1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "opto/addnode.hpp"
34 #include "opto/arraycopynode.hpp"
35 #include "opto/c2compiler.hpp"
36 #include "opto/callGenerator.hpp"
37 #include "opto/castnode.hpp"
38 #include "opto/cfgnode.hpp"
39 #include "opto/convertnode.hpp"
40 #include "opto/countbitsnode.hpp"
41 #include "opto/intrinsicnode.hpp"
42 #include "opto/idealKit.hpp"
43 #include "opto/mathexactnode.hpp"
44 #include "opto/movenode.hpp"
45 #include "opto/mulnode.hpp"
46 #include "opto/narrowptrnode.hpp"
47 #include "opto/opaquenode.hpp"
48 #include "opto/parse.hpp"
49 #include "opto/runtime.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/subnode.hpp"
52 #include "prims/nativeLookup.hpp"
53 #include "prims/unsafe.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #ifdef TRACE_HAVE_INTRINSICS
56 #include "trace/traceMacros.hpp"
57 #endif
58
59 class LibraryIntrinsic : public InlineCallGenerator {
60 // Extend the set of intrinsics known to the runtime:
61 public:
62 private:
63 bool _is_virtual;
64 bool _does_virtual_dispatch;
65 int8_t _predicates_count; // Intrinsic is predicated by several conditions
66 int8_t _last_predicate; // Last generated predicate
67 vmIntrinsics::ID _intrinsic_id;
68
69 public:
70 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
71 : InlineCallGenerator(m),
72 _is_virtual(is_virtual),
73 _does_virtual_dispatch(does_virtual_dispatch),
74 _predicates_count((int8_t)predicates_count),
75 _last_predicate((int8_t)-1),
76 _intrinsic_id(id)
77 {
78 }
79 virtual bool is_intrinsic() const { return true; }
80 virtual bool is_virtual() const { return _is_virtual; }
81 virtual bool is_predicated() const { return _predicates_count > 0; }
82 virtual int predicates_count() const { return _predicates_count; }
83 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
84 virtual JVMState* generate(JVMState* jvms);
85 virtual Node* generate_predicate(JVMState* jvms, int predicate);
86 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
87 };
88
89
90 // Local helper class for LibraryIntrinsic:
91 class LibraryCallKit : public GraphKit {
92 private:
93 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
94 Node* _result; // the result node, if any
95 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
96
97 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
98
99 public:
100 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
101 : GraphKit(jvms),
102 _intrinsic(intrinsic),
103 _result(NULL)
104 {
105 // Check if this is a root compile. In that case we don't have a caller.
106 if (!jvms->has_method()) {
107 _reexecute_sp = sp();
108 } else {
109 // Find out how many arguments the interpreter needs when deoptimizing
110 // and save the stack pointer value so it can used by uncommon_trap.
111 // We find the argument count by looking at the declared signature.
112 bool ignored_will_link;
113 ciSignature* declared_signature = NULL;
114 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
115 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
116 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
117 }
118 }
119
120 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
121
122 ciMethod* caller() const { return jvms()->method(); }
123 int bci() const { return jvms()->bci(); }
124 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
125 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
126 ciMethod* callee() const { return _intrinsic->method(); }
127
128 bool try_to_inline(int predicate);
129 Node* try_to_predicate(int predicate);
130
131 void push_result() {
132 // Push the result onto the stack.
133 if (!stopped() && result() != NULL) {
134 BasicType bt = result()->bottom_type()->basic_type();
135 push_node(bt, result());
136 }
137 }
138
139 private:
140 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
141 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
142 }
143
144 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
145 void set_result(RegionNode* region, PhiNode* value);
146 Node* result() { return _result; }
147
148 virtual int reexecute_sp() { return _reexecute_sp; }
149
150 // Helper functions to inline natives
151 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
152 Node* generate_slow_guard(Node* test, RegionNode* region);
153 Node* generate_fair_guard(Node* test, RegionNode* region);
154 Node* generate_negative_guard(Node* index, RegionNode* region,
155 // resulting CastII of index:
156 Node* *pos_index = NULL);
157 Node* generate_limit_guard(Node* offset, Node* subseq_length,
158 Node* array_length,
159 RegionNode* region);
160 void generate_string_range_check(Node* array, Node* offset,
161 Node* length, bool char_count);
162 Node* generate_current_thread(Node* &tls_output);
163 Node* load_mirror_from_klass(Node* klass);
164 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
165 RegionNode* region, int null_path,
166 int offset);
167 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path) {
169 int offset = java_lang_Class::klass_offset_in_bytes();
170 return load_klass_from_mirror_common(mirror, never_see_null,
171 region, null_path,
172 offset);
173 }
174 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
175 RegionNode* region, int null_path) {
176 int offset = java_lang_Class::array_klass_offset_in_bytes();
177 return load_klass_from_mirror_common(mirror, never_see_null,
178 region, null_path,
179 offset);
180 }
181 Node* generate_access_flags_guard(Node* kls,
182 int modifier_mask, int modifier_bits,
183 RegionNode* region);
184 Node* generate_interface_guard(Node* kls, RegionNode* region);
185 Node* generate_array_guard(Node* kls, RegionNode* region) {
186 return generate_array_guard_common(kls, region, false, false);
187 }
188 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
189 return generate_array_guard_common(kls, region, false, true);
190 }
191 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
192 return generate_array_guard_common(kls, region, true, false);
193 }
194 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
195 return generate_array_guard_common(kls, region, true, true);
196 }
197 Node* generate_array_guard_common(Node* kls, RegionNode* region,
198 bool obj_array, bool not_array);
199 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
200 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
201 bool is_virtual = false, bool is_static = false);
202 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
203 return generate_method_call(method_id, false, true);
204 }
205 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
206 return generate_method_call(method_id, true, false);
207 }
208 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
209 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
210
211 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
213 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
214 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
215 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
216 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOfChar();
218 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
219 bool inline_string_toBytesU();
220 bool inline_string_getCharsU();
221 bool inline_string_copy(bool compress);
222 bool inline_string_char_access(bool is_store);
223 Node* round_double_node(Node* n);
224 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
225 bool inline_math_native(vmIntrinsics::ID id);
226 bool inline_math(vmIntrinsics::ID id);
227 template <typename OverflowOp>
228 bool inline_math_overflow(Node* arg1, Node* arg2);
229 void inline_math_mathExact(Node* math, Node* test);
230 bool inline_math_addExactI(bool is_increment);
231 bool inline_math_addExactL(bool is_increment);
232 bool inline_math_multiplyExactI();
233 bool inline_math_multiplyExactL();
234 bool inline_math_negateExactI();
235 bool inline_math_negateExactL();
236 bool inline_math_subtractExactI(bool is_decrement);
237 bool inline_math_subtractExactL(bool is_decrement);
238 bool inline_min_max(vmIntrinsics::ID id);
239 bool inline_notify(vmIntrinsics::ID id);
240 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
241 // This returns Type::AnyPtr, RawPtr, or OopPtr.
242 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
243 Node* make_unsafe_address(Node*& base, Node* offset, BasicType type);
244 // Helper for inline_unsafe_access.
245 // Generates the guards that check whether the result of
246 // Unsafe.getObject should be recorded in an SATB log buffer.
247 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
248
249 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
250 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
251 static bool klass_needs_init_guard(Node* kls);
252 bool inline_unsafe_allocate();
253 bool inline_unsafe_newArray(bool uninitialized);
254 bool inline_unsafe_copyMemory();
255 bool inline_native_currentThread();
256
257 bool inline_native_time_funcs(address method, const char* funcName);
258 #ifdef TRACE_HAVE_INTRINSICS
259 bool inline_native_classID();
260 bool inline_native_getBufferWriter();
261 #endif
262 bool inline_native_isInterrupted();
263 bool inline_native_Class_query(vmIntrinsics::ID id);
264 bool inline_native_subtype_check();
265 bool inline_native_getLength();
266 bool inline_array_copyOf(bool is_copyOfRange);
267 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
268 bool inline_preconditions_checkIndex();
269 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
270 bool inline_native_clone(bool is_virtual);
271 bool inline_native_Reflection_getCallerClass();
272 // Helper function for inlining native object hash method
273 bool inline_native_hashcode(bool is_virtual, bool is_static);
274 bool inline_native_getClass();
275
276 // Helper functions for inlining arraycopy
277 bool inline_arraycopy();
278 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
279 RegionNode* slow_region);
280 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
281 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
282
283 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
284 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
285 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
286 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
287 bool inline_unsafe_fence(vmIntrinsics::ID id);
288 bool inline_onspinwait();
289 bool inline_fp_conversions(vmIntrinsics::ID id);
290 bool inline_number_methods(vmIntrinsics::ID id);
291 bool inline_reference_get();
292 bool inline_Class_cast();
293 bool inline_aescrypt_Block(vmIntrinsics::ID id);
294 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
295 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
296 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
297 Node* inline_counterMode_AESCrypt_predicate();
298 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
299 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
300 bool inline_ghash_processBlocks();
301 bool inline_sha_implCompress(vmIntrinsics::ID id);
302 bool inline_digestBase_implCompressMB(int predicate);
303 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
304 bool long_state, address stubAddr, const char *stubName,
305 Node* src_start, Node* ofs, Node* limit);
306 Node* get_state_from_sha_object(Node *sha_object);
307 Node* get_state_from_sha5_object(Node *sha_object);
308 Node* inline_digestBase_implCompressMB_predicate(int predicate);
309 bool inline_encodeISOArray();
310 bool inline_updateCRC32();
311 bool inline_updateBytesCRC32();
312 bool inline_updateByteBufferCRC32();
313 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
314 bool inline_updateBytesCRC32C();
315 bool inline_updateDirectByteBufferCRC32C();
316 bool inline_updateBytesAdler32();
317 bool inline_updateByteBufferAdler32();
318 bool inline_multiplyToLen();
319 bool inline_hasNegatives();
320 bool inline_squareToLen();
321 bool inline_mulAdd();
322 bool inline_montgomeryMultiply();
323 bool inline_montgomerySquare();
324 bool inline_vectorizedMismatch();
325 bool inline_fma(vmIntrinsics::ID id);
326
327 bool inline_profileBoolean();
328 bool inline_isCompileConstant();
329 };
330
331 //---------------------------make_vm_intrinsic----------------------------
332 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
333 vmIntrinsics::ID id = m->intrinsic_id();
334 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
335
336 if (!m->is_loaded()) {
337 // Do not attempt to inline unloaded methods.
338 return NULL;
339 }
340
341 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
342 bool is_available = false;
343
344 {
345 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
346 // the compiler must transition to '_thread_in_vm' state because both
347 // methods access VM-internal data.
348 VM_ENTRY_MARK;
349 methodHandle mh(THREAD, m->get_Method());
350 is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
351 !C->directive()->is_intrinsic_disabled(mh) &&
352 !vmIntrinsics::is_disabled_by_flags(mh);
353
354 }
355
356 if (is_available) {
357 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
358 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
359 return new LibraryIntrinsic(m, is_virtual,
360 vmIntrinsics::predicates_needed(id),
361 vmIntrinsics::does_virtual_dispatch(id),
362 (vmIntrinsics::ID) id);
363 } else {
364 return NULL;
365 }
366 }
367
368 //----------------------register_library_intrinsics-----------------------
369 // Initialize this file's data structures, for each Compile instance.
370 void Compile::register_library_intrinsics() {
371 // Nothing to do here.
372 }
373
374 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
375 LibraryCallKit kit(jvms, this);
376 Compile* C = kit.C;
377 int nodes = C->unique();
378 #ifndef PRODUCT
379 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
380 char buf[1000];
381 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
382 tty->print_cr("Intrinsic %s", str);
383 }
384 #endif
385 ciMethod* callee = kit.callee();
386 const int bci = kit.bci();
387
388 // Try to inline the intrinsic.
389 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
390 kit.try_to_inline(_last_predicate)) {
391 if (C->print_intrinsics() || C->print_inlining()) {
392 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
393 }
394 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
395 if (C->log()) {
396 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
397 vmIntrinsics::name_at(intrinsic_id()),
398 (is_virtual() ? " virtual='1'" : ""),
399 C->unique() - nodes);
400 }
401 // Push the result from the inlined method onto the stack.
402 kit.push_result();
403 C->print_inlining_update(this);
404 return kit.transfer_exceptions_into_jvms();
405 }
406
407 // The intrinsic bailed out
408 if (C->print_intrinsics() || C->print_inlining()) {
409 if (jvms->has_method()) {
410 // Not a root compile.
411 const char* msg;
412 if (callee->intrinsic_candidate()) {
413 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
414 } else {
415 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
416 : "failed to inline (intrinsic), method not annotated";
417 }
418 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
419 } else {
420 // Root compile
421 tty->print("Did not generate intrinsic %s%s at bci:%d in",
422 vmIntrinsics::name_at(intrinsic_id()),
423 (is_virtual() ? " (virtual)" : ""), bci);
424 }
425 }
426 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
427 C->print_inlining_update(this);
428 return NULL;
429 }
430
431 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
432 LibraryCallKit kit(jvms, this);
433 Compile* C = kit.C;
434 int nodes = C->unique();
435 _last_predicate = predicate;
436 #ifndef PRODUCT
437 assert(is_predicated() && predicate < predicates_count(), "sanity");
438 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
439 char buf[1000];
440 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
441 tty->print_cr("Predicate for intrinsic %s", str);
442 }
443 #endif
444 ciMethod* callee = kit.callee();
445 const int bci = kit.bci();
446
447 Node* slow_ctl = kit.try_to_predicate(predicate);
448 if (!kit.failing()) {
449 if (C->print_intrinsics() || C->print_inlining()) {
450 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
451 }
452 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
453 if (C->log()) {
454 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
455 vmIntrinsics::name_at(intrinsic_id()),
456 (is_virtual() ? " virtual='1'" : ""),
457 C->unique() - nodes);
458 }
459 return slow_ctl; // Could be NULL if the check folds.
460 }
461
462 // The intrinsic bailed out
463 if (C->print_intrinsics() || C->print_inlining()) {
464 if (jvms->has_method()) {
465 // Not a root compile.
466 const char* msg = "failed to generate predicate for intrinsic";
467 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
468 } else {
469 // Root compile
470 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
471 vmIntrinsics::name_at(intrinsic_id()),
472 (is_virtual() ? " (virtual)" : ""), bci);
473 }
474 }
475 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
476 return NULL;
477 }
478
479 bool LibraryCallKit::try_to_inline(int predicate) {
480 // Handle symbolic names for otherwise undistinguished boolean switches:
481 const bool is_store = true;
482 const bool is_compress = true;
483 const bool is_static = true;
484 const bool is_volatile = true;
485
486 if (!jvms()->has_method()) {
487 // Root JVMState has a null method.
488 assert(map()->memory()->Opcode() == Op_Parm, "");
489 // Insert the memory aliasing node
490 set_all_memory(reset_memory());
491 }
492 assert(merged_memory(), "");
493
494
495 switch (intrinsic_id()) {
496 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
497 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
498 case vmIntrinsics::_getClass: return inline_native_getClass();
499
500 case vmIntrinsics::_dsin:
501 case vmIntrinsics::_dcos:
502 case vmIntrinsics::_dtan:
503 case vmIntrinsics::_dabs:
504 case vmIntrinsics::_datan2:
505 case vmIntrinsics::_dsqrt:
506 case vmIntrinsics::_dexp:
507 case vmIntrinsics::_dlog:
508 case vmIntrinsics::_dlog10:
509 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
510
511 case vmIntrinsics::_min:
512 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
513
514 case vmIntrinsics::_notify:
515 case vmIntrinsics::_notifyAll:
516 if (InlineNotify) {
517 return inline_notify(intrinsic_id());
518 }
519 return false;
520
521 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
522 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
523 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
524 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
525 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
526 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
527 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
528 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
529 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
530 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
531 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
532 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
533
534 case vmIntrinsics::_arraycopy: return inline_arraycopy();
535
536 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
537 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
538 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
539 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
540
541 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
542 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
543 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
544 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
545 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
546 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
547 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
548
549 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
550 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
551
552 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
553 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
554 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
555 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
556
557 case vmIntrinsics::_compressStringC:
558 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
559 case vmIntrinsics::_inflateStringC:
560 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
561
562 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
563 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
564 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
565 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
566 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
567 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
568 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
569 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
570 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
571
572 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
573 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
574 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
575 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
576 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
577 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
578 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
579 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
580 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
581
582 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
583 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
584 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
585 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
586 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
587 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
588 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
589 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
590 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
591
592 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
593 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
594 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
595 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
596 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
597 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
598 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
599 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
600 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
601
602 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
603 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
604 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
605 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
606
607 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
608 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
609 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
610 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
611
612 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
613 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
614 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
615 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
616 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
617 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
618 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
619 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
620 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
621
622 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
623 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
624 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
625 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
626 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
627 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
628 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
629 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
630 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
631
632 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
633 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
634 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
635 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
636 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
637 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
638 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
639 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
640 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
641
642 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
643 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
644 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
645 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
646 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
647 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
648 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
649 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
650 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
651
652 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
653 case vmIntrinsics::_compareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
654 case vmIntrinsics::_compareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
655 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
656 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
657
658 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
659 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
660 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
661 case vmIntrinsics::_weakCompareAndSwapObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
662 case vmIntrinsics::_weakCompareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
663 case vmIntrinsics::_weakCompareAndSwapByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
664 case vmIntrinsics::_weakCompareAndSwapByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
665 case vmIntrinsics::_weakCompareAndSwapByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
666 case vmIntrinsics::_weakCompareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
667 case vmIntrinsics::_weakCompareAndSwapShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
668 case vmIntrinsics::_weakCompareAndSwapShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
669 case vmIntrinsics::_weakCompareAndSwapShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
670 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
671 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
672 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
673 case vmIntrinsics::_weakCompareAndSwapIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
674 case vmIntrinsics::_weakCompareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
675 case vmIntrinsics::_weakCompareAndSwapLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
676 case vmIntrinsics::_weakCompareAndSwapLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
677 case vmIntrinsics::_weakCompareAndSwapLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
678
679 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
680 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
681 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
682 case vmIntrinsics::_compareAndExchangeByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
683 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
684 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
685 case vmIntrinsics::_compareAndExchangeShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
686 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
687 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
688 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
689 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
690 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
691 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
692 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
693 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
694
695 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
696 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
697 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
698 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
699
700 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
701 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
702 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
703 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
704 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
705
706 case vmIntrinsics::_loadFence:
707 case vmIntrinsics::_storeFence:
708 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
709
710 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
711
712 case vmIntrinsics::_currentThread: return inline_native_currentThread();
713 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
714
715 #ifdef TRACE_HAVE_INTRINSICS
716 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
717 case vmIntrinsics::_getClassId: return inline_native_classID();
718 case vmIntrinsics::_getBufferWriter: return inline_native_getBufferWriter();
719 #endif
720 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
721 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
722 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
723 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
724 case vmIntrinsics::_getLength: return inline_native_getLength();
725 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
726 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
727 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
728 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
729 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
730 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
731
732 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
733 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
734
735 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
736
737 case vmIntrinsics::_isInstance:
738 case vmIntrinsics::_getModifiers:
739 case vmIntrinsics::_isInterface:
740 case vmIntrinsics::_isArray:
741 case vmIntrinsics::_isPrimitive:
742 case vmIntrinsics::_getSuperclass:
743 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
744
745 case vmIntrinsics::_floatToRawIntBits:
746 case vmIntrinsics::_floatToIntBits:
747 case vmIntrinsics::_intBitsToFloat:
748 case vmIntrinsics::_doubleToRawLongBits:
749 case vmIntrinsics::_doubleToLongBits:
750 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
751
752 case vmIntrinsics::_numberOfLeadingZeros_i:
753 case vmIntrinsics::_numberOfLeadingZeros_l:
754 case vmIntrinsics::_numberOfTrailingZeros_i:
755 case vmIntrinsics::_numberOfTrailingZeros_l:
756 case vmIntrinsics::_bitCount_i:
757 case vmIntrinsics::_bitCount_l:
758 case vmIntrinsics::_reverseBytes_i:
759 case vmIntrinsics::_reverseBytes_l:
760 case vmIntrinsics::_reverseBytes_s:
761 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
762
763 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
764
765 case vmIntrinsics::_Reference_get: return inline_reference_get();
766
767 case vmIntrinsics::_Class_cast: return inline_Class_cast();
768
769 case vmIntrinsics::_aescrypt_encryptBlock:
770 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
771
772 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
773 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
774 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
775
776 case vmIntrinsics::_counterMode_AESCrypt:
777 return inline_counterMode_AESCrypt(intrinsic_id());
778
779 case vmIntrinsics::_sha_implCompress:
780 case vmIntrinsics::_sha2_implCompress:
781 case vmIntrinsics::_sha5_implCompress:
782 return inline_sha_implCompress(intrinsic_id());
783
784 case vmIntrinsics::_digestBase_implCompressMB:
785 return inline_digestBase_implCompressMB(predicate);
786
787 case vmIntrinsics::_multiplyToLen:
788 return inline_multiplyToLen();
789
790 case vmIntrinsics::_squareToLen:
791 return inline_squareToLen();
792
793 case vmIntrinsics::_mulAdd:
794 return inline_mulAdd();
795
796 case vmIntrinsics::_montgomeryMultiply:
797 return inline_montgomeryMultiply();
798 case vmIntrinsics::_montgomerySquare:
799 return inline_montgomerySquare();
800
801 case vmIntrinsics::_vectorizedMismatch:
802 return inline_vectorizedMismatch();
803
804 case vmIntrinsics::_ghash_processBlocks:
805 return inline_ghash_processBlocks();
806
807 case vmIntrinsics::_encodeISOArray:
808 case vmIntrinsics::_encodeByteISOArray:
809 return inline_encodeISOArray();
810
811 case vmIntrinsics::_updateCRC32:
812 return inline_updateCRC32();
813 case vmIntrinsics::_updateBytesCRC32:
814 return inline_updateBytesCRC32();
815 case vmIntrinsics::_updateByteBufferCRC32:
816 return inline_updateByteBufferCRC32();
817
818 case vmIntrinsics::_updateBytesCRC32C:
819 return inline_updateBytesCRC32C();
820 case vmIntrinsics::_updateDirectByteBufferCRC32C:
821 return inline_updateDirectByteBufferCRC32C();
822
823 case vmIntrinsics::_updateBytesAdler32:
824 return inline_updateBytesAdler32();
825 case vmIntrinsics::_updateByteBufferAdler32:
826 return inline_updateByteBufferAdler32();
827
828 case vmIntrinsics::_profileBoolean:
829 return inline_profileBoolean();
830 case vmIntrinsics::_isCompileConstant:
831 return inline_isCompileConstant();
832
833 case vmIntrinsics::_hasNegatives:
834 return inline_hasNegatives();
835
836 case vmIntrinsics::_fmaD:
837 case vmIntrinsics::_fmaF:
838 return inline_fma(intrinsic_id());
839
840 default:
841 // If you get here, it may be that someone has added a new intrinsic
842 // to the list in vmSymbols.hpp without implementing it here.
843 #ifndef PRODUCT
844 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
845 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
846 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
847 }
848 #endif
849 return false;
850 }
851 }
852
853 Node* LibraryCallKit::try_to_predicate(int predicate) {
854 if (!jvms()->has_method()) {
855 // Root JVMState has a null method.
856 assert(map()->memory()->Opcode() == Op_Parm, "");
857 // Insert the memory aliasing node
858 set_all_memory(reset_memory());
859 }
860 assert(merged_memory(), "");
861
862 switch (intrinsic_id()) {
863 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
864 return inline_cipherBlockChaining_AESCrypt_predicate(false);
865 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
866 return inline_cipherBlockChaining_AESCrypt_predicate(true);
867 case vmIntrinsics::_counterMode_AESCrypt:
868 return inline_counterMode_AESCrypt_predicate();
869 case vmIntrinsics::_digestBase_implCompressMB:
870 return inline_digestBase_implCompressMB_predicate(predicate);
871
872 default:
873 // If you get here, it may be that someone has added a new intrinsic
874 // to the list in vmSymbols.hpp without implementing it here.
875 #ifndef PRODUCT
876 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
877 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
878 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
879 }
880 #endif
881 Node* slow_ctl = control();
882 set_control(top()); // No fast path instrinsic
883 return slow_ctl;
884 }
885 }
886
887 //------------------------------set_result-------------------------------
888 // Helper function for finishing intrinsics.
889 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
890 record_for_igvn(region);
891 set_control(_gvn.transform(region));
892 set_result( _gvn.transform(value));
893 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
894 }
895
896 //------------------------------generate_guard---------------------------
897 // Helper function for generating guarded fast-slow graph structures.
898 // The given 'test', if true, guards a slow path. If the test fails
899 // then a fast path can be taken. (We generally hope it fails.)
900 // In all cases, GraphKit::control() is updated to the fast path.
901 // The returned value represents the control for the slow path.
902 // The return value is never 'top'; it is either a valid control
903 // or NULL if it is obvious that the slow path can never be taken.
904 // Also, if region and the slow control are not NULL, the slow edge
905 // is appended to the region.
906 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
907 if (stopped()) {
908 // Already short circuited.
909 return NULL;
910 }
911
912 // Build an if node and its projections.
913 // If test is true we take the slow path, which we assume is uncommon.
914 if (_gvn.type(test) == TypeInt::ZERO) {
915 // The slow branch is never taken. No need to build this guard.
916 return NULL;
917 }
918
919 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
920
921 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
922 if (if_slow == top()) {
923 // The slow branch is never taken. No need to build this guard.
924 return NULL;
925 }
926
927 if (region != NULL)
928 region->add_req(if_slow);
929
930 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
931 set_control(if_fast);
932
933 return if_slow;
934 }
935
936 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
937 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
938 }
939 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
940 return generate_guard(test, region, PROB_FAIR);
941 }
942
943 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
944 Node* *pos_index) {
945 if (stopped())
946 return NULL; // already stopped
947 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
948 return NULL; // index is already adequately typed
949 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
950 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
951 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
952 if (is_neg != NULL && pos_index != NULL) {
953 // Emulate effect of Parse::adjust_map_after_if.
