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 "oops/objArrayKlass.hpp"
32 #include "opto/addnode.hpp"
33 #include "opto/arraycopynode.hpp"
34 #include "opto/c2compiler.hpp"
35 #include "opto/callGenerator.hpp"
36 #include "opto/castnode.hpp"
37 #include "opto/cfgnode.hpp"
38 #include "opto/convertnode.hpp"
39 #include "opto/countbitsnode.hpp"
40 #include "opto/intrinsicnode.hpp"
41 #include "opto/idealKit.hpp"
42 #include "opto/mathexactnode.hpp"
43 #include "opto/movenode.hpp"
44 #include "opto/mulnode.hpp"
45 #include "opto/narrowptrnode.hpp"
46 #include "opto/opaquenode.hpp"
47 #include "opto/parse.hpp"
48 #include "opto/runtime.hpp"
49 #include "opto/subnode.hpp"
50 #include "prims/nativeLookup.hpp"
51 #include "prims/unsafe.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #ifdef TRACE_HAVE_INTRINSICS
54 #include "trace/traceMacros.hpp"
55 #endif
56
57 class LibraryIntrinsic : public InlineCallGenerator {
58 // Extend the set of intrinsics known to the runtime:
59 public:
60 private:
61 bool _is_virtual;
62 bool _does_virtual_dispatch;
63 int8_t _predicates_count; // Intrinsic is predicated by several conditions
64 int8_t _last_predicate; // Last generated predicate
65 vmIntrinsics::ID _intrinsic_id;
66
67 public:
68 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
69 : InlineCallGenerator(m),
70 _is_virtual(is_virtual),
71 _does_virtual_dispatch(does_virtual_dispatch),
72 _predicates_count((int8_t)predicates_count),
73 _last_predicate((int8_t)-1),
74 _intrinsic_id(id)
75 {
76 }
77 virtual bool is_intrinsic() const { return true; }
78 virtual bool is_virtual() const { return _is_virtual; }
79 virtual bool is_predicated() const { return _predicates_count > 0; }
80 virtual int predicates_count() const { return _predicates_count; }
81 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
82 virtual JVMState* generate(JVMState* jvms);
83 virtual Node* generate_predicate(JVMState* jvms, int predicate);
84 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
85 };
86
87
88 // Local helper class for LibraryIntrinsic:
89 class LibraryCallKit : public GraphKit {
90 private:
91 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
92 Node* _result; // the result node, if any
93 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
94
95 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr = false);
96
97 public:
98 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
99 : GraphKit(jvms),
100 _intrinsic(intrinsic),
101 _result(NULL)
102 {
103 // Check if this is a root compile. In that case we don't have a caller.
104 if (!jvms->has_method()) {
105 _reexecute_sp = sp();
106 } else {
107 // Find out how many arguments the interpreter needs when deoptimizing
108 // and save the stack pointer value so it can used by uncommon_trap.
109 // We find the argument count by looking at the declared signature.
110 bool ignored_will_link;
111 ciSignature* declared_signature = NULL;
112 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
113 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
114 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
115 }
116 }
117
118 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
119
120 ciMethod* caller() const { return jvms()->method(); }
121 int bci() const { return jvms()->bci(); }
122 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
123 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
124 ciMethod* callee() const { return _intrinsic->method(); }
125
126 bool try_to_inline(int predicate);
127 Node* try_to_predicate(int predicate);
128
129 void push_result() {
130 // Push the result onto the stack.
131 if (!stopped() && result() != NULL) {
132 BasicType bt = result()->bottom_type()->basic_type();
133 push_node(bt, result());
134 }
135 }
136
137 private:
138 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
139 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
140 }
141
142 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
143 void set_result(RegionNode* region, PhiNode* value);
144 Node* result() { return _result; }
145
146 virtual int reexecute_sp() { return _reexecute_sp; }
147
148 // Helper functions to inline natives
149 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
150 Node* generate_slow_guard(Node* test, RegionNode* region);
151 Node* generate_fair_guard(Node* test, RegionNode* region);
152 Node* generate_negative_guard(Node* index, RegionNode* region,
153 // resulting CastII of index:
154 Node* *pos_index = NULL);
155 Node* generate_limit_guard(Node* offset, Node* subseq_length,
156 Node* array_length,
157 RegionNode* region);
158 void generate_string_range_check(Node* array, Node* offset,
159 Node* length, bool char_count);
160 Node* generate_current_thread(Node* &tls_output);
161 Node* load_mirror_from_klass(Node* klass);
162 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
163 RegionNode* region, int null_path,
164 int offset);
165 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
166 RegionNode* region, int null_path) {
167 int offset = java_lang_Class::klass_offset_in_bytes();
168 return load_klass_from_mirror_common(mirror, never_see_null,
169 region, null_path,
170 offset);
171 }
172 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
173 RegionNode* region, int null_path) {
174 int offset = java_lang_Class::array_klass_offset_in_bytes();
175 return load_klass_from_mirror_common(mirror, never_see_null,
176 region, null_path,
177 offset);
178 }
179 Node* generate_access_flags_guard(Node* kls,
180 int modifier_mask, int modifier_bits,
181 RegionNode* region);
182 Node* generate_interface_guard(Node* kls, RegionNode* region);
183 Node* generate_array_guard(Node* kls, RegionNode* region) {
184 return generate_array_guard_common(kls, region, false, false);
185 }
186 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
187 return generate_array_guard_common(kls, region, false, true);
188 }
189 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
190 return generate_array_guard_common(kls, region, true, false);
191 }
192 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
193 return generate_array_guard_common(kls, region, true, true);
194 }
195 Node* generate_array_guard_common(Node* kls, RegionNode* region,
196 bool obj_array, bool not_array);
197 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
