrinsic(cg->method(), cg->is_virtual()) == cg, "registration worked"); | |
159 } | |
160 | |
161 CallGenerator* Compile::find_intrinsic(ciMethod* m, bool is_virtual) { | |
162 assert(m->is_loaded(), "don't try this on unloaded methods"); | |
163 if (_intrinsics != NULL) { | |
164 int index = intrinsic_insertion_index(m, is_virtual); | |
165 if (index < _intrinsics->length() | |
166 && _intrinsics->at(index)->method() == m | |
167 && _intrinsics->at(index)->is_virtual() == is_virtual) { | |
168 return _intrinsics->at(index); | |
169 } | |
170 } | |
171 // Lazily create intrinsics for intrinsic IDs well-known in the runtime. | |
856
75596850f863
6862576: vmIntrinsics needs cleanup in order to support JSR 292 intrinsics
jrose
parents:
647
diff
changeset
|
172 if (m->intrinsic_id() != vmIntrinsics::_none && |
75596850f863
6862576: vmIntrinsics needs cleanup in order to support JSR 292 intrinsics
jrose
parents:
647
diff
changeset
|
173 m->intrinsic_id() <= vmIntrinsics::LAST_COMPILER_INLINE) { |
0 | 174 CallGenerator* cg = make_vm_intrinsic(m, is_virtual); |
175 if (cg != NULL) { | |
176 // Save it for next time: | |
177 register_intrinsic(cg); | |
178 return cg; | |
179 } else { | |
180 gather_intrinsic_statistics(m->intrinsic_id(), is_virtual, _intrinsic_disabled); | |
181 } | |
182 } | |
183 return NULL; | |
184 } | |
185 | |
186 // Compile:: register_library_intrinsics and make_vm_intrinsic are defined | |
187 // in library_call.cpp. | |
188 | |
189 | |
190 #ifndef PRODUCT | |
191 // statistics gathering... | |
192 | |
193 juint Compile::_intrinsic_hist_count[vmIntrinsics::ID_LIMIT] = {0}; | |
194 jubyte Compile::_intrinsic_hist_flags[vmIntrinsics::ID_LIMIT] = {0}; | |
195 | |
196 bool Compile::gather_intrinsic_statistics(vmIntrinsics::ID id, bool is_virtual, int flags) { | |
197 assert(id > vmIntrinsics::_none && id < vmIntrinsics::ID_LIMIT, "oob"); | |
198 int oflags = _intrinsic_hist_flags[id]; | |
199 assert(flags != 0, "what happened?"); | |
200 if (is_virtual) { | |
201 flags |= _intrinsic_virtual; | |
202 } | |
203 bool changed = (flags != oflags); | |
204 if ((flags & _intrinsic_worked) != 0) { | |
205 juint count = (_intrinsic_hist_count[id] += 1); | |
206 if (count == 1) { | |
207 changed = true; // first time | |
208 } | |
209 // increment the overall count also: | |
210 _intrinsic_hist_count[vmIntrinsics::_none] += 1; | |
211 } | |
212 if (changed) { | |
213 if (((oflags ^ flags) & _intrinsic_virtual) != 0) { | |
214 // Something changed about the intrinsic's virtuality. | |
215 if ((flags & _intrinsic_virtual) != 0) { | |
216 // This is the first use of this intrinsic as a virtual call. | |
217 if (oflags != 0) { | |
218 // We already saw it as a non-virtual, so note both cases. | |
219 flags |= _intrinsic_both; | |
220 } | |
221 } else if ((oflags & _intrinsic_both) == 0) { | |
222 // This is the first use of this intrinsic as a non-virtual | |
223 flags |= _intrinsic_both; | |
224 } | |
225 } | |
226 _intrinsic_hist_flags[id] = (jubyte) (oflags | flags); | |
227 } | |
228 // update the overall flags also: | |
229 _intrinsic_hist_flags[vmIntrinsics::_none] |= (jubyte) flags; | |
230 return changed; | |
231 } | |
232 | |
233 static char* format_flags(int flags, char* buf) { | |
234 buf[0] = 0; | |
235 if ((flags & Compile::_intrinsic_worked) != 0) strcat(buf, ",worked"); | |
236 if ((flags & Compile::_intrinsic_failed) != 0) strcat(buf, ",failed"); | |
237 if ((flags & Compile::_intrinsic_disabled) != 0) strcat(buf, ",disabled"); | |
238 if ((flags & Compile::_intrinsic_virtual) != 0) strcat(buf, ",virtual"); | |
239 if ((flags & Compile::_intrinsic_both) != 0) strcat(buf, ",nonvirtual"); | |
240 if (buf[0] == 0) strcat(buf, ","); | |
241 assert(buf[0] == ',', "must be"); | |
242 return &buf[1]; | |
243 } | |
244 | |
245 void Compile::print_intrinsic_statistics() { | |
246 char flagsbuf[100]; | |
247 ttyLocker ttyl; | |
248 if (xtty != NULL) xtty->head("statistics type='intrinsic'"); | |
249 tty->print_cr("Compiler intrinsic usage:"); | |
250 juint total = _intrinsic_hist_count[vmIntrinsics::_none]; | |
251 if (total == 0) total = 1; // avoid div0 in case of no successes | |
252 #define PRINT_STAT_LINE(name, c, f) \ | |
253 tty->print_cr(" %4d (%4.1f%%) %s (%s)", (int)(c), ((c) * 100.0) / total, name, f); | |
254 for (int index = 1 + (int)vmIntrinsics::_none; index < (int)vmIntrinsics::ID_LIMIT; index++) { | |
255 vmIntrinsics::ID id = (vmIntrinsics::ID) index; | |
256 int flags = _intrinsic_hist_flags[id]; | |
257 juint count = _intrinsic_hist_count[id]; | |
258 if ((flags | count) != 0) { | |
259 PRINT_STAT_LINE(vmIntrinsics::name_at(id), count, format_flags(flags, flagsbuf)); | |
260 } | |
261 } | |
262 PRINT_STAT_LINE("total", total, format_flags(_intrinsic_hist_flags[vmIntrinsics::_none], flagsbuf)); | |
263 if (xtty != NULL) xtty->tail("statistics"); | |
264 } | |
265 | |
266 void Compile::print_statistics() { | |
267 { ttyLocker ttyl; | |
268 if (xtty != NULL) xtty->head("statistics type='opto'"); | |
269 Parse::print_statistics(); | |
270 PhaseCCP::print_statistics(); | |
271 PhaseRegAlloc::print_statistics(); | |
272 Scheduling::print_statistics(); | |
273 PhasePeephole::print_statistics(); | |
274 PhaseIdealLoop::print_statistics(); | |
275 if (xtty != NULL) xtty->tail("statistics"); | |
276 } | |
277 if (_intrinsic_hist_flags[vmIntrinsics::_none] != 0) { | |
278 // put this under its own <statistics> element. | |
279 print_intrinsic_statistics(); | |
280 } | |
281 } | |
282 #endif //PRODUCT | |
283 | |
284 // Support for bundling info | |
285 Bundle* Compile::node_bundling(const Node *n) { | |
286 assert(valid_bundle_info(n), "oob"); | |
287 return &_node_bundling_base[n->_idx]; | |
288 } | |
289 | |
290 bool Compile::valid_bundle_info(const Node *n) { | |
291 return (_node_bundling_limit > n->_idx); | |
292 } | |
293 | |
294 | |
1080
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
295 void Compile::gvn_replace_by(Node* n, Node* nn) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
296 for (DUIterator_Last imin, i = n->last_outs(imin); i >= imin; ) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
297 Node* use = n->last_out(i); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
298 bool is_in_table = initial_gvn()->hash_delete(use); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
299 uint uses_found = 0; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
300 for (uint j = 0; j < use->len(); j++) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
301 if (use->in(j) == n) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
302 if (j < use->req()) |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
303 use->set_req(j, nn); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
304 else |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
305 use->set_prec(j, nn); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
306 uses_found++; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
307 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
308 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
309 if (is_in_table) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
310 // reinsert into table |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
311 initial_gvn()->hash_find_insert(use); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
312 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
313 record_for_igvn(use); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
314 i -= uses_found; // we deleted 1 or more copies of this edge |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
315 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
316 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
317 |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
318 |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
319 |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
320 |
0 | 321 // Identify all nodes that are reachable from below, useful. |
322 // Use breadth-first pass that records state in a Unique_Node_List, | |
323 // recursive traversal is slower. | |
324 void Compile::identify_useful_nodes(Unique_Node_List &useful) { | |
325 int estimated_worklist_size = unique(); | |
326 useful.map( estimated_worklist_size, NULL ); // preallocate space | |
327 | |
328 // Initialize worklist | |
329 if (root() != NULL) { useful.push(root()); } | |
330 // If 'top' is cached, declare it useful to preserve cached node | |
331 if( cached_top_node() ) { useful.push(cached_top_node()); } | |
332 | |
333 // Push all useful nodes onto the list, breadthfirst | |
334 for( uint next = 0; next < useful.size(); ++next ) { | |
335 assert( next < unique(), "Unique useful nodes < total nodes"); | |
336 Node *n = useful.at(next); | |
337 uint max = n->len(); | |
338 for( uint i = 0; i < max; ++i ) { | |
339 Node *m = n->in(i); | |
340 if( m == NULL ) continue; | |
341 useful.push(m); | |
342 } | |
343 } | |
344 } | |
345 | |
346 // Disconnect all useless nodes by disconnecting those at the boundary. | |
347 void Compile::remove_useless_nodes(Unique_Node_List &useful) { | |
348 uint next = 0; | |
349 while( next < useful.size() ) { | |
350 Node *n = useful.at(next++); | |
351 // Use raw traversal of out edges since this code removes out edges | |
352 int max = n->outcnt(); | |
353 for (int j = 0; j < max; ++j ) { | |
354 Node* child = n->raw_out(j); | |
355 if( ! useful.member(child) ) { | |
356 assert( !child->is_top() || child != top(), | |
357 "If top is cached in Compile object it is in useful list"); | |
358 // Only need to remove this out-edge to the useless node | |
359 n->raw_del_out(j); | |
360 --j; | |
361 --max; | |
362 } | |
363 } | |
364 if (n->outcnt() == 1 && n->has_special_unique_user()) { | |
365 record_for_igvn( n->unique_out() ); | |
366 } | |
367 } | |
368 debug_only(verify_graph_edges(true/*check for no_dead_code*/);) | |
369 } | |
370 | |
371 //------------------------------frame_size_in_words----------------------------- | |
372 // frame_slots in units of words | |
373 int Compile::frame_size_in_words() const { | |
374 // shift is 0 in LP32 and 1 in LP64 | |
375 const int shift = (LogBytesPerWord - LogBytesPerInt); | |
376 int words = _frame_slots >> shift; | |
377 assert( words << shift == _frame_slots, "frame size must be properly aligned in LP64" ); | |
378 return words; | |
379 } | |
380 | |
381 // ============================================================================ | |
382 //------------------------------CompileWrapper--------------------------------- | |
383 class CompileWrapper : public StackObj { | |
384 Compile *const _compile; | |
385 public: | |
386 CompileWrapper(Compile* compile); | |
387 | |
388 ~CompileWrapper(); | |
389 }; | |
390 | |
391 CompileWrapper::CompileWrapper(Compile* compile) : _compile(compile) { | |
392 // the Compile* pointer is stored in the current ciEnv: | |
393 ciEnv* env = compile->env(); | |
394 assert(env == ciEnv::current(), "must already be a ciEnv active"); | |
395 assert(env->compiler_data() == NULL, "compile already active?"); | |
396 env->set_compiler_data(compile); | |
397 assert(compile == Compile::current(), "sanity"); | |
398 | |
399 compile->set_type_dict(NULL); | |
400 compile->set_type_hwm(NULL); | |
401 compile->set_type_last_size(0); | |
402 compile->set_last_tf(NULL, NULL); | |
403 compile->set_indexSet_arena(NULL); | |
404 compile->set_indexSet_free_block_list(NULL); | |
405 compile->init_type_arena(); | |
406 Type::Initialize(compile); | |
407 _compile->set_scratch_buffer_blob(NULL); | |
408 _compile->begin_method(); | |
409 } | |
410 CompileWrapper::~CompileWrapper() { | |
411 _compile->end_method(); | |
412 if (_compile->scratch_buffer_blob() != NULL) | |
413 BufferBlob::free(_compile->scratch_buffer_blob()); | |
414 _compile->env()->set_compiler_data(NULL); | |
415 } | |
416 | |
417 | |
418 //----------------------------print_compile_messages--------------------------- | |
419 void Compile::print_compile_messages() { | |
420 #ifndef PRODUCT | |
421 // Check if recompiling | |
422 if (_subsume_loads == false && PrintOpto) { | |
423 // Recompiling without allowing machine instructions to subsume loads | |
424 tty->print_cr("*********************************************************"); | |
425 tty->print_cr("** Bailout: Recompile without subsuming loads **"); | |
426 tty->print_cr("*********************************************************"); | |
427 } | |
38
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
428 if (_do_escape_analysis != DoEscapeAnalysis && PrintOpto) { |
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
429 // Recompiling without escape analysis |
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
430 tty->print_cr("*********************************************************"); |
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
431 tty->print_cr("** Bailout: Recompile without escape analysis **"); |
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
432 tty->print_cr("*********************************************************"); |
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
433 } |
0 | 434 if (env()->break_at_compile()) { |
605 | 435 // Open the debugger when compiling this method. |
0 | 436 tty->print("### Breaking when compiling: "); |
437 method()->print_short_name(); | |
438 tty->cr(); | |
439 BREAKPOINT; | |
440 } | |
441 | |
442 if( PrintOpto ) { | |
443 if (is_osr_compilation()) { | |
444 tty->print("[OSR]%3d", _compile_id); | |
445 } else { | |
446 tty->print("%3d", _compile_id); | |
447 } | |
448 } | |
449 #endif | |
450 } | |
451 | |
452 | |
2091
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
453 //-----------------------init_scratch_buffer_blob------------------------------ |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
454 // Construct a temporary BufferBlob and cache it for this compile. |
2008 | 455 void Compile::init_scratch_buffer_blob(int const_size) { |
2091
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
456 // If there is already a scratch buffer blob allocated and the |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
457 // constant section is big enough, use it. Otherwise free the |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
458 // current and allocate a new one. |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
459 BufferBlob* blob = scratch_buffer_blob(); |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
460 if ((blob != NULL) && (const_size <= _scratch_const_size)) { |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
461 // Use the current blob. |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
462 } else { |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
463 if (blob != NULL) { |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
464 BufferBlob::free(blob); |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
465 } |
0 | 466 |
2091
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
467 ResourceMark rm; |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
468 _scratch_const_size = const_size; |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
469 int size = (MAX_inst_size + MAX_stubs_size + _scratch_const_size); |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
470 blob = BufferBlob::create("Compile::scratch_buffer", size); |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
471 // Record the buffer blob for next time. |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
472 set_scratch_buffer_blob(blob); |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
473 // Have we run out of code space? |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
474 if (scratch_buffer_blob() == NULL) { |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
475 // Let CompilerBroker disable further compilations. |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
476 record_failure("Not enough space for scratch buffer in CodeCache"); |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
477 return; |
51bd2d261853
7008325: CodeCache exhausted on sparc starting from hs20b04
kvn
parents:
2008
diff
changeset
|
478 } |
163 | 479 } |
0 | 480 |
481 // Initialize the relocation buffers | |
1748 | 482 relocInfo* locs_buf = (relocInfo*) blob->content_end() - MAX_locs_size; |
0 | 483 set_scratch_locs_memory(locs_buf); |
484 } | |
485 | |
486 | |
487 //-----------------------scratch_emit_size------------------------------------- | |
488 // Helper function that computes size by emitting code | |
489 uint Compile::scratch_emit_size(const Node* n) { | |
2008 | 490 // Start scratch_emit_size section. |
491 set_in_scratch_emit_size(true); | |
492 | |
0 | 493 // Emit into a trash buffer and count bytes emitted. |
494 // This is a pretty expensive way to compute a size, | |
495 // but it works well enough if seldom used. | |
496 // All common fixed-size instructions are given a size | |
497 // method by the AD file. | |
498 // Note that the scratch buffer blob and locs memory are | |
499 // allocated at the beginning of the compile task, and | |
500 // may be shared by several calls to scratch_emit_size. | |
501 // The allocation of the scratch buffer blob is particularly | |
502 // expensive, since it has to grab the code cache lock. | |
503 BufferBlob* blob = this->scratch_buffer_blob(); | |
504 assert(blob != NULL, "Initialize BufferBlob at start"); | |
505 assert(blob->size() > MAX_inst_size, "sanity"); | |
506 relocInfo* locs_buf = scratch_locs_memory(); | |
1748 | 507 address blob_begin = blob->content_begin(); |
0 | 508 address blob_end = (address)locs_buf; |
1748 | 509 assert(blob->content_contains(blob_end), "sanity"); |
0 | 510 CodeBuffer buf(blob_begin, blob_end - blob_begin); |
2008 | 511 buf.initialize_consts_size(_scratch_const_size); |
0 | 512 buf.initialize_stubs_size(MAX_stubs_size); |
513 assert(locs_buf != NULL, "sanity"); | |
2008 | 514 int lsize = MAX_locs_size / 3; |
515 buf.consts()->initialize_shared_locs(&locs_buf[lsize * 0], lsize); | |
516 buf.insts()->initialize_shared_locs( &locs_buf[lsize * 1], lsize); | |