954 Node* ccast = new CastIINode(index, TypeInt::POS);
955 ccast->set_req(0, control());
956 (*pos_index) = _gvn.transform(ccast);
957 }
958 return is_neg;
959 }
960
961 // Make sure that 'position' is a valid limit index, in [0..length].
962 // There are two equivalent plans for checking this:
963 // A. (offset + copyLength) unsigned<= arrayLength
964 // B. offset <= (arrayLength - copyLength)
965 // We require that all of the values above, except for the sum and
966 // difference, are already known to be non-negative.
967 // Plan A is robust in the face of overflow, if offset and copyLength
968 // are both hugely positive.
969 //
970 // Plan B is less direct and intuitive, but it does not overflow at
971 // all, since the difference of two non-negatives is always
972 // representable. Whenever Java methods must perform the equivalent
973 // check they generally use Plan B instead of Plan A.
974 // For the moment we use Plan A.
975 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
976 Node* subseq_length,
977 Node* array_length,
978 RegionNode* region) {
979 if (stopped())
980 return NULL; // already stopped
981 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
982 if (zero_offset && subseq_length->eqv_uncast(array_length))
983 return NULL; // common case of whole-array copy
984 Node* last = subseq_length;
985 if (!zero_offset) // last += offset
986 last = _gvn.transform(new AddINode(last, offset));
987 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
988 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
989 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
990 return is_over;
991 }
992
993 // Emit range checks for the given String.value byte array
994 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
995 if (stopped()) {
996 return; // already stopped
997 }
998 RegionNode* bailout = new RegionNode(1);
999 record_for_igvn(bailout);
1000 if (char_count) {
1001 // Convert char count to byte count
1002 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1003 }
1004
1005 // Offset and count must not be negative
1006 generate_negative_guard(offset, bailout);
1007 generate_negative_guard(count, bailout);
1008 // Offset + count must not exceed length of array
1009 generate_limit_guard(offset, count, load_array_length(array), bailout);
1010
1011 if (bailout->req() > 1) {
1012 PreserveJVMState pjvms(this);
1013 set_control(_gvn.transform(bailout));
1014 uncommon_trap(Deoptimization::Reason_intrinsic,
1015 Deoptimization::Action_maybe_recompile);
1016 }
1017 }
1018
1019 //--------------------------generate_current_thread--------------------
1020 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1021 ciKlass* thread_klass = env()->Thread_klass();
1022 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1023 Node* thread = _gvn.transform(new ThreadLocalNode());
1024 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1025 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1026 tls_output = thread;
1027 return threadObj;
1028 }
1029
1030
1031 //------------------------------make_string_method_node------------------------
1032 // Helper method for String intrinsic functions. This version is called with
1033 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1034 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1035 // containing the lengths of str1 and str2.
1036 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1037 Node* result = NULL;
1038 switch (opcode) {
1039 case Op_StrIndexOf:
1040 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1041 str1_start, cnt1, str2_start, cnt2, ae);
1042 break;
1043 case Op_StrComp:
1044 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1045 str1_start, cnt1, str2_start, cnt2, ae);
1046 break;
1047 case Op_StrEquals:
1048 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1049 // Use the constant length if there is one because optimized match rule may exist.
1050 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1051 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1052 break;
1053 default:
1054 ShouldNotReachHere();
1055 return NULL;
1056 }
1057
1058 // All these intrinsics have checks.
1059 C->set_has_split_ifs(true); // Has chance for split-if optimization
1060
1061 return _gvn.transform(result);
1062 }
1063
1064 //------------------------------inline_string_compareTo------------------------
1065 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1066 Node* arg1 = argument(0);
1067 Node* arg2 = argument(1);
1068
1069 // Get start addr and length of first argument
1070 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1071 Node* arg1_cnt = load_array_length(arg1);
1072
1073 // Get start addr and length of second argument
1074 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1075 Node* arg2_cnt = load_array_length(arg2);
1076
1077 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1078 set_result(result);
1079 return true;
1080 }
1081
1082 //------------------------------inline_string_equals------------------------
1083 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1084 Node* arg1 = argument(0);
1085 Node* arg2 = argument(1);
1086
1087 // paths (plus control) merge
1088 RegionNode* region = new RegionNode(3);
1089 Node* phi = new PhiNode(region, TypeInt::BOOL);
1090
1091 if (!stopped()) {
1092 // Get start addr and length of first argument
1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1094 Node* arg1_cnt = load_array_length(arg1);
1095
1096 // Get start addr and length of second argument
1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1098 Node* arg2_cnt = load_array_length(arg2);
1099
1100 // Check for arg1_cnt != arg2_cnt
1101 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1102 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1103 Node* if_ne = generate_slow_guard(bol, NULL);
1104 if (if_ne != NULL) {
1105 phi->init_req(2, intcon(0));
1106 region->init_req(2, if_ne);
1107 }
1108
1109 // Check for count == 0 is done by assembler code for StrEquals.
1110
1111 if (!stopped()) {
1112 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1113 phi->init_req(1, equals);
1114 region->init_req(1, control());
1115 }
1116 }
1117
1118 // post merge
1119 set_control(_gvn.transform(region));
1120 record_for_igvn(region);
1121
1122 set_result(_gvn.transform(phi));
1123 return true;
1124 }
1125
1126 //------------------------------inline_array_equals----------------------------
1127 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1128 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1129 Node* arg1 = argument(0);
1130 Node* arg2 = argument(1);
1131
1132 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1133 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1134 return true;
1135 }
1136
1137 //------------------------------inline_hasNegatives------------------------------
1138 bool LibraryCallKit::inline_hasNegatives() {
1139 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1140 return false;
1141 }
1142
1143 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1144 // no receiver since it is static method
1145 Node* ba = argument(0);
1146 Node* offset = argument(1);
1147 Node* len = argument(2);
1148
1149 // Range checks
1150 generate_string_range_check(ba, offset, len, false);
1151 if (stopped()) {
1152 return true;
1153 }
1154 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1155 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1156 set_result(_gvn.transform(result));
1157 return true;
1158 }
1159
1160 bool LibraryCallKit::inline_preconditions_checkIndex() {
1161 Node* index = argument(0);
1162 Node* length = argument(1);
1163 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1164 return false;
1165 }
1166
1167 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1168 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1169
1170 {
1171 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1172 uncommon_trap(Deoptimization::Reason_intrinsic,
1173 Deoptimization::Action_make_not_entrant);
1174 }
1175
1176 if (stopped()) {
1177 return false;
1178 }
1179
1180 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1181 BoolTest::mask btest = BoolTest::lt;
1182 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1183 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1184 _gvn.set_type(rc, rc->Value(&_gvn));
1185 if (!rc_bool->is_Con()) {
1186 record_for_igvn(rc);
1187 }
1188 set_control(_gvn.transform(new IfTrueNode(rc)));
1189 {
1190 PreserveJVMState pjvms(this);
1191 set_control(_gvn.transform(new IfFalseNode(rc)));
1192 uncommon_trap(Deoptimization::Reason_range_check,
1193 Deoptimization::Action_make_not_entrant);
1194 }
1195
1196 if (stopped()) {
1197 return false;
1198 }
1199
1200 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1201 result->set_req(0, control());
1202 result = _gvn.transform(result);
1203 set_result(result);
1204 replace_in_map(index, result);
1205 return true;
1206 }
1207
1208 //------------------------------inline_string_indexOf------------------------
1209 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1210 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1211 return false;
1212 }
1213 Node* src = argument(0);
1214 Node* tgt = argument(1);
1215
1216 // Make the merge point
1217 RegionNode* result_rgn = new RegionNode(4);
1218 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1219
1220 // Get start addr and length of source string
1221 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1222 Node* src_count = load_array_length(src);
1223
1224 // Get start addr and length of substring
1225 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1226 Node* tgt_count = load_array_length(tgt);
1227
1228 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1229 // Divide src size by 2 if String is UTF16 encoded
1230 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1231 }
1232 if (ae == StrIntrinsicNode::UU) {
1233 // Divide substring size by 2 if String is UTF16 encoded
1234 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1235 }
1236
1237 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1238 if (result != NULL) {
1239 result_phi->init_req(3, result);
1240 result_rgn->init_req(3, control());
1241 }
1242 set_control(_gvn.transform(result_rgn));
1243 record_for_igvn(result_rgn);
1244 set_result(_gvn.transform(result_phi));
1245
1246 return true;
1247 }
1248
1249 //-----------------------------inline_string_indexOf-----------------------
1250 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1251 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1252 return false;
1253 }
1254 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1255 return false;
1256 }
1257 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1258 Node* src = argument(0); // byte[]
1259 Node* src_count = argument(1); // char count
1260 Node* tgt = argument(2); // byte[]
1261 Node* tgt_count = argument(3); // char count
1262 Node* from_index = argument(4); // char index
1263
1264 // Multiply byte array index by 2 if String is UTF16 encoded
1265 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1266 src_count = _gvn.transform(new SubINode(src_count, from_index));
1267 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1268 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1269
1270 // Range checks
1271 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1272 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1273 if (stopped()) {
1274 return true;
1275 }
1276
1277 RegionNode* region = new RegionNode(5);
1278 Node* phi = new PhiNode(region, TypeInt::INT);
1279
1280 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1281 if (result != NULL) {
1282 // The result is index relative to from_index if substring was found, -1 otherwise.
1283 // Generate code which will fold into cmove.
1284 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1285 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1286
1287 Node* if_lt = generate_slow_guard(bol, NULL);
1288 if (if_lt != NULL) {
1289 // result == -1
1290 phi->init_req(3, result);
1291 region->init_req(3, if_lt);
1292 }
1293 if (!stopped()) {
1294 result = _gvn.transform(new AddINode(result, from_index));
1295 phi->init_req(4, result);
1296 region->init_req(4, control());
1297 }
1298 }
1299
1300 set_control(_gvn.transform(region));
1301 record_for_igvn(region);
1302 set_result(_gvn.transform(phi));
1303
1304 return true;
1305 }
1306
1307 // Create StrIndexOfNode with fast path checks
1308 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1309 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1310 // Check for substr count > string count
1311 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1312 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1313 Node* if_gt = generate_slow_guard(bol, NULL);
1314 if (if_gt != NULL) {
1315 phi->init_req(1, intcon(-1));
1316 region->init_req(1, if_gt);
1317 }
1318 if (!stopped()) {
1319 // Check for substr count == 0
1320 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1321 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1322 Node* if_zero = generate_slow_guard(bol, NULL);
1323 if (if_zero != NULL) {
1324 phi->init_req(2, intcon(0));
1325 region->init_req(2, if_zero);
1326 }
1327 }
1328 if (!stopped()) {
1329 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1330 }
1331 return NULL;
1332 }
1333
1334 //-----------------------------inline_string_indexOfChar-----------------------
1335 bool LibraryCallKit::inline_string_indexOfChar() {
1336 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1337 return false;
1338 }
1339 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1340 return false;
1341 }
1342 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1343 Node* src = argument(0); // byte[]
1344 Node* tgt = argument(1); // tgt is int ch
1345 Node* from_index = argument(2);
1346 Node* max = argument(3);
1347
1348 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1349 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1350 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1351
1352 // Range checks
1353 generate_string_range_check(src, src_offset, src_count, true);
1354 if (stopped()) {
1355 return true;
1356 }
1357
1358 RegionNode* region = new RegionNode(3);
1359 Node* phi = new PhiNode(region, TypeInt::INT);
1360
1361 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1362 C->set_has_split_ifs(true); // Has chance for split-if optimization
1363 _gvn.transform(result);
1364
1365 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1366 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1367
1368 Node* if_lt = generate_slow_guard(bol, NULL);
1369 if (if_lt != NULL) {
1370 // result == -1
1371 phi->init_req(2, result);
1372 region->init_req(2, if_lt);
1373 }
1374 if (!stopped()) {
1375 result = _gvn.transform(new AddINode(result, from_index));
1376 phi->init_req(1, result);
1377 region->init_req(1, control());
1378 }
1379 set_control(_gvn.transform(region));
1380 record_for_igvn(region);
1381 set_result(_gvn.transform(phi));
1382
1383 return true;
1384 }
1385 //---------------------------inline_string_copy---------------------
1386 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1387 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1388 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1389 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1390 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1391 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1392 bool LibraryCallKit::inline_string_copy(bool compress) {
1393 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1394 return false;
1395 }
1396 int nargs = 5; // 2 oops, 3 ints
1397 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1398
1399 Node* src = argument(0);
1400 Node* src_offset = argument(1);
1401 Node* dst = argument(2);
1402 Node* dst_offset = argument(3);
1403 Node* length = argument(4);
1404
1405 // Check for allocation before we add nodes that would confuse
1406 // tightly_coupled_allocation()
1407 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1408
1409 // Figure out the size and type of the elements we will be copying.
1410 const Type* src_type = src->Value(&_gvn);
1411 const Type* dst_type = dst->Value(&_gvn);
1412 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1413 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1414 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1415 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1416 "Unsupported array types for inline_string_copy");
1417
1418 // Convert char[] offsets to byte[] offsets
1419 bool convert_src = (compress && src_elem == T_BYTE);
1420 bool convert_dst = (!compress && dst_elem == T_BYTE);
1421 if (convert_src) {
1422 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1423 } else if (convert_dst) {
1424 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1425 }
1426
1427 // Range checks
1428 generate_string_range_check(src, src_offset, length, convert_src);
1429 generate_string_range_check(dst, dst_offset, length, convert_dst);
1430 if (stopped()) {
1431 return true;
1432 }
1433
1434 Node* src_start = array_element_address(src, src_offset, src_elem);
1435 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1436 // 'src_start' points to src array + scaled offset
1437 // 'dst_start' points to dst array + scaled offset
1438 Node* count = NULL;
1439 if (compress) {
1440 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1441 } else {
1442 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1443 }
1444
1445 if (alloc != NULL) {
1446 if (alloc->maybe_set_complete(&_gvn)) {
1447 // "You break it, you buy it."
1448 InitializeNode* init = alloc->initialization();
1449 assert(init->is_complete(), "we just did this");
1450 init->set_complete_with_arraycopy();
1451 assert(dst->is_CheckCastPP(), "sanity");
1452 assert(dst->in(0)->in(0) == init, "dest pinned");
1453 }
1454 // Do not let stores that initialize this object be reordered with
1455 // a subsequent store that would make this object accessible by
1456 // other threads.
1457 // Record what AllocateNode this StoreStore protects so that
1458 // escape analysis can go from the MemBarStoreStoreNode to the
1459 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1460 // based on the escape status of the AllocateNode.
1461 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1462 }
1463 if (compress) {
1464 set_result(_gvn.transform(count));
1465 }
1466 return true;
1467 }
1468
1469 #ifdef _LP64
1470 #define XTOP ,top() /*additional argument*/
1471 #else //_LP64
1472 #define XTOP /*no additional argument*/
1473 #endif //_LP64
1474
1475 //------------------------inline_string_toBytesU--------------------------
1476 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1477 bool LibraryCallKit::inline_string_toBytesU() {
1478 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1479 return false;
1480 }
1481 // Get the arguments.
1482 Node* value = argument(0);
1483 Node* offset = argument(1);
1484 Node* length = argument(2);
1485
1486 Node* newcopy = NULL;
1487
1488 // Set the original stack and the reexecute bit for the interpreter to reexecute
1489 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1490 { PreserveReexecuteState preexecs(this);
1491 jvms()->set_should_reexecute(true);
1492
1493 // Check if a null path was taken unconditionally.
1494 value = null_check(value);
1495
1496 RegionNode* bailout = new RegionNode(1);
1497 record_for_igvn(bailout);
1498
1499 // Range checks
1500 generate_negative_guard(offset, bailout);
1501 generate_negative_guard(length, bailout);
1502 generate_limit_guard(offset, length, load_array_length(value), bailout);
1503 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1504 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1505
1506 if (bailout->req() > 1) {
1507 PreserveJVMState pjvms(this);
1508 set_control(_gvn.transform(bailout));
1509 uncommon_trap(Deoptimization::Reason_intrinsic,
1510 Deoptimization::Action_maybe_recompile);
1511 }
1512 if (stopped()) {
1513 return true;
1514 }
1515
1516 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1517 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1518 newcopy = new_array(klass_node, size, 0); // no arguments to push
1519 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1520
1521 // Calculate starting addresses.
1522 Node* src_start = array_element_address(value, offset, T_CHAR);
1523 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1524
1525 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1526 const TypeInt* toffset = gvn().type(offset)->is_int();
1527 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1528
1529 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1530 const char* copyfunc_name = "arraycopy";
1531 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1532 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1533 OptoRuntime::fast_arraycopy_Type(),
1534 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1535 src_start, dst_start, ConvI2X(length) XTOP);
1536 // Do not let reads from the cloned object float above the arraycopy.
1537 if (alloc != NULL) {
1538 if (alloc->maybe_set_complete(&_gvn)) {
1539 // "You break it, you buy it."
1540 InitializeNode* init = alloc->initialization();
1541 assert(init->is_complete(), "we just did this");
1542 init->set_complete_with_arraycopy();
1543 assert(newcopy->is_CheckCastPP(), "sanity");
1544 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1545 }
1546 // Do not let stores that initialize this object be reordered with
1547 // a subsequent store that would make this object accessible by
1548 // other threads.
1549 // Record what AllocateNode this StoreStore protects so that
1550 // escape analysis can go from the MemBarStoreStoreNode to the
1551 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1552 // based on the escape status of the AllocateNode.
1553 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1554 } else {
1555 insert_mem_bar(Op_MemBarCPUOrder);
1556 }
1557 } // original reexecute is set back here
1558
1559 C->set_has_split_ifs(true); // Has chance for split-if optimization
1560 if (!stopped()) {
1561 set_result(newcopy);
1562 }
1563 return true;
1564 }
1565
1566 //------------------------inline_string_getCharsU--------------------------
1567 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1568 bool LibraryCallKit::inline_string_getCharsU() {
1569 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1570 return false;
1571 }
1572
1573 // Get the arguments.
1574 Node* src = argument(0);
1575 Node* src_begin = argument(1);
1576 Node* src_end = argument(2); // exclusive offset (i < src_end)
1577 Node* dst = argument(3);
1578 Node* dst_begin = argument(4);
1579
1580 // Check for allocation before we add nodes that would confuse
1581 // tightly_coupled_allocation()
1582 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1583
1584 // Check if a null path was taken unconditionally.
1585 src = null_check(src);
1586 dst = null_check(dst);
1587 if (stopped()) {
1588 return true;
1589 }
1590
1591 // Get length and convert char[] offset to byte[] offset
1592 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1593 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1594
1595 // Range checks
1596 generate_string_range_check(src, src_begin, length, true);
1597 generate_string_range_check(dst, dst_begin, length, false);
1598 if (stopped()) {
1599 return true;
1600 }
1601
1602 if (!stopped()) {
1603 // Calculate starting addresses.
1604 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1605 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1606
1607 // Check if array addresses are aligned to HeapWordSize
1608 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1609 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1610 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1611 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1612
1613 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1614 const char* copyfunc_name = "arraycopy";
1615 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1616 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1617 OptoRuntime::fast_arraycopy_Type(),
1618 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1619 src_start, dst_start, ConvI2X(length) XTOP);
1620 // Do not let reads from the cloned object float above the arraycopy.
1621 if (alloc != NULL) {
1622 if (alloc->maybe_set_complete(&_gvn)) {
1623 // "You break it, you buy it."
1624 InitializeNode* init = alloc->initialization();
1625 assert(init->is_complete(), "we just did this");
1626 init->set_complete_with_arraycopy();
1627 assert(dst->is_CheckCastPP(), "sanity");
1628 assert(dst->in(0)->in(0) == init, "dest pinned");
1629 }
1630 // Do not let stores that initialize this object be reordered with
1631 // a subsequent store that would make this object accessible by
1632 // other threads.
1633 // Record what AllocateNode this StoreStore protects so that
1634 // escape analysis can go from the MemBarStoreStoreNode to the
1635 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1636 // based on the escape status of the AllocateNode.
1637 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1638 } else {
1639 insert_mem_bar(Op_MemBarCPUOrder);
1640 }
1641 }
1642
1643 C->set_has_split_ifs(true); // Has chance for split-if optimization
1644 return true;
1645 }
1646
1647 //----------------------inline_string_char_access----------------------------
1648 // Store/Load char to/from byte[] array.
1649 // static void StringUTF16.putChar(byte[] val, int index, int c)
1650 // static char StringUTF16.getChar(byte[] val, int index)
1651 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1652 Node* value = argument(0);
1653 Node* index = argument(1);
1654 Node* ch = is_store ? argument(2) : NULL;
1655
1656 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1657 // correctly requires matched array shapes.
1658 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1659 "sanity: byte[] and char[] bases agree");
1660 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1661 "sanity: byte[] and char[] scales agree");
1662
1663 // Bail when getChar over constants is requested: constant folding would
1664 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1665 // Java method would constant fold nicely instead.
1666 if (!is_store && value->is_Con() && index->is_Con()) {
1667 return false;
1668 }
1669
1670 Node* adr = array_element_address(value, index, T_CHAR);
1671 if (is_store) {
1672 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1673 false, false, true /* mismatched */);
1674 } else {
1675 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1676 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1677 set_result(ch);
1678 }
1679 return true;
1680 }
1681
1682 //--------------------------round_double_node--------------------------------
1683 // Round a double node if necessary.
1684 Node* LibraryCallKit::round_double_node(Node* n) {
1685 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1686 n = _gvn.transform(new RoundDoubleNode(0, n));
1687 return n;
1688 }
1689
1690 //------------------------------inline_math-----------------------------------
1691 // public static double Math.abs(double)
1692 // public static double Math.sqrt(double)
1693 // public static double Math.log(double)
1694 // public static double Math.log10(double)
1695 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1696 Node* arg = round_double_node(argument(0));
1697 Node* n = NULL;
1698 switch (id) {
1699 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1700 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1701 default: fatal_unexpected_iid(id); break;
1702 }
1703 set_result(_gvn.transform(n));
1704 return true;
1705 }
1706
1707 //------------------------------runtime_math-----------------------------
1708 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1709 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1710 "must be (DD)D or (D)D type");
1711
1712 // Inputs
1713 Node* a = round_double_node(argument(0));
1714 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1715
1716 const TypePtr* no_memory_effects = NULL;
1717 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1718 no_memory_effects,
1719 a, top(), b, b ? top() : NULL);
1720 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1721 #ifdef ASSERT
1722 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1723 assert(value_top == top(), "second value must be top");
1724 #endif
1725
1726 set_result(value);
1727 return true;
1728 }
1729
1730 //------------------------------inline_math_native-----------------------------
1731 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1732 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1733 switch (id) {
1734 // These intrinsics are not properly supported on all hardware
1735 case vmIntrinsics::_dsin:
1736 return StubRoutines::dsin() != NULL ?
1737 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1738 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1739 case vmIntrinsics::_dcos:
1740 return StubRoutines::dcos() != NULL ?
1741 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1742 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1743 case vmIntrinsics::_dtan:
1744 return StubRoutines::dtan() != NULL ?
1745 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1746 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1747 case vmIntrinsics::_dlog:
1748 return StubRoutines::dlog() != NULL ?
1749 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1750 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1751 case vmIntrinsics::_dlog10:
1752 return StubRoutines::dlog10() != NULL ?
1753 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1754 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1755
1756 // These intrinsics are supported on all hardware
1757 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1758 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1759
1760 case vmIntrinsics::_dexp:
1761 return StubRoutines::dexp() != NULL ?
1762 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1763 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1764 case vmIntrinsics::_dpow:
1765 return StubRoutines::dpow() != NULL ?
1766 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1767 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1768 #undef FN_PTR
1769
1770 // These intrinsics are not yet correctly implemented
1771 case vmIntrinsics::_datan2:
1772 return false;
1773
1774 default:
1775 fatal_unexpected_iid(id);
1776 return false;
1777 }
1778 }
1779
1780 static bool is_simple_name(Node* n) {
1781 return (n->req() == 1 // constant
1782 || (n->is_Type() && n->as_Type()->type()->singleton())
1783 || n->is_Proj() // parameter or return value
1784 || n->is_Phi() // local of some sort
1785 );
1786 }
1787
1788 //----------------------------inline_notify-----------------------------------*
1789 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1790 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1791 address func;
1792 if (id == vmIntrinsics::_notify) {
1793 func = OptoRuntime::monitor_notify_Java();
1794 } else {
1795 func = OptoRuntime::monitor_notifyAll_Java();
1796 }
1797 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1798 make_slow_call_ex(call, env()->Throwable_klass(), false);
1799 return true;
1800 }
1801
1802
1803 //----------------------------inline_min_max-----------------------------------
1804 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1805 set_result(generate_min_max(id, argument(0), argument(1)));
1806 return true;
1807 }
1808
1809 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1810 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1811 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1812 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1813 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1814
1815 {
1816 PreserveJVMState pjvms(this);
1817 PreserveReexecuteState preexecs(this);
1818 jvms()->set_should_reexecute(true);
1819
1820 set_control(slow_path);
1821 set_i_o(i_o());
1822
1823 uncommon_trap(Deoptimization::Reason_intrinsic,
1824 Deoptimization::Action_none);
1825 }
1826
1827 set_control(fast_path);
1828 set_result(math);
1829 }
1830
1831 template <typename OverflowOp>
1832 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1833 typedef typename OverflowOp::MathOp MathOp;
1834
1835 MathOp* mathOp = new MathOp(arg1, arg2);
1836 Node* operation = _gvn.transform( mathOp );
1837 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1838 inline_math_mathExact(operation, ofcheck);
1839 return true;
1840 }
1841
1842 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1843 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1844 }
1845
1846 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1847 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1848 }
1849
1850 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1851 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1852 }
1853
1854 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1855 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1856 }
1857
1858 bool LibraryCallKit::inline_math_negateExactI() {
1859 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1860 }
1861
1862 bool LibraryCallKit::inline_math_negateExactL() {
1863 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1864 }
1865
1866 bool LibraryCallKit::inline_math_multiplyExactI() {
1867 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1868 }
1869
1870 bool LibraryCallKit::inline_math_multiplyExactL() {
1871 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1872 }
1873
1874 Node*
1875 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1876 // These are the candidate return value:
1877 Node* xvalue = x0;
1878 Node* yvalue = y0;
1879
1880 if (xvalue == yvalue) {
1881 return xvalue;
1882 }
1883
1884 bool want_max = (id == vmIntrinsics::_max);
1885
1886 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1887 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1888 if (txvalue == NULL || tyvalue == NULL) return top();
1889 // This is not really necessary, but it is consistent with a
1890 // hypothetical MaxINode::Value method:
1891 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1892
1893 // %%% This folding logic should (ideally) be in a different place.