198 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
199 bool is_virtual = false, bool is_static = false);
200 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
201 return generate_method_call(method_id, false, true);
202 }
203 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
204 return generate_method_call(method_id, true, false);
205 }
206 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
207 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
208
209 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
210 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
211 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
213 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
214 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
215 bool inline_string_indexOfChar();
216 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_toBytesU();
218 bool inline_string_getCharsU();
219 bool inline_string_copy(bool compress);
220 bool inline_string_char_access(bool is_store);
221 Node* round_double_node(Node* n);
222 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
223 bool inline_math_native(vmIntrinsics::ID id);
224 bool inline_trig(vmIntrinsics::ID id);
225 bool inline_math(vmIntrinsics::ID id);
226 template <typename OverflowOp>
227 bool inline_math_overflow(Node* arg1, Node* arg2);
228 void inline_math_mathExact(Node* math, Node* test);
229 bool inline_math_addExactI(bool is_increment);
230 bool inline_math_addExactL(bool is_increment);
231 bool inline_math_multiplyExactI();
232 bool inline_math_multiplyExactL();
233 bool inline_math_negateExactI();
234 bool inline_math_negateExactL();
235 bool inline_math_subtractExactI(bool is_decrement);
236 bool inline_math_subtractExactL(bool is_decrement);
237 bool inline_min_max(vmIntrinsics::ID id);
238 bool inline_notify(vmIntrinsics::ID id);
239 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
240 // This returns Type::AnyPtr, RawPtr, or OopPtr.
241 int classify_unsafe_addr(Node* &base, Node* &offset);
242 Node* make_unsafe_address(Node* base, Node* offset);
243 // Helper for inline_unsafe_access.
244 // Generates the guards that check whether the result of
245 // Unsafe.getObject should be recorded in an SATB log buffer.
246 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
247
248 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
249 bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
250 static bool klass_needs_init_guard(Node* kls);
251 bool inline_unsafe_allocate();
252 bool inline_unsafe_newArray(bool uninitialized);
253 bool inline_unsafe_copyMemory();
254 bool inline_native_currentThread();
255
256 bool inline_native_time_funcs(address method, const char* funcName);
257 bool inline_native_isInterrupted();
258 bool inline_native_Class_query(vmIntrinsics::ID id);
259 bool inline_native_subtype_check();
260 bool inline_native_getLength();
261 bool inline_array_copyOf(bool is_copyOfRange);
262 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
263 bool inline_objects_checkIndex();
264 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
265 bool inline_native_clone(bool is_virtual);
266 bool inline_native_Reflection_getCallerClass();
267 // Helper function for inlining native object hash method
268 bool inline_native_hashcode(bool is_virtual, bool is_static);
269 bool inline_native_getClass();
270
271 // Helper functions for inlining arraycopy
272 bool inline_arraycopy();
273 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
274 RegionNode* slow_region);
275 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
276 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
277
278 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
279 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
280 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
281 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
282 bool inline_unsafe_fence(vmIntrinsics::ID id);
283 bool inline_fp_conversions(vmIntrinsics::ID id);
284 bool inline_number_methods(vmIntrinsics::ID id);
285 bool inline_reference_get();
286 bool inline_Class_cast();
287 bool inline_aescrypt_Block(vmIntrinsics::ID id);
288 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
289 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
290 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
291 Node* inline_counterMode_AESCrypt_predicate();
292 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
293 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
294 bool inline_ghash_processBlocks();
295 bool inline_sha_implCompress(vmIntrinsics::ID id);
296 bool inline_digestBase_implCompressMB(int predicate);
297 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
298 bool long_state, address stubAddr, const char *stubName,
299 Node* src_start, Node* ofs, Node* limit);
300 Node* get_state_from_sha_object(Node *sha_object);
301 Node* get_state_from_sha5_object(Node *sha_object);
302 Node* inline_digestBase_implCompressMB_predicate(int predicate);
303 bool inline_encodeISOArray();
304 bool inline_updateCRC32();
305 bool inline_updateBytesCRC32();
306 bool inline_updateByteBufferCRC32();
307 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
308 bool inline_updateBytesCRC32C();
309 bool inline_updateDirectByteBufferCRC32C();
310 bool inline_updateBytesAdler32();
311 bool inline_updateByteBufferAdler32();
312 bool inline_multiplyToLen();
313 bool inline_hasNegatives();
314 bool inline_squareToLen();
315 bool inline_mulAdd();
316 bool inline_montgomeryMultiply();
317 bool inline_montgomerySquare();
318 bool inline_vectorizedMismatch();
319
320 bool inline_profileBoolean();
321 bool inline_isCompileConstant();
322 };
323
324 //---------------------------make_vm_intrinsic----------------------------
325 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
326 vmIntrinsics::ID id = m->intrinsic_id();
327 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
328
329 if (!m->is_loaded()) {
330 // Do not attempt to inline unloaded methods.
331 return NULL;
332 }
333
334 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
335 bool is_available = false;
336
337 {
338 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
339 // the compiler must transition to '_thread_in_vm' state because both
340 // methods access VM-internal data.