517 buf.stubs()->initialize_shared_locs( &locs_buf[lsize * 2], lsize); | |
518 | |
519 // Do the emission. | |
0 | 520 n->emit(buf, this->regalloc()); |
2008 | 521 |
522 // End scratch_emit_size section. | |
523 set_in_scratch_emit_size(false); | |
524 | |
1748 | 525 return buf.insts_size(); |
0 | 526 } |
527 | |
528 | |
529 // ============================================================================ | |
530 //------------------------------Compile standard------------------------------- | |
531 debug_only( int Compile::_debug_idx = 100000; ) | |
532 | |
533 // Compile a method. entry_bci is -1 for normal compilations and indicates | |
534 // the continuation bci for on stack replacement. | |
535 | |
536 | |
38
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
537 Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr_bci, bool subsume_loads, bool do_escape_analysis ) |
0 | 538 : Phase(Compiler), |
539 _env(ci_env), | |
540 _log(ci_env->log()), | |
541 _compile_id(ci_env->compile_id()), | |
542 _save_argument_registers(false), | |
543 _stub_name(NULL), | |
544 _stub_function(NULL), | |
545 _stub_entry_point(NULL), | |
546 _method(target), | |
547 _entry_bci(osr_bci), | |
548 _initial_gvn(NULL), | |
549 _for_igvn(NULL), | |
550 _warm_calls(NULL), | |
551 _subsume_loads(subsume_loads), | |
38
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
552 _do_escape_analysis(do_escape_analysis), |
0 | 553 _failure_reason(NULL), |
554 _code_buffer("Compile::Fill_buffer"), | |
555 _orig_pc_slot(0), | |
556 _orig_pc_slot_offset_in_bytes(0), | |
1265 | 557 _has_method_handle_invokes(false), |
2008 | 558 _mach_constant_base_node(NULL), |
0 | 559 _node_bundling_limit(0), |
560 _node_bundling_base(NULL), | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
561 _java_calls(0), |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
562 _inner_loops(0), |
2008 | 563 _scratch_const_size(-1), |
564 _in_scratch_emit_size(false), | |
0 | 565 #ifndef PRODUCT |
566 _trace_opto_output(TraceOptoOutput || method()->has_option("TraceOptoOutput")), | |
567 _printer(IdealGraphPrinter::printer()), | |
568 #endif | |
569 _congraph(NULL) { | |
570 C = this; | |
571 | |
572 CompileWrapper cw(this); | |
573 #ifndef PRODUCT | |
574 if (TimeCompiler2) { | |
575 tty->print(" "); | |
576 target->holder()->name()->print(); | |
577 tty->print("."); | |
578 target->print_short_name(); | |
579 tty->print(" "); | |
580 } | |
581 TraceTime t1("Total compilation time", &_t_totalCompilation, TimeCompiler, TimeCompiler2); | |
582 TraceTime t2(NULL, &_t_methodCompilation, TimeCompiler, false); | |
100
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
583 bool print_opto_assembly = PrintOptoAssembly || _method->has_option("PrintOptoAssembly"); |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
584 if (!print_opto_assembly) { |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
585 bool print_assembly = (PrintAssembly || _method->should_print_assembly()); |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
586 if (print_assembly && !Disassembler::can_decode()) { |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
587 tty->print_cr("PrintAssembly request changed to PrintOptoAssembly"); |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
588 print_opto_assembly = true; |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
589 } |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
590 } |
c7c777385a15
6667042: PrintAssembly option does not work without special plugin
jrose
parents:
65
diff
changeset
|
591 set_print_assembly(print_opto_assembly); |
367
194b8e3a2fc4
6384206: Phis which are later unneeded are impairing our ability to inline based on static types
never
parents:
333
diff
changeset
|
592 set_parsed_irreducible_loop(false); |
0 | 593 #endif |
594 | |
595 if (ProfileTraps) { | |
596 // Make sure the method being compiled gets its own MDO, | |
597 // so we can at least track the decompile_count(). | |
2007
5ddfcf4b079e
7003554: (tiered) assert(is_null_object() || handle() != NULL) failed: cannot embed null pointer
iveresov
parents:
1972
diff
changeset
|
598 method()->ensure_method_data(); |
0 | 599 } |
600 | |
601 Init(::AliasLevel); | |
602 | |
603 | |
604 print_compile_messages(); | |
605 | |
606 if (UseOldInlining || PrintCompilation NOT_PRODUCT( || PrintOpto) ) | |
607 _ilt = InlineTree::build_inline_tree_root(); | |
608 else | |
609 _ilt = NULL; | |
610 | |
611 // Even if NO memory addresses are used, MergeMem nodes must have at least 1 slice | |
612 assert(num_alias_types() >= AliasIdxRaw, ""); | |
613 | |
614 #define MINIMUM_NODE_HASH 1023 | |
615 // Node list that Iterative GVN will start with | |
616 Unique_Node_List for_igvn(comp_arena()); | |
617 set_for_igvn(&for_igvn); | |
618 | |
619 // GVN that will be run immediately on new nodes | |
620 uint estimated_size = method()->code_size()*4+64; | |
621 estimated_size = (estimated_size < MINIMUM_NODE_HASH ? MINIMUM_NODE_HASH : estimated_size); | |
622 PhaseGVN gvn(node_arena(), estimated_size); | |
623 set_initial_gvn(&gvn); | |
624 | |
625 { // Scope for timing the parser | |
626 TracePhase t3("parse", &_t_parser, true); | |
627 | |
628 // Put top into the hash table ASAP. | |
629 initial_gvn()->transform_no_reclaim(top()); | |
630 | |
631 // Set up tf(), start(), and find a CallGenerator. | |
632 CallGenerator* cg; | |
633 if (is_osr_compilation()) { | |
634 const TypeTuple *domain = StartOSRNode::osr_domain(); | |
635 const TypeTuple *range = TypeTuple::make_range(method()->signature()); | |
636 init_tf(TypeFunc::make(domain, range)); | |
637 StartNode* s = new (this, 2) StartOSRNode(root(), domain); | |
638 initial_gvn()->set_type_bottom(s); | |
639 init_start(s); | |
640 cg = CallGenerator::for_osr(method(), entry_bci()); | |
641 } else { | |
642 // Normal case. | |
643 init_tf(TypeFunc::make(method())); | |
644 StartNode* s = new (this, 2) StartNode(root(), tf()->domain()); | |
645 initial_gvn()->set_type_bottom(s); | |
646 init_start(s); | |
647 float past_uses = method()->interpreter_invocation_count(); | |
648 float expected_uses = past_uses; | |
649 cg = CallGenerator::for_inline(method(), expected_uses); | |
650 } | |
651 if (failing()) return; | |
652 if (cg == NULL) { | |
653 record_method_not_compilable_all_tiers("cannot parse method"); | |
654 return; | |
655 } | |
656 JVMState* jvms = build_start_state(start(), tf()); | |
657 if ((jvms = cg->generate(jvms)) == NULL) { | |
658 record_method_not_compilable("method parse failed"); | |
659 return; | |
660 } | |
661 GraphKit kit(jvms); | |
662 | |
663 if (!kit.stopped()) { | |
664 // Accept return values, and transfer control we know not where. | |
665 // This is done by a special, unique ReturnNode bound to root. | |
666 return_values(kit.jvms()); | |
667 } | |
668 | |
669 if (kit.has_exceptions()) { | |
670 // Any exceptions that escape from this call must be rethrown | |
671 // to whatever caller is dynamically above us on the stack. | |
672 // This is done by a special, unique RethrowNode bound to root. | |
673 rethrow_exceptions(kit.transfer_exceptions_into_jvms()); | |
674 } | |
675 | |
1080
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
676 if (!failing() && has_stringbuilder()) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
677 { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
678 // remove useless nodes to make the usage analysis simpler |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
679 ResourceMark rm; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
680 PhaseRemoveUseless pru(initial_gvn(), &for_igvn); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
681 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
682 |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
683 { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
684 ResourceMark rm; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
685 print_method("Before StringOpts", 3); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
686 PhaseStringOpts pso(initial_gvn(), &for_igvn); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
687 print_method("After StringOpts", 3); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
688 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
689 |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
690 // now inline anything that we skipped the first time around |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
691 while (_late_inlines.length() > 0) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
692 CallGenerator* cg = _late_inlines.pop(); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
693 cg->do_late_inline(); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
694 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
695 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
696 assert(_late_inlines.length() == 0, "should have been processed"); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
697 |
417 | 698 print_method("Before RemoveUseless", 3); |
367
194b8e3a2fc4
6384206: Phis which are later unneeded are impairing our ability to inline based on static types
never
parents:
333
diff
changeset
|
699 |
0 | 700 // Remove clutter produced by parsing. |
701 if (!failing()) { | |
702 ResourceMark rm; | |
703 PhaseRemoveUseless pru(initial_gvn(), &for_igvn); | |
704 } | |
705 } | |
706 | |
707 // Note: Large methods are capped off in do_one_bytecode(). | |
708 if (failing()) return; | |
709 | |
710 // After parsing, node notes are no longer automagic. | |
711 // They must be propagated by register_new_node_with_optimizer(), | |
712 // clone(), or the like. | |
713 set_default_node_notes(NULL); | |
714 | |
715 for (;;) { | |
716 int successes = Inline_Warm(); | |
717 if (failing()) return; | |
718 if (successes == 0) break; | |
719 } | |
720 | |
721 // Drain the list. | |
722 Finish_Warm(); | |
723 #ifndef PRODUCT | |
724 if (_printer) { | |
725 _printer->print_inlining(this); | |
726 } | |
727 #endif | |
728 | |
729 if (failing()) return; | |
730 NOT_PRODUCT( verify_graph_edges(); ) | |
731 | |
732 // Now optimize | |
733 Optimize(); | |
734 if (failing()) return; | |
735 NOT_PRODUCT( verify_graph_edges(); ) | |
736 | |
737 #ifndef PRODUCT | |
738 if (PrintIdeal) { | |
739 ttyLocker ttyl; // keep the following output all in one block | |
740 // This output goes directly to the tty, not the compiler log. | |
741 // To enable tools to match it up with the compilation activity, | |
742 // be sure to tag this tty output with the compile ID. | |
743 if (xtty != NULL) { | |
744 xtty->head("ideal compile_id='%d'%s", compile_id(), | |
745 is_osr_compilation() ? " compile_kind='osr'" : | |
746 ""); | |
747 } | |
748 root()->dump(9999); | |
749 if (xtty != NULL) { | |
750 xtty->tail("ideal"); | |
751 } | |
752 } | |
753 #endif | |
754 | |
755 // Now that we know the size of all the monitors we can add a fixed slot | |
756 // for the original deopt pc. | |
757 | |
758 _orig_pc_slot = fixed_slots(); | |
759 int next_slot = _orig_pc_slot + (sizeof(address) / VMRegImpl::stack_slot_size); | |
760 set_fixed_slots(next_slot); | |
761 | |
762 // Now generate code | |
763 Code_Gen(); | |
764 if (failing()) return; | |
765 | |
766 // Check if we want to skip execution of all compiled code. | |
767 { | |
768 #ifndef PRODUCT | |
769 if (OptoNoExecute) { | |
770 record_method_not_compilable("+OptoNoExecute"); // Flag as failed | |
771 return; | |
772 } | |
773 TracePhase t2("install_code", &_t_registerMethod, TimeCompiler); | |
774 #endif | |
775 | |
776 if (is_osr_compilation()) { | |
777 _code_offsets.set_value(CodeOffsets::Verified_Entry, 0); | |
778 _code_offsets.set_value(CodeOffsets::OSR_Entry, _first_block_size); | |
779 } else { | |
780 _code_offsets.set_value(CodeOffsets::Verified_Entry, _first_block_size); | |
781 _code_offsets.set_value(CodeOffsets::OSR_Entry, 0); | |
782 } | |
783 | |
784 env()->register_method(_method, _entry_bci, | |
785 &_code_offsets, | |
786 _orig_pc_slot_offset_in_bytes, | |
787 code_buffer(), | |
788 frame_size_in_words(), _oop_map_set, | |
789 &_handler_table, &_inc_table, | |
790 compiler, | |
791 env()->comp_level(), | |
792 true, /*has_debug_info*/ | |
793 has_unsafe_access() | |
794 ); | |
795 } | |
796 } | |
797 | |
798 //------------------------------Compile---------------------------------------- | |
799 // Compile a runtime stub | |
800 Compile::Compile( ciEnv* ci_env, | |
801 TypeFunc_generator generator, | |
802 address stub_function, | |
803 const char *stub_name, | |
804 int is_fancy_jump, | |
805 bool pass_tls, | |
806 bool save_arg_registers, | |
807 bool return_pc ) | |
808 : Phase(Compiler), | |
809 _env(ci_env), | |
810 _log(ci_env->log()), | |
811 _compile_id(-1), | |
812 _save_argument_registers(save_arg_registers), | |
813 _method(NULL), | |
814 _stub_name(stub_name), | |
815 _stub_function(stub_function), | |
816 _stub_entry_point(NULL), | |
817 _entry_bci(InvocationEntryBci), | |
818 _initial_gvn(NULL), | |
819 _for_igvn(NULL), | |
820 _warm_calls(NULL), | |
821 _orig_pc_slot(0), | |
822 _orig_pc_slot_offset_in_bytes(0), | |
823 _subsume_loads(true), | |
38
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
824 _do_escape_analysis(false), |
0 | 825 _failure_reason(NULL), |
826 _code_buffer("Compile::Fill_buffer"), | |
1265 | 827 _has_method_handle_invokes(false), |
2008 | 828 _mach_constant_base_node(NULL), |
0 | 829 _node_bundling_limit(0), |
830 _node_bundling_base(NULL), | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
831 _java_calls(0), |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
832 _inner_loops(0), |
0 | 833 #ifndef PRODUCT |
834 _trace_opto_output(TraceOptoOutput), | |
835 _printer(NULL), | |
836 #endif | |
837 _congraph(NULL) { | |
838 C = this; | |
839 | |
840 #ifndef PRODUCT | |
841 TraceTime t1(NULL, &_t_totalCompilation, TimeCompiler, false); | |
842 TraceTime t2(NULL, &_t_stubCompilation, TimeCompiler, false); | |
843 set_print_assembly(PrintFrameConverterAssembly); | |
367
194b8e3a2fc4
6384206: Phis which are later unneeded are impairing our ability to inline based on static types
never
parents:
333
diff
changeset
|
844 set_parsed_irreducible_loop(false); |
0 | 845 #endif |
846 CompileWrapper cw(this); | |
847 Init(/*AliasLevel=*/ 0); | |
848 init_tf((*generator)()); | |
849 | |
850 { | |
851 // The following is a dummy for the sake of GraphKit::gen_stub | |
852 Unique_Node_List for_igvn(comp_arena()); | |
853 set_for_igvn(&for_igvn); // not used, but some GraphKit guys push on this | |
854 PhaseGVN gvn(Thread::current()->resource_area(),255); | |
855 set_initial_gvn(&gvn); // not significant, but GraphKit guys use it pervasively | |
856 gvn.transform_no_reclaim(top()); | |
857 | |
858 GraphKit kit; | |
859 kit.gen_stub(stub_function, stub_name, is_fancy_jump, pass_tls, return_pc); | |
860 } | |
861 | |
862 NOT_PRODUCT( verify_graph_edges(); ) | |
863 Code_Gen(); | |
864 if (failing()) return; | |
865 | |
866 | |
867 // Entry point will be accessed using compile->stub_entry_point(); | |
868 if (code_buffer() == NULL) { | |
869 Matcher::soft_match_failure(); | |
870 } else { | |
871 if (PrintAssembly && (WizardMode || Verbose)) | |
872 tty->print_cr("### Stub::%s", stub_name); | |
873 | |
874 if (!failing()) { | |
875 assert(_fixed_slots == 0, "no fixed slots used for runtime stubs"); | |
876 | |
877 // Make the NMethod | |
878 // For now we mark the frame as never safe for profile stackwalking | |
879 RuntimeStub *rs = RuntimeStub::new_runtime_stub(stub_name, | |
880 code_buffer(), | |
881 CodeOffsets::frame_never_safe, | |
882 // _code_offsets.value(CodeOffsets::Frame_Complete), | |
883 frame_size_in_words(), | |
884 _oop_map_set, | |
885 save_arg_registers); | |
886 assert(rs != NULL && rs->is_runtime_stub(), "sanity check"); | |
887 | |
888 _stub_entry_point = rs->entry_point(); | |
889 } | |
890 } | |
891 } | |
892 | |
893 #ifndef PRODUCT | |
894 void print_opto_verbose_signature( const TypeFunc *j_sig, const char *stub_name ) { | |
895 if(PrintOpto && Verbose) { | |
896 tty->print("%s ", stub_name); j_sig->print_flattened(); tty->cr(); | |
897 } | |
898 } | |
899 #endif | |
900 | |
901 void Compile::print_codes() { | |
902 } | |
903 | |
904 //------------------------------Init------------------------------------------- | |
905 // Prepare for a single compilation | |
906 void Compile::Init(int aliaslevel) { | |
907 _unique = 0; | |
908 _regalloc = NULL; | |
909 | |
910 _tf = NULL; // filled in later | |
911 _top = NULL; // cached later | |
912 _matcher = NULL; // filled in later | |
913 _cfg = NULL; // filled in later | |
914 | |
915 set_24_bit_selection_and_mode(Use24BitFP, false); | |
916 | |
917 _node_note_array = NULL; | |
918 _default_node_notes = NULL; | |
919 | |
920 _immutable_memory = NULL; // filled in at first inquiry | |
921 | |
922 // Globally visible Nodes | |
923 // First set TOP to NULL to give safe behavior during creation of RootNode | |
924 set_cached_top_node(NULL); | |
925 set_root(new (this, 3) RootNode()); | |
926 // Now that you have a Root to point to, create the real TOP | |
927 set_cached_top_node( new (this, 1) ConNode(Type::TOP) ); | |
928 set_recent_alloc(NULL, NULL); | |
929 | |
930 // Create Debug Information Recorder to record scopes, oopmaps, etc. | |
931 env()->set_oop_recorder(new OopRecorder(comp_arena())); | |
932 env()->set_debug_info(new DebugInformationRecorder(env()->oop_recorder())); | |
933 env()->set_dependencies(new Dependencies(env())); | |
934 | |
935 _fixed_slots = 0; | |
936 set_has_split_ifs(false); | |
937 set_has_loops(has_method() && method()->has_loops()); // first approximation | |
1080
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
938 set_has_stringbuilder(false); |
0 | 939 _trap_can_recompile = false; // no traps emitted yet |
940 _major_progress = true; // start out assuming good things will happen | |
941 set_has_unsafe_access(false); | |
942 Copy::zero_to_bytes(_trap_hist, sizeof(_trap_hist)); | |
943 set_decompile_count(0); | |
944 | |