1894 // Some should be inside IfNode, and there to be a more reliable
1895 // transformation of ?: style patterns into cmoves. We also want
1896 // more powerful optimizations around cmove and min/max.
1897
1898 // Try to find a dominating comparison of these guys.
1899 // It can simplify the index computation for Arrays.copyOf
1900 // and similar uses of System.arraycopy.
1901 // First, compute the normalized version of CmpI(x, y).
1902 int cmp_op = Op_CmpI;
1903 Node* xkey = xvalue;
1904 Node* ykey = yvalue;
1905 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1906 if (ideal_cmpxy->is_Cmp()) {
1907 // E.g., if we have CmpI(length - offset, count),
1908 // it might idealize to CmpI(length, count + offset)
1909 cmp_op = ideal_cmpxy->Opcode();
1910 xkey = ideal_cmpxy->in(1);
1911 ykey = ideal_cmpxy->in(2);
1912 }
1913
1914 // Start by locating any relevant comparisons.
1915 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1916 Node* cmpxy = NULL;
1917 Node* cmpyx = NULL;
1918 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1919 Node* cmp = start_from->fast_out(k);
1920 if (cmp->outcnt() > 0 && // must have prior uses
1921 cmp->in(0) == NULL && // must be context-independent
1922 cmp->Opcode() == cmp_op) { // right kind of compare
1923 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1924 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1925 }
1926 }
1927
1928 const int NCMPS = 2;
1929 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1930 int cmpn;
1931 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1932 if (cmps[cmpn] != NULL) break; // find a result
1933 }
1934 if (cmpn < NCMPS) {
1935 // Look for a dominating test that tells us the min and max.
1936 int depth = 0; // Limit search depth for speed
1937 Node* dom = control();
1938 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1939 if (++depth >= 100) break;
1940 Node* ifproj = dom;
1941 if (!ifproj->is_Proj()) continue;
1942 Node* iff = ifproj->in(0);
1943 if (!iff->is_If()) continue;
1944 Node* bol = iff->in(1);
1945 if (!bol->is_Bool()) continue;
1946 Node* cmp = bol->in(1);
1947 if (cmp == NULL) continue;
1948 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1949 if (cmps[cmpn] == cmp) break;
1950 if (cmpn == NCMPS) continue;
1951 BoolTest::mask btest = bol->as_Bool()->_test._test;
1952 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1953 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1954 // At this point, we know that 'x btest y' is true.
1955 switch (btest) {
1956 case BoolTest::eq:
1957 // They are proven equal, so we can collapse the min/max.
1958 // Either value is the answer. Choose the simpler.
1959 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1960 return yvalue;
1961 return xvalue;
1962 case BoolTest::lt: // x < y
1963 case BoolTest::le: // x <= y
1964 return (want_max ? yvalue : xvalue);
1965 case BoolTest::gt: // x > y
1966 case BoolTest::ge: // x >= y
1967 return (want_max ? xvalue : yvalue);
1968 }
1969 }
1970 }
1971
1972 // We failed to find a dominating test.
1973 // Let's pick a test that might GVN with prior tests.
1974 Node* best_bol = NULL;
1975 BoolTest::mask best_btest = BoolTest::illegal;
1976 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1977 Node* cmp = cmps[cmpn];
1978 if (cmp == NULL) continue;
1979 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
1980 Node* bol = cmp->fast_out(j);
1981 if (!bol->is_Bool()) continue;
1982 BoolTest::mask btest = bol->as_Bool()->_test._test;
1983 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
1984 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1985 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
1986 best_bol = bol->as_Bool();
1987 best_btest = btest;
1988 }
1989 }
1990 }
1991
1992 Node* answer_if_true = NULL;
1993 Node* answer_if_false = NULL;
1994 switch (best_btest) {
1995 default:
1996 if (cmpxy == NULL)
1997 cmpxy = ideal_cmpxy;
1998 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
1999 // and fall through:
2000 case BoolTest::lt: // x < y
2001 case BoolTest::le: // x <= y
2002 answer_if_true = (want_max ? yvalue : xvalue);
2003 answer_if_false = (want_max ? xvalue : yvalue);
2004 break;
2005 case BoolTest::gt: // x > y
2006 case BoolTest::ge: // x >= y
2007 answer_if_true = (want_max ? xvalue : yvalue);
2008 answer_if_false = (want_max ? yvalue : xvalue);
2009 break;
2010 }
2011
2012 jint hi, lo;
2013 if (want_max) {
2014 // We can sharpen the minimum.
2015 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2016 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2017 } else {
2018 // We can sharpen the maximum.
2019 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2020 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2021 }
2022
2023 // Use a flow-free graph structure, to avoid creating excess control edges
2024 // which could hinder other optimizations.
2025 // Since Math.min/max is often used with arraycopy, we want
2026 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2027 Node* cmov = CMoveNode::make(NULL, best_bol,
2028 answer_if_false, answer_if_true,
2029 TypeInt::make(lo, hi, widen));
2030
2031 return _gvn.transform(cmov);
2032
2033 /*
2034 // This is not as desirable as it may seem, since Min and Max
2035 // nodes do not have a full set of optimizations.
2036 // And they would interfere, anyway, with 'if' optimizations
2037 // and with CMoveI canonical forms.
2038 switch (id) {
2039 case vmIntrinsics::_min:
2040 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2041 case vmIntrinsics::_max:
2042 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2043 default:
2044 ShouldNotReachHere();
2045 }
2046 */
2047 }
2048
2049 inline int
2050 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2051 const TypePtr* base_type = TypePtr::NULL_PTR;
2052 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2053 if (base_type == NULL) {
2054 // Unknown type.
2055 return Type::AnyPtr;
2056 } else if (base_type == TypePtr::NULL_PTR) {
2057 // Since this is a NULL+long form, we have to switch to a rawptr.
2058 base = _gvn.transform(new CastX2PNode(offset));
2059 offset = MakeConX(0);
2060 return Type::RawPtr;
2061 } else if (base_type->base() == Type::RawPtr) {
2062 return Type::RawPtr;
2063 } else if (base_type->isa_oopptr()) {
2064 // Base is never null => always a heap address.
2065 if (base_type->ptr() == TypePtr::NotNull) {
2066 return Type::OopPtr;
2067 }
2068 // Offset is small => always a heap address.
2069 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2070 if (offset_type != NULL &&
2071 base_type->offset() == 0 && // (should always be?)
2072 offset_type->_lo >= 0 &&
2073 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2074 return Type::OopPtr;
2075 } else if (type == T_OBJECT) {
2076 // off heap access to an oop doesn't make any sense. Has to be on
2077 // heap.
2078 return Type::OopPtr;
2079 }
2080 // Otherwise, it might either be oop+off or NULL+addr.
2081 return Type::AnyPtr;
2082 } else {
2083 // No information:
2084 return Type::AnyPtr;
2085 }
2086 }
2087
2088 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type) {
2089 int kind = classify_unsafe_addr(base, offset, type);
2090 if (kind == Type::RawPtr) {
2091 return basic_plus_adr(top(), base, offset);
2092 } else if (kind == Type::AnyPtr) {
2093 // We don't know if it's an on heap or off heap access. Fall back
2094 // to raw memory access.
2095 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2096 return basic_plus_adr(top(), raw, offset);
2097 } else {
2098 // We know it's an on heap access so base can't be null
2099 base = must_be_not_null(base, true);
2100 return basic_plus_adr(base, offset);
2101 }
2102 }
2103
2104 //--------------------------inline_number_methods-----------------------------
2105 // inline int Integer.numberOfLeadingZeros(int)
2106 // inline int Long.numberOfLeadingZeros(long)
2107 //
2108 // inline int Integer.numberOfTrailingZeros(int)
2109 // inline int Long.numberOfTrailingZeros(long)
2110 //
2111 // inline int Integer.bitCount(int)
2112 // inline int Long.bitCount(long)
2113 //
2114 // inline char Character.reverseBytes(char)
2115 // inline short Short.reverseBytes(short)
2116 // inline int Integer.reverseBytes(int)
2117 // inline long Long.reverseBytes(long)
2118 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2119 Node* arg = argument(0);
2120 Node* n = NULL;
2121 switch (id) {
2122 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2123 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2124 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2125 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2126 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2127 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2128 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2129 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2130 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2131 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2132 default: fatal_unexpected_iid(id); break;
2133 }
2134 set_result(_gvn.transform(n));
2135 return true;
2136 }
2137
2138 //----------------------------inline_unsafe_access----------------------------
2139
2140 // Helper that guards and inserts a pre-barrier.
2141 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2142 Node* pre_val, bool need_mem_bar) {
2143 // We could be accessing the referent field of a reference object. If so, when G1
2144 // is enabled, we need to log the value in the referent field in an SATB buffer.
2145 // This routine performs some compile time filters and generates suitable
2146 // runtime filters that guard the pre-barrier code.
2147 // Also add memory barrier for non volatile load from the referent field
2148 // to prevent commoning of loads across safepoint.
2149 if (!UseG1GC && !need_mem_bar)
2150 return;
2151
2152 // Some compile time checks.
2153
2154 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2155 const TypeX* otype = offset->find_intptr_t_type();
2156 if (otype != NULL && otype->is_con() &&
2157 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2158 // Constant offset but not the reference_offset so just return
2159 return;
2160 }
2161
2162 // We only need to generate the runtime guards for instances.
2163 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2164 if (btype != NULL) {
2165 if (btype->isa_aryptr()) {
2166 // Array type so nothing to do
2167 return;
2168 }
2169
2170 const TypeInstPtr* itype = btype->isa_instptr();
2171 if (itype != NULL) {
2172 // Can the klass of base_oop be statically determined to be
2173 // _not_ a sub-class of Reference and _not_ Object?
2174 ciKlass* klass = itype->klass();
2175 if ( klass->is_loaded() &&
2176 !klass->is_subtype_of(env()->Reference_klass()) &&
2177 !env()->Object_klass()->is_subtype_of(klass)) {
2178 return;
2179 }
2180 }
2181 }
2182
2183 // The compile time filters did not reject base_oop/offset so
2184 // we need to generate the following runtime filters
2185 //
2186 // if (offset == java_lang_ref_Reference::_reference_offset) {
2187 // if (instance_of(base, java.lang.ref.Reference)) {
2188 // pre_barrier(_, pre_val, ...);
2189 // }
2190 // }
2191
2192 float likely = PROB_LIKELY( 0.999);
2193 float unlikely = PROB_UNLIKELY(0.999);
2194
2195 IdealKit ideal(this);
2196 #define __ ideal.
2197
2198 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2199
2200 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2201 // Update graphKit memory and control from IdealKit.
2202 sync_kit(ideal);
2203
2204 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2205 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2206
2207 // Update IdealKit memory and control from graphKit.
2208 __ sync_kit(this);
2209
2210 Node* one = __ ConI(1);
2211 // is_instof == 0 if base_oop == NULL
2212 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2213
2214 // Update graphKit from IdeakKit.
2215 sync_kit(ideal);
2216
2217 // Use the pre-barrier to record the value in the referent field
2218 pre_barrier(false /* do_load */,
2219 __ ctrl(),
2220 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2221 pre_val /* pre_val */,
2222 T_OBJECT);
2223 if (need_mem_bar) {
2224 // Add memory barrier to prevent commoning reads from this field
2225 // across safepoint since GC can change its value.
2226 insert_mem_bar(Op_MemBarCPUOrder);
2227 }
2228 // Update IdealKit from graphKit.
2229 __ sync_kit(this);
2230
2231 } __ end_if(); // _ref_type != ref_none
2232 } __ end_if(); // offset == referent_offset
2233
2234 // Final sync IdealKit and GraphKit.
2235 final_sync(ideal);
2236 #undef __
2237 }
2238
2239
2240 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2241 // Attempt to infer a sharper value type from the offset and base type.
2242 ciKlass* sharpened_klass = NULL;
2243
2244 // See if it is an instance field, with an object type.
2245 if (alias_type->field() != NULL) {
2246 if (alias_type->field()->type()->is_klass()) {
2247 sharpened_klass = alias_type->field()->type()->as_klass();
2248 }
2249 }
2250
2251 // See if it is a narrow oop array.
2252 if (adr_type->isa_aryptr()) {
2253 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2254 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2255 if (elem_type != NULL) {
2256 sharpened_klass = elem_type->klass();
2257 }
2258 }
2259 }
2260
2261 // The sharpened class might be unloaded if there is no class loader
2262 // contraint in place.
2263 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2264 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2265
2266 #ifndef PRODUCT
2267 if (C->print_intrinsics() || C->print_inlining()) {
2268 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2269 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2270 }
2271 #endif
2272 // Sharpen the value type.
2273 return tjp;
2274 }
2275 return NULL;
2276 }
2277
2278 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2279 if (callee()->is_static()) return false; // caller must have the capability!
2280 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2281 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2282 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2283
2284 #ifndef PRODUCT
2285 {
2286 ResourceMark rm;
2287 // Check the signatures.
2288 ciSignature* sig = callee()->signature();
2289 #ifdef ASSERT
2290 if (!is_store) {
2291 // Object getObject(Object base, int/long offset), etc.
2292 BasicType rtype = sig->return_type()->basic_type();
2293 assert(rtype == type, "getter must return the expected value");
2294 assert(sig->count() == 2, "oop getter has 2 arguments");
2295 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2296 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2297 } else {
2298 // void putObject(Object base, int/long offset, Object x), etc.
2299 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2300 assert(sig->count() == 3, "oop putter has 3 arguments");
2301 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2302 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2303 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2304 assert(vtype == type, "putter must accept the expected value");
2305 }
2306 #endif // ASSERT
2307 }
2308 #endif //PRODUCT
2309
2310 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2311
2312 Node* receiver = argument(0); // type: oop
2313
2314 // Build address expression.
2315 Node* adr;
2316 Node* heap_base_oop = top();
2317 Node* offset = top();
2318 Node* val;
2319
2320 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2321 Node* base = argument(1); // type: oop
2322 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2323 offset = argument(2); // type: long
2324 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2325 // to be plain byte offsets, which are also the same as those accepted
2326 // by oopDesc::field_base.
2327 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2328 "fieldOffset must be byte-scaled");
2329 // 32-bit machines ignore the high half!
2330 offset = ConvL2X(offset);
2331 adr = make_unsafe_address(base, offset, type);
2332 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2333 heap_base_oop = base;
2334 } else if (type == T_OBJECT) {
2335 return false; // off-heap oop accesses are not supported
2336 }
2337
2338 // Can base be NULL? Otherwise, always on-heap access.
2339 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2340
2341 val = is_store ? argument(4) : NULL;
2342
2343 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2344
2345 // Try to categorize the address.
2346 Compile::AliasType* alias_type = C->alias_type(adr_type);
2347 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2348
2349 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2350 alias_type->adr_type() == TypeAryPtr::RANGE) {
2351 return false; // not supported
2352 }
2353
2354 bool mismatched = false;
2355 BasicType bt = alias_type->basic_type();
2356 if (bt != T_ILLEGAL) {
2357 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2358 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2359 // Alias type doesn't differentiate between byte[] and boolean[]).
2360 // Use address type to get the element type.
2361 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2362 }
2363 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2364 // accessing an array field with getObject is not a mismatch
2365 bt = T_OBJECT;
2366 }
2367 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2368 // Don't intrinsify mismatched object accesses
2369 return false;
2370 }
2371 mismatched = (bt != type);
2372 } else if (alias_type->adr_type()->isa_oopptr()) {
2373 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2374 }
2375
2376 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2377
2378 // First guess at the value type.
2379 const Type *value_type = Type::get_const_basic_type(type);
2380
2381 // We will need memory barriers unless we can determine a unique
2382 // alias category for this reference. (Note: If for some reason
2383 // the barriers get omitted and the unsafe reference begins to "pollute"
2384 // the alias analysis of the rest of the graph, either Compile::can_alias
2385 // or Compile::must_alias will throw a diagnostic assert.)
2386 bool need_mem_bar;
2387 switch (kind) {
2388 case Relaxed:
2389 need_mem_bar = mismatched || can_access_non_heap;
2390 break;
2391 case Opaque:
2392 // Opaque uses CPUOrder membars for protection against code movement.
2393 case Acquire:
2394 case Release:
2395 case Volatile:
2396 need_mem_bar = true;
2397 break;
2398 default:
2399 ShouldNotReachHere();
2400 }
2401
2402 // Some accesses require access atomicity for all types, notably longs and doubles.
2403 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2404 bool requires_atomic_access = false;
2405 switch (kind) {
2406 case Relaxed:
2407 requires_atomic_access = AlwaysAtomicAccesses;
2408 break;
2409 case Opaque:
2410 // Opaque accesses are atomic.
2411 case Acquire:
2412 case Release:
2413 case Volatile:
2414 requires_atomic_access = true;
2415 break;
2416 default:
2417 ShouldNotReachHere();
2418 }
2419
2420 // Figure out the memory ordering.
2421 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2422 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2423
2424 // If we are reading the value of the referent field of a Reference
2425 // object (either by using Unsafe directly or through reflection)
2426 // then, if G1 is enabled, we need to record the referent in an
2427 // SATB log buffer using the pre-barrier mechanism.
2428 // Also we need to add memory barrier to prevent commoning reads
2429 // from this field across safepoint since GC can change its value.
2430 bool need_read_barrier = !is_store &&
2431 offset != top() && heap_base_oop != top();
2432
2433 if (!is_store && type == T_OBJECT) {
2434 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2435 if (tjp != NULL) {
2436 value_type = tjp;
2437 }
2438 }
2439
2440 receiver = null_check(receiver);
2441 if (stopped()) {
2442 return true;
2443 }
2444 // Heap pointers get a null-check from the interpreter,
2445 // as a courtesy. However, this is not guaranteed by Unsafe,
2446 // and it is not possible to fully distinguish unintended nulls
2447 // from intended ones in this API.
2448
2449 // We need to emit leading and trailing CPU membars (see below) in
2450 // addition to memory membars for special access modes. This is a little
2451 // too strong, but avoids the need to insert per-alias-type
2452 // volatile membars (for stores; compare Parse::do_put_xxx), which
2453 // we cannot do effectively here because we probably only have a
2454 // rough approximation of type.
2455
2456 switch(kind) {
2457 case Relaxed:
2458 case Opaque:
2459 case Acquire:
2460 break;
2461 case Release:
2462 case Volatile:
2463 if (is_store) {
2464 insert_mem_bar(Op_MemBarRelease);
2465 } else {
2466 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2467 insert_mem_bar(Op_MemBarVolatile);
2468 }
2469 }
2470 break;
2471 default:
2472 ShouldNotReachHere();
2473 }
2474
2475 // Memory barrier to prevent normal and 'unsafe' accesses from
2476 // bypassing each other. Happens after null checks, so the
2477 // exception paths do not take memory state from the memory barrier,
2478 // so there's no problems making a strong assert about mixing users
2479 // of safe & unsafe memory.
2480 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2481
2482 if (!is_store) {
2483 Node* p = NULL;
2484 // Try to constant fold a load from a constant field
2485 ciField* field = alias_type->field();
2486 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2487 // final or stable field
2488 p = make_constant_from_field(field, heap_base_oop);
2489 }
2490 if (p == NULL) {
2491 // To be valid, unsafe loads may depend on other conditions than
2492 // the one that guards them: pin the Load node
2493 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2494 // load value
2495 switch (type) {
2496 case T_BOOLEAN:
2497 {
2498 // Normalize the value returned by getBoolean in the following cases
2499 if (mismatched ||
2500 heap_base_oop == top() || // - heap_base_oop is NULL or
2501 (can_access_non_heap && alias_type->field() == NULL) // - heap_base_oop is potentially NULL
2502 // and the unsafe access is made to large offset
2503 // (i.e., larger than the maximum offset necessary for any
2504 // field access)
2505 ) {
2506 IdealKit ideal = IdealKit(this);
2507 #define __ ideal.
2508 IdealVariable normalized_result(ideal);
2509 __ declarations_done();
2510 __ set(normalized_result, p);
2511 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2512 __ set(normalized_result, ideal.ConI(1));
2513 ideal.end_if();
2514 final_sync(ideal);
2515 p = __ value(normalized_result);
2516 #undef __
2517 }
2518 }
2519 case T_CHAR:
2520 case T_BYTE:
2521 case T_SHORT:
2522 case T_INT:
2523 case T_LONG:
2524 case T_FLOAT:
2525 case T_DOUBLE:
2526 break;
2527 case T_OBJECT:
2528 if (need_read_barrier) {
2529 // We do not require a mem bar inside pre_barrier if need_mem_bar
2530 // is set: the barriers would be emitted by us.
2531 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2532 }
2533 break;
2534 case T_ADDRESS:
2535 // Cast to an int type.
2536 p = _gvn.transform(new CastP2XNode(NULL, p));
2537 p = ConvX2UL(p);
2538 break;
2539 default:
2540 fatal("unexpected type %d: %s", type, type2name(type));
2541 break;
2542 }
2543 }
2544 // The load node has the control of the preceding MemBarCPUOrder. All
2545 // following nodes will have the control of the MemBarCPUOrder inserted at
2546 // the end of this method. So, pushing the load onto the stack at a later
2547 // point is fine.
2548 set_result(p);
2549 } else {
2550 // place effect of store into memory
2551 switch (type) {
2552 case T_DOUBLE:
2553 val = dstore_rounding(val);
2554 break;
2555 case T_ADDRESS:
2556 // Repackage the long as a pointer.
2557 val = ConvL2X(val);
2558 val = _gvn.transform(new CastX2PNode(val));
2559 break;
2560 }
2561
2562 if (type == T_OBJECT) {
2563 store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2564 } else {
2565 store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2566 }
2567 }
2568
2569 switch(kind) {
2570 case Relaxed:
2571 case Opaque:
2572 case Release:
2573 break;
2574 case Acquire:
2575 case Volatile:
2576 if (!is_store) {
2577 insert_mem_bar(Op_MemBarAcquire);
2578 } else {
2579 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2580 insert_mem_bar(Op_MemBarVolatile);
2581 }
2582 }
2583 break;
2584 default:
2585 ShouldNotReachHere();
2586 }
2587
2588 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2589
2590 return true;
2591 }
2592
2593 //----------------------------inline_unsafe_load_store----------------------------
2594 // This method serves a couple of different customers (depending on LoadStoreKind):
2595 //
2596 // LS_cmp_swap:
2597 //
2598 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2599 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2600 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2601 //
2602 // LS_cmp_swap_weak:
2603 //
2604 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2605 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2606 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2607 //
2608 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2609 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2610 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2611 //
2612 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2613 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2614 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2615 //
2616 // LS_cmp_exchange:
2617 //
2618 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2620 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2621 //
2622 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2623 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2624 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2625 //
2626 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2627 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2628 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2629 //
2630 // LS_get_add:
2631 //
2632 // int getAndAddInt( Object o, long offset, int delta)
2633 // long getAndAddLong(Object o, long offset, long delta)
2634 //
2635 // LS_get_set:
2636 //
2637 // int getAndSet(Object o, long offset, int newValue)
2638 // long getAndSet(Object o, long offset, long newValue)
2639 // Object getAndSet(Object o, long offset, Object newValue)
2640 //
2641 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2642 // This basic scheme here is the same as inline_unsafe_access, but
2643 // differs in enough details that combining them would make the code
2644 // overly confusing. (This is a true fact! I originally combined
2645 // them, but even I was confused by it!) As much code/comments as
2646 // possible are retained from inline_unsafe_access though to make
2647 // the correspondences clearer. - dl
2648
2649 if (callee()->is_static()) return false; // caller must have the capability!
2650
2651 #ifndef PRODUCT
2652 BasicType rtype;
2653 {
2654 ResourceMark rm;
2655 // Check the signatures.
2656 ciSignature* sig = callee()->signature();
2657 rtype = sig->return_type()->basic_type();
2658 switch(kind) {
2659 case LS_get_add:
2660 case LS_get_set: {
2661 // Check the signatures.
2662 #ifdef ASSERT
2663 assert(rtype == type, "get and set must return the expected type");
2664 assert(sig->count() == 3, "get and set has 3 arguments");
2665 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2666 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2667 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2668 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2669 #endif // ASSERT
2670 break;
2671 }
2672 case LS_cmp_swap:
2673 case LS_cmp_swap_weak: {
2674 // Check the signatures.
2675 #ifdef ASSERT
2676 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2677 assert(sig->count() == 4, "CAS has 4 arguments");
2678 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2679 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2680 #endif // ASSERT
2681 break;
2682 }
2683 case LS_cmp_exchange: {
2684 // Check the signatures.
2685 #ifdef ASSERT
2686 assert(rtype == type, "CAS must return the expected type");
2687 assert(sig->count() == 4, "CAS has 4 arguments");
2688 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2689 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2690 #endif // ASSERT
2691 break;
2692 }
2693 default:
2694 ShouldNotReachHere();
2695 }
2696 }
2697 #endif //PRODUCT
2698
2699 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2700
2701 // Get arguments:
2702 Node* receiver = NULL;
2703 Node* base = NULL;
2704 Node* offset = NULL;
2705 Node* oldval = NULL;
2706 Node* newval = NULL;
2707 switch(kind) {
2708 case LS_cmp_swap:
2709 case LS_cmp_swap_weak:
2710 case LS_cmp_exchange: {
2711 const bool two_slot_type = type2size[type] == 2;
2712 receiver = argument(0); // type: oop
2713 base = argument(1); // type: oop
2714 offset = argument(2); // type: long
2715 oldval = argument(4); // type: oop, int, or long
2716 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2717 break;
2718 }
2719 case LS_get_add:
2720 case LS_get_set: {
2721 receiver = argument(0); // type: oop
2722 base = argument(1); // type: oop
2723 offset = argument(2); // type: long
2724 oldval = NULL;
2725 newval = argument(4); // type: oop, int, or long
2726 break;
2727 }
2728 default:
2729 ShouldNotReachHere();
2730 }
2731
2732 // Build field offset expression.