341 VM_ENTRY_MARK;
342 methodHandle mh(THREAD, m->get_Method());
343 is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
344 !C->directive()->is_intrinsic_disabled(mh) &&
345 !vmIntrinsics::is_disabled_by_flags(mh);
346
347 }
348
349 if (is_available) {
350 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
351 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
352 return new LibraryIntrinsic(m, is_virtual,
353 vmIntrinsics::predicates_needed(id),
354 vmIntrinsics::does_virtual_dispatch(id),
355 (vmIntrinsics::ID) id);
356 } else {
357 return NULL;
358 }
359 }
360
361 //----------------------register_library_intrinsics-----------------------
362 // Initialize this file's data structures, for each Compile instance.
363 void Compile::register_library_intrinsics() {
364 // Nothing to do here.
365 }
366
367 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
368 LibraryCallKit kit(jvms, this);
369 Compile* C = kit.C;
370 int nodes = C->unique();
371 #ifndef PRODUCT
372 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
373 char buf[1000];
374 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
375 tty->print_cr("Intrinsic %s", str);
376 }
377 #endif
378 ciMethod* callee = kit.callee();
379 const int bci = kit.bci();
380
381 // Try to inline the intrinsic.
382 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
383 kit.try_to_inline(_last_predicate)) {
384 if (C->print_intrinsics() || C->print_inlining()) {
385 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
386 }
387 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
388 if (C->log()) {
389 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
390 vmIntrinsics::name_at(intrinsic_id()),
391 (is_virtual() ? " virtual='1'" : ""),
392 C->unique() - nodes);
393 }
394 // Push the result from the inlined method onto the stack.
395 kit.push_result();
396 C->print_inlining_update(this);
397 return kit.transfer_exceptions_into_jvms();
398 }
399
400 // The intrinsic bailed out
401 if (C->print_intrinsics() || C->print_inlining()) {
402 if (jvms->has_method()) {
403 // Not a root compile.
404 const char* msg;
405 if (callee->intrinsic_candidate()) {
406 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
407 } else {
408 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
409 : "failed to inline (intrinsic), method not annotated";
410 }
411 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
412 } else {
413 // Root compile
414 tty->print("Did not generate intrinsic %s%s at bci:%d in",
415 vmIntrinsics::name_at(intrinsic_id()),
416 (is_virtual() ? " (virtual)" : ""), bci);
417 }
418 }
419 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
420 C->print_inlining_update(this);
421 return NULL;
422 }
423
424 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
425 LibraryCallKit kit(jvms, this);
426 Compile* C = kit.C;
427 int nodes = C->unique();
428 _last_predicate = predicate;
429 #ifndef PRODUCT
430 assert(is_predicated() && predicate < predicates_count(), "sanity");
431 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
432 char buf[1000];
433 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
434 tty->print_cr("Predicate for intrinsic %s", str);
435 }
436 #endif
437 ciMethod* callee = kit.callee();
438 const int bci = kit.bci();
439
440 Node* slow_ctl = kit.try_to_predicate(predicate);
441 if (!kit.failing()) {
442 if (C->print_intrinsics() || C->print_inlining()) {
443 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
444 }
445 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
446 if (C->log()) {
447 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
448 vmIntrinsics::name_at(intrinsic_id()),
449 (is_virtual() ? " virtual='1'" : ""),
450 C->unique() - nodes);
451 }
452 return slow_ctl; // Could be NULL if the check folds.
453 }
454
455 // The intrinsic bailed out
456 if (C->print_intrinsics() || C->print_inlining()) {
457 if (jvms->has_method()) {
458 // Not a root compile.
459 const char* msg = "failed to generate predicate for intrinsic";
460 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
461 } else {
462 // Root compile
463 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
464 vmIntrinsics::name_at(intrinsic_id()),
465 (is_virtual() ? " (virtual)" : ""), bci);
466 }
467 }
468 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
469 return NULL;
470 }
471
472 bool LibraryCallKit::try_to_inline(int predicate) {
473 // Handle symbolic names for otherwise undistinguished boolean switches:
474 const bool is_store = true;
475 const bool is_compress = true;
476 const bool is_native_ptr = true;
477 const bool is_static = true;
478 const bool is_volatile = true;
479
480 if (!jvms()->has_method()) {
481 // Root JVMState has a null method.