418 | 945 set_do_freq_based_layout(BlockLayoutByFrequency || method_has_option("BlockLayoutByFrequency")); |
1783 | 946 set_num_loop_opts(LoopOptsCount); |
947 set_do_inlining(Inline); | |
948 set_max_inline_size(MaxInlineSize); | |
949 set_freq_inline_size(FreqInlineSize); | |
950 set_do_scheduling(OptoScheduling); | |
951 set_do_count_invocations(false); | |
952 set_do_method_data_update(false); | |
0 | 953 |
954 if (debug_info()->recording_non_safepoints()) { | |
955 set_node_note_array(new(comp_arena()) GrowableArray<Node_Notes*> | |
956 (comp_arena(), 8, 0, NULL)); | |
957 set_default_node_notes(Node_Notes::make(this)); | |
958 } | |
959 | |
960 // // -- Initialize types before each compile -- | |
961 // // Update cached type information | |
962 // if( _method && _method->constants() ) | |
963 // Type::update_loaded_types(_method, _method->constants()); | |
964 | |
965 // Init alias_type map. | |
38
b789bcaf2dd9
6667610: (Escape Analysis) retry compilation without EA if it fails
kvn
parents:
0
diff
changeset
|
966 if (!_do_escape_analysis && aliaslevel == 3) |
0 | 967 aliaslevel = 2; // No unique types without escape analysis |
968 _AliasLevel = aliaslevel; | |
969 const int grow_ats = 16; | |
970 _max_alias_types = grow_ats; | |
971 _alias_types = NEW_ARENA_ARRAY(comp_arena(), AliasType*, grow_ats); | |
972 AliasType* ats = NEW_ARENA_ARRAY(comp_arena(), AliasType, grow_ats); | |
973 Copy::zero_to_bytes(ats, sizeof(AliasType)*grow_ats); | |
974 { | |
975 for (int i = 0; i < grow_ats; i++) _alias_types[i] = &ats[i]; | |
976 } | |
977 // Initialize the first few types. | |
978 _alias_types[AliasIdxTop]->Init(AliasIdxTop, NULL); | |
979 _alias_types[AliasIdxBot]->Init(AliasIdxBot, TypePtr::BOTTOM); | |
980 _alias_types[AliasIdxRaw]->Init(AliasIdxRaw, TypeRawPtr::BOTTOM); | |
981 _num_alias_types = AliasIdxRaw+1; | |
982 // Zero out the alias type cache. | |
983 Copy::zero_to_bytes(_alias_cache, sizeof(_alias_cache)); | |
984 // A NULL adr_type hits in the cache right away. Preload the right answer. | |
985 probe_alias_cache(NULL)->_index = AliasIdxTop; | |
986 | |
987 _intrinsics = NULL; | |
1685 | 988 _macro_nodes = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL); |
989 _predicate_opaqs = new(comp_arena()) GrowableArray<Node*>(comp_arena(), 8, 0, NULL); | |
0 | 990 register_library_intrinsics(); |
991 } | |
992 | |
993 //---------------------------init_start---------------------------------------- | |
994 // Install the StartNode on this compile object. | |
995 void Compile::init_start(StartNode* s) { | |
996 if (failing()) | |
997 return; // already failing | |
998 assert(s == start(), ""); | |
999 } | |
1000 | |
1001 StartNode* Compile::start() const { | |
1002 assert(!failing(), ""); | |
1003 for (DUIterator_Fast imax, i = root()->fast_outs(imax); i < imax; i++) { | |
1004 Node* start = root()->fast_out(i); | |
1005 if( start->is_Start() ) | |
1006 return start->as_Start(); | |
1007 } | |
1008 ShouldNotReachHere(); | |
1009 return NULL; | |
1010 } | |
1011 | |
1012 //-------------------------------immutable_memory------------------------------------- | |
1013 // Access immutable memory | |
1014 Node* Compile::immutable_memory() { | |
1015 if (_immutable_memory != NULL) { | |
1016 return _immutable_memory; | |
1017 } | |
1018 StartNode* s = start(); | |
1019 for (DUIterator_Fast imax, i = s->fast_outs(imax); true; i++) { | |
1020 Node *p = s->fast_out(i); | |
1021 if (p != s && p->as_Proj()->_con == TypeFunc::Memory) { | |
1022 _immutable_memory = p; | |
1023 return _immutable_memory; | |
1024 } | |
1025 } | |
1026 ShouldNotReachHere(); | |
1027 return NULL; | |
1028 } | |
1029 | |
1030 //----------------------set_cached_top_node------------------------------------ | |
1031 // Install the cached top node, and make sure Node::is_top works correctly. | |
1032 void Compile::set_cached_top_node(Node* tn) { | |
1033 if (tn != NULL) verify_top(tn); | |
1034 Node* old_top = _top; | |
1035 _top = tn; | |
1036 // Calling Node::setup_is_top allows the nodes the chance to adjust | |
1037 // their _out arrays. | |
1038 if (_top != NULL) _top->setup_is_top(); | |
1039 if (old_top != NULL) old_top->setup_is_top(); | |
1040 assert(_top == NULL || top()->is_top(), ""); | |
1041 } | |
1042 | |
1043 #ifndef PRODUCT | |
1044 void Compile::verify_top(Node* tn) const { | |
1045 if (tn != NULL) { | |
1046 assert(tn->is_Con(), "top node must be a constant"); | |
1047 assert(((ConNode*)tn)->type() == Type::TOP, "top node must have correct type"); | |
1048 assert(tn->in(0) != NULL, "must have live top node"); | |
1049 } | |
1050 } | |
1051 #endif | |
1052 | |
1053 | |
1054 ///-------------------Managing Per-Node Debug & Profile Info------------------- | |
1055 | |
1056 void Compile::grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by) { | |
1057 guarantee(arr != NULL, ""); | |
1058 int num_blocks = arr->length(); | |
1059 if (grow_by < num_blocks) grow_by = num_blocks; | |
1060 int num_notes = grow_by * _node_notes_block_size; | |
1061 Node_Notes* notes = NEW_ARENA_ARRAY(node_arena(), Node_Notes, num_notes); | |
1062 Copy::zero_to_bytes(notes, num_notes * sizeof(Node_Notes)); | |
1063 while (num_notes > 0) { | |
1064 arr->append(notes); | |
1065 notes += _node_notes_block_size; | |
1066 num_notes -= _node_notes_block_size; | |
1067 } | |
1068 assert(num_notes == 0, "exact multiple, please"); | |
1069 } | |
1070 | |
1071 bool Compile::copy_node_notes_to(Node* dest, Node* source) { | |
1072 if (source == NULL || dest == NULL) return false; | |
1073 | |
1074 if (dest->is_Con()) | |
1075 return false; // Do not push debug info onto constants. | |
1076 | |
1077 #ifdef ASSERT | |
1078 // Leave a bread crumb trail pointing to the original node: | |
1079 if (dest != NULL && dest != source && dest->debug_orig() == NULL) { | |
1080 dest->set_debug_orig(source); | |
1081 } | |
1082 #endif | |
1083 | |
1084 if (node_note_array() == NULL) | |
1085 return false; // Not collecting any notes now. | |
1086 | |
1087 // This is a copy onto a pre-existing node, which may already have notes. | |
1088 // If both nodes have notes, do not overwrite any pre-existing notes. | |
1089 Node_Notes* source_notes = node_notes_at(source->_idx); | |
1090 if (source_notes == NULL || source_notes->is_clear()) return false; | |
1091 Node_Notes* dest_notes = node_notes_at(dest->_idx); | |
1092 if (dest_notes == NULL || dest_notes->is_clear()) { | |
1093 return set_node_notes_at(dest->_idx, source_notes); | |
1094 } | |
1095 | |
1096 Node_Notes merged_notes = (*source_notes); | |
1097 // The order of operations here ensures that dest notes will win... | |
1098 merged_notes.update_from(dest_notes); | |
1099 return set_node_notes_at(dest->_idx, &merged_notes); | |
1100 } | |
1101 | |
1102 | |
1103 //--------------------------allow_range_check_smearing------------------------- | |
1104 // Gating condition for coalescing similar range checks. | |
1105 // Sometimes we try 'speculatively' replacing a series of a range checks by a | |
1106 // single covering check that is at least as strong as any of them. | |
1107 // If the optimization succeeds, the simplified (strengthened) range check | |
1108 // will always succeed. If it fails, we will deopt, and then give up | |
1109 // on the optimization. | |
1110 bool Compile::allow_range_check_smearing() const { | |
1111 // If this method has already thrown a range-check, | |
1112 // assume it was because we already tried range smearing | |
1113 // and it failed. | |
1114 uint already_trapped = trap_count(Deoptimization::Reason_range_check); | |
1115 return !already_trapped; | |
1116 } | |
1117 | |
1118 | |
1119 //------------------------------flatten_alias_type----------------------------- | |
1120 const TypePtr *Compile::flatten_alias_type( const TypePtr *tj ) const { | |
1121 int offset = tj->offset(); | |
1122 TypePtr::PTR ptr = tj->ptr(); | |
1123 | |
247 | 1124 // Known instance (scalarizable allocation) alias only with itself. |
1125 bool is_known_inst = tj->isa_oopptr() != NULL && | |
1126 tj->is_oopptr()->is_known_instance(); | |
1127 | |
0 | 1128 // Process weird unsafe references. |
1129 if (offset == Type::OffsetBot && (tj->isa_instptr() /*|| tj->isa_klassptr()*/)) { | |
1130 assert(InlineUnsafeOps, "indeterminate pointers come only from unsafe ops"); | |
247 | 1131 assert(!is_known_inst, "scalarizable allocation should not have unsafe references"); |
0 | 1132 tj = TypeOopPtr::BOTTOM; |
1133 ptr = tj->ptr(); | |
1134 offset = tj->offset(); | |
1135 } | |
1136 | |
1137 // Array pointers need some flattening | |
1138 const TypeAryPtr *ta = tj->isa_aryptr(); | |
247 | 1139 if( ta && is_known_inst ) { |
1140 if ( offset != Type::OffsetBot && | |
1141 offset > arrayOopDesc::length_offset_in_bytes() ) { | |
1142 offset = Type::OffsetBot; // Flatten constant access into array body only | |
1143 tj = ta = TypeAryPtr::make(ptr, ta->ary(), ta->klass(), true, offset, ta->instance_id()); | |
1144 } | |
1145 } else if( ta && _AliasLevel >= 2 ) { | |
0 | 1146 // For arrays indexed by constant indices, we flatten the alias |
1147 // space to include all of the array body. Only the header, klass | |
1148 // and array length can be accessed un-aliased. | |
1149 if( offset != Type::OffsetBot ) { | |
1150 if( ta->const_oop() ) { // methodDataOop or methodOop | |
1151 offset = Type::OffsetBot; // Flatten constant access into array body | |
247 | 1152 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),ta->ary(),ta->klass(),false,offset); |
0 | 1153 } else if( offset == arrayOopDesc::length_offset_in_bytes() ) { |
1154 // range is OK as-is. | |
1155 tj = ta = TypeAryPtr::RANGE; | |
1156 } else if( offset == oopDesc::klass_offset_in_bytes() ) { | |
1157 tj = TypeInstPtr::KLASS; // all klass loads look alike | |
1158 ta = TypeAryPtr::RANGE; // generic ignored junk | |
1159 ptr = TypePtr::BotPTR; | |
1160 } else if( offset == oopDesc::mark_offset_in_bytes() ) { | |
1161 tj = TypeInstPtr::MARK; | |
1162 ta = TypeAryPtr::RANGE; // generic ignored junk | |
1163 ptr = TypePtr::BotPTR; | |
1164 } else { // Random constant offset into array body | |
1165 offset = Type::OffsetBot; // Flatten constant access into array body | |
247 | 1166 tj = ta = TypeAryPtr::make(ptr,ta->ary(),ta->klass(),false,offset); |
0 | 1167 } |
1168 } | |
1169 // Arrays of fixed size alias with arrays of unknown size. | |
1170 if (ta->size() != TypeInt::POS) { | |
1171 const TypeAry *tary = TypeAry::make(ta->elem(), TypeInt::POS); | |
247 | 1172 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,ta->klass(),false,offset); |
0 | 1173 } |
1174 // Arrays of known objects become arrays of unknown objects. | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1175 if (ta->elem()->isa_narrowoop() && ta->elem() != TypeNarrowOop::BOTTOM) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1176 const TypeAry *tary = TypeAry::make(TypeNarrowOop::BOTTOM, ta->size()); |
247 | 1177 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,NULL,false,offset); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1178 } |
0 | 1179 if (ta->elem()->isa_oopptr() && ta->elem() != TypeInstPtr::BOTTOM) { |
1180 const TypeAry *tary = TypeAry::make(TypeInstPtr::BOTTOM, ta->size()); | |
247 | 1181 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,NULL,false,offset); |
0 | 1182 } |
1183 // Arrays of bytes and of booleans both use 'bastore' and 'baload' so | |
1184 // cannot be distinguished by bytecode alone. | |
1185 if (ta->elem() == TypeInt::BOOL) { | |
1186 const TypeAry *tary = TypeAry::make(TypeInt::BYTE, ta->size()); | |
1187 ciKlass* aklass = ciTypeArrayKlass::make(T_BYTE); | |
247 | 1188 tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,aklass,false,offset); |
0 | 1189 } |
1190 // During the 2nd round of IterGVN, NotNull castings are removed. | |
1191 // Make sure the Bottom and NotNull variants alias the same. | |
1192 // Also, make sure exact and non-exact variants alias the same. | |
1193 if( ptr == TypePtr::NotNull || ta->klass_is_exact() ) { | |
1194 if (ta->const_oop()) { | |
1195 tj = ta = TypeAryPtr::make(TypePtr::Constant,ta->const_oop(),ta->ary(),ta->klass(),false,offset); | |
1196 } else { | |
1197 tj = ta = TypeAryPtr::make(TypePtr::BotPTR,ta->ary(),ta->klass(),false,offset); | |
1198 } | |
1199 } | |
1200 } | |
1201 | |
1202 // Oop pointers need some flattening | |
1203 const TypeInstPtr *to = tj->isa_instptr(); | |
1204 if( to && _AliasLevel >= 2 && to != TypeOopPtr::BOTTOM ) { | |
1205 if( ptr == TypePtr::Constant ) { | |
1206 // No constant oop pointers (such as Strings); they alias with | |
1207 // unknown strings. | |
247 | 1208 assert(!is_known_inst, "not scalarizable allocation"); |
0 | 1209 tj = to = TypeInstPtr::make(TypePtr::BotPTR,to->klass(),false,0,offset); |
247 | 1210 } else if( is_known_inst ) { |
163 | 1211 tj = to; // Keep NotNull and klass_is_exact for instance type |
0 | 1212 } else if( ptr == TypePtr::NotNull || to->klass_is_exact() ) { |
1213 // During the 2nd round of IterGVN, NotNull castings are removed. | |
1214 // Make sure the Bottom and NotNull variants alias the same. | |
1215 // Also, make sure exact and non-exact variants alias the same. | |
247 | 1216 tj = to = TypeInstPtr::make(TypePtr::BotPTR,to->klass(),false,0,offset); |
0 | 1217 } |
1218 // Canonicalize the holder of this field | |
1219 ciInstanceKlass *k = to->klass()->as_instance_klass(); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1220 if (offset >= 0 && offset < instanceOopDesc::base_offset_in_bytes()) { |
0 | 1221 // First handle header references such as a LoadKlassNode, even if the |
1222 // object's klass is unloaded at compile time (4965979). | |
247 | 1223 if (!is_known_inst) { // Do it only for non-instance types |
1224 tj = to = TypeInstPtr::make(TypePtr::BotPTR, env()->Object_klass(), false, NULL, offset); | |
1225 } | |
0 | 1226 } else if (offset < 0 || offset >= k->size_helper() * wordSize) { |
1227 to = NULL; | |
1228 tj = TypeOopPtr::BOTTOM; | |
1229 offset = tj->offset(); | |
1230 } else { | |
1231 ciInstanceKlass *canonical_holder = k->get_canonical_holder(offset); | |
1232 if (!k->equals(canonical_holder) || tj->offset() != offset) { | |
247 | 1233 if( is_known_inst ) { |
1234 tj = to = TypeInstPtr::make(to->ptr(), canonical_holder, true, NULL, offset, to->instance_id()); | |
1235 } else { | |
1236 tj = to = TypeInstPtr::make(to->ptr(), canonical_holder, false, NULL, offset); | |
1237 } | |
0 | 1238 } |
1239 } | |
1240 } | |
1241 | |
1242 // Klass pointers to object array klasses need some flattening | |
1243 const TypeKlassPtr *tk = tj->isa_klassptr(); | |
1244 if( tk ) { | |
1245 // If we are referencing a field within a Klass, we need | |
1246 // to assume the worst case of an Object. Both exact and | |
1247 // inexact types must flatten to the same alias class. | |
1248 // Since the flattened result for a klass is defined to be | |
1249 // precisely java.lang.Object, use a constant ptr. | |
1250 if ( offset == Type::OffsetBot || (offset >= 0 && (size_t)offset < sizeof(Klass)) ) { | |
1251 | |
1252 tj = tk = TypeKlassPtr::make(TypePtr::Constant, | |
1253 TypeKlassPtr::OBJECT->klass(), | |
1254 offset); | |
1255 } | |
1256 | |
1257 ciKlass* klass = tk->klass(); | |
1258 if( klass->is_obj_array_klass() ) { | |
1259 ciKlass* k = TypeAryPtr::OOPS->klass(); | |
1260 if( !k || !k->is_loaded() ) // Only fails for some -Xcomp runs | |
1261 k = TypeInstPtr::BOTTOM->klass(); | |
1262 tj = tk = TypeKlassPtr::make( TypePtr::NotNull, k, offset ); | |
1263 } | |
1264 | |
1265 // Check for precise loads from the primary supertype array and force them | |
1266 // to the supertype cache alias index. Check for generic array loads from | |
1267 // the primary supertype array and also force them to the supertype cache | |
1268 // alias index. Since the same load can reach both, we need to merge | |
1269 // these 2 disparate memories into the same alias class. Since the | |
1270 // primary supertype array is read-only, there's no chance of confusion | |
1271 // where we bypass an array load and an array store. | |
1272 uint off2 = offset - Klass::primary_supers_offset_in_bytes(); | |
1273 if( offset == Type::OffsetBot || | |
1274 off2 < Klass::primary_super_limit()*wordSize ) { | |
1275 offset = sizeof(oopDesc) +Klass::secondary_super_cache_offset_in_bytes(); | |
1276 tj = tk = TypeKlassPtr::make( TypePtr::NotNull, tk->klass(), offset ); | |
1277 } | |
1278 } | |
1279 | |
1280 // Flatten all Raw pointers together. | |
1281 if (tj->base() == Type::RawPtr) | |
1282 tj = TypeRawPtr::BOTTOM; | |
1283 | |
1284 if (tj->base() == Type::AnyPtr) | |
1285 tj = TypePtr::BOTTOM; // An error, which the caller must check for. | |
1286 | |
1287 // Flatten all to bottom for now | |
1288 switch( _AliasLevel ) { | |
1289 case 0: | |
1290 tj = TypePtr::BOTTOM; | |
1291 break; | |
1292 case 1: // Flatten to: oop, static, field or array | |
1293 switch (tj->base()) { | |
1294 //case Type::AryPtr: tj = TypeAryPtr::RANGE; break; | |
1295 case Type::RawPtr: tj = TypeRawPtr::BOTTOM; break; | |
1296 case Type::AryPtr: // do not distinguish arrays at all | |
1297 case Type::InstPtr: tj = TypeInstPtr::BOTTOM; break; | |
1298 case Type::KlassPtr: tj = TypeKlassPtr::OBJECT; break; | |
1299 case Type::AnyPtr: tj = TypePtr::BOTTOM; break; // caller checks it | |
1300 default: ShouldNotReachHere(); | |
1301 } | |
1302 break; | |
605 | 1303 case 2: // No collapsing at level 2; keep all splits |
1304 case 3: // No collapsing at level 3; keep all splits | |
0 | 1305 break; |
1306 default: | |
1307 Unimplemented(); | |
1308 } | |
1309 | |
1310 offset = tj->offset(); | |
1311 assert( offset != Type::OffsetTop, "Offset has fallen from constant" ); | |
1312 | |
1313 assert( (offset != Type::OffsetBot && tj->base() != Type::AryPtr) || | |
1314 (offset == Type::OffsetBot && tj->base() == Type::AryPtr) || | |
1315 (offset == Type::OffsetBot && tj == TypeOopPtr::BOTTOM) || | |
1316 (offset == Type::OffsetBot && tj == TypePtr::BOTTOM) || | |