2733 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2734 // to be plain byte offsets, which are also the same as those accepted
2735 // by oopDesc::field_base.
2736 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2737 // 32-bit machines ignore the high half of long offsets
2738 offset = ConvL2X(offset);
2739 Node* adr = make_unsafe_address(base, offset, type);
2740 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2741
2742 Compile::AliasType* alias_type = C->alias_type(adr_type);
2743 BasicType bt = alias_type->basic_type();
2744 if (bt != T_ILLEGAL &&
2745 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2746 // Don't intrinsify mismatched object accesses.
2747 return false;
2748 }
2749
2750 // For CAS, unlike inline_unsafe_access, there seems no point in
2751 // trying to refine types. Just use the coarse types here.
2752 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2753 const Type *value_type = Type::get_const_basic_type(type);
2754
2755 switch (kind) {
2756 case LS_get_set:
2757 case LS_cmp_exchange: {
2758 if (type == T_OBJECT) {
2759 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2760 if (tjp != NULL) {
2761 value_type = tjp;
2762 }
2763 }
2764 break;
2765 }
2766 case LS_cmp_swap:
2767 case LS_cmp_swap_weak:
2768 case LS_get_add:
2769 break;
2770 default:
2771 ShouldNotReachHere();
2772 }
2773
2774 // Null check receiver.
2775 receiver = null_check(receiver);
2776 if (stopped()) {
2777 return true;
2778 }
2779
2780 int alias_idx = C->get_alias_index(adr_type);
2781
2782 // Memory-model-wise, a LoadStore acts like a little synchronized
2783 // block, so needs barriers on each side. These don't translate
2784 // into actual barriers on most machines, but we still need rest of
2785 // compiler to respect ordering.
2786
2787 switch (access_kind) {
2788 case Relaxed:
2789 case Acquire:
2790 break;
2791 case Release:
2792 insert_mem_bar(Op_MemBarRelease);
2793 break;
2794 case Volatile:
2795 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2796 insert_mem_bar(Op_MemBarVolatile);
2797 } else {
2798 insert_mem_bar(Op_MemBarRelease);
2799 }
2800 break;
2801 default:
2802 ShouldNotReachHere();
2803 }
2804 insert_mem_bar(Op_MemBarCPUOrder);
2805
2806 // Figure out the memory ordering.
2807 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2808
2809 // 4984716: MemBars must be inserted before this
2810 // memory node in order to avoid a false
2811 // dependency which will confuse the scheduler.
2812 Node *mem = memory(alias_idx);
2813
2814 // For now, we handle only those cases that actually exist: ints,
2815 // longs, and Object. Adding others should be straightforward.
2816 Node* load_store = NULL;
2817 switch(type) {
2818 case T_BYTE:
2819 switch(kind) {
2820 case LS_get_add:
2821 load_store = _gvn.transform(new GetAndAddBNode(control(), mem, adr, newval, adr_type));
2822 break;
2823 case LS_get_set:
2824 load_store = _gvn.transform(new GetAndSetBNode(control(), mem, adr, newval, adr_type));
2825 break;
2826 case LS_cmp_swap_weak:
2827 load_store = _gvn.transform(new WeakCompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2828 break;
2829 case LS_cmp_swap:
2830 load_store = _gvn.transform(new CompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2831 break;
2832 case LS_cmp_exchange:
2833 load_store = _gvn.transform(new CompareAndExchangeBNode(control(), mem, adr, newval, oldval, adr_type, mo));
2834 break;
2835 default:
2836 ShouldNotReachHere();
2837 }
2838 break;
2839 case T_SHORT:
2840 switch(kind) {
2841 case LS_get_add:
2842 load_store = _gvn.transform(new GetAndAddSNode(control(), mem, adr, newval, adr_type));
2843 break;
2844 case LS_get_set:
2845 load_store = _gvn.transform(new GetAndSetSNode(control(), mem, adr, newval, adr_type));
2846 break;
2847 case LS_cmp_swap_weak:
2848 load_store = _gvn.transform(new WeakCompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2849 break;
2850 case LS_cmp_swap:
2851 load_store = _gvn.transform(new CompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2852 break;
2853 case LS_cmp_exchange:
2854 load_store = _gvn.transform(new CompareAndExchangeSNode(control(), mem, adr, newval, oldval, adr_type, mo));
2855 break;
2856 default:
2857 ShouldNotReachHere();
2858 }
2859 break;
2860 case T_INT:
2861 switch(kind) {
2862 case LS_get_add:
2863 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2864 break;
2865 case LS_get_set:
2866 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2867 break;
2868 case LS_cmp_swap_weak:
2869 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2870 break;
2871 case LS_cmp_swap:
2872 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2873 break;
2874 case LS_cmp_exchange:
2875 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2876 break;
2877 default:
2878 ShouldNotReachHere();
2879 }
2880 break;
2881 case T_LONG:
2882 switch(kind) {
2883 case LS_get_add:
2884 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2885 break;
2886 case LS_get_set:
2887 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2888 break;
2889 case LS_cmp_swap_weak:
2890 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2891 break;
2892 case LS_cmp_swap:
2893 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2894 break;
2895 case LS_cmp_exchange:
2896 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2897 break;
2898 default:
2899 ShouldNotReachHere();
2900 }
2901 break;
2902 case T_OBJECT:
2903 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2904 // could be delayed during Parse (for example, in adjust_map_after_if()).
2905 // Execute transformation here to avoid barrier generation in such case.
2906 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2907 newval = _gvn.makecon(TypePtr::NULL_PTR);
2908
2909 // Reference stores need a store barrier.
2910 switch(kind) {
2911 case LS_get_set: {
2912 // If pre-barrier must execute before the oop store, old value will require do_load here.
2913 if (!can_move_pre_barrier()) {
2914 pre_barrier(true /* do_load*/,
2915 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2916 NULL /* pre_val*/,
2917 T_OBJECT);
2918 } // Else move pre_barrier to use load_store value, see below.
2919 break;
2920 }
2921 case LS_cmp_swap_weak:
2922 case LS_cmp_swap:
2923 case LS_cmp_exchange: {
2924 // Same as for newval above:
2925 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2926 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2927 }
2928 // The only known value which might get overwritten is oldval.
2929 pre_barrier(false /* do_load */,
2930 control(), NULL, NULL, max_juint, NULL, NULL,
2931 oldval /* pre_val */,
2932 T_OBJECT);
2933 break;
2934 }
2935 default:
2936 ShouldNotReachHere();
2937 }
2938
2939 #ifdef _LP64
2940 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2941 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2942
2943 switch(kind) {
2944 case LS_get_set:
2945 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2946 break;
2947 case LS_cmp_swap_weak: {
2948 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2949 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2950 break;
2951 }
2952 case LS_cmp_swap: {
2953 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2954 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2955 break;
2956 }
2957 case LS_cmp_exchange: {
2958 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2959 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2960 break;
2961 }
2962 default:
2963 ShouldNotReachHere();
2964 }
2965 } else
2966 #endif
2967 switch (kind) {
2968 case LS_get_set:
2969 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2970 break;
2971 case LS_cmp_swap_weak:
2972 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2973 break;
2974 case LS_cmp_swap:
2975 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2976 break;
2977 case LS_cmp_exchange:
2978 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
2979 break;
2980 default:
2981 ShouldNotReachHere();
2982 }
2983
2984 // Emit the post barrier only when the actual store happened. This makes sense
2985 // to check only for LS_cmp_* that can fail to set the value.
2986 // LS_cmp_exchange does not produce any branches by default, so there is no
2987 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
2988 // CompareAndExchange and move branches here, it would make sense to conditionalize
2989 // post_barriers for LS_cmp_exchange as well.
2990 //
2991 // CAS success path is marked more likely since we anticipate this is a performance
2992 // critical path, while CAS failure path can use the penalty for going through unlikely
2993 // path as backoff. Which is still better than doing a store barrier there.
2994 switch (kind) {
2995 case LS_get_set:
2996 case LS_cmp_exchange: {
2997 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2998 break;
2999 }
3000 case LS_cmp_swap_weak:
3001 case LS_cmp_swap: {
3002 IdealKit ideal(this);
3003 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3004 sync_kit(ideal);
3005 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3006 ideal.sync_kit(this);
3007 } ideal.end_if();
3008 final_sync(ideal);
3009 break;
3010 }
3011 default:
3012 ShouldNotReachHere();
3013 }
3014 break;
3015 default:
3016 fatal("unexpected type %d: %s", type, type2name(type));
3017 break;
3018 }
3019
3020 // SCMemProjNodes represent the memory state of a LoadStore. Their
3021 // main role is to prevent LoadStore nodes from being optimized away
3022 // when their results aren't used.
3023 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3024 set_memory(proj, alias_idx);
3025
3026 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3027 #ifdef _LP64
3028 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3029 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3030 }
3031 #endif
3032 if (can_move_pre_barrier()) {
3033 // Don't need to load pre_val. The old value is returned by load_store.
3034 // The pre_barrier can execute after the xchg as long as no safepoint
3035 // gets inserted between them.
3036 pre_barrier(false /* do_load */,
3037 control(), NULL, NULL, max_juint, NULL, NULL,
3038 load_store /* pre_val */,
3039 T_OBJECT);
3040 }
3041 }
3042
3043 // Add the trailing membar surrounding the access
3044 insert_mem_bar(Op_MemBarCPUOrder);
3045
3046 switch (access_kind) {
3047 case Relaxed:
3048 case Release:
3049 break; // do nothing
3050 case Acquire:
3051 case Volatile:
3052 insert_mem_bar(Op_MemBarAcquire);
3053 // !support_IRIW_for_not_multiple_copy_atomic_cpu handled in platform code
3054 break;
3055 default:
3056 ShouldNotReachHere();
3057 }
3058
3059 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3060 set_result(load_store);
3061 return true;
3062 }
3063
3064 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3065 MemNode::MemOrd mo = MemNode::unset;
3066 switch(kind) {
3067 case Opaque:
3068 case Relaxed: mo = MemNode::unordered; break;
3069 case Acquire: mo = MemNode::acquire; break;
3070 case Release: mo = MemNode::release; break;
3071 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3072 default:
3073 ShouldNotReachHere();
3074 }
3075 guarantee(mo != MemNode::unset, "Should select memory ordering");
3076 return mo;
3077 }
3078
3079 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3080 MemNode::MemOrd mo = MemNode::unset;
3081 switch(kind) {
3082 case Opaque:
3083 case Relaxed: mo = MemNode::unordered; break;
3084 case Acquire: mo = MemNode::acquire; break;
3085 case Release: mo = MemNode::release; break;
3086 case Volatile: mo = MemNode::seqcst; break;
3087 default:
3088 ShouldNotReachHere();
3089 }
3090 guarantee(mo != MemNode::unset, "Should select memory ordering");
3091 return mo;
3092 }
3093
3094 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3095 // Regardless of form, don't allow previous ld/st to move down,
3096 // then issue acquire, release, or volatile mem_bar.
3097 insert_mem_bar(Op_MemBarCPUOrder);
3098 switch(id) {
3099 case vmIntrinsics::_loadFence:
3100 insert_mem_bar(Op_LoadFence);
3101 return true;
3102 case vmIntrinsics::_storeFence:
3103 insert_mem_bar(Op_StoreFence);
3104 return true;
3105 case vmIntrinsics::_fullFence:
3106 insert_mem_bar(Op_MemBarVolatile);
3107 return true;
3108 default:
3109 fatal_unexpected_iid(id);
3110 return false;
3111 }
3112 }
3113
3114 bool LibraryCallKit::inline_onspinwait() {
3115 insert_mem_bar(Op_OnSpinWait);
3116 return true;
3117 }
3118
3119 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3120 if (!kls->is_Con()) {
3121 return true;
3122 }
3123 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3124 if (klsptr == NULL) {
3125 return true;
3126 }
3127 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3128 // don't need a guard for a klass that is already initialized
3129 return !ik->is_initialized();
3130 }
3131
3132 //----------------------------inline_unsafe_allocate---------------------------
3133 // public native Object Unsafe.allocateInstance(Class<?> cls);
3134 bool LibraryCallKit::inline_unsafe_allocate() {
3135 if (callee()->is_static()) return false; // caller must have the capability!
3136
3137 null_check_receiver(); // null-check, then ignore
3138 Node* cls = null_check(argument(1));
3139 if (stopped()) return true;
3140
3141 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3142 kls = null_check(kls);
3143 if (stopped()) return true; // argument was like int.class
3144
3145 Node* test = NULL;
3146 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3147 // Note: The argument might still be an illegal value like
3148 // Serializable.class or Object[].class. The runtime will handle it.
3149 // But we must make an explicit check for initialization.
3150 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3151 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3152 // can generate code to load it as unsigned byte.
3153 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3154 Node* bits = intcon(InstanceKlass::fully_initialized);
3155 test = _gvn.transform(new SubINode(inst, bits));
3156 // The 'test' is non-zero if we need to take a slow path.
3157 }
3158
3159 Node* obj = new_instance(kls, test);
3160 set_result(obj);
3161 return true;
3162 }
3163
3164 //------------------------inline_native_time_funcs--------------
3165 // inline code for System.currentTimeMillis() and System.nanoTime()
3166 // these have the same type and signature
3167 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3168 const TypeFunc* tf = OptoRuntime::void_long_Type();
3169 const TypePtr* no_memory_effects = NULL;
3170 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3171 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3172 #ifdef ASSERT
3173 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3174 assert(value_top == top(), "second value must be top");
3175 #endif
3176 set_result(value);
3177 return true;
3178 }
3179
3180 #ifdef TRACE_HAVE_INTRINSICS
3181
3182 /*
3183 * oop -> myklass
3184 * myklass->trace_id |= USED
3185 * return myklass->trace_id & ~0x3
3186 */
3187 bool LibraryCallKit::inline_native_classID() {
3188 Node* cls = null_check(argument(0), T_OBJECT);
3189 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3190 kls = null_check(kls, T_OBJECT);
3191
3192 ByteSize offset = TRACE_KLASS_TRACE_ID_OFFSET;
3193 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3194 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3195
3196 Node* clsused = longcon(0x01l); // set the class bit
3197 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3198 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3199 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3200
3201 #ifdef TRACE_ID_META_BITS
3202 Node* mbits = longcon(~TRACE_ID_META_BITS);
3203 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
3204 #endif
3205 #ifdef TRACE_ID_CLASS_SHIFT
3206 Node* cbits = intcon(TRACE_ID_CLASS_SHIFT);
3207 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
3208 #endif
3209
3210 set_result(tvalue);
3211 return true;
3212
3213 }
3214
3215 bool LibraryCallKit::inline_native_getBufferWriter() {
3216 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3217
3218 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3219 in_bytes(TRACE_THREAD_DATA_WRITER_OFFSET)
3220 );
3221
3222 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3223
3224 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3225 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3226
3227 IfNode* iff_jobj_null =
3228 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3229
3230 enum { _normal_path = 1,
3231 _null_path = 2,
3232 PATH_LIMIT };
3233
3234 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3235 PhiNode* result_val = new PhiNode(result_rgn, TypePtr::BOTTOM);
3236
3237 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3238 result_rgn->init_req(_null_path, jobj_is_null);
3239 result_val->init_req(_null_path, null());
3240
3241 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3242 result_rgn->init_req(_normal_path, jobj_is_not_null);
3243
3244 Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
3245 result_val->init_req(_normal_path, res);
3246
3247 set_result(result_rgn, result_val);
3248
3249 return true;
3250 }
3251
3252 #endif
3253
3254 //------------------------inline_native_currentThread------------------
3255 bool LibraryCallKit::inline_native_currentThread() {
3256 Node* junk = NULL;
3257 set_result(generate_current_thread(junk));
3258 return true;
3259 }
3260
3261 //------------------------inline_native_isInterrupted------------------
3262 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3263 bool LibraryCallKit::inline_native_isInterrupted() {
3264 // Add a fast path to t.isInterrupted(clear_int):
3265 // (t == Thread.current() &&
3266 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3267 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3268 // So, in the common case that the interrupt bit is false,
3269 // we avoid making a call into the VM. Even if the interrupt bit
3270 // is true, if the clear_int argument is false, we avoid the VM call.
3271 // However, if the receiver is not currentThread, we must call the VM,
3272 // because there must be some locking done around the operation.
3273
3274 // We only go to the fast case code if we pass two guards.
3275 // Paths which do not pass are accumulated in the slow_region.
3276
3277 enum {
3278 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3279 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3280 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3281 PATH_LIMIT
3282 };
3283
3284 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3285 // out of the function.
3286 insert_mem_bar(Op_MemBarCPUOrder);
3287
3288 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3289 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3290
3291 RegionNode* slow_region = new RegionNode(1);
3292 record_for_igvn(slow_region);
3293
3294 // (a) Receiving thread must be the current thread.
3295 Node* rec_thr = argument(0);
3296 Node* tls_ptr = NULL;
3297 Node* cur_thr = generate_current_thread(tls_ptr);
3298 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3299 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3300
3301 generate_slow_guard(bol_thr, slow_region);
3302
3303 // (b) Interrupt bit on TLS must be false.
3304 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3305 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3306 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3307
3308 // Set the control input on the field _interrupted read to prevent it floating up.
3309 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3310 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3311 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3312
3313 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3314
3315 // First fast path: if (!TLS._interrupted) return false;
3316 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3317 result_rgn->init_req(no_int_result_path, false_bit);
3318 result_val->init_req(no_int_result_path, intcon(0));
3319
3320 // drop through to next case
3321 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3322
3323 #ifndef _WINDOWS
3324 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3325 Node* clr_arg = argument(1);
3326 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3327 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3328 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3329
3330 // Second fast path: ... else if (!clear_int) return true;
3331 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3332 result_rgn->init_req(no_clear_result_path, false_arg);
3333 result_val->init_req(no_clear_result_path, intcon(1));
3334
3335 // drop through to next case
3336 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3337 #else
3338 // To return true on Windows you must read the _interrupted field
3339 // and check the event state i.e. take the slow path.
3340 #endif // _WINDOWS
3341
3342 // (d) Otherwise, go to the slow path.
3343 slow_region->add_req(control());
3344 set_control( _gvn.transform(slow_region));
3345
3346 if (stopped()) {
3347 // There is no slow path.
3348 result_rgn->init_req(slow_result_path, top());
3349 result_val->init_req(slow_result_path, top());
3350 } else {
3351 // non-virtual because it is a private non-static
3352 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3353
3354 Node* slow_val = set_results_for_java_call(slow_call);
3355 // this->control() comes from set_results_for_java_call
3356
3357 Node* fast_io = slow_call->in(TypeFunc::I_O);
3358 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3359
3360 // These two phis are pre-filled with copies of of the fast IO and Memory
3361 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3362 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3363
3364 result_rgn->init_req(slow_result_path, control());
3365 result_io ->init_req(slow_result_path, i_o());
3366 result_mem->init_req(slow_result_path, reset_memory());
3367 result_val->init_req(slow_result_path, slow_val);
3368
3369 set_all_memory(_gvn.transform(result_mem));
3370 set_i_o( _gvn.transform(result_io));
3371 }
3372
3373 C->set_has_split_ifs(true); // Has chance for split-if optimization
3374 set_result(result_rgn, result_val);
3375 return true;
3376 }
3377
3378 //---------------------------load_mirror_from_klass----------------------------
3379 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3380 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3381 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3382 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3383 }
3384
3385 //-----------------------load_klass_from_mirror_common-------------------------
3386 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3387 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3388 // and branch to the given path on the region.
3389 // If never_see_null, take an uncommon trap on null, so we can optimistically
3390 // compile for the non-null case.
3391 // If the region is NULL, force never_see_null = true.
3392 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3393 bool never_see_null,
3394 RegionNode* region,
3395 int null_path,
3396 int offset) {
3397 if (region == NULL) never_see_null = true;
3398 Node* p = basic_plus_adr(mirror, offset);
3399 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3400 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3401 Node* null_ctl = top();
3402 kls = null_check_oop(kls, &null_ctl, never_see_null);
3403 if (region != NULL) {
3404 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3405 region->init_req(null_path, null_ctl);
3406 } else {
3407 assert(null_ctl == top(), "no loose ends");
3408 }
3409 return kls;
3410 }
3411
3412 //--------------------(inline_native_Class_query helpers)---------------------
3413 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3414 // Fall through if (mods & mask) == bits, take the guard otherwise.
3415 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3416 // Branch around if the given klass has the given modifier bit set.
3417 // Like generate_guard, adds a new path onto the region.
3418 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3419 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3420 Node* mask = intcon(modifier_mask);
3421 Node* bits = intcon(modifier_bits);
3422 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3423 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3424 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3425 return generate_fair_guard(bol, region);
3426 }
3427 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3428 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3429 }
3430
3431 //-------------------------inline_native_Class_query-------------------
3432 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3433 const Type* return_type = TypeInt::BOOL;
3434 Node* prim_return_value = top(); // what happens if it's a primitive class?
3435 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3436 bool expect_prim = false; // most of these guys expect to work on refs
3437
3438 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3439
3440 Node* mirror = argument(0);
3441 Node* obj = top();
3442
3443 switch (id) {
3444 case vmIntrinsics::_isInstance:
3445 // nothing is an instance of a primitive type
3446 prim_return_value = intcon(0);
3447 obj = argument(1);
3448 break;
3449 case vmIntrinsics::_getModifiers:
3450 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3451 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3452 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3453 break;
3454 case vmIntrinsics::_isInterface:
3455 prim_return_value = intcon(0);
3456 break;
3457 case vmIntrinsics::_isArray:
3458 prim_return_value = intcon(0);
3459 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3460 break;
3461 case vmIntrinsics::_isPrimitive:
3462 prim_return_value = intcon(1);
3463 expect_prim = true; // obviously
3464 break;
3465 case vmIntrinsics::_getSuperclass:
3466 prim_return_value = null();
3467 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3468 break;
3469 case vmIntrinsics::_getClassAccessFlags:
3470 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3471 return_type = TypeInt::INT; // not bool! 6297094
3472 break;
3473 default:
3474 fatal_unexpected_iid(id);
3475 break;
3476 }
3477
3478 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3479 if (mirror_con == NULL) return false; // cannot happen?
3480
3481 #ifndef PRODUCT
3482 if (C->print_intrinsics() || C->print_inlining()) {
3483 ciType* k = mirror_con->java_mirror_type();
3484 if (k) {
3485 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3486 k->print_name();
3487 tty->cr();
3488 }
3489 }
3490 #endif
3491
3492 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3493 RegionNode* region = new RegionNode(PATH_LIMIT);
3494 record_for_igvn(region);
3495 PhiNode* phi = new PhiNode(region, return_type);
3496
3497 // The mirror will never be null of Reflection.getClassAccessFlags, however
3498 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3499 // if it is. See bug 4774291.
3500
3501 // For Reflection.getClassAccessFlags(), the null check occurs in
3502 // the wrong place; see inline_unsafe_access(), above, for a similar
3503 // situation.
3504 mirror = null_check(mirror);
3505 // If mirror or obj is dead, only null-path is taken.
3506 if (stopped()) return true;
3507
3508 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3509
3510 // Now load the mirror's klass metaobject, and null-check it.
3511 // Side-effects region with the control path if the klass is null.
3512 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3513 // If kls is null, we have a primitive mirror.
3514 phi->init_req(_prim_path, prim_return_value);
3515 if (stopped()) { set_result(region, phi); return true; }
3516 bool safe_for_replace = (region->in(_prim_path) == top());
3517
3518 Node* p; // handy temp
3519 Node* null_ctl;
3520
3521 // Now that we have the non-null klass, we can perform the real query.
3522 // For constant classes, the query will constant-fold in LoadNode::Value.
3523 Node* query_value = top();
3524 switch (id) {
3525 case vmIntrinsics::_isInstance:
3526 // nothing is an instance of a primitive type
3527 query_value = gen_instanceof(obj, kls, safe_for_replace);
3528 break;
3529
3530 case vmIntrinsics::_getModifiers:
3531 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3532 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3533 break;
3534
3535 case vmIntrinsics::_isInterface:
3536 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3537 if (generate_interface_guard(kls, region) != NULL)
3538 // A guard was added. If the guard is taken, it was an interface.
3539 phi->add_req(intcon(1));
3540 // If we fall through, it's a plain class.
3541 query_value = intcon(0);
3542 break;
3543
3544 case vmIntrinsics::_isArray:
3545 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3546 if (generate_array_guard(kls, region) != NULL)
3547 // A guard was added. If the guard is taken, it was an array.
3548 phi->add_req(intcon(1));
3549 // If we fall through, it's a plain class.
3550 query_value = intcon(0);
3551 break;
3552
3553 case vmIntrinsics::_isPrimitive:
3554 query_value = intcon(0); // "normal" path produces false
3555 break;
3556
3557 case vmIntrinsics::_getSuperclass:
3558 // The rules here are somewhat unfortunate, but we can still do better
3559 // with random logic than with a JNI call.
3560 // Interfaces store null or Object as _super, but must report null.
3561 // Arrays store an intermediate super as _super, but must report Object.
3562 // Other types can report the actual _super.
3563 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3564 if (generate_interface_guard(kls, region) != NULL)
3565 // A guard was added. If the guard is taken, it was an interface.
3566 phi->add_req(null());
3567 if (generate_array_guard(kls, region) != NULL)
3568 // A guard was added. If the guard is taken, it was an array.
3569 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3570 // If we fall through, it's a plain class. Get its _super.
3571 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3572 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3573 null_ctl = top();
3574 kls = null_check_oop(kls, &null_ctl);
3575 if (null_ctl != top()) {
3576 // If the guard is taken, Object.superClass is null (both klass and mirror).
3577 region->add_req(null_ctl);
3578 phi ->add_req(null());
3579 }
3580 if (!stopped()) {
3581 query_value = load_mirror_from_klass(kls);
3582 }
3583 break;
3584
3585 case vmIntrinsics::_getClassAccessFlags:
3586 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3587 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3588 break;
3589
3590 default:
3591 fatal_unexpected_iid(id);
3592 break;
3593 }
3594
3595 // Fall-through is the normal case of a query to a real class.