482 assert(map()->memory()->Opcode() == Op_Parm, "");
483 // Insert the memory aliasing node
484 set_all_memory(reset_memory());
485 }
486 assert(merged_memory(), "");
487
488
489 switch (intrinsic_id()) {
490 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
491 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
492 case vmIntrinsics::_getClass: return inline_native_getClass();
493
494 case vmIntrinsics::_dsin:
495 case vmIntrinsics::_dcos:
496 case vmIntrinsics::_dtan:
497 case vmIntrinsics::_dabs:
498 case vmIntrinsics::_datan2:
499 case vmIntrinsics::_dsqrt:
500 case vmIntrinsics::_dexp:
501 case vmIntrinsics::_dlog:
502 case vmIntrinsics::_dlog10:
503 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
504
505 case vmIntrinsics::_min:
506 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
507
508 case vmIntrinsics::_notify:
509 case vmIntrinsics::_notifyAll:
510 if (InlineNotify) {
511 return inline_notify(intrinsic_id());
512 }
513 return false;
514
515 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
516 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
517 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
518 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
519 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
520 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
521 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
522 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
523 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
524 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
525 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
526 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
527
528 case vmIntrinsics::_arraycopy: return inline_arraycopy();
529
530 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
531 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
532 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
533 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
534
535 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
536 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
537 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
538 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
539 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
540 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
541 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
542
543 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
544 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
545
546 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
547 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
548 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
549 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
550
551 case vmIntrinsics::_compressStringC:
552 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
553 case vmIntrinsics::_inflateStringC:
554 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
555
556 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Relaxed, false);
557 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Relaxed, false);
558 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Relaxed, false);
559 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, false);
560 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, false);
561 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, false);
562 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, false);
563 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
564 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
565
566 case vmIntrinsics::_putObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Relaxed, false);
567 case vmIntrinsics::_putBoolean: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Relaxed, false);
568 case vmIntrinsics::_putByte: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Relaxed, false);
569 case vmIntrinsics::_putShort: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, false);
570 case vmIntrinsics::_putChar: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, false);
571 case vmIntrinsics::_putInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, false);
572 case vmIntrinsics::_putLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, false);
573 case vmIntrinsics::_putFloat: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Relaxed, false);
574 case vmIntrinsics::_putDouble: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
575
576 case vmIntrinsics::_getByte_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE, Relaxed, false);
577 case vmIntrinsics::_getShort_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT, Relaxed, false);
578 case vmIntrinsics::_getChar_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR, Relaxed, false);
579 case vmIntrinsics::_getInt_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_INT, Relaxed, false);
580 case vmIntrinsics::_getLong_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_LONG, Relaxed, false);
581 case vmIntrinsics::_getFloat_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
582 case vmIntrinsics::_getDouble_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
583 case vmIntrinsics::_getAddress_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS, Relaxed, false);
584
585 case vmIntrinsics::_putByte_raw: return inline_unsafe_access( is_native_ptr, is_store, T_BYTE, Relaxed, false);
586 case vmIntrinsics::_putShort_raw: return inline_unsafe_access( is_native_ptr, is_store, T_SHORT, Relaxed, false);
587 case vmIntrinsics::_putChar_raw: return inline_unsafe_access( is_native_ptr, is_store, T_CHAR, Relaxed, false);
588 case vmIntrinsics::_putInt_raw: return inline_unsafe_access( is_native_ptr, is_store, T_INT, Relaxed, false);
589 case vmIntrinsics::_putLong_raw: return inline_unsafe_access( is_native_ptr, is_store, T_LONG, Relaxed, false);
590 case vmIntrinsics::_putFloat_raw: return inline_unsafe_access( is_native_ptr, is_store, T_FLOAT, Relaxed, false);
591 case vmIntrinsics::_putDouble_raw: return inline_unsafe_access( is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
592 case vmIntrinsics::_putAddress_raw: return inline_unsafe_access( is_native_ptr, is_store, T_ADDRESS, Relaxed, false);
593
594 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Volatile, false);
595 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Volatile, false);
596 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Volatile, false);
597 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Volatile, false);
598 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Volatile, false);
599 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Volatile, false);
600 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Volatile, false);
601 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Volatile, false);
602 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Volatile, false);
603
604 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Volatile, false);
605 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Volatile, false);
606 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Volatile, false);
607 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Volatile, false);
608 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Volatile, false);
609 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Volatile, false);
610 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Volatile, false);
611 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Volatile, false);
612 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Volatile, false);
613
614 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, true);
615 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, true);
616 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, true);
617 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, true);
618
619 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, true);
620 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, true);
621 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, true);
622 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, true);
623
624 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Acquire, false);
625 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Acquire, false);
626 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Acquire, false);
627 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Acquire, false);
628 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Acquire, false);
629 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Acquire, false);
630 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Acquire, false);
631 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Acquire, false);
632 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Acquire, false);
633
634 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false);
635 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Release, false);
636 case vmIntrinsics::_putByteRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Release, false);
637 case vmIntrinsics::_putShortRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Release, false);
638 case vmIntrinsics::_putCharRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Release, false);
639 case vmIntrinsics::_putIntRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false);
640 case vmIntrinsics::_putLongRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false);
641 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Release, false);
642 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Release, false);
643