1317 (offset == oopDesc::mark_offset_in_bytes() && tj->base() == Type::AryPtr) || | |
1318 (offset == oopDesc::klass_offset_in_bytes() && tj->base() == Type::AryPtr) || | |
1319 (offset == arrayOopDesc::length_offset_in_bytes() && tj->base() == Type::AryPtr) , | |
1320 "For oops, klasses, raw offset must be constant; for arrays the offset is never known" ); | |
1321 assert( tj->ptr() != TypePtr::TopPTR && | |
1322 tj->ptr() != TypePtr::AnyNull && | |
1323 tj->ptr() != TypePtr::Null, "No imprecise addresses" ); | |
1324 // assert( tj->ptr() != TypePtr::Constant || | |
1325 // tj->base() == Type::RawPtr || | |
1326 // tj->base() == Type::KlassPtr, "No constant oop addresses" ); | |
1327 | |
1328 return tj; | |
1329 } | |
1330 | |
1331 void Compile::AliasType::Init(int i, const TypePtr* at) { | |
1332 _index = i; | |
1333 _adr_type = at; | |
1334 _field = NULL; | |
1335 _is_rewritable = true; // default | |
1336 const TypeOopPtr *atoop = (at != NULL) ? at->isa_oopptr() : NULL; | |
223 | 1337 if (atoop != NULL && atoop->is_known_instance()) { |
1338 const TypeOopPtr *gt = atoop->cast_to_instance_id(TypeOopPtr::InstanceBot); | |
0 | 1339 _general_index = Compile::current()->get_alias_index(gt); |
1340 } else { | |
1341 _general_index = 0; | |
1342 } | |
1343 } | |
1344 | |
1345 //---------------------------------print_on------------------------------------ | |
1346 #ifndef PRODUCT | |
1347 void Compile::AliasType::print_on(outputStream* st) { | |
1348 if (index() < 10) | |
1349 st->print("@ <%d> ", index()); | |
1350 else st->print("@ <%d>", index()); | |
1351 st->print(is_rewritable() ? " " : " RO"); | |
1352 int offset = adr_type()->offset(); | |
1353 if (offset == Type::OffsetBot) | |
1354 st->print(" +any"); | |
1355 else st->print(" +%-3d", offset); | |
1356 st->print(" in "); | |
1357 adr_type()->dump_on(st); | |
1358 const TypeOopPtr* tjp = adr_type()->isa_oopptr(); | |
1359 if (field() != NULL && tjp) { | |
1360 if (tjp->klass() != field()->holder() || | |
1361 tjp->offset() != field()->offset_in_bytes()) { | |
1362 st->print(" != "); | |
1363 field()->print(); | |
1364 st->print(" ***"); | |
1365 } | |
1366 } | |
1367 } | |
1368 | |
1369 void print_alias_types() { | |
1370 Compile* C = Compile::current(); | |
1371 tty->print_cr("--- Alias types, AliasIdxBot .. %d", C->num_alias_types()-1); | |
1372 for (int idx = Compile::AliasIdxBot; idx < C->num_alias_types(); idx++) { | |
1373 C->alias_type(idx)->print_on(tty); | |
1374 tty->cr(); | |
1375 } | |
1376 } | |
1377 #endif | |
1378 | |
1379 | |
1380 //----------------------------probe_alias_cache-------------------------------- | |
1381 Compile::AliasCacheEntry* Compile::probe_alias_cache(const TypePtr* adr_type) { | |
1382 intptr_t key = (intptr_t) adr_type; | |
1383 key ^= key >> logAliasCacheSize; | |
1384 return &_alias_cache[key & right_n_bits(logAliasCacheSize)]; | |
1385 } | |
1386 | |
1387 | |
1388 //-----------------------------grow_alias_types-------------------------------- | |
1389 void Compile::grow_alias_types() { | |
1390 const int old_ats = _max_alias_types; // how many before? | |
1391 const int new_ats = old_ats; // how many more? | |
1392 const int grow_ats = old_ats+new_ats; // how many now? | |
1393 _max_alias_types = grow_ats; | |
1394 _alias_types = REALLOC_ARENA_ARRAY(comp_arena(), AliasType*, _alias_types, old_ats, grow_ats); | |
1395 AliasType* ats = NEW_ARENA_ARRAY(comp_arena(), AliasType, new_ats); | |
1396 Copy::zero_to_bytes(ats, sizeof(AliasType)*new_ats); | |
1397 for (int i = 0; i < new_ats; i++) _alias_types[old_ats+i] = &ats[i]; | |
1398 } | |
1399 | |
1400 | |
1401 //--------------------------------find_alias_type------------------------------ | |
1402 Compile::AliasType* Compile::find_alias_type(const TypePtr* adr_type, bool no_create) { | |
1403 if (_AliasLevel == 0) | |
1404 return alias_type(AliasIdxBot); | |
1405 | |
1406 AliasCacheEntry* ace = probe_alias_cache(adr_type); | |
1407 if (ace->_adr_type == adr_type) { | |
1408 return alias_type(ace->_index); | |
1409 } | |
1410 | |
1411 // Handle special cases. | |
1412 if (adr_type == NULL) return alias_type(AliasIdxTop); | |
1413 if (adr_type == TypePtr::BOTTOM) return alias_type(AliasIdxBot); | |
1414 | |
1415 // Do it the slow way. | |
1416 const TypePtr* flat = flatten_alias_type(adr_type); | |
1417 | |
1418 #ifdef ASSERT | |
1419 assert(flat == flatten_alias_type(flat), "idempotent"); | |
1420 assert(flat != TypePtr::BOTTOM, "cannot alias-analyze an untyped ptr"); | |
1421 if (flat->isa_oopptr() && !flat->isa_klassptr()) { | |
1422 const TypeOopPtr* foop = flat->is_oopptr(); | |
247 | 1423 // Scalarizable allocations have exact klass always. |
1424 bool exact = !foop->klass_is_exact() || foop->is_known_instance(); | |
1425 const TypePtr* xoop = foop->cast_to_exactness(exact)->is_ptr(); | |
0 | 1426 assert(foop == flatten_alias_type(xoop), "exactness must not affect alias type"); |
1427 } | |
1428 assert(flat == flatten_alias_type(flat), "exact bit doesn't matter"); | |
1429 #endif | |
1430 | |
1431 int idx = AliasIdxTop; | |
1432 for (int i = 0; i < num_alias_types(); i++) { | |
1433 if (alias_type(i)->adr_type() == flat) { | |
1434 idx = i; | |
1435 break; | |
1436 } | |
1437 } | |
1438 | |
1439 if (idx == AliasIdxTop) { | |
1440 if (no_create) return NULL; | |
1441 // Grow the array if necessary. | |
1442 if (_num_alias_types == _max_alias_types) grow_alias_types(); | |
1443 // Add a new alias type. | |
1444 idx = _num_alias_types++; | |
1445 _alias_types[idx]->Init(idx, flat); | |
1446 if (flat == TypeInstPtr::KLASS) alias_type(idx)->set_rewritable(false); | |
1447 if (flat == TypeAryPtr::RANGE) alias_type(idx)->set_rewritable(false); | |
1448 if (flat->isa_instptr()) { | |
1449 if (flat->offset() == java_lang_Class::klass_offset_in_bytes() | |
1450 && flat->is_instptr()->klass() == env()->Class_klass()) | |
1451 alias_type(idx)->set_rewritable(false); | |
1452 } | |
1453 if (flat->isa_klassptr()) { | |
1454 if (flat->offset() == Klass::super_check_offset_offset_in_bytes() + (int)sizeof(oopDesc)) | |
1455 alias_type(idx)->set_rewritable(false); | |
1456 if (flat->offset() == Klass::modifier_flags_offset_in_bytes() + (int)sizeof(oopDesc)) | |
1457 alias_type(idx)->set_rewritable(false); | |
1458 if (flat->offset() == Klass::access_flags_offset_in_bytes() + (int)sizeof(oopDesc)) | |
1459 alias_type(idx)->set_rewritable(false); | |
1460 if (flat->offset() == Klass::java_mirror_offset_in_bytes() + (int)sizeof(oopDesc)) | |
1461 alias_type(idx)->set_rewritable(false); | |
1462 } | |
1463 // %%% (We would like to finalize JavaThread::threadObj_offset(), | |
1464 // but the base pointer type is not distinctive enough to identify | |
1465 // references into JavaThread.) | |
1466 | |
1467 // Check for final instance fields. | |
1468 const TypeInstPtr* tinst = flat->isa_instptr(); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1469 if (tinst && tinst->offset() >= instanceOopDesc::base_offset_in_bytes()) { |
0 | 1470 ciInstanceKlass *k = tinst->klass()->as_instance_klass(); |
1471 ciField* field = k->get_field_by_offset(tinst->offset(), false); | |
1472 // Set field() and is_rewritable() attributes. | |
1473 if (field != NULL) alias_type(idx)->set_field(field); | |
1474 } | |
1475 const TypeKlassPtr* tklass = flat->isa_klassptr(); | |
1476 // Check for final static fields. | |
1477 if (tklass && tklass->klass()->is_instance_klass()) { | |
1478 ciInstanceKlass *k = tklass->klass()->as_instance_klass(); | |
1479 ciField* field = k->get_field_by_offset(tklass->offset(), true); | |
1480 // Set field() and is_rewritable() attributes. | |
1481 if (field != NULL) alias_type(idx)->set_field(field); | |
1482 } | |
1483 } | |
1484 | |
1485 // Fill the cache for next time. | |
1486 ace->_adr_type = adr_type; | |
1487 ace->_index = idx; | |
1488 assert(alias_type(adr_type) == alias_type(idx), "type must be installed"); | |
1489 | |
1490 // Might as well try to fill the cache for the flattened version, too. | |
1491 AliasCacheEntry* face = probe_alias_cache(flat); | |
1492 if (face->_adr_type == NULL) { | |
1493 face->_adr_type = flat; | |
1494 face->_index = idx; | |
1495 assert(alias_type(flat) == alias_type(idx), "flat type must work too"); | |
1496 } | |
1497 | |
1498 return alias_type(idx); | |
1499 } | |
1500 | |
1501 | |
1502 Compile::AliasType* Compile::alias_type(ciField* field) { | |
1503 const TypeOopPtr* t; | |
1504 if (field->is_static()) | |
1505 t = TypeKlassPtr::make(field->holder()); | |
1506 else | |
1507 t = TypeOopPtr::make_from_klass_raw(field->holder()); | |
1508 AliasType* atp = alias_type(t->add_offset(field->offset_in_bytes())); | |
1509 assert(field->is_final() == !atp->is_rewritable(), "must get the rewritable bits correct"); | |
1510 return atp; | |
1511 } | |
1512 | |
1513 | |
1514 //------------------------------have_alias_type-------------------------------- | |
1515 bool Compile::have_alias_type(const TypePtr* adr_type) { | |
1516 AliasCacheEntry* ace = probe_alias_cache(adr_type); | |
1517 if (ace->_adr_type == adr_type) { | |
1518 return true; | |
1519 } | |
1520 | |
1521 // Handle special cases. | |
1522 if (adr_type == NULL) return true; | |
1523 if (adr_type == TypePtr::BOTTOM) return true; | |
1524 | |
1525 return find_alias_type(adr_type, true) != NULL; | |
1526 } | |
1527 | |
1528 //-----------------------------must_alias-------------------------------------- | |
1529 // True if all values of the given address type are in the given alias category. | |
1530 bool Compile::must_alias(const TypePtr* adr_type, int alias_idx) { | |
1531 if (alias_idx == AliasIdxBot) return true; // the universal category | |
1532 if (adr_type == NULL) return true; // NULL serves as TypePtr::TOP | |
1533 if (alias_idx == AliasIdxTop) return false; // the empty category | |
1534 if (adr_type->base() == Type::AnyPtr) return false; // TypePtr::BOTTOM or its twins | |
1535 | |
1536 // the only remaining possible overlap is identity | |
1537 int adr_idx = get_alias_index(adr_type); | |
1538 assert(adr_idx != AliasIdxBot && adr_idx != AliasIdxTop, ""); | |
1539 assert(adr_idx == alias_idx || | |
1540 (alias_type(alias_idx)->adr_type() != TypeOopPtr::BOTTOM | |
1541 && adr_type != TypeOopPtr::BOTTOM), | |
1542 "should not be testing for overlap with an unsafe pointer"); | |
1543 return adr_idx == alias_idx; | |
1544 } | |
1545 | |
1546 //------------------------------can_alias-------------------------------------- | |
1547 // True if any values of the given address type are in the given alias category. | |
1548 bool Compile::can_alias(const TypePtr* adr_type, int alias_idx) { | |
1549 if (alias_idx == AliasIdxTop) return false; // the empty category | |
1550 if (adr_type == NULL) return false; // NULL serves as TypePtr::TOP | |
1551 if (alias_idx == AliasIdxBot) return true; // the universal category | |
1552 if (adr_type->base() == Type::AnyPtr) return true; // TypePtr::BOTTOM or its twins | |
1553 | |
1554 // the only remaining possible overlap is identity | |
1555 int adr_idx = get_alias_index(adr_type); | |
1556 assert(adr_idx != AliasIdxBot && adr_idx != AliasIdxTop, ""); | |
1557 return adr_idx == alias_idx; | |
1558 } | |
1559 | |
1560 | |
1561 | |
1562 //---------------------------pop_warm_call------------------------------------- | |
1563 WarmCallInfo* Compile::pop_warm_call() { | |
1564 WarmCallInfo* wci = _warm_calls; | |
1565 if (wci != NULL) _warm_calls = wci->remove_from(wci); | |
1566 return wci; | |
1567 } | |
1568 | |
1569 //----------------------------Inline_Warm-------------------------------------- | |
1570 int Compile::Inline_Warm() { | |
1571 // If there is room, try to inline some more warm call sites. | |
1572 // %%% Do a graph index compaction pass when we think we're out of space? | |
1573 if (!InlineWarmCalls) return 0; | |
1574 | |
1575 int calls_made_hot = 0; | |
1576 int room_to_grow = NodeCountInliningCutoff - unique(); | |
1577 int amount_to_grow = MIN2(room_to_grow, (int)NodeCountInliningStep); | |
1578 int amount_grown = 0; | |
1579 WarmCallInfo* call; | |
1580 while (amount_to_grow > 0 && (call = pop_warm_call()) != NULL) { | |
1581 int est_size = (int)call->size(); | |
1582 if (est_size > (room_to_grow - amount_grown)) { | |
1583 // This one won't fit anyway. Get rid of it. | |
1584 call->make_cold(); | |
1585 continue; | |
1586 } | |
1587 call->make_hot(); | |
1588 calls_made_hot++; | |
1589 amount_grown += est_size; | |
1590 amount_to_grow -= est_size; | |
1591 } | |
1592 | |
1593 if (calls_made_hot > 0) set_major_progress(); | |
1594 return calls_made_hot; | |
1595 } | |
1596 | |
1597 | |
1598 //----------------------------Finish_Warm-------------------------------------- | |
1599 void Compile::Finish_Warm() { | |
1600 if (!InlineWarmCalls) return; | |
1601 if (failing()) return; | |
1602 if (warm_calls() == NULL) return; | |
1603 | |
1604 // Clean up loose ends, if we are out of space for inlining. | |
1605 WarmCallInfo* call; | |
1606 while ((call = pop_warm_call()) != NULL) { | |
1607 call->make_cold(); | |
1608 } | |
1609 } | |
1610 | |
1172 | 1611 //---------------------cleanup_loop_predicates----------------------- |
1612 // Remove the opaque nodes that protect the predicates so that all unused | |
1613 // checks and uncommon_traps will be eliminated from the ideal graph | |
1614 void Compile::cleanup_loop_predicates(PhaseIterGVN &igvn) { | |
1615 if (predicate_count()==0) return; | |
1616 for (int i = predicate_count(); i > 0; i--) { | |
1617 Node * n = predicate_opaque1_node(i-1); | |
1618 assert(n->Opcode() == Op_Opaque1, "must be"); | |
1619 igvn.replace_node(n, n->in(1)); | |
1620 } | |
1621 assert(predicate_count()==0, "should be clean!"); | |
1622 igvn.optimize(); | |
1623 } | |
0 | 1624 |
1625 //------------------------------Optimize--------------------------------------- | |
1626 // Given a graph, optimize it. | |
1627 void Compile::Optimize() { | |
1628 TracePhase t1("optimizer", &_t_optimizer, true); | |
1629 | |
1630 #ifndef PRODUCT | |
1631 if (env()->break_at_compile()) { | |
1632 BREAKPOINT; | |
1633 } | |
1634 | |
1635 #endif | |
1636 | |
1637 ResourceMark rm; | |
1638 int loop_opts_cnt; | |
1639 | |
1640 NOT_PRODUCT( verify_graph_edges(); ) | |
1641 | |
222 | 1642 print_method("After Parsing"); |
0 | 1643 |
1644 { | |
1645 // Iterative Global Value Numbering, including ideal transforms | |
1646 // Initialize IterGVN with types and values from parse-time GVN | |
1647 PhaseIterGVN igvn(initial_gvn()); | |
1648 { | |
1649 NOT_PRODUCT( TracePhase t2("iterGVN", &_t_iterGVN, TimeCompiler); ) | |
1650 igvn.optimize(); | |
1651 } | |
1652 | |
1653 print_method("Iter GVN 1", 2); | |
1654 | |
1655 if (failing()) return; | |
1656 | |
1634
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1657 // Perform escape analysis |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1658 if (_do_escape_analysis && ConnectionGraph::has_candidates(this)) { |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1659 TracePhase t2("escapeAnalysis", &_t_escapeAnalysis, true); |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1660 ConnectionGraph::do_analysis(this, &igvn); |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1661 |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1662 if (failing()) return; |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1663 |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1664 igvn.optimize(); |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1665 print_method("Iter GVN 3", 2); |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1666 |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1667 if (failing()) return; |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1668 |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1669 } |
60a14ad85270
6966411: escape.cpp:450 assert(base->Opcode() == Op_ConP
kvn
parents:
1609
diff
changeset
|
1670 |
0 | 1671 // Loop transforms on the ideal graph. Range Check Elimination, |
1672 // peeling, unrolling, etc. | |
1673 | |
1674 // Set loop opts counter | |
1675 loop_opts_cnt = num_loop_opts(); | |
1676 if((loop_opts_cnt > 0) && (has_loops() || has_split_ifs())) { | |
1677 { | |
1678 TracePhase t2("idealLoop", &_t_idealLoop, true); | |
1172 | 1679 PhaseIdealLoop ideal_loop( igvn, true, UseLoopPredicate); |
0 | 1680 loop_opts_cnt--; |
1681 if (major_progress()) print_method("PhaseIdealLoop 1", 2); | |
1682 if (failing()) return; | |
1683 } | |
1684 // Loop opts pass if partial peeling occurred in previous pass | |
1685 if(PartialPeelLoop && major_progress() && (loop_opts_cnt > 0)) { | |
1686 TracePhase t3("idealLoop", &_t_idealLoop, true); | |
1172 | 1687 PhaseIdealLoop ideal_loop( igvn, false, UseLoopPredicate); |
0 | 1688 loop_opts_cnt--; |
1689 if (major_progress()) print_method("PhaseIdealLoop 2", 2); | |
1690 if (failing()) return; | |
1691 } | |
1692 // Loop opts pass for loop-unrolling before CCP | |
1693 if(major_progress() && (loop_opts_cnt > 0)) { | |
1694 TracePhase t4("idealLoop", &_t_idealLoop, true); | |
1172 | 1695 PhaseIdealLoop ideal_loop( igvn, false, UseLoopPredicate); |
0 | 1696 loop_opts_cnt--; |
1697 if (major_progress()) print_method("PhaseIdealLoop 3", 2); | |
1698 } | |
921
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1699 if (!failing()) { |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1700 // Verify that last round of loop opts produced a valid graph |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1701 NOT_PRODUCT( TracePhase t2("idealLoopVerify", &_t_idealLoopVerify, TimeCompiler); ) |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1702 PhaseIdealLoop::verify(igvn); |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1703 } |
0 | 1704 } |
1705 if (failing()) return; | |
1706 | |
1707 // Conditional Constant Propagation; | |
1708 PhaseCCP ccp( &igvn ); | |
1709 assert( true, "Break here to ccp.dump_nodes_and_types(_root,999,1)"); | |
1710 { | |
1711 TracePhase t2("ccp", &_t_ccp, true); | |
1712 ccp.do_transform(); | |
1713 } | |
1714 print_method("PhaseCPP 1", 2); | |
1715 | |