3596 phi->init_req(1, query_value);
3597 region->init_req(1, control());
3598
3599 C->set_has_split_ifs(true); // Has chance for split-if optimization
3600 set_result(region, phi);
3601 return true;
3602 }
3603
3604 //-------------------------inline_Class_cast-------------------
3605 bool LibraryCallKit::inline_Class_cast() {
3606 Node* mirror = argument(0); // Class
3607 Node* obj = argument(1);
3608 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3609 if (mirror_con == NULL) {
3610 return false; // dead path (mirror->is_top()).
3611 }
3612 if (obj == NULL || obj->is_top()) {
3613 return false; // dead path
3614 }
3615 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3616
3617 // First, see if Class.cast() can be folded statically.
3618 // java_mirror_type() returns non-null for compile-time Class constants.
3619 ciType* tm = mirror_con->java_mirror_type();
3620 if (tm != NULL && tm->is_klass() &&
3621 tp != NULL && tp->klass() != NULL) {
3622 if (!tp->klass()->is_loaded()) {
3623 // Don't use intrinsic when class is not loaded.
3624 return false;
3625 } else {
3626 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3627 if (static_res == Compile::SSC_always_true) {
3628 // isInstance() is true - fold the code.
3629 set_result(obj);
3630 return true;
3631 } else if (static_res == Compile::SSC_always_false) {
3632 // Don't use intrinsic, have to throw ClassCastException.
3633 // If the reference is null, the non-intrinsic bytecode will
3634 // be optimized appropriately.
3635 return false;
3636 }
3637 }
3638 }
3639
3640 // Bailout intrinsic and do normal inlining if exception path is frequent.
3641 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3642 return false;
3643 }
3644
3645 // Generate dynamic checks.
3646 // Class.cast() is java implementation of _checkcast bytecode.
3647 // Do checkcast (Parse::do_checkcast()) optimizations here.
3648
3649 mirror = null_check(mirror);
3650 // If mirror is dead, only null-path is taken.
3651 if (stopped()) {
3652 return true;
3653 }
3654
3655 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3656 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3657 RegionNode* region = new RegionNode(PATH_LIMIT);
3658 record_for_igvn(region);
3659
3660 // Now load the mirror's klass metaobject, and null-check it.
3661 // If kls is null, we have a primitive mirror and
3662 // nothing is an instance of a primitive type.
3663 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3664
3665 Node* res = top();
3666 if (!stopped()) {
3667 Node* bad_type_ctrl = top();
3668 // Do checkcast optimizations.
3669 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3670 region->init_req(_bad_type_path, bad_type_ctrl);
3671 }
3672 if (region->in(_prim_path) != top() ||
3673 region->in(_bad_type_path) != top()) {
3674 // Let Interpreter throw ClassCastException.
3675 PreserveJVMState pjvms(this);
3676 set_control(_gvn.transform(region));
3677 uncommon_trap(Deoptimization::Reason_intrinsic,
3678 Deoptimization::Action_maybe_recompile);
3679 }
3680 if (!stopped()) {
3681 set_result(res);
3682 }
3683 return true;
3684 }
3685
3686
3687 //--------------------------inline_native_subtype_check------------------------
3688 // This intrinsic takes the JNI calls out of the heart of
3689 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3690 bool LibraryCallKit::inline_native_subtype_check() {
3691 // Pull both arguments off the stack.
3692 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3693 args[0] = argument(0);
3694 args[1] = argument(1);
3695 Node* klasses[2]; // corresponding Klasses: superk, subk
3696 klasses[0] = klasses[1] = top();
3697
3698 enum {
3699 // A full decision tree on {superc is prim, subc is prim}:
3700 _prim_0_path = 1, // {P,N} => false
3701 // {P,P} & superc!=subc => false
3702 _prim_same_path, // {P,P} & superc==subc => true
3703 _prim_1_path, // {N,P} => false
3704 _ref_subtype_path, // {N,N} & subtype check wins => true
3705 _both_ref_path, // {N,N} & subtype check loses => false
3706 PATH_LIMIT
3707 };
3708
3709 RegionNode* region = new RegionNode(PATH_LIMIT);
3710 Node* phi = new PhiNode(region, TypeInt::BOOL);
3711 record_for_igvn(region);
3712
3713 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3714 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3715 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3716
3717 // First null-check both mirrors and load each mirror's klass metaobject.
3718 int which_arg;
3719 for (which_arg = 0; which_arg <= 1; which_arg++) {
3720 Node* arg = args[which_arg];
3721 arg = null_check(arg);
3722 if (stopped()) break;
3723 args[which_arg] = arg;
3724
3725 Node* p = basic_plus_adr(arg, class_klass_offset);
3726 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3727 klasses[which_arg] = _gvn.transform(kls);
3728 }
3729
3730 // Having loaded both klasses, test each for null.
3731 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3732 for (which_arg = 0; which_arg <= 1; which_arg++) {
3733 Node* kls = klasses[which_arg];
3734 Node* null_ctl = top();
3735 kls = null_check_oop(kls, &null_ctl, never_see_null);
3736 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3737 region->init_req(prim_path, null_ctl);
3738 if (stopped()) break;
3739 klasses[which_arg] = kls;
3740 }
3741
3742 if (!stopped()) {
3743 // now we have two reference types, in klasses[0..1]
3744 Node* subk = klasses[1]; // the argument to isAssignableFrom
3745 Node* superk = klasses[0]; // the receiver
3746 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3747 // now we have a successful reference subtype check
3748 region->set_req(_ref_subtype_path, control());
3749 }
3750
3751 // If both operands are primitive (both klasses null), then
3752 // we must return true when they are identical primitives.
3753 // It is convenient to test this after the first null klass check.
3754 set_control(region->in(_prim_0_path)); // go back to first null check
3755 if (!stopped()) {
3756 // Since superc is primitive, make a guard for the superc==subc case.
3757 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3758 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3759 generate_guard(bol_eq, region, PROB_FAIR);
3760 if (region->req() == PATH_LIMIT+1) {
3761 // A guard was added. If the added guard is taken, superc==subc.
3762 region->swap_edges(PATH_LIMIT, _prim_same_path);
3763 region->del_req(PATH_LIMIT);
3764 }
3765 region->set_req(_prim_0_path, control()); // Not equal after all.
3766 }
3767
3768 // these are the only paths that produce 'true':
3769 phi->set_req(_prim_same_path, intcon(1));
3770 phi->set_req(_ref_subtype_path, intcon(1));
3771
3772 // pull together the cases:
3773 assert(region->req() == PATH_LIMIT, "sane region");
3774 for (uint i = 1; i < region->req(); i++) {
3775 Node* ctl = region->in(i);
3776 if (ctl == NULL || ctl == top()) {
3777 region->set_req(i, top());
3778 phi ->set_req(i, top());
3779 } else if (phi->in(i) == NULL) {
3780 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3781 }
3782 }
3783
3784 set_control(_gvn.transform(region));
3785 set_result(_gvn.transform(phi));
3786 return true;
3787 }
3788
3789 //---------------------generate_array_guard_common------------------------
3790 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3791 bool obj_array, bool not_array) {
3792
3793 if (stopped()) {
3794 return NULL;
3795 }
3796
3797 // If obj_array/non_array==false/false:
3798 // Branch around if the given klass is in fact an array (either obj or prim).
3799 // If obj_array/non_array==false/true:
3800 // Branch around if the given klass is not an array klass of any kind.
3801 // If obj_array/non_array==true/true:
3802 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3803 // If obj_array/non_array==true/false:
3804 // Branch around if the kls is an oop array (Object[] or subtype)
3805 //
3806 // Like generate_guard, adds a new path onto the region.
3807 jint layout_con = 0;
3808 Node* layout_val = get_layout_helper(kls, layout_con);
3809 if (layout_val == NULL) {
3810 bool query = (obj_array
3811 ? Klass::layout_helper_is_objArray(layout_con)
3812 : Klass::layout_helper_is_array(layout_con));
3813 if (query == not_array) {
3814 return NULL; // never a branch
3815 } else { // always a branch
3816 Node* always_branch = control();
3817 if (region != NULL)
3818 region->add_req(always_branch);
3819 set_control(top());
3820 return always_branch;
3821 }
3822 }
3823 // Now test the correct condition.
3824 jint nval = (obj_array
3825 ? (jint)(Klass::_lh_array_tag_type_value
3826 << Klass::_lh_array_tag_shift)
3827 : Klass::_lh_neutral_value);
3828 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3829 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3830 // invert the test if we are looking for a non-array
3831 if (not_array) btest = BoolTest(btest).negate();
3832 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3833 return generate_fair_guard(bol, region);
3834 }
3835
3836
3837 //-----------------------inline_native_newArray--------------------------
3838 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3839 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3840 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3841 Node* mirror;
3842 Node* count_val;
3843 if (uninitialized) {
3844 mirror = argument(1);
3845 count_val = argument(2);
3846 } else {
3847 mirror = argument(0);
3848 count_val = argument(1);
3849 }
3850
3851 mirror = null_check(mirror);
3852 // If mirror or obj is dead, only null-path is taken.
3853 if (stopped()) return true;
3854
3855 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3856 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3857 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3858 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3859 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3860
3861 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3862 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3863 result_reg, _slow_path);
3864 Node* normal_ctl = control();
3865 Node* no_array_ctl = result_reg->in(_slow_path);
3866
3867 // Generate code for the slow case. We make a call to newArray().
3868 set_control(no_array_ctl);
3869 if (!stopped()) {
3870 // Either the input type is void.class, or else the
3871 // array klass has not yet been cached. Either the
3872 // ensuing call will throw an exception, or else it
3873 // will cache the array klass for next time.
3874 PreserveJVMState pjvms(this);
3875 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3876 Node* slow_result = set_results_for_java_call(slow_call);
3877 // this->control() comes from set_results_for_java_call
3878 result_reg->set_req(_slow_path, control());
3879 result_val->set_req(_slow_path, slow_result);
3880 result_io ->set_req(_slow_path, i_o());
3881 result_mem->set_req(_slow_path, reset_memory());
3882 }
3883
3884 set_control(normal_ctl);
3885 if (!stopped()) {
3886 // Normal case: The array type has been cached in the java.lang.Class.
3887 // The following call works fine even if the array type is polymorphic.
3888 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3889 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3890 result_reg->init_req(_normal_path, control());
3891 result_val->init_req(_normal_path, obj);
3892 result_io ->init_req(_normal_path, i_o());
3893 result_mem->init_req(_normal_path, reset_memory());
3894
3895 if (uninitialized) {
3896 // Mark the allocation so that zeroing is skipped
3897 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3898 alloc->maybe_set_complete(&_gvn);
3899 }
3900 }
3901
3902 // Return the combined state.
3903 set_i_o( _gvn.transform(result_io) );
3904 set_all_memory( _gvn.transform(result_mem));
3905
3906 C->set_has_split_ifs(true); // Has chance for split-if optimization
3907 set_result(result_reg, result_val);
3908 return true;
3909 }
3910
3911 //----------------------inline_native_getLength--------------------------
3912 // public static native int java.lang.reflect.Array.getLength(Object array);
3913 bool LibraryCallKit::inline_native_getLength() {
3914 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3915
3916 Node* array = null_check(argument(0));
3917 // If array is dead, only null-path is taken.
3918 if (stopped()) return true;
3919
3920 // Deoptimize if it is a non-array.
3921 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3922
3923 if (non_array != NULL) {
3924 PreserveJVMState pjvms(this);
3925 set_control(non_array);
3926 uncommon_trap(Deoptimization::Reason_intrinsic,
3927 Deoptimization::Action_maybe_recompile);
3928 }
3929
3930 // If control is dead, only non-array-path is taken.
3931 if (stopped()) return true;
3932
3933 // The works fine even if the array type is polymorphic.
3934 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3935 Node* result = load_array_length(array);
3936
3937 C->set_has_split_ifs(true); // Has chance for split-if optimization
3938 set_result(result);
3939 return true;
3940 }
3941
3942 //------------------------inline_array_copyOf----------------------------
3943 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3944 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3945 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3946 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3947
3948 // Get the arguments.
3949 Node* original = argument(0);
3950 Node* start = is_copyOfRange? argument(1): intcon(0);
3951 Node* end = is_copyOfRange? argument(2): argument(1);
3952 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3953
3954 Node* newcopy = NULL;
3955
3956 // Set the original stack and the reexecute bit for the interpreter to reexecute
3957 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3958 { PreserveReexecuteState preexecs(this);
3959 jvms()->set_should_reexecute(true);
3960
3961 array_type_mirror = null_check(array_type_mirror);
3962 original = null_check(original);
3963
3964 // Check if a null path was taken unconditionally.
3965 if (stopped()) return true;
3966
3967 Node* orig_length = load_array_length(original);
3968
3969 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3970 klass_node = null_check(klass_node);
3971
3972 RegionNode* bailout = new RegionNode(1);
3973 record_for_igvn(bailout);
3974
3975 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3976 // Bail out if that is so.
3977 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3978 if (not_objArray != NULL) {
3979 // Improve the klass node's type from the new optimistic assumption:
3980 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3981 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3982 Node* cast = new CastPPNode(klass_node, akls);
3983 cast->init_req(0, control());
3984 klass_node = _gvn.transform(cast);
3985 }
3986
3987 // Bail out if either start or end is negative.
3988 generate_negative_guard(start, bailout, &start);
3989 generate_negative_guard(end, bailout, &end);
3990
3991 Node* length = end;
3992 if (_gvn.type(start) != TypeInt::ZERO) {
3993 length = _gvn.transform(new SubINode(end, start));
3994 }
3995
3996 // Bail out if length is negative.
3997 // Without this the new_array would throw
3998 // NegativeArraySizeException but IllegalArgumentException is what
3999 // should be thrown
4000 generate_negative_guard(length, bailout, &length);
4001
4002 if (bailout->req() > 1) {
4003 PreserveJVMState pjvms(this);
4004 set_control(_gvn.transform(bailout));
4005 uncommon_trap(Deoptimization::Reason_intrinsic,
4006 Deoptimization::Action_maybe_recompile);
4007 }
4008
4009 if (!stopped()) {
4010 // How many elements will we copy from the original?
4011 // The answer is MinI(orig_length - start, length).
4012 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4013 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4014
4015 // Generate a direct call to the right arraycopy function(s).
4016 // We know the copy is disjoint but we might not know if the
4017 // oop stores need checking.
4018 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4019 // This will fail a store-check if x contains any non-nulls.
4020
4021 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4022 // loads/stores but it is legal only if we're sure the
4023 // Arrays.copyOf would succeed. So we need all input arguments
4024 // to the copyOf to be validated, including that the copy to the
4025 // new array won't trigger an ArrayStoreException. That subtype
4026 // check can be optimized if we know something on the type of
4027 // the input array from type speculation.
4028 if (_gvn.type(klass_node)->singleton()) {
4029 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
4030 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4031
4032 int test = C->static_subtype_check(superk, subk);
4033 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4034 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4035 if (t_original->speculative_type() != NULL) {
4036 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4037 }
4038 }
4039 }
4040
4041 bool validated = false;
4042 // Reason_class_check rather than Reason_intrinsic because we
4043 // want to intrinsify even if this traps.
4044 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4045 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
4046 klass_node);
4047
4048 if (not_subtype_ctrl != top()) {
4049 PreserveJVMState pjvms(this);
4050 set_control(not_subtype_ctrl);
4051 uncommon_trap(Deoptimization::Reason_class_check,
4052 Deoptimization::Action_make_not_entrant);
4053 assert(stopped(), "Should be stopped");
4054 }
4055 validated = true;
4056 }
4057
4058 if (!stopped()) {
4059 newcopy = new_array(klass_node, length, 0); // no arguments to push
4060
4061 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4062 load_object_klass(original), klass_node);
4063 if (!is_copyOfRange) {
4064 ac->set_copyof(validated);
4065 } else {
4066 ac->set_copyofrange(validated);
4067 }
4068 Node* n = _gvn.transform(ac);
4069 if (n == ac) {
4070 ac->connect_outputs(this);
4071 } else {
4072 assert(validated, "shouldn't transform if all arguments not validated");
4073 set_all_memory(n);
4074 }
4075 }
4076 }
4077 } // original reexecute is set back here
4078
4079 C->set_has_split_ifs(true); // Has chance for split-if optimization
4080 if (!stopped()) {
4081 set_result(newcopy);
4082 }
4083 return true;
4084 }
4085
4086
4087 //----------------------generate_virtual_guard---------------------------
4088 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4089 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4090 RegionNode* slow_region) {
4091 ciMethod* method = callee();
4092 int vtable_index = method->vtable_index();
4093 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4094 "bad index %d", vtable_index);
4095 // Get the Method* out of the appropriate vtable entry.
4096 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4097 vtable_index*vtableEntry::size_in_bytes() +
4098 vtableEntry::method_offset_in_bytes();
4099 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4100 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4101
4102 // Compare the target method with the expected method (e.g., Object.hashCode).
4103 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4104
4105 Node* native_call = makecon(native_call_addr);
4106 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4107 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4108
4109 return generate_slow_guard(test_native, slow_region);
4110 }
4111
4112 //-----------------------generate_method_call----------------------------
4113 // Use generate_method_call to make a slow-call to the real
4114 // method if the fast path fails. An alternative would be to
4115 // use a stub like OptoRuntime::slow_arraycopy_Java.
4116 // This only works for expanding the current library call,
4117 // not another intrinsic. (E.g., don't use this for making an
4118 // arraycopy call inside of the copyOf intrinsic.)
4119 CallJavaNode*
4120 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4121 // When compiling the intrinsic method itself, do not use this technique.
4122 guarantee(callee() != C->method(), "cannot make slow-call to self");
4123
4124 ciMethod* method = callee();
4125 // ensure the JVMS we have will be correct for this call
4126 guarantee(method_id == method->intrinsic_id(), "must match");
4127
4128 const TypeFunc* tf = TypeFunc::make(method);
4129 CallJavaNode* slow_call;
4130 if (is_static) {
4131 assert(!is_virtual, "");
4132 slow_call = new CallStaticJavaNode(C, tf,
4133 SharedRuntime::get_resolve_static_call_stub(),
4134 method, bci());
4135 } else if (is_virtual) {
4136 null_check_receiver();
4137 int vtable_index = Method::invalid_vtable_index;
4138 if (UseInlineCaches) {
4139 // Suppress the vtable call
4140 } else {
4141 // hashCode and clone are not a miranda methods,
4142 // so the vtable index is fixed.
4143 // No need to use the linkResolver to get it.
4144 vtable_index = method->vtable_index();
4145 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4146 "bad index %d", vtable_index);
4147 }
4148 slow_call = new CallDynamicJavaNode(tf,
4149 SharedRuntime::get_resolve_virtual_call_stub(),
4150 method, vtable_index, bci());
4151 } else { // neither virtual nor static: opt_virtual
4152 null_check_receiver();
4153 slow_call = new CallStaticJavaNode(C, tf,
4154 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4155 method, bci());
4156 slow_call->set_optimized_virtual(true);
4157 }
4158 set_arguments_for_java_call(slow_call);
4159 set_edges_for_java_call(slow_call);
4160 return slow_call;
4161 }
4162
4163
4164 /**
4165 * Build special case code for calls to hashCode on an object. This call may
4166 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4167 * slightly different code.
4168 */
4169 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4170 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4171 assert(!(is_virtual && is_static), "either virtual, special, or static");
4172
4173 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4174
4175 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4176 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4177 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4178 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4179 Node* obj = NULL;
4180 if (!is_static) {
4181 // Check for hashing null object
4182 obj = null_check_receiver();
4183 if (stopped()) return true; // unconditionally null
4184 result_reg->init_req(_null_path, top());
4185 result_val->init_req(_null_path, top());
4186 } else {
4187 // Do a null check, and return zero if null.
4188 // System.identityHashCode(null) == 0
4189 obj = argument(0);
4190 Node* null_ctl = top();
4191 obj = null_check_oop(obj, &null_ctl);
4192 result_reg->init_req(_null_path, null_ctl);
4193 result_val->init_req(_null_path, _gvn.intcon(0));
4194 }
4195
4196 // Unconditionally null? Then return right away.
4197 if (stopped()) {
4198 set_control( result_reg->in(_null_path));
4199 if (!stopped())
4200 set_result(result_val->in(_null_path));
4201 return true;
4202 }
4203
4204 // We only go to the fast case code if we pass a number of guards. The
4205 // paths which do not pass are accumulated in the slow_region.
4206 RegionNode* slow_region = new RegionNode(1);
4207 record_for_igvn(slow_region);
4208
4209 // If this is a virtual call, we generate a funny guard. We pull out
4210 // the vtable entry corresponding to hashCode() from the target object.
4211 // If the target method which we are calling happens to be the native
4212 // Object hashCode() method, we pass the guard. We do not need this
4213 // guard for non-virtual calls -- the caller is known to be the native
4214 // Object hashCode().
4215 if (is_virtual) {
4216 // After null check, get the object's klass.
4217 Node* obj_klass = load_object_klass(obj);
4218 generate_virtual_guard(obj_klass, slow_region);
4219 }
4220
4221 // Get the header out of the object, use LoadMarkNode when available
4222 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4223 // The control of the load must be NULL. Otherwise, the load can move before
4224 // the null check after castPP removal.
4225 Node* no_ctrl = NULL;
4226 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4227
4228 // Test the header to see if it is unlocked.
4229 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4230 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4231 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4232 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4233 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4234
4235 generate_slow_guard(test_unlocked, slow_region);
4236
4237 // Get the hash value and check to see that it has been properly assigned.
4238 // We depend on hash_mask being at most 32 bits and avoid the use of
4239 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4240 // vm: see markOop.hpp.
4241 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4242 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4243 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4244 // This hack lets the hash bits live anywhere in the mark object now, as long
4245 // as the shift drops the relevant bits into the low 32 bits. Note that
4246 // Java spec says that HashCode is an int so there's no point in capturing
4247 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4248 hshifted_header = ConvX2I(hshifted_header);
4249 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4250
4251 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4252 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4253 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4254
4255 generate_slow_guard(test_assigned, slow_region);
4256
4257 Node* init_mem = reset_memory();
4258 // fill in the rest of the null path:
4259 result_io ->init_req(_null_path, i_o());
4260 result_mem->init_req(_null_path, init_mem);
4261
4262 result_val->init_req(_fast_path, hash_val);
4263 result_reg->init_req(_fast_path, control());
4264 result_io ->init_req(_fast_path, i_o());
4265 result_mem->init_req(_fast_path, init_mem);
4266
4267 // Generate code for the slow case. We make a call to hashCode().
4268 set_control(_gvn.transform(slow_region));
4269 if (!stopped()) {
4270 // No need for PreserveJVMState, because we're using up the present state.
4271 set_all_memory(init_mem);
4272 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4273 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4274 Node* slow_result = set_results_for_java_call(slow_call);
4275 // this->control() comes from set_results_for_java_call
4276 result_reg->init_req(_slow_path, control());
4277 result_val->init_req(_slow_path, slow_result);
4278 result_io ->set_req(_slow_path, i_o());
4279 result_mem ->set_req(_slow_path, reset_memory());
4280 }
4281
4282 // Return the combined state.
4283 set_i_o( _gvn.transform(result_io) );
4284 set_all_memory( _gvn.transform(result_mem));
4285
4286 set_result(result_reg, result_val);
4287 return true;
4288 }
4289
4290 //---------------------------inline_native_getClass----------------------------
4291 // public final native Class<?> java.lang.Object.getClass();
4292 //
4293 // Build special case code for calls to getClass on an object.
4294 bool LibraryCallKit::inline_native_getClass() {
4295 Node* obj = null_check_receiver();
4296 if (stopped()) return true;
4297 set_result(load_mirror_from_klass(load_object_klass(obj)));
4298 return true;
4299 }
4300
4301 //-----------------inline_native_Reflection_getCallerClass---------------------
4302 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4303 //
4304 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4305 //
4306 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4307 // in that it must skip particular security frames and checks for
4308 // caller sensitive methods.
4309 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4310 #ifndef PRODUCT
4311 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4312 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4313 }
4314 #endif
4315
4316 if (!jvms()->has_method()) {
4317 #ifndef PRODUCT
4318 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4319 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4320 }
4321 #endif
4322 return false;
4323 }
4324
4325 // Walk back up the JVM state to find the caller at the required
4326 // depth.
4327 JVMState* caller_jvms = jvms();
4328
4329 // Cf. JVM_GetCallerClass
4330 // NOTE: Start the loop at depth 1 because the current JVM state does
4331 // not include the Reflection.getCallerClass() frame.
4332 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4333 ciMethod* m = caller_jvms->method();
4334 switch (n) {
4335 case 0:
4336 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4337 break;
4338 case 1:
4339 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4340 if (!m->caller_sensitive()) {
4341 #ifndef PRODUCT
4342 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4343 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4344 }
4345 #endif
4346 return false; // bail-out; let JVM_GetCallerClass do the work
4347 }
4348 break;
4349 default:
4350 if (!m->is_ignored_by_security_stack_walk()) {
4351 // We have reached the desired frame; return the holder class.
4352 // Acquire method holder as java.lang.Class and push as constant.
4353 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4354 ciInstance* caller_mirror = caller_klass->java_mirror();
4355 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4356
4357 #ifndef PRODUCT
4358 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4359 tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth());
4360 tty->print_cr(" JVM state at this point:");
4361 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4362 ciMethod* m = jvms()->of_depth(i)->method();
4363 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4364 }
4365 }
4366 #endif
4367 return true;
4368 }
4369 break;
4370 }
4371 }
4372
4373 #ifndef PRODUCT
4374 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4375 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4376 tty->print_cr(" JVM state at this point:");
4377 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4378 ciMethod* m = jvms()->of_depth(i)->method();
4379 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4380 }
4381 }
4382 #endif
4383
4384 return false; // bail-out; let JVM_GetCallerClass do the work
4385 }
4386
4387 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4388 Node* arg = argument(0);
4389 Node* result = NULL;
4390
4391 switch (id) {
4392 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4393 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4394 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4395 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4396
4397 case vmIntrinsics::_doubleToLongBits: {
4398 // two paths (plus control) merge in a wood
4399 RegionNode *r = new RegionNode(3);
4400 Node *phi = new PhiNode(r, TypeLong::LONG);
4401
4402 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4403 // Build the boolean node
4404 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4405
4406 // Branch either way.