644 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Opaque, false);
645 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Opaque, false);
646 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Opaque, false);
647 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Opaque, false);
648 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Opaque, false);
649 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Opaque, false);
650 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Opaque, false);
651 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Opaque, false);
652 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Opaque, false);
653
654 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Opaque, false);
655 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Opaque, false);
656 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Opaque, false);
657 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Opaque, false);
658 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Opaque, false);
659 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Opaque, false);
660 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Opaque, false);
661 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Opaque, false);
662 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Opaque, false);
663
664 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
665 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
666 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
667
668 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
669 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
670 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
671 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
672 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
673 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
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
678 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
679 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
680 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
681 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
682 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
683 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
684 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
685 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
686 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
687
688 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
689 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
690 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
691 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
692 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
693
694 case vmIntrinsics::_loadFence:
695 case vmIntrinsics::_storeFence:
696 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
697
698 case vmIntrinsics::_currentThread: return inline_native_currentThread();
699 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
700
701 #ifdef TRACE_HAVE_INTRINSICS
702 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
703 #endif
704 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
705 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
706 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
707 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
708 case vmIntrinsics::_getLength: return inline_native_getLength();
709 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
710 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
711 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
712 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
713 case vmIntrinsics::_Objects_checkIndex: return inline_objects_checkIndex();
714 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
715
716 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
717 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
718
719 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
720
721 case vmIntrinsics::_isInstance:
722 case vmIntrinsics::_getModifiers:
723 case vmIntrinsics::_isInterface:
724 case vmIntrinsics::_isArray:
725 case vmIntrinsics::_isPrimitive:
726 case vmIntrinsics::_getSuperclass:
727 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
728
729 case vmIntrinsics::_floatToRawIntBits:
730 case vmIntrinsics::_floatToIntBits:
731 case vmIntrinsics::_intBitsToFloat:
732 case vmIntrinsics::_doubleToRawLongBits:
733 case vmIntrinsics::_doubleToLongBits:
734 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
735
736 case vmIntrinsics::_numberOfLeadingZeros_i:
737 case vmIntrinsics::_numberOfLeadingZeros_l:
738 case vmIntrinsics::_numberOfTrailingZeros_i:
739 case vmIntrinsics::_numberOfTrailingZeros_l:
740 case vmIntrinsics::_bitCount_i:
741 case vmIntrinsics::_bitCount_l:
742 case vmIntrinsics::_reverseBytes_i:
743 case vmIntrinsics::_reverseBytes_l:
744 case vmIntrinsics::_reverseBytes_s:
745 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
746
747 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
748
749 case vmIntrinsics::_Reference_get: return inline_reference_get();
750
751 case vmIntrinsics::_Class_cast: return inline_Class_cast();
752
753 case vmIntrinsics::_aescrypt_encryptBlock:
754 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
755
756 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
757 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
758 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
759
760 case vmIntrinsics::_counterMode_AESCrypt:
761 return inline_counterMode_AESCrypt(intrinsic_id());
762
763 case vmIntrinsics::_sha_implCompress:
764 case vmIntrinsics::_sha2_implCompress:
765 case vmIntrinsics::_sha5_implCompress:
766 return inline_sha_implCompress(intrinsic_id());
767
768 case vmIntrinsics::_digestBase_implCompressMB:
769 return inline_digestBase_implCompressMB(predicate);
770
771 case vmIntrinsics::_multiplyToLen:
772 return inline_multiplyToLen();
773
774 case vmIntrinsics::_squareToLen:
775 return inline_squareToLen();
776
777 case vmIntrinsics::_mulAdd:
778 return inline_mulAdd();
779
780 case vmIntrinsics::_montgomeryMultiply:
781 return inline_montgomeryMultiply();
782 case vmIntrinsics::_montgomerySquare:
783 return inline_montgomerySquare();
784
785 case vmIntrinsics::_vectorizedMismatch:
786 return inline_vectorizedMismatch();
787
788 case vmIntrinsics::_ghash_processBlocks:
789 return inline_ghash_processBlocks();
790
791 case vmIntrinsics::_encodeISOArray:
792 case vmIntrinsics::_encodeByteISOArray:
793 return inline_encodeISOArray();
794
795 case vmIntrinsics::_updateCRC32:
796 return inline_updateCRC32();
797 case vmIntrinsics::_updateBytesCRC32:
798 return inline_updateBytesCRC32();
799 case vmIntrinsics::_updateByteBufferCRC32:
800 return inline_updateByteBufferCRC32();
801
802 case vmIntrinsics::_updateBytesCRC32C:
803 return inline_updateBytesCRC32C();
804 case vmIntrinsics::_updateDirectByteBufferCRC32C:
805 return inline_updateDirectByteBufferCRC32C();
806
807 case vmIntrinsics::_updateBytesAdler32:
808 return inline_updateBytesAdler32();
809 case vmIntrinsics::_updateByteBufferAdler32:
810 return inline_updateByteBufferAdler32();
811
812 case vmIntrinsics::_profileBoolean:
813 return inline_profileBoolean();
814 case vmIntrinsics::_isCompileConstant:
815 return inline_isCompileConstant();
816
817 case vmIntrinsics::_hasNegatives:
818 return inline_hasNegatives();
819
820 default:
821 // If you get here, it may be that someone has added a new intrinsic
822 // to the list in vmSymbols.hpp without implementing it here.
823 #ifndef PRODUCT
824 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
825 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
826 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
827 }
828 #endif
829 return false;
830 }
831 }
832
833 Node* LibraryCallKit::try_to_predicate(int predicate) {
834 if (!jvms()->has_method()) {
835 // Root JVMState has a null method.
836 assert(map()->memory()->Opcode() == Op_Parm, "");
837 // Insert the memory aliasing node
838 set_all_memory(reset_memory());
839 }
840 assert(merged_memory(), "");
841
842 switch (intrinsic_id()) {
843 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
844 return inline_cipherBlockChaining_AESCrypt_predicate(false);
845 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
846 return inline_cipherBlockChaining_AESCrypt_predicate(true);
847 case vmIntrinsics::_counterMode_AESCrypt:
848 return inline_counterMode_AESCrypt_predicate();
849 case vmIntrinsics::_digestBase_implCompressMB:
850 return inline_digestBase_implCompressMB_predicate(predicate);
851
852 default:
853 // If you get here, it may be that someone has added a new intrinsic
854 // to the list in vmSymbols.hpp without implementing it here.
855 #ifndef PRODUCT
856 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
857 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
858 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
859 }
860 #endif
861 Node* slow_ctl = control();
862 set_control(top()); // No fast path instrinsic
863 return slow_ctl;
864 }
865 }
866
867 //------------------------------set_result-------------------------------
868 // Helper function for finishing intrinsics.
869 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
870 record_for_igvn(region);
871 set_control(_gvn.transform(region));
872 set_result( _gvn.transform(value));
873 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
874 }
875
876 //------------------------------generate_guard---------------------------
877 // Helper function for generating guarded fast-slow graph structures.