1716 assert( true, "Break here to ccp.dump_old2new_map()"); | |
1717 | |
1718 // Iterative Global Value Numbering, including ideal transforms | |
1719 { | |
1720 NOT_PRODUCT( TracePhase t2("iterGVN2", &_t_iterGVN2, TimeCompiler); ) | |
1721 igvn = ccp; | |
1722 igvn.optimize(); | |
1723 } | |
1724 | |
1725 print_method("Iter GVN 2", 2); | |
1726 | |
1727 if (failing()) return; | |
1728 | |
1729 // Loop transforms on the ideal graph. Range Check Elimination, | |
1730 // peeling, unrolling, etc. | |
1731 if(loop_opts_cnt > 0) { | |
1732 debug_only( int cnt = 0; ); | |
1172 | 1733 bool loop_predication = UseLoopPredicate; |
0 | 1734 while(major_progress() && (loop_opts_cnt > 0)) { |
1735 TracePhase t2("idealLoop", &_t_idealLoop, true); | |
1736 assert( cnt++ < 40, "infinite cycle in loop optimization" ); | |
1172 | 1737 PhaseIdealLoop ideal_loop( igvn, true, loop_predication); |
0 | 1738 loop_opts_cnt--; |
1739 if (major_progress()) print_method("PhaseIdealLoop iterations", 2); | |
1740 if (failing()) return; | |
1172 | 1741 // Perform loop predication optimization during first iteration after CCP. |
1742 // After that switch it off and cleanup unused loop predicates. | |
1743 if (loop_predication) { | |
1744 loop_predication = false; | |
1745 cleanup_loop_predicates(igvn); | |
1746 if (failing()) return; | |
1747 } | |
0 | 1748 } |
1749 } | |
921
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1750 |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1751 { |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1752 // Verify that all previous optimizations produced a valid graph |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1753 // at least to this point, even if no loop optimizations were done. |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1754 NOT_PRODUCT( TracePhase t2("idealLoopVerify", &_t_idealLoopVerify, TimeCompiler); ) |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1755 PhaseIdealLoop::verify(igvn); |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1756 } |
046932b72aa2
6862956: PhaseIdealLoop should have a CFG verification mode
never
parents:
859
diff
changeset
|
1757 |
0 | 1758 { |
1759 NOT_PRODUCT( TracePhase t2("macroExpand", &_t_macroExpand, TimeCompiler); ) | |
1760 PhaseMacroExpand mex(igvn); | |
1761 if (mex.expand_macro_nodes()) { | |
1762 assert(failing(), "must bail out w/ explicit message"); | |
1763 return; | |
1764 } | |
1765 } | |
1766 | |
1767 } // (End scope of igvn; run destructor if necessary for asserts.) | |
1768 | |
1769 // A method with only infinite loops has no edges entering loops from root | |
1770 { | |
1771 NOT_PRODUCT( TracePhase t2("graphReshape", &_t_graphReshaping, TimeCompiler); ) | |
1772 if (final_graph_reshaping()) { | |
1773 assert(failing(), "must bail out w/ explicit message"); | |
1774 return; | |
1775 } | |
1776 } | |
1777 | |
1778 print_method("Optimize finished", 2); | |
1779 } | |
1780 | |
1781 | |
1782 //------------------------------Code_Gen--------------------------------------- | |
1783 // Given a graph, generate code for it | |
1784 void Compile::Code_Gen() { | |
1785 if (failing()) return; | |
1786 | |
1787 // Perform instruction selection. You might think we could reclaim Matcher | |
1788 // memory PDQ, but actually the Matcher is used in generating spill code. | |
1789 // Internals of the Matcher (including some VectorSets) must remain live | |
1790 // for awhile - thus I cannot reclaim Matcher memory lest a VectorSet usage | |
1791 // set a bit in reclaimed memory. | |
1792 | |
1793 // In debug mode can dump m._nodes.dump() for mapping of ideal to machine | |
1794 // nodes. Mapping is only valid at the root of each matched subtree. | |
1795 NOT_PRODUCT( verify_graph_edges(); ) | |
1796 | |
1797 Node_List proj_list; | |
1798 Matcher m(proj_list); | |
1799 _matcher = &m; | |
1800 { | |
1801 TracePhase t2("matcher", &_t_matcher, true); | |
1802 m.match(); | |
1803 } | |
1804 // In debug mode can dump m._nodes.dump() for mapping of ideal to machine | |
1805 // nodes. Mapping is only valid at the root of each matched subtree. | |
1806 NOT_PRODUCT( verify_graph_edges(); ) | |
1807 | |
1808 // If you have too many nodes, or if matching has failed, bail out | |
1809 check_node_count(0, "out of nodes matching instructions"); | |
1810 if (failing()) return; | |
1811 | |
1812 // Build a proper-looking CFG | |
1813 PhaseCFG cfg(node_arena(), root(), m); | |
1814 _cfg = &cfg; | |
1815 { | |
1816 NOT_PRODUCT( TracePhase t2("scheduler", &_t_scheduler, TimeCompiler); ) | |
1817 cfg.Dominators(); | |
1818 if (failing()) return; | |
1819 | |
1820 NOT_PRODUCT( verify_graph_edges(); ) | |
1821 | |
1822 cfg.Estimate_Block_Frequency(); | |
1823 cfg.GlobalCodeMotion(m,unique(),proj_list); | |
1824 | |
1825 print_method("Global code motion", 2); | |
1826 | |
1827 if (failing()) return; | |
1828 NOT_PRODUCT( verify_graph_edges(); ) | |
1829 | |
1830 debug_only( cfg.verify(); ) | |
1831 } | |
1832 NOT_PRODUCT( verify_graph_edges(); ) | |
1833 | |
1834 PhaseChaitin regalloc(unique(),cfg,m); | |
1835 _regalloc = ®alloc; | |
1836 { | |
1837 TracePhase t2("regalloc", &_t_registerAllocation, true); | |
1838 // Perform any platform dependent preallocation actions. This is used, | |
1839 // for example, to avoid taking an implicit null pointer exception | |
1840 // using the frame pointer on win95. | |
1841 _regalloc->pd_preallocate_hook(); | |
1842 | |
1843 // Perform register allocation. After Chaitin, use-def chains are | |
1844 // no longer accurate (at spill code) and so must be ignored. | |
1845 // Node->LRG->reg mappings are still accurate. | |
1846 _regalloc->Register_Allocate(); | |
1847 | |
1848 // Bail out if the allocator builds too many nodes | |
1849 if (failing()) return; | |
1850 } | |
1851 | |
1852 // Prior to register allocation we kept empty basic blocks in case the | |
1853 // the allocator needed a place to spill. After register allocation we | |
1854 // are not adding any new instructions. If any basic block is empty, we | |
1855 // can now safely remove it. | |
1856 { | |
418 | 1857 NOT_PRODUCT( TracePhase t2("blockOrdering", &_t_blockOrdering, TimeCompiler); ) |
1858 cfg.remove_empty(); | |
1859 if (do_freq_based_layout()) { | |
1860 PhaseBlockLayout layout(cfg); | |
1861 } else { | |
1862 cfg.set_loop_alignment(); | |
1863 } | |
1864 cfg.fixup_flow(); | |
0 | 1865 } |
1866 | |
1867 // Perform any platform dependent postallocation verifications. | |
1868 debug_only( _regalloc->pd_postallocate_verify_hook(); ) | |
1869 | |
1870 // Apply peephole optimizations | |
1871 if( OptoPeephole ) { | |
1872 NOT_PRODUCT( TracePhase t2("peephole", &_t_peephole, TimeCompiler); ) | |
1873 PhasePeephole peep( _regalloc, cfg); | |
1874 peep.do_transform(); | |
1875 } | |
1876 | |
1877 // Convert Nodes to instruction bits in a buffer | |
1878 { | |
1879 // %%%% workspace merge brought two timers together for one job | |
1880 TracePhase t2a("output", &_t_output, true); | |
1881 NOT_PRODUCT( TraceTime t2b(NULL, &_t_codeGeneration, TimeCompiler, false); ) | |
1882 Output(); | |
1883 } | |
1884 | |
222 | 1885 print_method("Final Code"); |
0 | 1886 |
1887 // He's dead, Jim. | |
1888 _cfg = (PhaseCFG*)0xdeadbeef; | |
1889 _regalloc = (PhaseChaitin*)0xdeadbeef; | |
1890 } | |
1891 | |
1892 | |
1893 //------------------------------dump_asm--------------------------------------- | |
1894 // Dump formatted assembly | |
1895 #ifndef PRODUCT | |
1896 void Compile::dump_asm(int *pcs, uint pc_limit) { | |
1897 bool cut_short = false; | |
1898 tty->print_cr("#"); | |
1899 tty->print("# "); _tf->dump(); tty->cr(); | |
1900 tty->print_cr("#"); | |
1901 | |
1902 // For all blocks | |
1903 int pc = 0x0; // Program counter | |
1904 char starts_bundle = ' '; | |
1905 _regalloc->dump_frame(); | |
1906 | |
1907 Node *n = NULL; | |
1908 for( uint i=0; i<_cfg->_num_blocks; i++ ) { | |
1909 if (VMThread::should_terminate()) { cut_short = true; break; } | |
1910 Block *b = _cfg->_blocks[i]; | |
1911 if (b->is_connector() && !Verbose) continue; | |
1912 n = b->_nodes[0]; | |
1913 if (pcs && n->_idx < pc_limit) | |
1914 tty->print("%3.3x ", pcs[n->_idx]); | |
1915 else | |
1916 tty->print(" "); | |
1917 b->dump_head( &_cfg->_bbs ); | |
1918 if (b->is_connector()) { | |
1919 tty->print_cr(" # Empty connector block"); | |
1920 } else if (b->num_preds() == 2 && b->pred(1)->is_CatchProj() && b->pred(1)->as_CatchProj()->_con == CatchProjNode::fall_through_index) { | |
1921 tty->print_cr(" # Block is sole successor of call"); | |
1922 } | |
1923 | |
1924 // For all instructions | |
1925 Node *delay = NULL; | |
1926 for( uint j = 0; j<b->_nodes.size(); j++ ) { | |
1927 if (VMThread::should_terminate()) { cut_short = true; break; } | |
1928 n = b->_nodes[j]; | |
1929 if (valid_bundle_info(n)) { | |
1930 Bundle *bundle = node_bundling(n); | |
1931 if (bundle->used_in_unconditional_delay()) { | |
1932 delay = n; | |
1933 continue; | |
1934 } | |
1935 if (bundle->starts_bundle()) | |
1936 starts_bundle = '+'; | |
1937 } | |
1938 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1939 if (WizardMode) n->dump(); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1940 |
0 | 1941 if( !n->is_Region() && // Dont print in the Assembly |
1942 !n->is_Phi() && // a few noisely useless nodes | |
1943 !n->is_Proj() && | |
1944 !n->is_MachTemp() && | |
1100
f96a1a986f7b
6895383: JCK test throws NPE for method compiled with Escape Analysis
kvn
parents:
1080
diff
changeset
|
1945 !n->is_SafePointScalarObject() && |
0 | 1946 !n->is_Catch() && // Would be nice to print exception table targets |
1947 !n->is_MergeMem() && // Not very interesting | |
1948 !n->is_top() && // Debug info table constants | |
1949 !(n->is_Con() && !n->is_Mach())// Debug info table constants | |
1950 ) { | |
1951 if (pcs && n->_idx < pc_limit) | |
1952 tty->print("%3.3x", pcs[n->_idx]); | |
1953 else | |
1954 tty->print(" "); | |
1955 tty->print(" %c ", starts_bundle); | |
1956 starts_bundle = ' '; | |
1957 tty->print("\t"); | |
1958 n->format(_regalloc, tty); | |
1959 tty->cr(); | |
1960 } | |
1961 | |
1962 // If we have an instruction with a delay slot, and have seen a delay, | |
1963 // then back up and print it | |
1964 if (valid_bundle_info(n) && node_bundling(n)->use_unconditional_delay()) { | |
1965 assert(delay != NULL, "no unconditional delay instruction"); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1966 if (WizardMode) delay->dump(); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
1967 |
0 | 1968 if (node_bundling(delay)->starts_bundle()) |
1969 starts_bundle = '+'; | |
1970 if (pcs && n->_idx < pc_limit) | |
1971 tty->print("%3.3x", pcs[n->_idx]); | |
1972 else | |
1973 tty->print(" "); | |
1974 tty->print(" %c ", starts_bundle); | |
1975 starts_bundle = ' '; | |
1976 tty->print("\t"); | |
1977 delay->format(_regalloc, tty); | |
1978 tty->print_cr(""); | |
1979 delay = NULL; | |
1980 } | |
1981 | |
1982 // Dump the exception table as well | |
1983 if( n->is_Catch() && (Verbose || WizardMode) ) { | |
1984 // Print the exception table for this offset | |
1985 _handler_table.print_subtable_for(pc); | |
1986 } | |
1987 } | |
1988 | |
1989 if (pcs && n->_idx < pc_limit) | |
1990 tty->print_cr("%3.3x", pcs[n->_idx]); | |
1991 else | |
1992 tty->print_cr(""); | |
1993 | |
1994 assert(cut_short || delay == NULL, "no unconditional delay branch"); | |
1995 | |
1996 } // End of per-block dump | |
1997 tty->print_cr(""); | |
1998 | |
1999 if (cut_short) tty->print_cr("*** disassembly is cut short ***"); | |
2000 } | |
2001 #endif | |
2002 | |
2003 //------------------------------Final_Reshape_Counts--------------------------- | |
2004 // This class defines counters to help identify when a method | |
2005 // may/must be executed using hardware with only 24-bit precision. | |
2006 struct Final_Reshape_Counts : public StackObj { | |
2007 int _call_count; // count non-inlined 'common' calls | |
2008 int _float_count; // count float ops requiring 24-bit precision | |
2009 int _double_count; // count double ops requiring more precision | |
2010 int _java_call_count; // count non-inlined 'java' calls | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2011 int _inner_loop_count; // count loops which need alignment |
0 | 2012 VectorSet _visited; // Visitation flags |
2013 Node_List _tests; // Set of IfNodes & PCTableNodes | |
2014 | |
2015 Final_Reshape_Counts() : | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2016 _call_count(0), _float_count(0), _double_count(0), |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2017 _java_call_count(0), _inner_loop_count(0), |
0 | 2018 _visited( Thread::current()->resource_area() ) { } |
2019 | |
2020 void inc_call_count () { _call_count ++; } | |
2021 void inc_float_count () { _float_count ++; } | |
2022 void inc_double_count() { _double_count++; } | |
2023 void inc_java_call_count() { _java_call_count++; } | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2024 void inc_inner_loop_count() { _inner_loop_count++; } |
0 | 2025 |
2026 int get_call_count () const { return _call_count ; } | |
2027 int get_float_count () const { return _float_count ; } | |
2028 int get_double_count() const { return _double_count; } | |
2029 int get_java_call_count() const { return _java_call_count; } | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2030 int get_inner_loop_count() const { return _inner_loop_count; } |
0 | 2031 }; |
2032 | |
2033 static bool oop_offset_is_sane(const TypeInstPtr* tp) { | |
2034 ciInstanceKlass *k = tp->klass()->as_instance_klass(); | |
2035 // Make sure the offset goes inside the instance layout. | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
2036 return k->contains_field_offset(tp->offset()); |
0 | 2037 // Note that OffsetBot and OffsetTop are very negative. |
2038 } | |
2039 | |
2040 //------------------------------final_graph_reshaping_impl---------------------- | |
2041 // Implement items 1-5 from final_graph_reshaping below. | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2042 static void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc ) { |
0 | 2043 |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2044 if ( n->outcnt() == 0 ) return; // dead node |
0 | 2045 uint nop = n->Opcode(); |
2046 | |
2047 // Check for 2-input instruction with "last use" on right input. | |
2048 // Swap to left input. Implements item (2). | |
2049 if( n->req() == 3 && // two-input instruction | |
2050 n->in(1)->outcnt() > 1 && // left use is NOT a last use | |
2051 (!n->in(1)->is_Phi() || n->in(1)->in(2) != n) && // it is not data loop | |
2052 n->in(2)->outcnt() == 1 &&// right use IS a last use | |
2053 !n->in(2)->is_Con() ) { // right use is not a constant | |
2054 // Check for commutative opcode | |
2055 switch( nop ) { | |
2056 case Op_AddI: case Op_AddF: case Op_AddD: case Op_AddL: | |
2057 case Op_MaxI: case Op_MinI: | |
2058 case Op_MulI: case Op_MulF: case Op_MulD: case Op_MulL: | |
2059 case Op_AndL: case Op_XorL: case Op_OrL: | |
2060 case Op_AndI: case Op_XorI: case Op_OrI: { | |
2061 // Move "last use" input to left by swapping inputs | |
2062 n->swap_edges(1, 2); | |
2063 break; | |
2064 } | |
2065 default: | |
2066 break; | |
2067 } | |
2068 } | |
2069 | |
1609 | 2070 #ifdef ASSERT |
2071 if( n->is_Mem() ) { | |
2072 Compile* C = Compile::current(); | |
2073 int alias_idx = C->get_alias_index(n->as_Mem()->adr_type()); | |
2074 assert( n->in(0) != NULL || alias_idx != Compile::AliasIdxRaw || | |
2075 // oop will be recorded in oop map if load crosses safepoint | |
2076 n->is_Load() && (n->as_Load()->bottom_type()->isa_oopptr() || | |
2077 LoadNode::is_immutable_value(n->in(MemNode::Address))), | |
2078 "raw memory operations should have control edge"); | |
2079 } | |
2080 #endif | |
0 | 2081 // Count FPU ops and common calls, implements item (3) |
2082 switch( nop ) { | |
2083 // Count all float operations that may use FPU | |
2084 case Op_AddF: | |
2085 case Op_SubF: | |
2086 case Op_MulF: | |
2087 case Op_DivF: | |
2088 case Op_NegF: | |
2089 case Op_ModF: | |
2090 case Op_ConvI2F: | |
2091 case Op_ConF: | |
2092 case Op_CmpF: | |
2093 case Op_CmpF3: | |
2094 // case Op_ConvL2F: // longs are split into 32-bit halves | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2095 frc.inc_float_count(); |
0 | 2096 break; |
2097 | |
2098 case Op_ConvF2D: | |
2099 case Op_ConvD2F: | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2100 frc.inc_float_count(); |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2101 frc.inc_double_count(); |
0 | 2102 break; |
2103 | |
2104 // Count all double operations that may use FPU | |
2105 case Op_AddD: | |
2106 case Op_SubD: | |
2107 case Op_MulD: | |
2108 case Op_DivD: | |
2109 case Op_NegD: | |
2110 case Op_ModD: | |
2111 case Op_ConvI2D: | |
2112 case Op_ConvD2I: | |
2113 // case Op_ConvL2D: // handled by leaf call | |
2114 // case Op_ConvD2L: // handled by leaf call | |
2115 case Op_ConD: | |
2116 case Op_CmpD: | |
2117 case Op_CmpD3: | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2118 frc.inc_double_count(); |
0 | 2119 break; |
2120 case Op_Opaque1: // Remove Opaque Nodes before matching | |
2121 case Op_Opaque2: // Remove Opaque Nodes before matching | |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2122 n->subsume_by(n->in(1)); |
0 | 2123 break; |
2124 case Op_CallStaticJava: | |
2125 case Op_CallJava: | |
2126 case Op_CallDynamicJava: | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2127 frc.inc_java_call_count(); // Count java call site; |
0 | 2128 case Op_CallRuntime: |
2129 case Op_CallLeaf: | |
2130 case Op_CallLeafNoFP: { | |
2131 assert( n->is_Call(), "" ); | |
2132 CallNode *call = n->as_Call(); | |
2133 // Count call sites where the FP mode bit would have to be flipped. | |
2134 // Do not count uncommon runtime calls: | |
2135 // uncommon_trap, _complete_monitor_locking, _complete_monitor_unlocking, | |
2136 // _new_Java, _new_typeArray, _new_objArray, _rethrow_Java, ... | |