4407 // NaN case is less traveled, which makes all the difference.
4408 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4409 Node *opt_isnan = _gvn.transform(ifisnan);
4410 assert( opt_isnan->is_If(), "Expect an IfNode");
4411 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4412 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4413
4414 set_control(iftrue);
4415
4416 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4417 Node *slow_result = longcon(nan_bits); // return NaN
4418 phi->init_req(1, _gvn.transform( slow_result ));
4419 r->init_req(1, iftrue);
4420
4421 // Else fall through
4422 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4423 set_control(iffalse);
4424
4425 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4426 r->init_req(2, iffalse);
4427
4428 // Post merge
4429 set_control(_gvn.transform(r));
4430 record_for_igvn(r);
4431
4432 C->set_has_split_ifs(true); // Has chance for split-if optimization
4433 result = phi;
4434 assert(result->bottom_type()->isa_long(), "must be");
4435 break;
4436 }
4437
4438 case vmIntrinsics::_floatToIntBits: {
4439 // two paths (plus control) merge in a wood
4440 RegionNode *r = new RegionNode(3);
4441 Node *phi = new PhiNode(r, TypeInt::INT);
4442
4443 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4444 // Build the boolean node
4445 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4446
4447 // Branch either way.
4448 // NaN case is less traveled, which makes all the difference.
4449 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4450 Node *opt_isnan = _gvn.transform(ifisnan);
4451 assert( opt_isnan->is_If(), "Expect an IfNode");
4452 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4453 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4454
4455 set_control(iftrue);
4456
4457 static const jint nan_bits = 0x7fc00000;
4458 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4459 phi->init_req(1, _gvn.transform( slow_result ));
4460 r->init_req(1, iftrue);
4461
4462 // Else fall through
4463 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4464 set_control(iffalse);
4465
4466 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4467 r->init_req(2, iffalse);
4468
4469 // Post merge
4470 set_control(_gvn.transform(r));
4471 record_for_igvn(r);
4472
4473 C->set_has_split_ifs(true); // Has chance for split-if optimization
4474 result = phi;
4475 assert(result->bottom_type()->isa_int(), "must be");
4476 break;
4477 }
4478
4479 default:
4480 fatal_unexpected_iid(id);
4481 break;
4482 }
4483 set_result(_gvn.transform(result));
4484 return true;
4485 }
4486
4487 //----------------------inline_unsafe_copyMemory-------------------------
4488 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4489 bool LibraryCallKit::inline_unsafe_copyMemory() {
4490 if (callee()->is_static()) return false; // caller must have the capability!
4491 null_check_receiver(); // null-check receiver
4492 if (stopped()) return true;
4493
4494 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4495
4496 Node* src_ptr = argument(1); // type: oop
4497 Node* src_off = ConvL2X(argument(2)); // type: long
4498 Node* dst_ptr = argument(4); // type: oop
4499 Node* dst_off = ConvL2X(argument(5)); // type: long
4500 Node* size = ConvL2X(argument(7)); // type: long
4501
4502 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4503 "fieldOffset must be byte-scaled");
4504
4505 Node* src = make_unsafe_address(src_ptr, src_off, T_ILLEGAL);
4506 Node* dst = make_unsafe_address(dst_ptr, dst_off, T_ILLEGAL);
4507
4508 // Conservatively insert a memory barrier on all memory slices.
4509 // Do not let writes of the copy source or destination float below the copy.
4510 insert_mem_bar(Op_MemBarCPUOrder);
4511
4512 // Call it. Note that the length argument is not scaled.
4513 make_runtime_call(RC_LEAF|RC_NO_FP,
4514 OptoRuntime::fast_arraycopy_Type(),
4515 StubRoutines::unsafe_arraycopy(),
4516 "unsafe_arraycopy",
4517 TypeRawPtr::BOTTOM,
4518 src, dst, size XTOP);
4519
4520 // Do not let reads of the copy destination float above the copy.
4521 insert_mem_bar(Op_MemBarCPUOrder);
4522
4523 return true;
4524 }
4525
4526 //------------------------clone_coping-----------------------------------
4527 // Helper function for inline_native_clone.
4528 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4529 assert(obj_size != NULL, "");
4530 Node* raw_obj = alloc_obj->in(1);
4531 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4532
4533 AllocateNode* alloc = NULL;
4534 if (ReduceBulkZeroing) {
4535 // We will be completely responsible for initializing this object -
4536 // mark Initialize node as complete.
4537 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4538 // The object was just allocated - there should be no any stores!
4539 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4540 // Mark as complete_with_arraycopy so that on AllocateNode
4541 // expansion, we know this AllocateNode is initialized by an array
4542 // copy and a StoreStore barrier exists after the array copy.
4543 alloc->initialization()->set_complete_with_arraycopy();
4544 }
4545
4546 // Copy the fastest available way.
4547 // TODO: generate fields copies for small objects instead.
4548 Node* src = obj;
4549 Node* dest = alloc_obj;
4550 Node* size = _gvn.transform(obj_size);
4551
4552 // Exclude the header but include array length to copy by 8 bytes words.
4553 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4554 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4555 instanceOopDesc::base_offset_in_bytes();
4556 // base_off:
4557 // 8 - 32-bit VM
4558 // 12 - 64-bit VM, compressed klass
4559 // 16 - 64-bit VM, normal klass
4560 if (base_off % BytesPerLong != 0) {
4561 assert(UseCompressedClassPointers, "");
4562 if (is_array) {
4563 // Exclude length to copy by 8 bytes words.
4564 base_off += sizeof(int);
4565 } else {
4566 // Include klass to copy by 8 bytes words.
4567 base_off = instanceOopDesc::klass_offset_in_bytes();
4568 }
4569 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4570 }
4571 src = basic_plus_adr(src, base_off);
4572 dest = basic_plus_adr(dest, base_off);
4573
4574 // Compute the length also, if needed:
4575 Node* countx = size;
4576 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4577 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4578
4579 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4580
4581 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false, false);
4582 ac->set_clonebasic();
4583 Node* n = _gvn.transform(ac);
4584 if (n == ac) {
4585 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4586 } else {
4587 set_all_memory(n);
4588 }
4589
4590 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4591 if (card_mark) {
4592 assert(!is_array, "");
4593 // Put in store barrier for any and all oops we are sticking
4594 // into this object. (We could avoid this if we could prove
4595 // that the object type contains no oop fields at all.)
4596 Node* no_particular_value = NULL;
4597 Node* no_particular_field = NULL;
4598 int raw_adr_idx = Compile::AliasIdxRaw;
4599 post_barrier(control(),
4600 memory(raw_adr_type),
4601 alloc_obj,
4602 no_particular_field,
4603 raw_adr_idx,
4604 no_particular_value,
4605 T_OBJECT,
4606 false);
4607 }
4608
4609 // Do not let reads from the cloned object float above the arraycopy.
4610 if (alloc != NULL) {
4611 // Do not let stores that initialize this object be reordered with
4612 // a subsequent store that would make this object accessible by
4613 // other threads.
4614 // Record what AllocateNode this StoreStore protects so that
4615 // escape analysis can go from the MemBarStoreStoreNode to the
4616 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4617 // based on the escape status of the AllocateNode.
4618 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4619 } else {
4620 insert_mem_bar(Op_MemBarCPUOrder);
4621 }
4622 }
4623
4624 //------------------------inline_native_clone----------------------------
4625 // protected native Object java.lang.Object.clone();
4626 //
4627 // Here are the simple edge cases:
4628 // null receiver => normal trap
4629 // virtual and clone was overridden => slow path to out-of-line clone
4630 // not cloneable or finalizer => slow path to out-of-line Object.clone
4631 //
4632 // The general case has two steps, allocation and copying.
4633 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4634 //
4635 // Copying also has two cases, oop arrays and everything else.
4636 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4637 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4638 //
4639 // These steps fold up nicely if and when the cloned object's klass
4640 // can be sharply typed as an object array, a type array, or an instance.
4641 //
4642 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4643 PhiNode* result_val;
4644
4645 // Set the reexecute bit for the interpreter to reexecute
4646 // the bytecode that invokes Object.clone if deoptimization happens.
4647 { PreserveReexecuteState preexecs(this);
4648 jvms()->set_should_reexecute(true);
4649
4650 Node* obj = null_check_receiver();
4651 if (stopped()) return true;
4652
4653 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4654
4655 // If we are going to clone an instance, we need its exact type to
4656 // know the number and types of fields to convert the clone to
4657 // loads/stores. Maybe a speculative type can help us.
4658 if (!obj_type->klass_is_exact() &&
4659 obj_type->speculative_type() != NULL &&
4660 obj_type->speculative_type()->is_instance_klass()) {
4661 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4662 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4663 !spec_ik->has_injected_fields()) {
4664 ciKlass* k = obj_type->klass();
4665 if (!k->is_instance_klass() ||
4666 k->as_instance_klass()->is_interface() ||
4667 k->as_instance_klass()->has_subklass()) {
4668 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4669 }
4670 }
4671 }
4672
4673 Node* obj_klass = load_object_klass(obj);
4674 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4675 const TypeOopPtr* toop = ((tklass != NULL)
4676 ? tklass->as_instance_type()
4677 : TypeInstPtr::NOTNULL);
4678
4679 // Conservatively insert a memory barrier on all memory slices.
4680 // Do not let writes into the original float below the clone.
4681 insert_mem_bar(Op_MemBarCPUOrder);
4682
4683 // paths into result_reg:
4684 enum {
4685 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4686 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4687 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4688 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4689 PATH_LIMIT
4690 };
4691 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4692 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4693 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4694 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4695 record_for_igvn(result_reg);
4696
4697 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4698 int raw_adr_idx = Compile::AliasIdxRaw;
4699
4700 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4701 if (array_ctl != NULL) {
4702 // It's an array.
4703 PreserveJVMState pjvms(this);
4704 set_control(array_ctl);
4705 Node* obj_length = load_array_length(obj);
4706 Node* obj_size = NULL;
4707 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4708
4709 if (!use_ReduceInitialCardMarks()) {
4710 // If it is an oop array, it requires very special treatment,
4711 // because card marking is required on each card of the array.
4712 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4713 if (is_obja != NULL) {
4714 PreserveJVMState pjvms2(this);
4715 set_control(is_obja);
4716 // Generate a direct call to the right arraycopy function(s).
4717 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4718 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4719 ac->set_cloneoop();
4720 Node* n = _gvn.transform(ac);
4721 assert(n == ac, "cannot disappear");
4722 ac->connect_outputs(this);
4723
4724 result_reg->init_req(_objArray_path, control());
4725 result_val->init_req(_objArray_path, alloc_obj);
4726 result_i_o ->set_req(_objArray_path, i_o());
4727 result_mem ->set_req(_objArray_path, reset_memory());
4728 }
4729 }
4730 // Otherwise, there are no card marks to worry about.
4731 // (We can dispense with card marks if we know the allocation
4732 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4733 // causes the non-eden paths to take compensating steps to
4734 // simulate a fresh allocation, so that no further
4735 // card marks are required in compiled code to initialize
4736 // the object.)
4737
4738 if (!stopped()) {
4739 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4740
4741 // Present the results of the copy.
4742 result_reg->init_req(_array_path, control());
4743 result_val->init_req(_array_path, alloc_obj);
4744 result_i_o ->set_req(_array_path, i_o());
4745 result_mem ->set_req(_array_path, reset_memory());
4746 }
4747 }
4748
4749 // We only go to the instance fast case code if we pass a number of guards.
4750 // The paths which do not pass are accumulated in the slow_region.
4751 RegionNode* slow_region = new RegionNode(1);
4752 record_for_igvn(slow_region);
4753 if (!stopped()) {
4754 // It's an instance (we did array above). Make the slow-path tests.
4755 // If this is a virtual call, we generate a funny guard. We grab
4756 // the vtable entry corresponding to clone() from the target object.
4757 // If the target method which we are calling happens to be the
4758 // Object clone() method, we pass the guard. We do not need this
4759 // guard for non-virtual calls; the caller is known to be the native
4760 // Object clone().
4761 if (is_virtual) {
4762 generate_virtual_guard(obj_klass, slow_region);
4763 }
4764
4765 // The object must be easily cloneable and must not have a finalizer.
4766 // Both of these conditions may be checked in a single test.
4767 // We could optimize the test further, but we don't care.
4768 generate_access_flags_guard(obj_klass,
4769 // Test both conditions:
4770 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4771 // Must be cloneable but not finalizer:
4772 JVM_ACC_IS_CLONEABLE_FAST,
4773 slow_region);
4774 }
4775
4776 if (!stopped()) {
4777 // It's an instance, and it passed the slow-path tests.
4778 PreserveJVMState pjvms(this);
4779 Node* obj_size = NULL;
4780 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4781 // is reexecuted if deoptimization occurs and there could be problems when merging
4782 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4783 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4784
4785 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4786
4787 // Present the results of the slow call.
4788 result_reg->init_req(_instance_path, control());
4789 result_val->init_req(_instance_path, alloc_obj);
4790 result_i_o ->set_req(_instance_path, i_o());
4791 result_mem ->set_req(_instance_path, reset_memory());
4792 }
4793
4794 // Generate code for the slow case. We make a call to clone().
4795 set_control(_gvn.transform(slow_region));
4796 if (!stopped()) {
4797 PreserveJVMState pjvms(this);
4798 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4799 Node* slow_result = set_results_for_java_call(slow_call);
4800 // this->control() comes from set_results_for_java_call
4801 result_reg->init_req(_slow_path, control());
4802 result_val->init_req(_slow_path, slow_result);
4803 result_i_o ->set_req(_slow_path, i_o());
4804 result_mem ->set_req(_slow_path, reset_memory());
4805 }
4806
4807 // Return the combined state.
4808 set_control( _gvn.transform(result_reg));
4809 set_i_o( _gvn.transform(result_i_o));
4810 set_all_memory( _gvn.transform(result_mem));
4811 } // original reexecute is set back here
4812
4813 set_result(_gvn.transform(result_val));
4814 return true;
4815 }
4816
4817 // If we have a tighly coupled allocation, the arraycopy may take care
4818 // of the array initialization. If one of the guards we insert between
4819 // the allocation and the arraycopy causes a deoptimization, an
4820 // unitialized array will escape the compiled method. To prevent that
4821 // we set the JVM state for uncommon traps between the allocation and
4822 // the arraycopy to the state before the allocation so, in case of
4823 // deoptimization, we'll reexecute the allocation and the
4824 // initialization.
4825 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4826 if (alloc != NULL) {
4827 ciMethod* trap_method = alloc->jvms()->method();
4828 int trap_bci = alloc->jvms()->bci();
4829
4830 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4831 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4832 // Make sure there's no store between the allocation and the
4833 // arraycopy otherwise visible side effects could be rexecuted
4834 // in case of deoptimization and cause incorrect execution.
4835 bool no_interfering_store = true;
4836 Node* mem = alloc->in(TypeFunc::Memory);
4837 if (mem->is_MergeMem()) {
4838 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4839 Node* n = mms.memory();
4840 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4841 assert(n->is_Store(), "what else?");
4842 no_interfering_store = false;
4843 break;
4844 }
4845 }
4846 } else {
4847 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4848 Node* n = mms.memory();
4849 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4850 assert(n->is_Store(), "what else?");
4851 no_interfering_store = false;
4852 break;
4853 }
4854 }
4855 }
4856
4857 if (no_interfering_store) {
4858 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4859 uint size = alloc->req();
4860 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4861 old_jvms->set_map(sfpt);
4862 for (uint i = 0; i < size; i++) {
4863 sfpt->init_req(i, alloc->in(i));
4864 }
4865 // re-push array length for deoptimization
4866 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4867 old_jvms->set_sp(old_jvms->sp()+1);
4868 old_jvms->set_monoff(old_jvms->monoff()+1);
4869 old_jvms->set_scloff(old_jvms->scloff()+1);
4870 old_jvms->set_endoff(old_jvms->endoff()+1);
4871 old_jvms->set_should_reexecute(true);
4872
4873 sfpt->set_i_o(map()->i_o());
4874 sfpt->set_memory(map()->memory());
4875 sfpt->set_control(map()->control());
4876
4877 JVMState* saved_jvms = jvms();
4878 saved_reexecute_sp = _reexecute_sp;
4879
4880 set_jvms(sfpt->jvms());
4881 _reexecute_sp = jvms()->sp();
4882
4883 return saved_jvms;
4884 }
4885 }
4886 }
4887 return NULL;
4888 }
4889
4890 // In case of a deoptimization, we restart execution at the
4891 // allocation, allocating a new array. We would leave an uninitialized
4892 // array in the heap that GCs wouldn't expect. Move the allocation
4893 // after the traps so we don't allocate the array if we
4894 // deoptimize. This is possible because tightly_coupled_allocation()
4895 // guarantees there's no observer of the allocated array at this point
4896 // and the control flow is simple enough.
4897 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4898 if (saved_jvms != NULL && !stopped()) {
4899 assert(alloc != NULL, "only with a tightly coupled allocation");
4900 // restore JVM state to the state at the arraycopy
4901 saved_jvms->map()->set_control(map()->control());
4902 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4903 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4904 // If we've improved the types of some nodes (null check) while
4905 // emitting the guards, propagate them to the current state
4906 map()->replaced_nodes().apply(saved_jvms->map());
4907 set_jvms(saved_jvms);
4908 _reexecute_sp = saved_reexecute_sp;
4909
4910 // Remove the allocation from above the guards
4911 CallProjections callprojs;
4912 alloc->extract_projections(&callprojs, true);
4913 InitializeNode* init = alloc->initialization();
4914 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4915 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4916 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4917 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4918
4919 // move the allocation here (after the guards)
4920 _gvn.hash_delete(alloc);
4921 alloc->set_req(TypeFunc::Control, control());
4922 alloc->set_req(TypeFunc::I_O, i_o());
4923 Node *mem = reset_memory();
4924 set_all_memory(mem);
4925 alloc->set_req(TypeFunc::Memory, mem);
4926 set_control(init->proj_out(TypeFunc::Control));
4927 set_i_o(callprojs.fallthrough_ioproj);
4928
4929 // Update memory as done in GraphKit::set_output_for_allocation()
4930 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4931 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4932 if (ary_type->isa_aryptr() && length_type != NULL) {
4933 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4934 }
4935 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4936 int elemidx = C->get_alias_index(telemref);
4937 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4938 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4939
4940 Node* allocx = _gvn.transform(alloc);
4941 assert(allocx == alloc, "where has the allocation gone?");
4942 assert(dest->is_CheckCastPP(), "not an allocation result?");
4943
4944 _gvn.hash_delete(dest);
4945 dest->set_req(0, control());
4946 Node* destx = _gvn.transform(dest);
4947 assert(destx == dest, "where has the allocation result gone?");
4948 }
4949 }
4950
4951
4952 //------------------------------inline_arraycopy-----------------------
4953 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4954 // Object dest, int destPos,
4955 // int length);
4956 bool LibraryCallKit::inline_arraycopy() {
4957 // Get the arguments.
4958 Node* src = argument(0); // type: oop
4959 Node* src_offset = argument(1); // type: int
4960 Node* dest = argument(2); // type: oop
4961 Node* dest_offset = argument(3); // type: int
4962 Node* length = argument(4); // type: int
4963
4964
4965 // Check for allocation before we add nodes that would confuse
4966 // tightly_coupled_allocation()
4967 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4968
4969 int saved_reexecute_sp = -1;
4970 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4971 // See arraycopy_restore_alloc_state() comment
4972 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4973 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4974 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4975 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4976
4977 // The following tests must be performed
4978 // (1) src and dest are arrays.
4979 // (2) src and dest arrays must have elements of the same BasicType
4980 // (3) src and dest must not be null.
4981 // (4) src_offset must not be negative.
4982 // (5) dest_offset must not be negative.
4983 // (6) length must not be negative.
4984 // (7) src_offset + length must not exceed length of src.
4985 // (8) dest_offset + length must not exceed length of dest.
4986 // (9) each element of an oop array must be assignable
4987
4988 // (3) src and dest must not be null.
4989 // always do this here because we need the JVM state for uncommon traps
4990 Node* null_ctl = top();
4991 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4992 assert(null_ctl->is_top(), "no null control here");
4993 dest = null_check(dest, T_ARRAY);
4994
4995 if (!can_emit_guards) {
4996 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4997 // guards but the arraycopy node could still take advantage of a
4998 // tightly allocated allocation. tightly_coupled_allocation() is
4999 // called again to make sure it takes the null check above into
5000 // account: the null check is mandatory and if it caused an
5001 // uncommon trap to be emitted then the allocation can't be
5002 // considered tightly coupled in this context.
5003 alloc = tightly_coupled_allocation(dest, NULL);
5004 }
5005
5006 bool validated = false;
5007
5008 const Type* src_type = _gvn.type(src);
5009 const Type* dest_type = _gvn.type(dest);
5010 const TypeAryPtr* top_src = src_type->isa_aryptr();
5011 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5012
5013 // Do we have the type of src?
5014 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5015 // Do we have the type of dest?
5016 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5017 // Is the type for src from speculation?
5018 bool src_spec = false;
5019 // Is the type for dest from speculation?
5020 bool dest_spec = false;
5021
5022 if ((!has_src || !has_dest) && can_emit_guards) {
5023 // We don't have sufficient type information, let's see if
5024 // speculative types can help. We need to have types for both src
5025 // and dest so that it pays off.
5026
5027 // Do we already have or could we have type information for src
5028 bool could_have_src = has_src;
5029 // Do we already have or could we have type information for dest
5030 bool could_have_dest = has_dest;
5031
5032 ciKlass* src_k = NULL;
5033 if (!has_src) {
5034 src_k = src_type->speculative_type_not_null();
5035 if (src_k != NULL && src_k->is_array_klass()) {
5036 could_have_src = true;
5037 }
5038 }
5039
5040 ciKlass* dest_k = NULL;
5041 if (!has_dest) {
5042 dest_k = dest_type->speculative_type_not_null();
5043 if (dest_k != NULL && dest_k->is_array_klass()) {
5044 could_have_dest = true;
5045 }
5046 }
5047
5048 if (could_have_src && could_have_dest) {
5049 // This is going to pay off so emit the required guards
5050 if (!has_src) {
5051 src = maybe_cast_profiled_obj(src, src_k, true);
5052 src_type = _gvn.type(src);
5053 top_src = src_type->isa_aryptr();
5054 has_src = (top_src != NULL && top_src->klass() != NULL);
5055 src_spec = true;
5056 }
5057 if (!has_dest) {
5058 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5059 dest_type = _gvn.type(dest);
5060 top_dest = dest_type->isa_aryptr();
5061 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5062 dest_spec = true;
5063 }
5064 }
5065 }
5066
5067 if (has_src && has_dest && can_emit_guards) {
5068 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5069 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5070 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5071 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5072
5073 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5074 // If both arrays are object arrays then having the exact types
5075 // for both will remove the need for a subtype check at runtime
5076 // before the call and may make it possible to pick a faster copy
5077 // routine (without a subtype check on every element)
5078 // Do we have the exact type of src?
5079 bool could_have_src = src_spec;
5080 // Do we have the exact type of dest?
5081 bool could_have_dest = dest_spec;
5082 ciKlass* src_k = top_src->klass();
5083 ciKlass* dest_k = top_dest->klass();
5084 if (!src_spec) {
5085 src_k = src_type->speculative_type_not_null();
5086 if (src_k != NULL && src_k->is_array_klass()) {
5087 could_have_src = true;
5088 }
5089 }
5090 if (!dest_spec) {
5091 dest_k = dest_type->speculative_type_not_null();
5092 if (dest_k != NULL && dest_k->is_array_klass()) {
5093 could_have_dest = true;
5094 }
5095 }
5096 if (could_have_src && could_have_dest) {
5097 // If we can have both exact types, emit the missing guards
5098 if (could_have_src && !src_spec) {
5099 src = maybe_cast_profiled_obj(src, src_k, true);
5100 }
5101 if (could_have_dest && !dest_spec) {
5102 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5103 }
5104 }
5105 }
5106 }
5107
5108 ciMethod* trap_method = method();
5109 int trap_bci = bci();
5110 if (saved_jvms != NULL) {
5111 trap_method = alloc->jvms()->method();
5112 trap_bci = alloc->jvms()->bci();
5113 }
5114
5115 bool negative_length_guard_generated = false;
5116
5117 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5118 can_emit_guards &&
5119 !src->is_top() && !dest->is_top()) {
5120 // validate arguments: enables transformation the ArrayCopyNode
5121 validated = true;
5122
5123 RegionNode* slow_region = new RegionNode(1);
5124 record_for_igvn(slow_region);
5125
5126 // (1) src and dest are arrays.
5127 generate_non_array_guard(load_object_klass(src), slow_region);
5128 generate_non_array_guard(load_object_klass(dest), slow_region);
5129
5130 // (2) src and dest arrays must have elements of the same BasicType
5131 // done at macro expansion or at Ideal transformation time
5132
5133 // (4) src_offset must not be negative.
5134 generate_negative_guard(src_offset, slow_region);
5135
5136 // (5) dest_offset must not be negative.
5137 generate_negative_guard(dest_offset, slow_region);
5138
5139 // (7) src_offset + length must not exceed length of src.
5140 generate_limit_guard(src_offset, length,
5141 load_array_length(src),
5142 slow_region);
5143
5144 // (8) dest_offset + length must not exceed length of dest.
5145 generate_limit_guard(dest_offset, length,
5146 load_array_length(dest),
5147 slow_region);
5148
5149 // (6) length must not be negative.
5150 // This is also checked in generate_arraycopy() during macro expansion, but
5151 // we also have to check it here for the case where the ArrayCopyNode will
5152 // be eliminated by Escape Analysis.