878 // The given 'test', if true, guards a slow path. If the test fails
879 // then a fast path can be taken. (We generally hope it fails.)
880 // In all cases, GraphKit::control() is updated to the fast path.
881 // The returned value represents the control for the slow path.
882 // The return value is never 'top'; it is either a valid control
883 // or NULL if it is obvious that the slow path can never be taken.
884 // Also, if region and the slow control are not NULL, the slow edge
885 // is appended to the region.
886 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
887 if (stopped()) {
888 // Already short circuited.
889 return NULL;
890 }
891
892 // Build an if node and its projections.
893 // If test is true we take the slow path, which we assume is uncommon.
894 if (_gvn.type(test) == TypeInt::ZERO) {
895 // The slow branch is never taken. No need to build this guard.
896 return NULL;
897 }
898
899 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
900
901 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
902 if (if_slow == top()) {
903 // The slow branch is never taken. No need to build this guard.
904 return NULL;
905 }
906
907 if (region != NULL)
908 region->add_req(if_slow);
909
910 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
911 set_control(if_fast);
912
913 return if_slow;
914 }
915
916 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
917 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
918 }
919 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
920 return generate_guard(test, region, PROB_FAIR);
921 }
922
923 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
924 Node* *pos_index) {
925 if (stopped())
926 return NULL; // already stopped
927 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
928 return NULL; // index is already adequately typed
929 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
930 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
931 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
932 if (is_neg != NULL && pos_index != NULL) {
933 // Emulate effect of Parse::adjust_map_after_if.
934 Node* ccast = new CastIINode(index, TypeInt::POS);
935 ccast->set_req(0, control());
936 (*pos_index) = _gvn.transform(ccast);
937 }
938 return is_neg;
939 }
940
941 // Make sure that 'position' is a valid limit index, in [0..length].
942 // There are two equivalent plans for checking this:
943 // A. (offset + copyLength) unsigned<= arrayLength
944 // B. offset <= (arrayLength - copyLength)
945 // We require that all of the values above, except for the sum and
946 // difference, are already known to be non-negative.
947 // Plan A is robust in the face of overflow, if offset and copyLength
948 // are both hugely positive.
949 //
950 // Plan B is less direct and intuitive, but it does not overflow at
951 // all, since the difference of two non-negatives is always
952 // representable. Whenever Java methods must perform the equivalent
953 // check they generally use Plan B instead of Plan A.
954 // For the moment we use Plan A.
955 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
956 Node* subseq_length,
957 Node* array_length,
958 RegionNode* region) {
959 if (stopped())
960 return NULL; // already stopped
961 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
962 if (zero_offset && subseq_length->eqv_uncast(array_length))
963 return NULL; // common case of whole-array copy
964 Node* last = subseq_length;
965 if (!zero_offset) // last += offset
966 last = _gvn.transform(new AddINode(last, offset));
967 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
968 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
969 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
970 return is_over;
971 }
972
973 // Emit range checks for the given String.value byte array
974 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
975 if (stopped()) {
976 return; // already stopped
977 }
978 RegionNode* bailout = new RegionNode(1);
979 record_for_igvn(bailout);
980 if (char_count) {
981 // Convert char count to byte count
982 count = _gvn.transform(new LShiftINode(count, intcon(1)));
983 }
984
985 // Offset and count must not be negative
986 generate_negative_guard(offset, bailout);
987 generate_negative_guard(count, bailout);
988 // Offset + count must not exceed length of array
989 generate_limit_guard(offset, count, load_array_length(array), bailout);
990
991 if (bailout->req() > 1) {
992 PreserveJVMState pjvms(this);
993 set_control(_gvn.transform(bailout));
994 uncommon_trap(Deoptimization::Reason_intrinsic,
995 Deoptimization::Action_maybe_recompile);
996 }
997 }
998
999 //--------------------------generate_current_thread--------------------
1000 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1001 ciKlass* thread_klass = env()->Thread_klass();
1002 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1003 Node* thread = _gvn.transform(new ThreadLocalNode());
1004 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1005 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1006 tls_output = thread;
1007 return threadObj;
1008 }
1009
1010
1011 //------------------------------make_string_method_node------------------------
1012 // Helper method for String intrinsic functions. This version is called with
1013 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1014 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1015 // containing the lengths of str1 and str2.
1016 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1017 Node* result = NULL;
1018 switch (opcode) {
1019 case Op_StrIndexOf:
1020 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1021 str1_start, cnt1, str2_start, cnt2, ae);
1022 break;
1023 case Op_StrComp:
1024 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1025 str1_start, cnt1, str2_start, cnt2, ae);
1026 break;
1027 case Op_StrEquals:
1028 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1029 // Use the constant length if there is one because optimized match rule may exist.
1030 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1031 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1032 break;
1033 default:
1034 ShouldNotReachHere();
1035 return NULL;
1036 }
1037
1038 // All these intrinsics have checks.