2137 if( !call->is_CallStaticJava() || !call->as_CallStaticJava()->_name ) { | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2138 frc.inc_call_count(); // Count the call site |
0 | 2139 } else { // See if uncommon argument is shared |
2140 Node *n = call->in(TypeFunc::Parms); | |
2141 int nop = n->Opcode(); | |
2142 // Clone shared simple arguments to uncommon calls, item (1). | |
2143 if( n->outcnt() > 1 && | |
2144 !n->is_Proj() && | |
2145 nop != Op_CreateEx && | |
2146 nop != Op_CheckCastPP && | |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2147 nop != Op_DecodeN && |
0 | 2148 !n->is_Mem() ) { |
2149 Node *x = n->clone(); | |
2150 call->set_req( TypeFunc::Parms, x ); | |
2151 } | |
2152 } | |
2153 break; | |
2154 } | |
2155 | |
2156 case Op_StoreD: | |
2157 case Op_LoadD: | |
2158 case Op_LoadD_unaligned: | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2159 frc.inc_double_count(); |
0 | 2160 goto handle_mem; |
2161 case Op_StoreF: | |
2162 case Op_LoadF: | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2163 frc.inc_float_count(); |
0 | 2164 goto handle_mem; |
2165 | |
2166 case Op_StoreB: | |
2167 case Op_StoreC: | |
2168 case Op_StoreCM: | |
2169 case Op_StorePConditional: | |
2170 case Op_StoreI: | |
2171 case Op_StoreL: | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
418
diff
changeset
|
2172 case Op_StoreIConditional: |
0 | 2173 case Op_StoreLConditional: |
2174 case Op_CompareAndSwapI: | |
2175 case Op_CompareAndSwapL: | |
2176 case Op_CompareAndSwapP: | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
2177 case Op_CompareAndSwapN: |
0 | 2178 case Op_StoreP: |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
2179 case Op_StoreN: |
0 | 2180 case Op_LoadB: |
624 | 2181 case Op_LoadUB: |
558
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
492
diff
changeset
|
2182 case Op_LoadUS: |
0 | 2183 case Op_LoadI: |
624 | 2184 case Op_LoadUI2L: |
0 | 2185 case Op_LoadKlass: |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2186 case Op_LoadNKlass: |
0 | 2187 case Op_LoadL: |
2188 case Op_LoadL_unaligned: | |
2189 case Op_LoadPLocked: | |
2190 case Op_LoadLLocked: | |
2191 case Op_LoadP: | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
100
diff
changeset
|
2192 case Op_LoadN: |
0 | 2193 case Op_LoadRange: |
2194 case Op_LoadS: { | |
2195 handle_mem: | |
2196 #ifdef ASSERT | |
2197 if( VerifyOptoOopOffsets ) { | |
2198 assert( n->is_Mem(), "" ); | |
2199 MemNode *mem = (MemNode*)n; | |
2200 // Check to see if address types have grounded out somehow. | |
2201 const TypeInstPtr *tp = mem->in(MemNode::Address)->bottom_type()->isa_instptr(); | |
2202 assert( !tp || oop_offset_is_sane(tp), "" ); | |
2203 } | |
2204 #endif | |
2205 break; | |
2206 } | |
2207 | |
2208 case Op_AddP: { // Assert sane base pointers | |
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2209 Node *addp = n->in(AddPNode::Address); |
0 | 2210 assert( !addp->is_AddP() || |
2211 addp->in(AddPNode::Base)->is_top() || // Top OK for allocation | |
2212 addp->in(AddPNode::Base) == n->in(AddPNode::Base), | |
2213 "Base pointers must match" ); | |
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2214 #ifdef _LP64 |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2215 if (UseCompressedOops && |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2216 addp->Opcode() == Op_ConP && |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2217 addp == n->in(AddPNode::Base) && |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2218 n->in(AddPNode::Offset)->is_Con()) { |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2219 // Use addressing with narrow klass to load with offset on x86. |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2220 // On sparc loading 32-bits constant and decoding it have less |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2221 // instructions (4) then load 64-bits constant (7). |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2222 // Do this transformation here since IGVN will convert ConN back to ConP. |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2223 const Type* t = addp->bottom_type(); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2224 if (t->isa_oopptr()) { |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2225 Node* nn = NULL; |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2226 |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2227 // Look for existing ConN node of the same exact type. |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2228 Compile* C = Compile::current(); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2229 Node* r = C->root(); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2230 uint cnt = r->outcnt(); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2231 for (uint i = 0; i < cnt; i++) { |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2232 Node* m = r->raw_out(i); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2233 if (m!= NULL && m->Opcode() == Op_ConN && |
221
1e026f8da827
6710487: More than half of JDI Regression tests hang with COOPs in -Xcomp mode
kvn
parents:
182
diff
changeset
|
2234 m->bottom_type()->make_ptr() == t) { |
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2235 nn = m; |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2236 break; |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2237 } |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2238 } |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2239 if (nn != NULL) { |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2240 // Decode a narrow oop to match address |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2241 // [R12 + narrow_oop_reg<<3 + offset] |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2242 nn = new (C, 2) DecodeNNode(nn, t); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2243 n->set_req(AddPNode::Base, nn); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2244 n->set_req(AddPNode::Address, nn); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2245 if (addp->outcnt() == 0) { |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2246 addp->disconnect_inputs(NULL); |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2247 } |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2248 } |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2249 } |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2250 } |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
168
diff
changeset
|
2251 #endif |
0 | 2252 break; |
2253 } | |
2254 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2255 #ifdef _LP64 |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2256 case Op_CastPP: |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2257 if (n->in(1)->is_DecodeN() && Matcher::gen_narrow_oop_implicit_null_checks()) { |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2258 Compile* C = Compile::current(); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2259 Node* in1 = n->in(1); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2260 const Type* t = n->bottom_type(); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2261 Node* new_in1 = in1->clone(); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2262 new_in1->as_DecodeN()->set_type(t); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2263 |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2264 if (!Matcher::narrow_oop_use_complex_address()) { |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2265 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2266 // x86, ARM and friends can handle 2 adds in addressing mode |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2267 // and Matcher can fold a DecodeN node into address by using |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2268 // a narrow oop directly and do implicit NULL check in address: |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2269 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2270 // [R12 + narrow_oop_reg<<3 + offset] |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2271 // NullCheck narrow_oop_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2272 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2273 // On other platforms (Sparc) we have to keep new DecodeN node and |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2274 // use it to do implicit NULL check in address: |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2275 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2276 // decode_not_null narrow_oop_reg, base_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2277 // [base_reg + offset] |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2278 // NullCheck base_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2279 // |
605 | 2280 // Pin the new DecodeN node to non-null path on these platform (Sparc) |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2281 // to keep the information to which NULL check the new DecodeN node |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2282 // corresponds to use it as value in implicit_null_check(). |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2283 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2284 new_in1->set_req(0, n->in(0)); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2285 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2286 |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2287 n->subsume_by(new_in1); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2288 if (in1->outcnt() == 0) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2289 in1->disconnect_inputs(NULL); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2290 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2291 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2292 break; |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2293 |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2294 case Op_CmpP: |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2295 // Do this transformation here to preserve CmpPNode::sub() and |
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2296 // other TypePtr related Ideal optimizations (for example, ptr nullness). |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2297 if (n->in(1)->is_DecodeN() || n->in(2)->is_DecodeN()) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2298 Node* in1 = n->in(1); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2299 Node* in2 = n->in(2); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2300 if (!in1->is_DecodeN()) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2301 in2 = in1; |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2302 in1 = n->in(2); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2303 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2304 assert(in1->is_DecodeN(), "sanity"); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2305 |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2306 Compile* C = Compile::current(); |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2307 Node* new_in2 = NULL; |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2308 if (in2->is_DecodeN()) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2309 new_in2 = in2->in(1); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2310 } else if (in2->Opcode() == Op_ConP) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2311 const Type* t = in2->bottom_type(); |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2312 if (t == TypePtr::NULL_PTR) { |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2313 // Don't convert CmpP null check into CmpN if compressed |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2314 // oops implicit null check is not generated. |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2315 // This will allow to generate normal oop implicit null check. |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2316 if (Matcher::gen_narrow_oop_implicit_null_checks()) |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2317 new_in2 = ConNode::make(C, TypeNarrowOop::NULL_PTR); |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2318 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2319 // This transformation together with CastPP transformation above |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2320 // will generated code for implicit NULL checks for compressed oops. |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2321 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2322 // The original code after Optimize() |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2323 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2324 // LoadN memory, narrow_oop_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2325 // decode narrow_oop_reg, base_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2326 // CmpP base_reg, NULL |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2327 // CastPP base_reg // NotNull |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2328 // Load [base_reg + offset], val_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2329 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2330 // after these transformations will be |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2331 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2332 // LoadN memory, narrow_oop_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2333 // CmpN narrow_oop_reg, NULL |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2334 // decode_not_null narrow_oop_reg, base_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2335 // Load [base_reg + offset], val_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2336 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2337 // and the uncommon path (== NULL) will use narrow_oop_reg directly |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2338 // since narrow oops can be used in debug info now (see the code in |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2339 // final_graph_reshaping_walk()). |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2340 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2341 // At the end the code will be matched to |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2342 // on x86: |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2343 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2344 // Load_narrow_oop memory, narrow_oop_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2345 // Load [R12 + narrow_oop_reg<<3 + offset], val_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2346 // NullCheck narrow_oop_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2347 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2348 // and on sparc: |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2349 // |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2350 // Load_narrow_oop memory, narrow_oop_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2351 // decode_not_null narrow_oop_reg, base_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2352 // Load [base_reg + offset], val_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2353 // NullCheck base_reg |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2354 // |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2355 } else if (t->isa_oopptr()) { |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2356 new_in2 = ConNode::make(C, t->make_narrowoop()); |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2357 } |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2358 } |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2359 if (new_in2 != NULL) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2360 Node* cmpN = new (C, 3) CmpNNode(in1->in(1), new_in2); |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2361 n->subsume_by( cmpN ); |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2362 if (in1->outcnt() == 0) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2363 in1->disconnect_inputs(NULL); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2364 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2365 if (in2->outcnt() == 0) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2366 in2->disconnect_inputs(NULL); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2367 } |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2368 } |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2369 } |
293
c3e045194476
6731641: assert(m->adr_type() == mach->adr_type(),"matcher should not change adr type")
kvn
parents:
253
diff
changeset
|
2370 break; |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2371 |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2372 case Op_DecodeN: |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2373 assert(!n->in(1)->is_EncodeP(), "should be optimized out"); |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2374 // DecodeN could be pinned when it can't be fold into |
492
5496e074077f
6787050: assert(n->in(0) == 0L,"no control") with UseCompressedOops on sparcv9
kvn
parents:
490
diff
changeset
|
2375 // an address expression, see the code for Op_CastPP above. |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2376 assert(n->in(0) == NULL || !Matcher::narrow_oop_use_complex_address(), "no control"); |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2377 break; |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2378 |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2379 case Op_EncodeP: { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2380 Node* in1 = n->in(1); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2381 if (in1->is_DecodeN()) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2382 n->subsume_by(in1->in(1)); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2383 } else if (in1->Opcode() == Op_ConP) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2384 Compile* C = Compile::current(); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2385 const Type* t = in1->bottom_type(); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2386 if (t == TypePtr::NULL_PTR) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2387 n->subsume_by(ConNode::make(C, TypeNarrowOop::NULL_PTR)); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2388 } else if (t->isa_oopptr()) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2389 n->subsume_by(ConNode::make(C, t->make_narrowoop())); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2390 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2391 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2392 if (in1->outcnt() == 0) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2393 in1->disconnect_inputs(NULL); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2394 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2395 break; |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2396 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2397 |