5153 if (EliminateAllocations) {
5154 generate_negative_guard(length, slow_region);
5155 negative_length_guard_generated = true;
5156 }
5157
5158 // (9) each element of an oop array must be assignable
5159 Node* src_klass = load_object_klass(src);
5160 Node* dest_klass = load_object_klass(dest);
5161 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5162
5163 if (not_subtype_ctrl != top()) {
5164 PreserveJVMState pjvms(this);
5165 set_control(not_subtype_ctrl);
5166 uncommon_trap(Deoptimization::Reason_intrinsic,
5167 Deoptimization::Action_make_not_entrant);
5168 assert(stopped(), "Should be stopped");
5169 }
5170 {
5171 PreserveJVMState pjvms(this);
5172 set_control(_gvn.transform(slow_region));
5173 uncommon_trap(Deoptimization::Reason_intrinsic,
5174 Deoptimization::Action_make_not_entrant);
5175 assert(stopped(), "Should be stopped");
5176 }
5177 }
5178
5179 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5180
5181 if (stopped()) {
5182 return true;
5183 }
5184
5185 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5186 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5187 // so the compiler has a chance to eliminate them: during macro expansion,
5188 // we have to set their control (CastPP nodes are eliminated).
5189 load_object_klass(src), load_object_klass(dest),
5190 load_array_length(src), load_array_length(dest));
5191
5192 ac->set_arraycopy(validated);
5193
5194 Node* n = _gvn.transform(ac);
5195 if (n == ac) {
5196 ac->connect_outputs(this);
5197 } else {
5198 assert(validated, "shouldn't transform if all arguments not validated");
5199 set_all_memory(n);
5200 }
5201
5202 return true;
5203 }
5204
5205
5206 // Helper function which determines if an arraycopy immediately follows
5207 // an allocation, with no intervening tests or other escapes for the object.
5208 AllocateArrayNode*
5209 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5210 RegionNode* slow_region) {
5211 if (stopped()) return NULL; // no fast path
5212 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5213
5214 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5215 if (alloc == NULL) return NULL;
5216
5217 Node* rawmem = memory(Compile::AliasIdxRaw);
5218 // Is the allocation's memory state untouched?
5219 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5220 // Bail out if there have been raw-memory effects since the allocation.
5221 // (Example: There might have been a call or safepoint.)
5222 return NULL;
5223 }
5224 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5225 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5226 return NULL;
5227 }
5228
5229 // There must be no unexpected observers of this allocation.
5230 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5231 Node* obs = ptr->fast_out(i);
5232 if (obs != this->map()) {
5233 return NULL;
5234 }
5235 }
5236
5237 // This arraycopy must unconditionally follow the allocation of the ptr.
5238 Node* alloc_ctl = ptr->in(0);
5239 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5240
5241 Node* ctl = control();
5242 while (ctl != alloc_ctl) {
5243 // There may be guards which feed into the slow_region.
5244 // Any other control flow means that we might not get a chance
5245 // to finish initializing the allocated object.
5246 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5247 IfNode* iff = ctl->in(0)->as_If();
5248 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5249 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5250 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5251 ctl = iff->in(0); // This test feeds the known slow_region.
5252 continue;
5253 }
5254 // One more try: Various low-level checks bottom out in
5255 // uncommon traps. If the debug-info of the trap omits
5256 // any reference to the allocation, as we've already
5257 // observed, then there can be no objection to the trap.
5258 bool found_trap = false;
5259 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5260 Node* obs = not_ctl->fast_out(j);
5261 if (obs->in(0) == not_ctl && obs->is_Call() &&
5262 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5263 found_trap = true; break;
5264 }
5265 }
5266 if (found_trap) {
5267 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5268 continue;
5269 }
5270 }
5271 return NULL;
5272 }
5273
5274 // If we get this far, we have an allocation which immediately
5275 // precedes the arraycopy, and we can take over zeroing the new object.
5276 // The arraycopy will finish the initialization, and provide
5277 // a new control state to which we will anchor the destination pointer.
5278
5279 return alloc;
5280 }
5281
5282 //-------------inline_encodeISOArray-----------------------------------
5283 // encode char[] to byte[] in ISO_8859_1
5284 bool LibraryCallKit::inline_encodeISOArray() {
5285 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5286 // no receiver since it is static method
5287 Node *src = argument(0);
5288 Node *src_offset = argument(1);
5289 Node *dst = argument(2);
5290 Node *dst_offset = argument(3);
5291 Node *length = argument(4);
5292
5293 const Type* src_type = src->Value(&_gvn);
5294 const Type* dst_type = dst->Value(&_gvn);
5295 const TypeAryPtr* top_src = src_type->isa_aryptr();
5296 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5297 if (top_src == NULL || top_src->klass() == NULL ||
5298 top_dest == NULL || top_dest->klass() == NULL) {
5299 // failed array check
5300 return false;
5301 }
5302
5303 // Figure out the size and type of the elements we will be copying.
5304 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5305 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5306 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5307 return false;
5308 }
5309
5310 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5311 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5312 // 'src_start' points to src array + scaled offset
5313 // 'dst_start' points to dst array + scaled offset
5314
5315 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5316 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5317 enc = _gvn.transform(enc);
5318 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5319 set_memory(res_mem, mtype);
5320 set_result(enc);
5321 return true;
5322 }
5323
5324 //-------------inline_multiplyToLen-----------------------------------
5325 bool LibraryCallKit::inline_multiplyToLen() {
5326 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5327
5328 address stubAddr = StubRoutines::multiplyToLen();
5329 if (stubAddr == NULL) {
5330 return false; // Intrinsic's stub is not implemented on this platform
5331 }
5332 const char* stubName = "multiplyToLen";
5333
5334 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5335
5336 // no receiver because it is a static method
5337 Node* x = argument(0);
5338 Node* xlen = argument(1);
5339 Node* y = argument(2);
5340 Node* ylen = argument(3);
5341 Node* z = argument(4);
5342
5343 const Type* x_type = x->Value(&_gvn);
5344 const Type* y_type = y->Value(&_gvn);
5345 const TypeAryPtr* top_x = x_type->isa_aryptr();
5346 const TypeAryPtr* top_y = y_type->isa_aryptr();
5347 if (top_x == NULL || top_x->klass() == NULL ||
5348 top_y == NULL || top_y->klass() == NULL) {
5349 // failed array check
5350 return false;
5351 }
5352
5353 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5354 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5355 if (x_elem != T_INT || y_elem != T_INT) {
5356 return false;
5357 }
5358
5359 // Set the original stack and the reexecute bit for the interpreter to reexecute
5360 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5361 // on the return from z array allocation in runtime.
5362 { PreserveReexecuteState preexecs(this);
5363 jvms()->set_should_reexecute(true);
5364
5365 Node* x_start = array_element_address(x, intcon(0), x_elem);
5366 Node* y_start = array_element_address(y, intcon(0), y_elem);
5367 // 'x_start' points to x array + scaled xlen
5368 // 'y_start' points to y array + scaled ylen
5369
5370 // Allocate the result array
5371 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5372 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5373 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5374
5375 IdealKit ideal(this);
5376
5377 #define __ ideal.
5378 Node* one = __ ConI(1);
5379 Node* zero = __ ConI(0);
5380 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5381 __ set(need_alloc, zero);
5382 __ set(z_alloc, z);
5383 __ if_then(z, BoolTest::eq, null()); {
5384 __ increment (need_alloc, one);
5385 } __ else_(); {
5386 // Update graphKit memory and control from IdealKit.
5387 sync_kit(ideal);
5388 Node* zlen_arg = load_array_length(z);
5389 // Update IdealKit memory and control from graphKit.
5390 __ sync_kit(this);
5391 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5392 __ increment (need_alloc, one);
5393 } __ end_if();
5394 } __ end_if();
5395
5396 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5397 // Update graphKit memory and control from IdealKit.
5398 sync_kit(ideal);
5399 Node * narr = new_array(klass_node, zlen, 1);
5400 // Update IdealKit memory and control from graphKit.
5401 __ sync_kit(this);
5402 __ set(z_alloc, narr);
5403 } __ end_if();
5404
5405 sync_kit(ideal);
5406 z = __ value(z_alloc);
5407 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5408 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5409 // Final sync IdealKit and GraphKit.
5410 final_sync(ideal);
5411 #undef __
5412
5413 Node* z_start = array_element_address(z, intcon(0), T_INT);
5414
5415 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5416 OptoRuntime::multiplyToLen_Type(),
5417 stubAddr, stubName, TypePtr::BOTTOM,
5418 x_start, xlen, y_start, ylen, z_start, zlen);
5419 } // original reexecute is set back here
5420
5421 C->set_has_split_ifs(true); // Has chance for split-if optimization
5422 set_result(z);
5423 return true;
5424 }
5425
5426 //-------------inline_squareToLen------------------------------------
5427 bool LibraryCallKit::inline_squareToLen() {
5428 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5429
5430 address stubAddr = StubRoutines::squareToLen();
5431 if (stubAddr == NULL) {
5432 return false; // Intrinsic's stub is not implemented on this platform
5433 }
5434 const char* stubName = "squareToLen";
5435
5436 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5437
5438 Node* x = argument(0);
5439 Node* len = argument(1);
5440 Node* z = argument(2);
5441 Node* zlen = argument(3);
5442
5443 const Type* x_type = x->Value(&_gvn);
5444 const Type* z_type = z->Value(&_gvn);
5445 const TypeAryPtr* top_x = x_type->isa_aryptr();
5446 const TypeAryPtr* top_z = z_type->isa_aryptr();
5447 if (top_x == NULL || top_x->klass() == NULL ||
5448 top_z == NULL || top_z->klass() == NULL) {
5449 // failed array check
5450 return false;
5451 }
5452
5453 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5454 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5455 if (x_elem != T_INT || z_elem != T_INT) {
5456 return false;
5457 }
5458
5459
5460 Node* x_start = array_element_address(x, intcon(0), x_elem);
5461 Node* z_start = array_element_address(z, intcon(0), z_elem);
5462
5463 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5464 OptoRuntime::squareToLen_Type(),
5465 stubAddr, stubName, TypePtr::BOTTOM,
5466 x_start, len, z_start, zlen);
5467
5468 set_result(z);
5469 return true;
5470 }
5471
5472 //-------------inline_mulAdd------------------------------------------
5473 bool LibraryCallKit::inline_mulAdd() {
5474 assert(UseMulAddIntrinsic, "not implemented on this platform");
5475
5476 address stubAddr = StubRoutines::mulAdd();
5477 if (stubAddr == NULL) {
5478 return false; // Intrinsic's stub is not implemented on this platform
5479 }
5480 const char* stubName = "mulAdd";
5481
5482 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5483
5484 Node* out = argument(0);
5485 Node* in = argument(1);
5486 Node* offset = argument(2);
5487 Node* len = argument(3);
5488 Node* k = argument(4);
5489
5490 const Type* out_type = out->Value(&_gvn);
5491 const Type* in_type = in->Value(&_gvn);
5492 const TypeAryPtr* top_out = out_type->isa_aryptr();
5493 const TypeAryPtr* top_in = in_type->isa_aryptr();
5494 if (top_out == NULL || top_out->klass() == NULL ||
5495 top_in == NULL || top_in->klass() == NULL) {
5496 // failed array check
5497 return false;
5498 }
5499
5500 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5501 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5502 if (out_elem != T_INT || in_elem != T_INT) {
5503 return false;
5504 }
5505
5506 Node* outlen = load_array_length(out);
5507 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5508 Node* out_start = array_element_address(out, intcon(0), out_elem);
5509 Node* in_start = array_element_address(in, intcon(0), in_elem);
5510
5511 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5512 OptoRuntime::mulAdd_Type(),
5513 stubAddr, stubName, TypePtr::BOTTOM,
5514 out_start,in_start, new_offset, len, k);
5515 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5516 set_result(result);
5517 return true;
5518 }
5519
5520 //-------------inline_montgomeryMultiply-----------------------------------
5521 bool LibraryCallKit::inline_montgomeryMultiply() {
5522 address stubAddr = StubRoutines::montgomeryMultiply();
5523 if (stubAddr == NULL) {
5524 return false; // Intrinsic's stub is not implemented on this platform
5525 }
5526
5527 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5528 const char* stubName = "montgomery_multiply";
5529
5530 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5531
5532 Node* a = argument(0);
5533 Node* b = argument(1);
5534 Node* n = argument(2);
5535 Node* len = argument(3);
5536 Node* inv = argument(4);
5537 Node* m = argument(6);
5538
5539 const Type* a_type = a->Value(&_gvn);
5540 const TypeAryPtr* top_a = a_type->isa_aryptr();
5541 const Type* b_type = b->Value(&_gvn);
5542 const TypeAryPtr* top_b = b_type->isa_aryptr();
5543 const Type* n_type = a->Value(&_gvn);
5544 const TypeAryPtr* top_n = n_type->isa_aryptr();
5545 const Type* m_type = a->Value(&_gvn);
5546 const TypeAryPtr* top_m = m_type->isa_aryptr();
5547 if (top_a == NULL || top_a->klass() == NULL ||
5548 top_b == NULL || top_b->klass() == NULL ||
5549 top_n == NULL || top_n->klass() == NULL ||
5550 top_m == NULL || top_m->klass() == NULL) {
5551 // failed array check
5552 return false;
5553 }
5554
5555 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5556 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5557 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5558 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5559 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5560 return false;
5561 }
5562
5563 // Make the call
5564 {
5565 Node* a_start = array_element_address(a, intcon(0), a_elem);
5566 Node* b_start = array_element_address(b, intcon(0), b_elem);
5567 Node* n_start = array_element_address(n, intcon(0), n_elem);
5568 Node* m_start = array_element_address(m, intcon(0), m_elem);
5569
5570 Node* call = make_runtime_call(RC_LEAF,
5571 OptoRuntime::montgomeryMultiply_Type(),
5572 stubAddr, stubName, TypePtr::BOTTOM,
5573 a_start, b_start, n_start, len, inv, top(),
5574 m_start);
5575 set_result(m);
5576 }
5577
5578 return true;
5579 }
5580
5581 bool LibraryCallKit::inline_montgomerySquare() {
5582 address stubAddr = StubRoutines::montgomerySquare();
5583 if (stubAddr == NULL) {
5584 return false; // Intrinsic's stub is not implemented on this platform
5585 }
5586
5587 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5588 const char* stubName = "montgomery_square";
5589
5590 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5591
5592 Node* a = argument(0);
5593 Node* n = argument(1);
5594 Node* len = argument(2);
5595 Node* inv = argument(3);
5596 Node* m = argument(5);
5597
5598 const Type* a_type = a->Value(&_gvn);
5599 const TypeAryPtr* top_a = a_type->isa_aryptr();
5600 const Type* n_type = a->Value(&_gvn);
5601 const TypeAryPtr* top_n = n_type->isa_aryptr();
5602 const Type* m_type = a->Value(&_gvn);
5603 const TypeAryPtr* top_m = m_type->isa_aryptr();
5604 if (top_a == NULL || top_a->klass() == NULL ||
5605 top_n == NULL || top_n->klass() == NULL ||
5606 top_m == NULL || top_m->klass() == NULL) {
5607 // failed array check
5608 return false;
5609 }
5610
5611 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5612 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5613 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5614 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5615 return false;
5616 }
5617
5618 // Make the call
5619 {
5620 Node* a_start = array_element_address(a, intcon(0), a_elem);
5621 Node* n_start = array_element_address(n, intcon(0), n_elem);
5622 Node* m_start = array_element_address(m, intcon(0), m_elem);
5623
5624 Node* call = make_runtime_call(RC_LEAF,
5625 OptoRuntime::montgomerySquare_Type(),
5626 stubAddr, stubName, TypePtr::BOTTOM,
5627 a_start, n_start, len, inv, top(),
5628 m_start);
5629 set_result(m);
5630 }
5631
5632 return true;
5633 }
5634
5635 //-------------inline_vectorizedMismatch------------------------------
5636 bool LibraryCallKit::inline_vectorizedMismatch() {
5637 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5638
5639 address stubAddr = StubRoutines::vectorizedMismatch();
5640 if (stubAddr == NULL) {
5641 return false; // Intrinsic's stub is not implemented on this platform
5642 }
5643 const char* stubName = "vectorizedMismatch";
5644 int size_l = callee()->signature()->size();
5645 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5646
5647 Node* obja = argument(0);
5648 Node* aoffset = argument(1);
5649 Node* objb = argument(3);
5650 Node* boffset = argument(4);
5651 Node* length = argument(6);
5652 Node* scale = argument(7);
5653
5654 const Type* a_type = obja->Value(&_gvn);
5655 const Type* b_type = objb->Value(&_gvn);
5656 const TypeAryPtr* top_a = a_type->isa_aryptr();
5657 const TypeAryPtr* top_b = b_type->isa_aryptr();
5658 if (top_a == NULL || top_a->klass() == NULL ||
5659 top_b == NULL || top_b->klass() == NULL) {
5660 // failed array check
5661 return false;
5662 }
5663
5664 Node* call;
5665 jvms()->set_should_reexecute(true);
5666
5667 Node* obja_adr = make_unsafe_address(obja, aoffset, T_ILLEGAL);
5668 Node* objb_adr = make_unsafe_address(objb, boffset, T_ILLEGAL);
5669
5670 call = make_runtime_call(RC_LEAF,
5671 OptoRuntime::vectorizedMismatch_Type(),
5672 stubAddr, stubName, TypePtr::BOTTOM,
5673 obja_adr, objb_adr, length, scale);
5674
5675 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5676 set_result(result);
5677 return true;
5678 }
5679
5680 /**
5681 * Calculate CRC32 for byte.
5682 * int java.util.zip.CRC32.update(int crc, int b)
5683 */
5684 bool LibraryCallKit::inline_updateCRC32() {
5685 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5686 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5687 // no receiver since it is static method
5688 Node* crc = argument(0); // type: int
5689 Node* b = argument(1); // type: int
5690
5691 /*
5692 * int c = ~ crc;
5693 * b = timesXtoThe32[(b ^ c) & 0xFF];
5694 * b = b ^ (c >>> 8);
5695 * crc = ~b;
5696 */
5697
5698 Node* M1 = intcon(-1);
5699 crc = _gvn.transform(new XorINode(crc, M1));
5700 Node* result = _gvn.transform(new XorINode(crc, b));
5701 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5702
5703 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5704 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5705 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5706 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5707
5708 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5709 result = _gvn.transform(new XorINode(crc, result));
5710 result = _gvn.transform(new XorINode(result, M1));
5711 set_result(result);
5712 return true;
5713 }
5714
5715 /**
5716 * Calculate CRC32 for byte[] array.
5717 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5718 */
5719 bool LibraryCallKit::inline_updateBytesCRC32() {
5720 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5721 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5722 // no receiver since it is static method
5723 Node* crc = argument(0); // type: int
5724 Node* src = argument(1); // type: oop
5725 Node* offset = argument(2); // type: int
5726 Node* length = argument(3); // type: int
5727
5728 const Type* src_type = src->Value(&_gvn);
5729 const TypeAryPtr* top_src = src_type->isa_aryptr();
5730 if (top_src == NULL || top_src->klass() == NULL) {
5731 // failed array check
5732 return false;
5733 }
5734
5735 // Figure out the size and type of the elements we will be copying.
5736 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5737 if (src_elem != T_BYTE) {
5738 return false;
5739 }
5740
5741 // 'src_start' points to src array + scaled offset
5742 Node* src_start = array_element_address(src, offset, src_elem);
5743
5744 // We assume that range check is done by caller.
5745 // TODO: generate range check (offset+length < src.length) in debug VM.
5746
5747 // Call the stub.
5748 address stubAddr = StubRoutines::updateBytesCRC32();
5749 const char *stubName = "updateBytesCRC32";
5750
5751 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5752 stubAddr, stubName, TypePtr::BOTTOM,
5753 crc, src_start, length);
5754 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5755 set_result(result);
5756 return true;
5757 }
5758
5759 /**
5760 * Calculate CRC32 for ByteBuffer.
5761 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5762 */
5763 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5764 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5765 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5766 // no receiver since it is static method
5767 Node* crc = argument(0); // type: int
5768 Node* src = argument(1); // type: long
5769 Node* offset = argument(3); // type: int
5770 Node* length = argument(4); // type: int
5771
5772 src = ConvL2X(src); // adjust Java long to machine word
5773 Node* base = _gvn.transform(new CastX2PNode(src));
5774 offset = ConvI2X(offset);
5775
5776 // 'src_start' points to src array + scaled offset
5777 Node* src_start = basic_plus_adr(top(), base, offset);
5778
5779 // Call the stub.
5780 address stubAddr = StubRoutines::updateBytesCRC32();
5781 const char *stubName = "updateBytesCRC32";
5782
5783 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5784 stubAddr, stubName, TypePtr::BOTTOM,
5785 crc, src_start, length);
5786 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5787 set_result(result);
5788 return true;
5789 }
5790
5791 //------------------------------get_table_from_crc32c_class-----------------------
5792 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5793 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5794 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5795
5796 return table;
5797 }
5798
5799 //------------------------------inline_updateBytesCRC32C-----------------------
5800 //
5801 // Calculate CRC32C for byte[] array.
5802 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5803 //
5804 bool LibraryCallKit::inline_updateBytesCRC32C() {
5805 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5806 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5807 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5808 // no receiver since it is a static method
5809 Node* crc = argument(0); // type: int
5810 Node* src = argument(1); // type: oop
5811 Node* offset = argument(2); // type: int
5812 Node* end = argument(3); // type: int
5813
5814 Node* length = _gvn.transform(new SubINode(end, offset));
5815
5816 const Type* src_type = src->Value(&_gvn);
5817 const TypeAryPtr* top_src = src_type->isa_aryptr();
5818 if (top_src == NULL || top_src->klass() == NULL) {
5819 // failed array check
5820 return false;
5821 }
5822
5823 // Figure out the size and type of the elements we will be copying.
5824 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5825 if (src_elem != T_BYTE) {
5826 return false;
5827 }
5828
5829 // 'src_start' points to src array + scaled offset
5830 Node* src_start = array_element_address(src, offset, src_elem);
5831
5832 // static final int[] byteTable in class CRC32C
5833 Node* table = get_table_from_crc32c_class(callee()->holder());
5834 Node* table_start = array_element_address(table, intcon(0), T_INT);
5835
5836 // We assume that range check is done by caller.
5837 // TODO: generate range check (offset+length < src.length) in debug VM.
5838
5839 // Call the stub.
5840 address stubAddr = StubRoutines::updateBytesCRC32C();
5841 const char *stubName = "updateBytesCRC32C";
5842
5843 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5844 stubAddr, stubName, TypePtr::BOTTOM,
5845 crc, src_start, length, table_start);
5846 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5847 set_result(result);
5848 return true;
5849 }
5850
5851 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5852 //
5853 // Calculate CRC32C for DirectByteBuffer.
5854 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5855 //
5856 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5857 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5858 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5859 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5860 // no receiver since it is a static method
5861 Node* crc = argument(0); // type: int
5862 Node* src = argument(1); // type: long
5863 Node* offset = argument(3); // type: int
5864 Node* end = argument(4); // type: int
5865
5866 Node* length = _gvn.transform(new SubINode(end, offset));
5867
5868 src = ConvL2X(src); // adjust Java long to machine word
5869 Node* base = _gvn.transform(new CastX2PNode(src));
5870 offset = ConvI2X(offset);
5871
5872 // 'src_start' points to src array + scaled offset
5873 Node* src_start = basic_plus_adr(top(), base, offset);
5874
5875 // static final int[] byteTable in class CRC32C
5876 Node* table = get_table_from_crc32c_class(callee()->holder());
5877 Node* table_start = array_element_address(table, intcon(0), T_INT);
5878
5879 // Call the stub.
5880 address stubAddr = StubRoutines::updateBytesCRC32C();
5881 const char *stubName = "updateBytesCRC32C";
5882
5883 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5884 stubAddr, stubName, TypePtr::BOTTOM,
5885 crc, src_start, length, table_start);
5886 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5887 set_result(result);
5888 return true;
5889 }
5890
5891 //------------------------------inline_updateBytesAdler32----------------------
5892 //
5893 // Calculate Adler32 checksum for byte[] array.
5894 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5895 //
5896 bool LibraryCallKit::inline_updateBytesAdler32() {
5897 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5898 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5899 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5900 // no receiver since it is static method
5901 Node* crc = argument(0); // type: int
5902 Node* src = argument(1); // type: oop
5903 Node* offset = argument(2); // type: int
5904 Node* length = argument(3); // type: int
5905
5906 const Type* src_type = src->Value(&_gvn);
5907 const TypeAryPtr* top_src = src_type->isa_aryptr();
5908 if (top_src == NULL || top_src->klass() == NULL) {
5909 // failed array check
5910 return false;
5911 }
5912
5913 // Figure out the size and type of the elements we will be copying.
5914 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5915 if (src_elem != T_BYTE) {
5916 return false;
5917 }
5918
5919 // 'src_start' points to src array + scaled offset
5920 Node* src_start = array_element_address(src, offset, src_elem);
5921
5922 // We assume that range check is done by caller.
5923 // TODO: generate range check (offset+length < src.length) in debug VM.
5924
5925 // Call the stub.
5926 address stubAddr = StubRoutines::updateBytesAdler32();
5927 const char *stubName = "updateBytesAdler32";
5928
5929 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5930 stubAddr, stubName, TypePtr::BOTTOM,
5931 crc, src_start, length);
5932 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5933 set_result(result);
5934 return true;
5935 }
5936
5937 //------------------------------inline_updateByteBufferAdler32---------------
5938 //
5939 // Calculate Adler32 checksum for DirectByteBuffer.
5940 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5941 //
5942 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5943 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5944 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5945 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5946 // no receiver since it is static method
5947 Node* crc = argument(0); // type: int
5948 Node* src = argument(1); // type: long
5949 Node* offset = argument(3); // type: int
5950 Node* length = argument(4); // type: int
5951
5952 src = ConvL2X(src); // adjust Java long to machine word
5953 Node* base = _gvn.transform(new CastX2PNode(src));
5954 offset = ConvI2X(offset);
5955
5956 // 'src_start' points to src array + scaled offset
5957 Node* src_start = basic_plus_adr(top(), base, offset);
5958
5959 // Call the stub.