1039 C->set_has_split_ifs(true); // Has chance for split-if optimization
1040
1041 return _gvn.transform(result);
1042 }
1043
1044 //------------------------------inline_string_compareTo------------------------
1045 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1046 Node* arg1 = argument(0);
1047 Node* arg2 = argument(1);
1048
1049 // Get start addr and length of first argument
1050 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1051 Node* arg1_cnt = load_array_length(arg1);
1052
1053 // Get start addr and length of second argument
1054 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1055 Node* arg2_cnt = load_array_length(arg2);
1056
1057 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1058 set_result(result);
1059 return true;
1060 }
1061
1062 //------------------------------inline_string_equals------------------------
1063 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1064 Node* arg1 = argument(0);
1065 Node* arg2 = argument(1);
1066
1067 // paths (plus control) merge
1068 RegionNode* region = new RegionNode(3);
1069 Node* phi = new PhiNode(region, TypeInt::BOOL);
1070
1071 if (!stopped()) {
1072 // Get start addr and length of first argument
1073 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1074 Node* arg1_cnt = load_array_length(arg1);
1075
1076 // Get start addr and length of second argument
1077 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1078 Node* arg2_cnt = load_array_length(arg2);
1079
1080 // Check for arg1_cnt != arg2_cnt
1081 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1082 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1083 Node* if_ne = generate_slow_guard(bol, NULL);
1084 if (if_ne != NULL) {
1085 phi->init_req(2, intcon(0));
1086 region->init_req(2, if_ne);
1087 }
1088
1089 // Check for count == 0 is done by assembler code for StrEquals.
1090
1091 if (!stopped()) {
1092 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1093 phi->init_req(1, equals);
1094 region->init_req(1, control());
1095 }
1096 }
1097
1098 // post merge
1099 set_control(_gvn.transform(region));
1100 record_for_igvn(region);
1101
1102 set_result(_gvn.transform(phi));
1103 return true;
1104 }
1105
1106 //------------------------------inline_array_equals----------------------------
1107 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1108 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1109 Node* arg1 = argument(0);
1110 Node* arg2 = argument(1);
1111
1112 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1113 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1114 return true;
1115 }
1116
1117 //------------------------------inline_hasNegatives------------------------------
1118 bool LibraryCallKit::inline_hasNegatives() {
1119 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1120 return false;
1121 }
1122
1123 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1124 // no receiver since it is static method
1125 Node* ba = argument(0);
1126 Node* offset = argument(1);
1127 Node* len = argument(2);
1128
1129 // Range checks
1130 generate_string_range_check(ba, offset, len, false);
1131 if (stopped()) {
1132 return true;
1133 }
1134 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1135 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1136 set_result(_gvn.transform(result));
1137 return true;
1138 }
1139
1140 bool LibraryCallKit::inline_objects_checkIndex() {
1141 Node* index = argument(0);
1142 Node* length = argument(1);
1143 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1144 return false;
1145 }
1146
1147 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1148 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1149
1150 {
1151 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1152 uncommon_trap(Deoptimization::Reason_intrinsic,
1153 Deoptimization::Action_make_not_entrant);
1154 }
1155
1156 if (stopped()) {
1157 return false;
1158 }
1159
1160 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1161 BoolTest::mask btest = BoolTest::lt;
1162 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1163 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1164 _gvn.set_type(rc, rc->Value(&_gvn));
1165 if (!rc_bool->is_Con()) {
1166 record_for_igvn(rc);
1167 }
1168 set_control(_gvn.transform(new IfTrueNode(rc)));
1169 {
1170 PreserveJVMState pjvms(this);
1171 set_control(_gvn.transform(new IfFalseNode(rc)));
1172 uncommon_trap(Deoptimization::Reason_range_check,
1173 Deoptimization::Action_make_not_entrant);
1174 }
1175
1176 if (stopped()) {
1177 return false;
1178 }
1179
1180 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1181 result->set_req(0, control());
1182 result = _gvn.transform(result);
1183 set_result(result);
1184 replace_in_map(index, result);
1185 return true;
1186 }
1187
1188 //------------------------------inline_string_indexOf------------------------
1189 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1190 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1191 return false;
1192 }
1193 Node* src = argument(0);
1194 Node* tgt = argument(1);
1195
1196 // Make the merge point
1197 RegionNode* result_rgn = new RegionNode(4);
1198 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1199
1200 // Get start addr and length of source string
1201 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1202 Node* src_count = load_array_length(src);
1203
1204 // Get start addr and length of substring
1205 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1206 Node* tgt_count = load_array_length(tgt);
1207
1208 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1209 // Divide src size by 2 if String is UTF16 encoded
1210 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1211 }
1212 if (ae == StrIntrinsicNode::UU) {
1213 // Divide substring size by 2 if String is UTF16 encoded
1214 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1215 }
1216
1217 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1218 if (result != NULL) {
1219 result_phi->init_req(3, result);
1220 result_rgn->init_req(3, control());
1221 }
1222 set_control(_gvn.transform(result_rgn));
1223 record_for_igvn(result_rgn);
1224 set_result(_gvn.transform(result_phi));
1225
1226 return true;
1227 }
1228
1229 //-----------------------------inline_string_indexOf-----------------------
1230 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1231 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1232 return false;
1233 }
1234 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1235 return false;
1236 }
1237 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1238 Node* src = argument(0); // byte[]
1239 Node* src_count = argument(1); // char count
1240 Node* tgt = argument(2); // byte[]
1241 Node* tgt_count = argument(3); // char count
1242 Node* from_index = argument(4); // char index
1243
1244 // Multiply byte array index by 2 if String is UTF16 encoded
1245 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1246 src_count = _gvn.transform(new SubINode(src_count, from_index));
1247 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1248 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1249
1250 // Range checks
1251 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1252 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1253 if (stopped()) {
1254 return true;
1255 }
1256
1257 RegionNode* region = new RegionNode(5);
1258 Node* phi = new PhiNode(region, TypeInt::INT);
1259
1260 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1261 if (result != NULL) {
1262 // The result is index relative to from_index if substring was found, -1 otherwise.