1080
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2398 case Op_Proj: { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2399 if (OptimizeStringConcat) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2400 ProjNode* p = n->as_Proj(); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2401 if (p->_is_io_use) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2402 // Separate projections were used for the exception path which |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2403 // are normally removed by a late inline. If it wasn't inlined |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2404 // then they will hang around and should just be replaced with |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2405 // the original one. |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2406 Node* proj = NULL; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2407 // Replace with just one |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2408 for (SimpleDUIterator i(p->in(0)); i.has_next(); i.next()) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2409 Node *use = i.get(); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2410 if (use->is_Proj() && p != use && use->as_Proj()->_con == p->_con) { |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2411 proj = use; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2412 break; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2413 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2414 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2415 assert(p != NULL, "must be found"); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2416 p->subsume_by(proj); |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2417 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2418 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2419 break; |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2420 } |
7c57aead6d3e
6892658: C2 should optimize some stringbuilder patterns
never
parents:
929
diff
changeset
|
2421 |
368
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2422 case Op_Phi: |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2423 if (n->as_Phi()->bottom_type()->isa_narrowoop()) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2424 // The EncodeP optimization may create Phi with the same edges |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2425 // for all paths. It is not handled well by Register Allocator. |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2426 Node* unique_in = n->in(1); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2427 assert(unique_in != NULL, ""); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2428 uint cnt = n->req(); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2429 for (uint i = 2; i < cnt; i++) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2430 Node* m = n->in(i); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2431 assert(m != NULL, ""); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2432 if (unique_in != m) |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2433 unique_in = NULL; |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2434 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2435 if (unique_in != NULL) { |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2436 n->subsume_by(unique_in); |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2437 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2438 } |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2439 break; |
36ccc817fca4
6747051: Improve code and implicit null check generation for compressed oops
kvn
parents:
367
diff
changeset
|
2440 |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2441 #endif |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
2442 |
0 | 2443 case Op_ModI: |
2444 if (UseDivMod) { | |
2445 // Check if a%b and a/b both exist | |
2446 Node* d = n->find_similar(Op_DivI); | |
2447 if (d) { | |
2448 // Replace them with a fused divmod if supported | |
2449 Compile* C = Compile::current(); | |
2450 if (Matcher::has_match_rule(Op_DivModI)) { | |
2451 DivModINode* divmod = DivModINode::make(C, n); | |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2452 d->subsume_by(divmod->div_proj()); |
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2453 n->subsume_by(divmod->mod_proj()); |
0 | 2454 } else { |
2455 // replace a%b with a-((a/b)*b) | |
2456 Node* mult = new (C, 3) MulINode(d, d->in(2)); | |
2457 Node* sub = new (C, 3) SubINode(d->in(1), mult); | |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2458 n->subsume_by( sub ); |
0 | 2459 } |
2460 } | |
2461 } | |
2462 break; | |
2463 | |
2464 case Op_ModL: | |
2465 if (UseDivMod) { | |
2466 // Check if a%b and a/b both exist | |
2467 Node* d = n->find_similar(Op_DivL); | |
2468 if (d) { | |
2469 // Replace them with a fused divmod if supported | |
2470 Compile* C = Compile::current(); | |
2471 if (Matcher::has_match_rule(Op_DivModL)) { | |
2472 DivModLNode* divmod = DivModLNode::make(C, n); | |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2473 d->subsume_by(divmod->div_proj()); |
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2474 n->subsume_by(divmod->mod_proj()); |
0 | 2475 } else { |
2476 // replace a%b with a-((a/b)*b) | |
2477 Node* mult = new (C, 3) MulLNode(d, d->in(2)); | |
2478 Node* sub = new (C, 3) SubLNode(d->in(1), mult); | |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2479 n->subsume_by( sub ); |
0 | 2480 } |
2481 } | |
2482 } | |
2483 break; | |
2484 | |
2485 case Op_Load16B: | |
2486 case Op_Load8B: | |
2487 case Op_Load4B: | |
2488 case Op_Load8S: | |
2489 case Op_Load4S: | |
2490 case Op_Load2S: | |
2491 case Op_Load8C: | |
2492 case Op_Load4C: | |
2493 case Op_Load2C: | |
2494 case Op_Load4I: | |
2495 case Op_Load2I: | |
2496 case Op_Load2L: | |
2497 case Op_Load4F: | |
2498 case Op_Load2F: | |
2499 case Op_Load2D: | |
2500 case Op_Store16B: | |
2501 case Op_Store8B: | |
2502 case Op_Store4B: | |
2503 case Op_Store8C: | |
2504 case Op_Store4C: | |
2505 case Op_Store2C: | |
2506 case Op_Store4I: | |
2507 case Op_Store2I: | |
2508 case Op_Store2L: | |
2509 case Op_Store4F: | |
2510 case Op_Store2F: | |
2511 case Op_Store2D: | |
2512 break; | |
2513 | |
2514 case Op_PackB: | |
2515 case Op_PackS: | |
2516 case Op_PackC: | |
2517 case Op_PackI: | |
2518 case Op_PackF: | |
2519 case Op_PackL: | |
2520 case Op_PackD: | |
2521 if (n->req()-1 > 2) { | |
2522 // Replace many operand PackNodes with a binary tree for matching | |
2523 PackNode* p = (PackNode*) n; | |
2524 Node* btp = p->binaryTreePack(Compile::current(), 1, n->req()); | |
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
164
diff
changeset
|
2525 n->subsume_by(btp); |
0 | 2526 } |
2527 break; | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2528 case Op_Loop: |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2529 case Op_CountedLoop: |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2530 if (n->as_Loop()->is_inner_loop()) { |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2531 frc.inc_inner_loop_count(); |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2532 } |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2533 break; |
0 | 2534 default: |
2535 assert( !n->is_Call(), "" ); | |
2536 assert( !n->is_Mem(), "" ); | |
2537 break; | |
2538 } | |
127 | 2539 |
2540 // Collect CFG split points | |
2541 if (n->is_MultiBranch()) | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2542 frc._tests.push(n); |
0 | 2543 } |
2544 | |
2545 //------------------------------final_graph_reshaping_walk--------------------- | |
2546 // Replacing Opaque nodes with their input in final_graph_reshaping_impl(), | |
2547 // requires that the walk visits a node's inputs before visiting the node. | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2548 static void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc ) { |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2549 ResourceArea *area = Thread::current()->resource_area(); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2550 Unique_Node_List sfpt(area); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2551 |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2552 frc._visited.set(root->_idx); // first, mark node as visited |
0 | 2553 uint cnt = root->req(); |
2554 Node *n = root; | |
2555 uint i = 0; | |
2556 while (true) { | |
2557 if (i < cnt) { | |
2558 // Place all non-visited non-null inputs onto stack | |
2559 Node* m = n->in(i); | |
2560 ++i; | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2561 if (m != NULL && !frc._visited.test_set(m->_idx)) { |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2562 if (m->is_SafePoint() && m->as_SafePoint()->jvms() != NULL) |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2563 sfpt.push(m); |
0 | 2564 cnt = m->req(); |
2565 nstack.push(n, i); // put on stack parent and next input's index | |
2566 n = m; | |
2567 i = 0; | |
2568 } | |
2569 } else { | |
2570 // Now do post-visit work | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2571 final_graph_reshaping_impl( n, frc ); |
0 | 2572 if (nstack.is_empty()) |
2573 break; // finished | |
2574 n = nstack.node(); // Get node from stack | |
2575 cnt = n->req(); | |
2576 i = nstack.index(); | |
2577 nstack.pop(); // Shift to the next node on stack | |
2578 } | |
2579 } | |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2580 |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2581 // Skip next transformation if compressed oops are not used. |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2582 if (!UseCompressedOops || !Matcher::gen_narrow_oop_implicit_null_checks()) |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2583 return; |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1397
diff
changeset
|
2584 |
331
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2585 // Go over safepoints nodes to skip DecodeN nodes for debug edges. |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2586 // It could be done for an uncommon traps or any safepoints/calls |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2587 // if the DecodeN node is referenced only in a debug info. |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2588 while (sfpt.size() > 0) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2589 n = sfpt.pop(); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2590 JVMState *jvms = n->as_SafePoint()->jvms(); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2591 assert(jvms != NULL, "sanity"); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2592 int start = jvms->debug_start(); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2593 int end = n->req(); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2594 bool is_uncommon = (n->is_CallStaticJava() && |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2595 n->as_CallStaticJava()->uncommon_trap_request() != 0); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2596 for (int j = start; j < end; j++) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2597 Node* in = n->in(j); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2598 if (in->is_DecodeN()) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2599 bool safe_to_skip = true; |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2600 if (!is_uncommon ) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2601 // Is it safe to skip? |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2602 for (uint i = 0; i < in->outcnt(); i++) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2603 Node* u = in->raw_out(i); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2604 if (!u->is_SafePoint() || |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2605 u->is_Call() && u->as_Call()->has_non_debug_use(n)) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2606 safe_to_skip = false; |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2607 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2608 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2609 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2610 if (safe_to_skip) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2611 n->set_req(j, in->in(1)); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2612 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2613 if (in->outcnt() == 0) { |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2614 in->disconnect_inputs(NULL); |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2615 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2616 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2617 } |
cecd8eb4e0ca
6706829: Compressed Oops: add debug info for narrow oops
kvn
parents:
293
diff
changeset
|
2618 } |
0 | 2619 } |
2620 | |
2621 //------------------------------final_graph_reshaping-------------------------- | |
2622 // Final Graph Reshaping. | |
2623 // | |
2624 // (1) Clone simple inputs to uncommon calls, so they can be scheduled late | |
2625 // and not commoned up and forced early. Must come after regular | |
2626 // optimizations to avoid GVN undoing the cloning. Clone constant | |
2627 // inputs to Loop Phis; these will be split by the allocator anyways. | |
2628 // Remove Opaque nodes. | |
2629 // (2) Move last-uses by commutative operations to the left input to encourage | |
2630 // Intel update-in-place two-address operations and better register usage | |
2631 // on RISCs. Must come after regular optimizations to avoid GVN Ideal | |
2632 // calls canonicalizing them back. | |
2633 // (3) Count the number of double-precision FP ops, single-precision FP ops | |
2634 // and call sites. On Intel, we can get correct rounding either by | |
2635 // forcing singles to memory (requires extra stores and loads after each | |
2636 // FP bytecode) or we can set a rounding mode bit (requires setting and | |
2637 // clearing the mode bit around call sites). The mode bit is only used | |
2638 // if the relative frequency of single FP ops to calls is low enough. | |
2639 // This is a key transform for SPEC mpeg_audio. | |
2640 // (4) Detect infinite loops; blobs of code reachable from above but not | |
2641 // below. Several of the Code_Gen algorithms fail on such code shapes, | |
2642 // so we simply bail out. Happens a lot in ZKM.jar, but also happens | |
2643 // from time to time in other codes (such as -Xcomp finalizer loops, etc). | |
2644 // Detection is by looking for IfNodes where only 1 projection is | |
2645 // reachable from below or CatchNodes missing some targets. | |
2646 // (5) Assert for insane oop offsets in debug mode. | |
2647 | |
2648 bool Compile::final_graph_reshaping() { | |
2649 // an infinite loop may have been eliminated by the optimizer, | |
2650 // in which case the graph will be empty. | |
2651 if (root()->req() == 1) { | |
2652 record_method_not_compilable("trivial infinite loop"); | |
2653 return true; | |
2654 } | |
2655 | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2656 Final_Reshape_Counts frc; |
0 | 2657 |
2658 // Visit everybody reachable! | |
2659 // Allocate stack of size C->unique()/2 to avoid frequent realloc | |
2660 Node_Stack nstack(unique() >> 1); | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2661 final_graph_reshaping_walk(nstack, root(), frc); |
0 | 2662 |
2663 // Check for unreachable (from below) code (i.e., infinite loops). | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2664 for( uint i = 0; i < frc._tests.size(); i++ ) { |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2665 MultiBranchNode *n = frc._tests[i]->as_MultiBranch(); |
127 | 2666 // Get number of CFG targets. |
0 | 2667 // Note that PCTables include exception targets after calls. |
127 | 2668 uint required_outcnt = n->required_outcnt(); |
2669 if (n->outcnt() != required_outcnt) { | |
0 | 2670 // Check for a few special cases. Rethrow Nodes never take the |
2671 // 'fall-thru' path, so expected kids is 1 less. | |
2672 if (n->is_PCTable() && n->in(0) && n->in(0)->in(0)) { | |
2673 if (n->in(0)->in(0)->is_Call()) { | |
2674 CallNode *call = n->in(0)->in(0)->as_Call(); | |
2675 if (call->entry_point() == OptoRuntime::rethrow_stub()) { | |
127 | 2676 required_outcnt--; // Rethrow always has 1 less kid |
0 | 2677 } else if (call->req() > TypeFunc::Parms && |
2678 call->is_CallDynamicJava()) { | |
2679 // Check for null receiver. In such case, the optimizer has | |
2680 // detected that the virtual call will always result in a null | |
2681 // pointer exception. The fall-through projection of this CatchNode | |
2682 // will not be populated. | |
2683 Node *arg0 = call->in(TypeFunc::Parms); | |
2684 if (arg0->is_Type() && | |
2685 arg0->as_Type()->type()->higher_equal(TypePtr::NULL_PTR)) { | |
127 | 2686 required_outcnt--; |
0 | 2687 } |
2688 } else if (call->entry_point() == OptoRuntime::new_array_Java() && | |
2689 call->req() > TypeFunc::Parms+1 && | |
2690 call->is_CallStaticJava()) { | |
2691 // Check for negative array length. In such case, the optimizer has | |
2692 // detected that the allocation attempt will always result in an | |
2693 // exception. There is no fall-through projection of this CatchNode . | |
2694 Node *arg1 = call->in(TypeFunc::Parms+1); | |
2695 if (arg1->is_Type() && | |
2696 arg1->as_Type()->type()->join(TypeInt::POS)->empty()) { | |