5960 address stubAddr = StubRoutines::updateBytesAdler32();
5961 const char *stubName = "updateBytesAdler32";
5962
5963 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5964 stubAddr, stubName, TypePtr::BOTTOM,
5965 crc, src_start, length);
5966
5967 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5968 set_result(result);
5969 return true;
5970 }
5971
5972 //----------------------------inline_reference_get----------------------------
5973 // public T java.lang.ref.Reference.get();
5974 bool LibraryCallKit::inline_reference_get() {
5975 const int referent_offset = java_lang_ref_Reference::referent_offset;
5976 guarantee(referent_offset > 0, "should have already been set");
5977
5978 // Get the argument:
5979 Node* reference_obj = null_check_receiver();
5980 if (stopped()) return true;
5981
5982 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5983
5984 ciInstanceKlass* klass = env()->Object_klass();
5985 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5986
5987 Node* no_ctrl = NULL;
5988 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5989
5990 // Use the pre-barrier to record the value in the referent field
5991 pre_barrier(false /* do_load */,
5992 control(),
5993 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5994 result /* pre_val */,
5995 T_OBJECT);
5996
5997 // Add memory barrier to prevent commoning reads from this field
5998 // across safepoint since GC can change its value.
5999 insert_mem_bar(Op_MemBarCPUOrder);
6000
6001 set_result(result);
6002 return true;
6003 }
6004
6005
6006 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6007 bool is_exact=true, bool is_static=false,
6008 ciInstanceKlass * fromKls=NULL) {
6009 if (fromKls == NULL) {
6010 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6011 assert(tinst != NULL, "obj is null");
6012 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6013 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6014 fromKls = tinst->klass()->as_instance_klass();
6015 } else {
6016 assert(is_static, "only for static field access");
6017 }
6018 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6019 ciSymbol::make(fieldTypeString),
6020 is_static);
6021
6022 assert (field != NULL, "undefined field");
6023 if (field == NULL) return (Node *) NULL;
6024
6025 if (is_static) {
6026 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6027 fromObj = makecon(tip);
6028 }
6029
6030 // Next code copied from Parse::do_get_xxx():
6031
6032 // Compute address and memory type.
6033 int offset = field->offset_in_bytes();
6034 bool is_vol = field->is_volatile();
6035 ciType* field_klass = field->type();
6036 assert(field_klass->is_loaded(), "should be loaded");
6037 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6038 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6039 BasicType bt = field->layout_type();
6040
6041 // Build the resultant type of the load
6042 const Type *type;
6043 if (bt == T_OBJECT) {
6044 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6045 } else {
6046 type = Type::get_const_basic_type(bt);
6047 }
6048
6049 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
6050 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
6051 }
6052 // Build the load.
6053 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
6054 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
6055 // If reference is volatile, prevent following memory ops from
6056 // floating up past the volatile read. Also prevents commoning
6057 // another volatile read.
6058 if (is_vol) {
6059 // Memory barrier includes bogus read of value to force load BEFORE membar
6060 insert_mem_bar(Op_MemBarAcquire, loadedField);
6061 }
6062 return loadedField;
6063 }
6064
6065 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6066 bool is_exact = true, bool is_static = false,
6067 ciInstanceKlass * fromKls = NULL) {
6068 if (fromKls == NULL) {
6069 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6070 assert(tinst != NULL, "obj is null");
6071 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6072 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6073 fromKls = tinst->klass()->as_instance_klass();
6074 }
6075 else {
6076 assert(is_static, "only for static field access");
6077 }
6078 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6079 ciSymbol::make(fieldTypeString),
6080 is_static);
6081
6082 assert(field != NULL, "undefined field");
6083 assert(!field->is_volatile(), "not defined for volatile fields");
6084
6085 if (is_static) {
6086 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6087 fromObj = makecon(tip);
6088 }
6089
6090 // Next code copied from Parse::do_get_xxx():
6091
6092 // Compute address and memory type.
6093 int offset = field->offset_in_bytes();
6094 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6095
6096 return adr;
6097 }
6098
6099 //------------------------------inline_aescrypt_Block-----------------------
6100 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6101 address stubAddr = NULL;
6102 const char *stubName;
6103 assert(UseAES, "need AES instruction support");
6104
6105 switch(id) {
6106 case vmIntrinsics::_aescrypt_encryptBlock:
6107 stubAddr = StubRoutines::aescrypt_encryptBlock();
6108 stubName = "aescrypt_encryptBlock";
6109 break;
6110 case vmIntrinsics::_aescrypt_decryptBlock:
6111 stubAddr = StubRoutines::aescrypt_decryptBlock();
6112 stubName = "aescrypt_decryptBlock";
6113 break;
6114 }
6115 if (stubAddr == NULL) return false;
6116
6117 Node* aescrypt_object = argument(0);
6118 Node* src = argument(1);
6119 Node* src_offset = argument(2);
6120 Node* dest = argument(3);
6121 Node* dest_offset = argument(4);
6122
6123 // (1) src and dest are arrays.
6124 const Type* src_type = src->Value(&_gvn);
6125 const Type* dest_type = dest->Value(&_gvn);
6126 const TypeAryPtr* top_src = src_type->isa_aryptr();
6127 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6128 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6129
6130 // for the quick and dirty code we will skip all the checks.
6131 // we are just trying to get the call to be generated.
6132 Node* src_start = src;
6133 Node* dest_start = dest;
6134 if (src_offset != NULL || dest_offset != NULL) {
6135 assert(src_offset != NULL && dest_offset != NULL, "");
6136 src_start = array_element_address(src, src_offset, T_BYTE);
6137 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6138 }
6139
6140 // now need to get the start of its expanded key array
6141 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6142 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6143 if (k_start == NULL) return false;
6144
6145 if (Matcher::pass_original_key_for_aes()) {
6146 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6147 // compatibility issues between Java key expansion and SPARC crypto instructions
6148 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6149 if (original_k_start == NULL) return false;
6150
6151 // Call the stub.
6152 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6153 stubAddr, stubName, TypePtr::BOTTOM,
6154 src_start, dest_start, k_start, original_k_start);
6155 } else {
6156 // Call the stub.
6157 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6158 stubAddr, stubName, TypePtr::BOTTOM,
6159 src_start, dest_start, k_start);
6160 }
6161
6162 return true;
6163 }
6164
6165 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6166 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6167 address stubAddr = NULL;
6168 const char *stubName = NULL;
6169
6170 assert(UseAES, "need AES instruction support");
6171
6172 switch(id) {
6173 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6174 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6175 stubName = "cipherBlockChaining_encryptAESCrypt";
6176 break;
6177 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6178 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6179 stubName = "cipherBlockChaining_decryptAESCrypt";
6180 break;
6181 }
6182 if (stubAddr == NULL) return false;
6183
6184 Node* cipherBlockChaining_object = argument(0);
6185 Node* src = argument(1);
6186 Node* src_offset = argument(2);
6187 Node* len = argument(3);
6188 Node* dest = argument(4);
6189 Node* dest_offset = argument(5);
6190
6191 // (1) src and dest are arrays.
6192 const Type* src_type = src->Value(&_gvn);
6193 const Type* dest_type = dest->Value(&_gvn);
6194 const TypeAryPtr* top_src = src_type->isa_aryptr();
6195 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6196 assert (top_src != NULL && top_src->klass() != NULL
6197 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6198
6199 // checks are the responsibility of the caller
6200 Node* src_start = src;
6201 Node* dest_start = dest;
6202 if (src_offset != NULL || dest_offset != NULL) {
6203 assert(src_offset != NULL && dest_offset != NULL, "");
6204 src_start = array_element_address(src, src_offset, T_BYTE);
6205 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6206 }
6207
6208 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6209 // (because of the predicated logic executed earlier).
6210 // so we cast it here safely.
6211 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6212
6213 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6214 if (embeddedCipherObj == NULL) return false;
6215
6216 // cast it to what we know it will be at runtime
6217 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6218 assert(tinst != NULL, "CBC obj is null");
6219 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6220 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6221 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6222
6223 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6224 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6225 const TypeOopPtr* xtype = aklass->as_instance_type();
6226 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6227 aescrypt_object = _gvn.transform(aescrypt_object);
6228
6229 // we need to get the start of the aescrypt_object's expanded key array
6230 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6231 if (k_start == NULL) return false;
6232
6233 // similarly, get the start address of the r vector
6234 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6235 if (objRvec == NULL) return false;
6236 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6237
6238 Node* cbcCrypt;
6239 if (Matcher::pass_original_key_for_aes()) {
6240 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6241 // compatibility issues between Java key expansion and SPARC crypto instructions
6242 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6243 if (original_k_start == NULL) return false;
6244
6245 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6246 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6247 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6248 stubAddr, stubName, TypePtr::BOTTOM,
6249 src_start, dest_start, k_start, r_start, len, original_k_start);
6250 } else {
6251 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6252 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6253 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6254 stubAddr, stubName, TypePtr::BOTTOM,
6255 src_start, dest_start, k_start, r_start, len);
6256 }
6257
6258 // return cipher length (int)
6259 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6260 set_result(retvalue);
6261 return true;
6262 }
6263
6264 //------------------------------inline_counterMode_AESCrypt-----------------------
6265 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6266 assert(UseAES, "need AES instruction support");
6267 if (!UseAESCTRIntrinsics) return false;
6268
6269 address stubAddr = NULL;
6270 const char *stubName = NULL;
6271 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6272 stubAddr = StubRoutines::counterMode_AESCrypt();
6273 stubName = "counterMode_AESCrypt";
6274 }
6275 if (stubAddr == NULL) return false;
6276
6277 Node* counterMode_object = argument(0);
6278 Node* src = argument(1);
6279 Node* src_offset = argument(2);
6280 Node* len = argument(3);
6281 Node* dest = argument(4);
6282 Node* dest_offset = argument(5);
6283
6284 // (1) src and dest are arrays.
6285 const Type* src_type = src->Value(&_gvn);
6286 const Type* dest_type = dest->Value(&_gvn);
6287 const TypeAryPtr* top_src = src_type->isa_aryptr();
6288 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6289 assert(top_src != NULL && top_src->klass() != NULL &&
6290 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6291
6292 // checks are the responsibility of the caller
6293 Node* src_start = src;
6294 Node* dest_start = dest;
6295 if (src_offset != NULL || dest_offset != NULL) {
6296 assert(src_offset != NULL && dest_offset != NULL, "");
6297 src_start = array_element_address(src, src_offset, T_BYTE);
6298 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6299 }
6300
6301 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6302 // (because of the predicated logic executed earlier).
6303 // so we cast it here safely.
6304 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6305 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6306 if (embeddedCipherObj == NULL) return false;
6307 // cast it to what we know it will be at runtime
6308 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6309 assert(tinst != NULL, "CTR obj is null");
6310 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6311 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6312 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6313 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6314 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6315 const TypeOopPtr* xtype = aklass->as_instance_type();
6316 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6317 aescrypt_object = _gvn.transform(aescrypt_object);
6318 // we need to get the start of the aescrypt_object's expanded key array
6319 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6320 if (k_start == NULL) return false;
6321 // similarly, get the start address of the r vector
6322 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6323 if (obj_counter == NULL) return false;
6324 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6325
6326 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6327 if (saved_encCounter == NULL) return false;
6328 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6329 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6330
6331 Node* ctrCrypt;
6332 if (Matcher::pass_original_key_for_aes()) {
6333 // no SPARC version for AES/CTR intrinsics now.
6334 return false;
6335 }
6336 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6337 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6338 OptoRuntime::counterMode_aescrypt_Type(),
6339 stubAddr, stubName, TypePtr::BOTTOM,
6340 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6341
6342 // return cipher length (int)
6343 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6344 set_result(retvalue);
6345 return true;
6346 }
6347
6348 //------------------------------get_key_start_from_aescrypt_object-----------------------
6349 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6350 #if defined(PPC64) || defined(S390)
6351 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6352 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6353 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6354 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6355 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6356 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6357 if (objSessionK == NULL) {
6358 return (Node *) NULL;
6359 }
6360 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6361 #else
6362 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6363 #endif // PPC64
6364 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6365 if (objAESCryptKey == NULL) return (Node *) NULL;
6366
6367 // now have the array, need to get the start address of the K array
6368 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6369 return k_start;
6370 }
6371
6372 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6373 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6374 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6375 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6376 if (objAESCryptKey == NULL) return (Node *) NULL;
6377
6378 // now have the array, need to get the start address of the lastKey array
6379 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6380 return original_k_start;
6381 }
6382
6383 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6384 // Return node representing slow path of predicate check.
6385 // the pseudo code we want to emulate with this predicate is:
6386 // for encryption:
6387 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6388 // for decryption:
6389 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6390 // note cipher==plain is more conservative than the original java code but that's OK
6391 //
6392 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6393 // The receiver was checked for NULL already.
6394 Node* objCBC = argument(0);
6395
6396 // Load embeddedCipher field of CipherBlockChaining object.
6397 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6398
6399 // get AESCrypt klass for instanceOf check
6400 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6401 // will have same classloader as CipherBlockChaining object
6402 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6403 assert(tinst != NULL, "CBCobj is null");
6404 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6405
6406 // we want to do an instanceof comparison against the AESCrypt class
6407 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6408 if (!klass_AESCrypt->is_loaded()) {
6409 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6410 Node* ctrl = control();
6411 set_control(top()); // no regular fast path
6412 return ctrl;
6413 }
6414 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6415
6416 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6417 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6418 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6419
6420 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6421
6422 // for encryption, we are done
6423 if (!decrypting)
6424 return instof_false; // even if it is NULL
6425
6426 // for decryption, we need to add a further check to avoid
6427 // taking the intrinsic path when cipher and plain are the same
6428 // see the original java code for why.
6429 RegionNode* region = new RegionNode(3);
6430 region->init_req(1, instof_false);
6431 Node* src = argument(1);
6432 Node* dest = argument(4);
6433 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6434 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6435 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6436 region->init_req(2, src_dest_conjoint);
6437
6438 record_for_igvn(region);
6439 return _gvn.transform(region);
6440 }
6441
6442 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6443 // Return node representing slow path of predicate check.
6444 // the pseudo code we want to emulate with this predicate is:
6445 // for encryption:
6446 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6447 // for decryption:
6448 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6449 // note cipher==plain is more conservative than the original java code but that's OK
6450 //
6451
6452 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6453 // The receiver was checked for NULL already.
6454 Node* objCTR = argument(0);
6455
6456 // Load embeddedCipher field of CipherBlockChaining object.
6457 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6458
6459 // get AESCrypt klass for instanceOf check
6460 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6461 // will have same classloader as CipherBlockChaining object
6462 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6463 assert(tinst != NULL, "CTRobj is null");
6464 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6465
6466 // we want to do an instanceof comparison against the AESCrypt class
6467 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6468 if (!klass_AESCrypt->is_loaded()) {
6469 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6470 Node* ctrl = control();
6471 set_control(top()); // no regular fast path
6472 return ctrl;
6473 }
6474
6475 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6476 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6477 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6478 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6479 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6480
6481 return instof_false; // even if it is NULL
6482 }
6483
6484 //------------------------------inline_ghash_processBlocks
6485 bool LibraryCallKit::inline_ghash_processBlocks() {
6486 address stubAddr;
6487 const char *stubName;
6488 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6489
6490 stubAddr = StubRoutines::ghash_processBlocks();
6491 stubName = "ghash_processBlocks";
6492
6493 Node* data = argument(0);
6494 Node* offset = argument(1);
6495 Node* len = argument(2);
6496 Node* state = argument(3);
6497 Node* subkeyH = argument(4);
6498
6499 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6500 assert(state_start, "state is NULL");
6501 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6502 assert(subkeyH_start, "subkeyH is NULL");
6503 Node* data_start = array_element_address(data, offset, T_BYTE);
6504 assert(data_start, "data is NULL");
6505
6506 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6507 OptoRuntime::ghash_processBlocks_Type(),
6508 stubAddr, stubName, TypePtr::BOTTOM,
6509 state_start, subkeyH_start, data_start, len);
6510 return true;
6511 }
6512
6513 //------------------------------inline_sha_implCompress-----------------------
6514 //
6515 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6516 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6517 //
6518 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6519 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6520 //
6521 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6522 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6523 //
6524 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6525 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6526
6527 Node* sha_obj = argument(0);
6528 Node* src = argument(1); // type oop
6529 Node* ofs = argument(2); // type int
6530
6531 const Type* src_type = src->Value(&_gvn);
6532 const TypeAryPtr* top_src = src_type->isa_aryptr();
6533 if (top_src == NULL || top_src->klass() == NULL) {
6534 // failed array check
6535 return false;
6536 }
6537 // Figure out the size and type of the elements we will be copying.
6538 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6539 if (src_elem != T_BYTE) {
6540 return false;
6541 }
6542 // 'src_start' points to src array + offset
6543 Node* src_start = array_element_address(src, ofs, src_elem);
6544 Node* state = NULL;
6545 address stubAddr;
6546 const char *stubName;
6547
6548 switch(id) {
6549 case vmIntrinsics::_sha_implCompress:
6550 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6551 state = get_state_from_sha_object(sha_obj);
6552 stubAddr = StubRoutines::sha1_implCompress();
6553 stubName = "sha1_implCompress";
6554 break;
6555 case vmIntrinsics::_sha2_implCompress:
6556 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6557 state = get_state_from_sha_object(sha_obj);
6558 stubAddr = StubRoutines::sha256_implCompress();
6559 stubName = "sha256_implCompress";
6560 break;
6561 case vmIntrinsics::_sha5_implCompress:
6562 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6563 state = get_state_from_sha5_object(sha_obj);
6564 stubAddr = StubRoutines::sha512_implCompress();
6565 stubName = "sha512_implCompress";
6566 break;
6567 default:
6568 fatal_unexpected_iid(id);
6569 return false;
6570 }
6571 if (state == NULL) return false;
6572
6573 // Call the stub.
6574 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6575 stubAddr, stubName, TypePtr::BOTTOM,
6576 src_start, state);
6577
6578 return true;
6579 }
6580
6581 //------------------------------inline_digestBase_implCompressMB-----------------------
6582 //
6583 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6584 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6585 //
6586 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6587 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6588 "need SHA1/SHA256/SHA512 instruction support");
6589 assert((uint)predicate < 3, "sanity");
6590 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6591
6592 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6593 Node* src = argument(1); // byte[] array
6594 Node* ofs = argument(2); // type int
6595 Node* limit = argument(3); // type int
6596
6597 const Type* src_type = src->Value(&_gvn);
6598 const TypeAryPtr* top_src = src_type->isa_aryptr();
6599 if (top_src == NULL || top_src->klass() == NULL) {
6600 // failed array check
6601 return false;
6602 }
6603 // Figure out the size and type of the elements we will be copying.
6604 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6605 if (src_elem != T_BYTE) {
6606 return false;
6607 }
6608 // 'src_start' points to src array + offset
6609 Node* src_start = array_element_address(src, ofs, src_elem);
6610
6611 const char* klass_SHA_name = NULL;
6612 const char* stub_name = NULL;
6613 address stub_addr = NULL;
6614 bool long_state = false;
6615
6616 switch (predicate) {
6617 case 0:
6618 if (UseSHA1Intrinsics) {
6619 klass_SHA_name = "sun/security/provider/SHA";
6620 stub_name = "sha1_implCompressMB";
6621 stub_addr = StubRoutines::sha1_implCompressMB();
6622 }
6623 break;
6624 case 1:
6625 if (UseSHA256Intrinsics) {
6626 klass_SHA_name = "sun/security/provider/SHA2";
6627 stub_name = "sha256_implCompressMB";
6628 stub_addr = StubRoutines::sha256_implCompressMB();
6629 }
6630 break;
6631 case 2:
6632 if (UseSHA512Intrinsics) {
6633 klass_SHA_name = "sun/security/provider/SHA5";
6634 stub_name = "sha512_implCompressMB";
6635 stub_addr = StubRoutines::sha512_implCompressMB();
6636 long_state = true;
6637 }
6638 break;
6639 default:
6640 fatal("unknown SHA intrinsic predicate: %d", predicate);
6641 }
6642 if (klass_SHA_name != NULL) {
6643 // get DigestBase klass to lookup for SHA klass
6644 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6645 assert(tinst != NULL, "digestBase_obj is not instance???");
6646 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6647
6648 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6649 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6650 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6651 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6652 }
6653 return false;
6654 }
6655 //------------------------------inline_sha_implCompressMB-----------------------
6656 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6657 bool long_state, address stubAddr, const char *stubName,
6658 Node* src_start, Node* ofs, Node* limit) {
6659 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6660 const TypeOopPtr* xtype = aklass->as_instance_type();
6661 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6662 sha_obj = _gvn.transform(sha_obj);
6663
6664 Node* state;
6665 if (long_state) {
6666 state = get_state_from_sha5_object(sha_obj);
6667 } else {
6668 state = get_state_from_sha_object(sha_obj);
6669 }
6670 if (state == NULL) return false;
6671
6672 // Call the stub.
6673 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6674 OptoRuntime::digestBase_implCompressMB_Type(),
6675 stubAddr, stubName, TypePtr::BOTTOM,
6676 src_start, state, ofs, limit);
6677 // return ofs (int)
6678 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6679 set_result(result);
6680
6681 return true;
6682 }
6683
6684 //------------------------------get_state_from_sha_object-----------------------
6685 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6686 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6687 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6688 if (sha_state == NULL) return (Node *) NULL;
6689
6690 // now have the array, need to get the start address of the state array
6691 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6692 return state;
6693 }
6694
6695 //------------------------------get_state_from_sha5_object-----------------------
6696 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6697 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6698 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6699 if (sha_state == NULL) return (Node *) NULL;
6700
6701 // now have the array, need to get the start address of the state array
6702 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6703 return state;
6704 }
6705
6706 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6707 // Return node representing slow path of predicate check.
6708 // the pseudo code we want to emulate with this predicate is:
6709 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6710 //
6711 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6712 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6713 "need SHA1/SHA256/SHA512 instruction support");
6714 assert((uint)predicate < 3, "sanity");
6715
6716 // The receiver was checked for NULL already.
6717 Node* digestBaseObj = argument(0);
6718
6719 // get DigestBase klass for instanceOf check
6720 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6721 assert(tinst != NULL, "digestBaseObj is null");
6722 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6723
6724 const char* klass_SHA_name = NULL;
6725 switch (predicate) {
6726 case 0:
6727 if (UseSHA1Intrinsics) {
6728 // we want to do an instanceof comparison against the SHA class
6729 klass_SHA_name = "sun/security/provider/SHA";
6730 }
6731 break;
6732 case 1:
6733 if (UseSHA256Intrinsics) {
6734 // we want to do an instanceof comparison against the SHA2 class
6735 klass_SHA_name = "sun/security/provider/SHA2";
6736 }
6737 break;
6738 case 2:
6739 if (UseSHA512Intrinsics) {
6740 // we want to do an instanceof comparison against the SHA5 class
6741 klass_SHA_name = "sun/security/provider/SHA5";
6742 }
6743 break;
6744 default:
6745 fatal("unknown SHA intrinsic predicate: %d", predicate);
6746 }
6747
6748 ciKlass* klass_SHA = NULL;
6749 if (klass_SHA_name != NULL) {
6750 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6751 }
6752 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6753 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6754 Node* ctrl = control();
6755 set_control(top()); // no intrinsic path
6756 return ctrl;
6757 }
6758 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6759
6760 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6761 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6762 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6763 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6764
6765 return instof_false; // even if it is NULL
6766 }
6767
6768 //-------------inline_fma-----------------------------------
6769 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6770 Node *a = NULL;
6771 Node *b = NULL;
6772 Node *c = NULL;
6773 Node* result = NULL;
6774 switch (id) {
6775 case vmIntrinsics::_fmaD:
6776 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6777 // no receiver since it is static method
6778 a = round_double_node(argument(0));
6779 b = round_double_node(argument(2));
6780 c = round_double_node(argument(4));
6781 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6782 break;
6783 case vmIntrinsics::_fmaF:
6784 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6785 a = argument(0);
6786 b = argument(1);
6787 c = argument(2);
6788 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6789 break;
6790 default:
6791 fatal_unexpected_iid(id); break;
6792 }
6793 set_result(result);
6794 return true;
6795 }
6796
6797 bool LibraryCallKit::inline_profileBoolean() {
6798 Node* counts = argument(1);
6799 const TypeAryPtr* ary = NULL;
6800 ciArray* aobj = NULL;
6801 if (counts->is_Con()
6802 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6803 && (aobj = ary->const_oop()->as_array()) != NULL
6804 && (aobj->length() == 2)) {
6805 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6806 jint false_cnt = aobj->element_value(0).as_int();
6807 jint true_cnt = aobj->element_value(1).as_int();
6808
6809 if (C->log() != NULL) {
6810 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6811 false_cnt, true_cnt);
6812 }
6813
6814 if (false_cnt + true_cnt == 0) {
6815 // According to profile, never executed.
6816 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6817 Deoptimization::Action_reinterpret);
6818 return true;
6819 }
6820
6821 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6822 // is a number of each value occurrences.
6823 Node* result = argument(0);
6824 if (false_cnt == 0 || true_cnt == 0) {
6825 // According to profile, one value has been never seen.
6826 int expected_val = (false_cnt == 0) ? 1 : 0;
6827
6828 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6829 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6830
6831 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6832 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6833 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6834
6835 { // Slow path: uncommon trap for never seen value and then reexecute
6836 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6837 // the value has been seen at least once.
6838 PreserveJVMState pjvms(this);
6839 PreserveReexecuteState preexecs(this);
6840 jvms()->set_should_reexecute(true);
6841
6842 set_control(slow_path);
6843 set_i_o(i_o());
6844
6845 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6846 Deoptimization::Action_reinterpret);
6847 }
6848 // The guard for never seen value enables sharpening of the result and
6849 // returning a constant. It allows to eliminate branches on the same value
6850 // later on.
6851 set_control(fast_path);
6852 result = intcon(expected_val);
6853 }
6854 // Stop profiling.
6855 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6856 // By replacing method body with profile data (represented as ProfileBooleanNode
6857 // on IR level) we effectively disable profiling.
6858 // It enables full speed execution once optimized code is generated.
6859 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6860 C->record_for_igvn(profile);
6861 set_result(profile);
6862 return true;
6863 } else {
6864 // Continue profiling.
6865 // Profile data isn't available at the moment. So, execute method's bytecode version.
6866 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6867 // is compiled and counters aren't available since corresponding MethodHandle
6868 // isn't a compile-time constant.
6869 return false;
6870 }
6871 }
6872
6873 bool LibraryCallKit::inline_isCompileConstant() {
6874 Node* n = argument(0);
6875 set_result(n->is_Con() ? intcon(1) : intcon(0));
6876 return true;
6877 }