127 | 2697 required_outcnt--; |
0 | 2698 } |
2699 } | |
2700 } | |
2701 } | |
127 | 2702 // Recheck with a better notion of 'required_outcnt' |
2703 if (n->outcnt() != required_outcnt) { | |
0 | 2704 record_method_not_compilable("malformed control flow"); |
2705 return true; // Not all targets reachable! | |
2706 } | |
2707 } | |
2708 // Check that I actually visited all kids. Unreached kids | |
2709 // must be infinite loops. | |
2710 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2711 if (!frc._visited.test(n->fast_out(j)->_idx)) { |
0 | 2712 record_method_not_compilable("infinite loop"); |
2713 return true; // Found unvisited kid; must be unreach | |
2714 } | |
2715 } | |
2716 | |
2717 // If original bytecodes contained a mixture of floats and doubles | |
2718 // check if the optimizer has made it homogenous, item (3). | |
929
cd18bd5e667c
6873777: FPU control word optimization still performed with SSE
never
parents:
921
diff
changeset
|
2719 if( Use24BitFPMode && Use24BitFP && UseSSE == 0 && |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2720 frc.get_float_count() > 32 && |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2721 frc.get_double_count() == 0 && |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2722 (10 * frc.get_call_count() < frc.get_float_count()) ) { |
0 | 2723 set_24_bit_selection_and_mode( false, true ); |
2724 } | |
2725 | |
859
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2726 set_java_calls(frc.get_java_call_count()); |
ea3f9723b5cf
6860599: nodes limit could be reached during Output phase
kvn
parents:
856
diff
changeset
|
2727 set_inner_loops(frc.get_inner_loop_count()); |
0 | 2728 |
2729 // No infinite loops, no reason to bail out. | |
2730 return false; | |
2731 } | |
2732 | |
2733 //-----------------------------too_many_traps---------------------------------- | |
2734 // Report if there are too many traps at the current method and bci. | |
2735 // Return true if there was a trap, and/or PerMethodTrapLimit is exceeded. | |
2736 bool Compile::too_many_traps(ciMethod* method, | |
2737 int bci, | |
2738 Deoptimization::DeoptReason reason) { | |
2739 ciMethodData* md = method->method_data(); | |
2740 if (md->is_empty()) { | |
2741 // Assume the trap has not occurred, or that it occurred only | |
2742 // because of a transient condition during start-up in the interpreter. | |
2743 return false; | |
2744 } | |
2745 if (md->has_trap_at(bci, reason) != 0) { | |
2746 // Assume PerBytecodeTrapLimit==0, for a more conservative heuristic. | |
2747 // Also, if there are multiple reasons, or if there is no per-BCI record, | |
2748 // assume the worst. | |
2749 if (log()) | |
2750 log()->elem("observe trap='%s' count='%d'", | |
2751 Deoptimization::trap_reason_name(reason), | |
2752 md->trap_count(reason)); | |
2753 return true; | |
2754 } else { | |
2755 // Ignore method/bci and see if there have been too many globally. | |
2756 return too_many_traps(reason, md); | |
2757 } | |
2758 } | |
2759 | |
2760 // Less-accurate variant which does not require a method and bci. | |
2761 bool Compile::too_many_traps(Deoptimization::DeoptReason reason, | |
2762 ciMethodData* logmd) { | |
2763 if (trap_count(reason) >= (uint)PerMethodTrapLimit) { | |
2764 // Too many traps globally. | |
2765 // Note that we use cumulative trap_count, not just md->trap_count. | |
2766 if (log()) { | |
2767 int mcount = (logmd == NULL)? -1: (int)logmd->trap_count(reason); | |
2768 log()->elem("observe trap='%s' count='0' mcount='%d' ccount='%d'", | |
2769 Deoptimization::trap_reason_name(reason), | |
2770 mcount, trap_count(reason)); | |
2771 } | |
2772 return true; | |
2773 } else { | |
2774 // The coast is clear. | |
2775 return false; | |
2776 } | |
2777 } | |
2778 | |
2779 //--------------------------too_many_recompiles-------------------------------- | |
2780 // Report if there are too many recompiles at the current method and bci. | |
2781 // Consults PerBytecodeRecompilationCutoff and PerMethodRecompilationCutoff. | |
2782 // Is not eager to return true, since this will cause the compiler to use | |
2783 // Action_none for a trap point, to avoid too many recompilations. | |
2784 bool Compile::too_many_recompiles(ciMethod* method, | |
2785 int bci, | |
2786 Deoptimization::DeoptReason reason) { | |
2787 ciMethodData* md = method->method_data(); | |
2788 if (md->is_empty()) { | |
2789 // Assume the trap has not occurred, or that it occurred only | |
2790 // because of a transient condition during start-up in the interpreter. | |
2791 return false; | |
2792 } | |
2793 // Pick a cutoff point well within PerBytecodeRecompilationCutoff. | |
2794 uint bc_cutoff = (uint) PerBytecodeRecompilationCutoff / 8; | |
2795 uint m_cutoff = (uint) PerMethodRecompilationCutoff / 2 + 1; // not zero | |
2796 Deoptimization::DeoptReason per_bc_reason | |
2797 = Deoptimization::reason_recorded_per_bytecode_if_any(reason); | |
2798 if ((per_bc_reason == Deoptimization::Reason_none | |
2799 || md->has_trap_at(bci, reason) != 0) | |
2800 // The trap frequency measure we care about is the recompile count: | |
2801 && md->trap_recompiled_at(bci) | |
2802 && md->overflow_recompile_count() >= bc_cutoff) { | |
2803 // Do not emit a trap here if it has already caused recompilations. | |
2804 // Also, if there are multiple reasons, or if there is no per-BCI record, | |
2805 // assume the worst. | |
2806 if (log()) | |
2807 log()->elem("observe trap='%s recompiled' count='%d' recompiles2='%d'", | |
2808 Deoptimization::trap_reason_name(reason), | |
2809 md->trap_count(reason), | |
2810 md->overflow_recompile_count()); | |
2811 return true; | |
2812 } else if (trap_count(reason) != 0 | |
2813 && decompile_count() >= m_cutoff) { | |
2814 // Too many recompiles globally, and we have seen this sort of trap. | |
2815 // Use cumulative decompile_count, not just md->decompile_count. | |
2816 if (log()) | |
2817 log()->elem("observe trap='%s' count='%d' mcount='%d' decompiles='%d' mdecompiles='%d'", | |
2818 Deoptimization::trap_reason_name(reason), | |
2819 md->trap_count(reason), trap_count(reason), | |
2820 md->decompile_count(), decompile_count()); | |
2821 return true; | |
2822 } else { | |
2823 // The coast is clear. | |
2824 return false; | |
2825 } | |
2826 } | |
2827 | |
2828 | |
2829 #ifndef PRODUCT | |
2830 //------------------------------verify_graph_edges--------------------------- | |
2831 // Walk the Graph and verify that there is a one-to-one correspondence | |
2832 // between Use-Def edges and Def-Use edges in the graph. | |
2833 void Compile::verify_graph_edges(bool no_dead_code) { | |
2834 if (VerifyGraphEdges) { | |
2835 ResourceArea *area = Thread::current()->resource_area(); | |
2836 Unique_Node_List visited(area); | |
2837 // Call recursive graph walk to check edges | |
2838 _root->verify_edges(visited); | |
2839 if (no_dead_code) { | |
2840 // Now make sure that no visited node is used by an unvisited node. | |
2841 bool dead_nodes = 0; | |
2842 Unique_Node_List checked(area); | |
2843 while (visited.size() > 0) { | |
2844 Node* n = visited.pop(); | |
2845 checked.push(n); | |
2846 for (uint i = 0; i < n->outcnt(); i++) { | |
2847 Node* use = n->raw_out(i); | |
2848 if (checked.member(use)) continue; // already checked | |
2849 if (visited.member(use)) continue; // already in the graph | |
2850 if (use->is_Con()) continue; // a dead ConNode is OK | |
2851 // At this point, we have found a dead node which is DU-reachable. | |
2852 if (dead_nodes++ == 0) | |
2853 tty->print_cr("*** Dead nodes reachable via DU edges:"); | |
2854 use->dump(2); | |
2855 tty->print_cr("---"); | |
2856 checked.push(use); // No repeats; pretend it is now checked. | |
2857 } | |
2858 } | |
2859 assert(dead_nodes == 0, "using nodes must be reachable from root"); | |
2860 } | |
2861 } | |
2862 } | |
2863 #endif | |
2864 | |
2865 // The Compile object keeps track of failure reasons separately from the ciEnv. | |
2866 // This is required because there is not quite a 1-1 relation between the | |
2867 // ciEnv and its compilation task and the Compile object. Note that one | |
2868 // ciEnv might use two Compile objects, if C2Compiler::compile_method decides | |
2869 // to backtrack and retry without subsuming loads. Other than this backtracking | |
2870 // behavior, the Compile's failure reason is quietly copied up to the ciEnv | |
2871 // by the logic in C2Compiler. | |
2872 void Compile::record_failure(const char* reason) { | |
2873 if (log() != NULL) { | |
2874 log()->elem("failure reason='%s' phase='compile'", reason); | |
2875 } | |
2876 if (_failure_reason == NULL) { | |
2877 // Record the first failure reason. | |
2878 _failure_reason = reason; | |
2879 } | |
222 | 2880 if (!C->failure_reason_is(C2Compiler::retry_no_subsuming_loads())) { |
2881 C->print_method(_failure_reason); | |
2882 } | |
0 | 2883 _root = NULL; // flush the graph, too |
2884 } | |
2885 | |
2886 Compile::TracePhase::TracePhase(const char* name, elapsedTimer* accumulator, bool dolog) | |
2887 : TraceTime(NULL, accumulator, false NOT_PRODUCT( || TimeCompiler ), false) | |
2888 { | |
2889 if (dolog) { | |
2890 C = Compile::current(); | |
2891 _log = C->log(); | |
2892 } else { | |
2893 C = NULL; | |
2894 _log = NULL; | |
2895 } | |
2896 if (_log != NULL) { | |
2897 _log->begin_head("phase name='%s' nodes='%d'", name, C->unique()); | |
2898 _log->stamp(); | |
2899 _log->end_head(); | |
2900 } | |
2901 } | |
2902 | |
2903 Compile::TracePhase::~TracePhase() { | |
2904 if (_log != NULL) { | |
2905 _log->done("phase nodes='%d'", C->unique()); | |
2906 } | |
2907 } | |
2008 | 2908 |
2909 //============================================================================= | |
2910 // Two Constant's are equal when the type and the value are equal. | |
2911 bool Compile::Constant::operator==(const Constant& other) { | |
2912 if (type() != other.type() ) return false; | |
2913 if (can_be_reused() != other.can_be_reused()) return false; | |
2914 // For floating point values we compare the bit pattern. | |
2915 switch (type()) { | |
2916 case T_FLOAT: return (_value.i == other._value.i); | |
2917 case T_LONG: | |
2918 case T_DOUBLE: return (_value.j == other._value.j); | |
2919 case T_OBJECT: | |
2920 case T_ADDRESS: return (_value.l == other._value.l); | |
2921 case T_VOID: return (_value.l == other._value.l); // jump-table entries | |
2922 default: ShouldNotReachHere(); | |
2923 } | |
2924 return false; | |
2925 } | |
2926 | |
2927 // Emit constants grouped in the following order: | |
2928 static BasicType type_order[] = { | |
2929 T_FLOAT, // 32-bit | |
2930 T_OBJECT, // 32 or 64-bit | |
2931 T_ADDRESS, // 32 or 64-bit | |
2932 T_DOUBLE, // 64-bit | |
2933 T_LONG, // 64-bit | |
2934 T_VOID, // 32 or 64-bit (jump-tables are at the end of the constant table for code emission reasons) | |
2935 T_ILLEGAL | |
2936 }; | |
2937 | |
2938 static int type_to_size_in_bytes(BasicType t) { | |
2939 switch (t) { | |
2940 case T_LONG: return sizeof(jlong ); | |
2941 case T_FLOAT: return sizeof(jfloat ); | |
2942 case T_DOUBLE: return sizeof(jdouble); | |
2943 // We use T_VOID as marker for jump-table entries (labels) which | |
2944 // need an interal word relocation. | |
2945 case T_VOID: | |
2946 case T_ADDRESS: | |
2947 case T_OBJECT: return sizeof(jobject); | |
2948 } | |
2949 | |
2950 ShouldNotReachHere(); | |
2951 return -1; | |
2952 } | |
2953 | |
2954 void Compile::ConstantTable::calculate_offsets_and_size() { | |
2955 int size = 0; | |
2956 for (int t = 0; type_order[t] != T_ILLEGAL; t++) { | |
2957 BasicType type = type_order[t]; | |
2958 | |
2959 for (int i = 0; i < _constants.length(); i++) { | |
2960 Constant con = _constants.at(i); | |
2961 if (con.type() != type) continue; // Skip other types. | |
2962 | |
2963 // Align size for type. | |
2964 int typesize = type_to_size_in_bytes(con.type()); | |
2965 size = align_size_up(size, typesize); | |
2966 | |
2967 // Set offset. | |
2968 con.set_offset(size); | |
2969 _constants.at_put(i, con); | |
2970 | |
2971 // Add type size. | |
2972 size = size + typesize; | |
2973 } | |
2974 } | |
2975 | |
2976 // Align size up to the next section start (which is insts; see | |
2977 // CodeBuffer::align_at_start). | |
2978 assert(_size == -1, "already set?"); | |
2979 _size = align_size_up(size, CodeEntryAlignment); | |
2980 | |
2981 if (Matcher::constant_table_absolute_addressing) { | |
2982 set_table_base_offset(0); // No table base offset required | |
2983 } else { | |
2984 if (UseRDPCForConstantTableBase) { | |
2985 // table base offset is set in MachConstantBaseNode::emit | |
2986 } else { | |
2987 // When RDPC is not used, the table base is set into the middle of | |
2988 // the constant table. | |
2989 int half_size = _size / 2; | |
2990 assert(half_size * 2 == _size, "sanity"); | |
2991 set_table_base_offset(-half_size); | |
2992 } | |
2993 } | |
2994 } | |
2995 | |
2996 void Compile::ConstantTable::emit(CodeBuffer& cb) { | |
2997 MacroAssembler _masm(&cb); | |
2998 for (int t = 0; type_order[t] != T_ILLEGAL; t++) { | |
2999 BasicType type = type_order[t]; | |
3000 | |
3001 for (int i = 0; i < _constants.length(); i++) { | |
3002 Constant con = _constants.at(i); | |
3003 if (con.type() != type) continue; // Skip other types. | |
3004 | |
3005 address constant_addr; | |
3006 switch (con.type()) { | |
3007 case T_LONG: constant_addr = _masm.long_constant( con.get_jlong() ); break; | |
3008 case T_FLOAT: constant_addr = _masm.float_constant( con.get_jfloat() ); break; | |
3009 case T_DOUBLE: constant_addr = _masm.double_constant(con.get_jdouble()); break; | |
3010 case T_OBJECT: { | |
3011 jobject obj = con.get_jobject(); | |
3012 int oop_index = _masm.oop_recorder()->find_index(obj); | |
3013 constant_addr = _masm.address_constant((address) obj, oop_Relocation::spec(oop_index)); | |
3014 break; | |
3015 } | |
3016 case T_ADDRESS: { | |
3017 address addr = (address) con.get_jobject(); | |
3018 constant_addr = _masm.address_constant(addr); | |
3019 break; | |
3020 } | |
3021 // We use T_VOID as marker for jump-table entries (labels) which | |
3022 // need an interal word relocation. | |
3023 case T_VOID: { | |
3024 // Write a dummy word. The real value is filled in later | |
3025 // in fill_jump_table_in_constant_table. | |
3026 address addr = (address) con.get_jobject(); | |
3027 constant_addr = _masm.address_constant(addr); | |
3028 break; | |
3029 } | |
3030 default: ShouldNotReachHere(); | |
3031 } | |
3032 assert(constant_addr != NULL, "consts section too small"); | |
3033 assert((constant_addr - _masm.code()->consts()->start()) == con.offset(), err_msg("must be: %d == %d", constant_addr - _masm.code()->consts()->start(), con.offset())); | |
3034 } | |
3035 } | |
3036 } | |
3037 | |
3038 int Compile::ConstantTable::find_offset(Constant& con) const { | |
3039 int idx = _constants.find(con); | |
3040 assert(idx != -1, "constant must be in constant table"); | |
3041 int offset = _constants.at(idx).offset(); | |
3042 assert(offset != -1, "constant table not emitted yet?"); | |
3043 return offset; | |
3044 } | |
3045 | |
3046 void Compile::ConstantTable::add(Constant& con) { | |
3047 if (con.can_be_reused()) { | |
3048 int idx = _constants.find(con); | |
3049 if (idx != -1 && _constants.at(idx).can_be_reused()) { | |
3050 return; | |
3051 } | |
3052 } | |
3053 (void) _constants.append(con); | |
3054 } | |
3055 | |
3056 Compile::Constant Compile::ConstantTable::add(BasicType type, jvalue value) { | |
3057 Constant con(type, value); | |
3058 add(con); | |
3059 return con; | |
3060 } | |
3061 | |
3062 Compile::Constant Compile::ConstantTable::add(MachOper* oper) { | |
3063 jvalue value; | |
3064 BasicType type = oper->type()->basic_type(); | |
3065 switch (type) { | |
3066 case T_LONG: value.j = oper->constantL(); break; | |
3067 case T_FLOAT: value.f = oper->constantF(); break; | |
3068 case T_DOUBLE: value.d = oper->constantD(); break; | |
3069 case T_OBJECT: | |
3070 case T_ADDRESS: value.l = (jobject) oper->constant(); break; | |
3071 default: ShouldNotReachHere(); | |
3072 } | |
3073 return add(type, value); | |
3074 } | |
3075 | |
3076 Compile::Constant Compile::ConstantTable::allocate_jump_table(MachConstantNode* n) { | |
3077 jvalue value; | |
3078 // We can use the node pointer here to identify the right jump-table | |
3079 // as this method is called from Compile::Fill_buffer right before | |
3080 // the MachNodes are emitted and the jump-table is filled (means the | |
3081 // MachNode pointers do not change anymore). | |
3082 value.l = (jobject) n; | |
3083 Constant con(T_VOID, value, false); // Labels of a jump-table cannot be reused. | |
3084 for (uint i = 0; i < n->outcnt(); i++) { | |
3085 add(con); | |
3086 } | |
3087 return con; | |
3088 } | |
3089 | |
3090 void Compile::ConstantTable::fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const { | |
3091 // If called from Compile::scratch_emit_size do nothing. | |
3092 if (Compile::current()->in_scratch_emit_size()) return; | |
3093 | |
3094 assert(labels.is_nonempty(), "must be"); | |
3095 assert((uint) labels.length() == n->outcnt(), err_msg("must be equal: %d == %d", labels.length(), n->outcnt())); | |
3096 | |
3097 // Since MachConstantNode::constant_offset() also contains | |
3098 // table_base_offset() we need to subtract the table_base_offset() | |
3099 // to get the plain offset into the constant table. | |
3100 int offset = n->constant_offset() - table_base_offset(); | |
3101 | |
3102 MacroAssembler _masm(&cb); | |
3103 address* jump_table_base = (address*) (_masm.code()->consts()->start() + offset); | |
3104 | |
3105 for (int i = 0; i < labels.length(); i++) { | |
3106 address* constant_addr = &jump_table_base[i]; | |
3107 assert(*constant_addr == (address) n, "all jump-table entries must contain node pointer"); | |
3108 *constant_addr = cb.consts()->target(*labels.at(i), (address) constant_addr); | |
3109 cb.consts()->relocate((address) constant_addr, relocInfo::internal_word_type); | |
3110 } | |
3111 } |