1263 // Generate code which will fold into cmove.
1264 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1265 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1266
1267 Node* if_lt = generate_slow_guard(bol, NULL);
1268 if (if_lt != NULL) {
1269 // result == -1
1270 phi->init_req(3, result);
1271 region->init_req(3, if_lt);
1272 }
1273 if (!stopped()) {
1274 result = _gvn.transform(new AddINode(result, from_index));
1275 phi->init_req(4, result);
1276 region->init_req(4, control());
1277 }
1278 }
1279
1280 set_control(_gvn.transform(region));
1281 record_for_igvn(region);
1282 set_result(_gvn.transform(phi));
1283
1284 return true;
1285 }
1286
1287 // Create StrIndexOfNode with fast path checks
1288 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1289 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1290 // Check for substr count > string count
1291 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1292 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1293 Node* if_gt = generate_slow_guard(bol, NULL);
1294 if (if_gt != NULL) {
1295 phi->init_req(1, intcon(-1));
1296 region->init_req(1, if_gt);
1297 }
1298 if (!stopped()) {
1299 // Check for substr count == 0
1300 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1301 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1302 Node* if_zero = generate_slow_guard(bol, NULL);
1303 if (if_zero != NULL) {
1304 phi->init_req(2, intcon(0));
1305 region->init_req(2, if_zero);
1306 }
1307 }
1308 if (!stopped()) {
1309 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1310 }
1311 return NULL;
1312 }
1313
1314 //-----------------------------inline_string_indexOfChar-----------------------
1315 bool LibraryCallKit::inline_string_indexOfChar() {
1316 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1317 return false;
1318 }
1319 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1320 return false;
1321 }
1322 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1323 Node* src = argument(0); // byte[]
1324 Node* tgt = argument(1); // tgt is int ch
1325 Node* from_index = argument(2);
1326 Node* max = argument(3);
1327
1328 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1329 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1330 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1331
1332 // Range checks
1333 generate_string_range_check(src, src_offset, src_count, true);
1334 if (stopped()) {
1335 return true;
1336 }
1337
1338 RegionNode* region = new RegionNode(3);
1339 Node* phi = new PhiNode(region, TypeInt::INT);
1340
1341 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1342 C->set_has_split_ifs(true); // Has chance for split-if optimization
1343 _gvn.transform(result);
1344
1345 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1346 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1347
1348 Node* if_lt = generate_slow_guard(bol, NULL);
1349 if (if_lt != NULL) {
1350 // result == -1
1351 phi->init_req(2, result);
1352 region->init_req(2, if_lt);
1353 }
1354 if (!stopped()) {
1355 result = _gvn.transform(new AddINode(result, from_index));
1356 phi->init_req(1, result);
1357 region->init_req(1, control());
1358 }
1359 set_control(_gvn.transform(region));
1360 record_for_igvn(region);
1361 set_result(_gvn.transform(phi));
1362
1363 return true;
1364 }
1365 //---------------------------inline_string_copy---------------------
1366 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1367 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1368 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1369 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1370 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1371 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1372 bool LibraryCallKit::inline_string_copy(bool compress) {
1373 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1374 return false;
1375 }
1376 int nargs = 5; // 2 oops, 3 ints
1377 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1378
1379 Node* src = argument(0);
1380 Node* src_offset = argument(1);
1381 Node* dst = argument(2);
1382 Node* dst_offset = argument(3);
1383 Node* length = argument(4);
1384
1385 // Check for allocation before we add nodes that would confuse
1386 // tightly_coupled_allocation()
1387 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1388
1389 // Figure out the size and type of the elements we will be copying.
1390 const Type* src_type = src->Value(&_gvn);
1391 const Type* dst_type = dst->Value(&_gvn);
1392 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1393 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1394 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1395 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1396 "Unsupported array types for inline_string_copy");
1397
1398 // Range checks
1399 generate_string_range_check(src, src_offset, length, compress && src_elem == T_BYTE);
1400 generate_string_range_check(dst, dst_offset, length, !compress && dst_elem == T_BYTE);
1401 if (stopped()) {
1402 return true;
1403 }
1404
1405 // Convert char[] offsets to byte[] offsets
1406 if (compress && src_elem == T_BYTE) {
1407 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1408 } else if (!compress && dst_elem == T_BYTE) {
1409 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1410 }
1411
1412 Node* src_start = array_element_address(src, src_offset, src_elem);
1413 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1414 // 'src_start' points to src array + scaled offset
1415 // 'dst_start' points to dst array + scaled offset
1416 Node* count = NULL;
1417 if (compress) {
1418 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1419 } else {
1420 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1421 }
1422
1423 if (alloc != NULL) {
1424 if (alloc->maybe_set_complete(&_gvn)) {
1425 // "You break it, you buy it."
1426 InitializeNode* init = alloc->initialization();
1427 assert(init->is_complete(), "we just did this");
1428 init->set_complete_with_arraycopy();
1429 assert(dst->is_CheckCastPP(), "sanity");
1430 assert(dst->in(0)->in(0) == init, "dest pinned");
1431 }
1432 // Do not let stores that initialize this object be reordered with
1433 // a subsequent store that would make this object accessible by
1434 // other threads.
1435 // Record what AllocateNode this StoreStore protects so that
1436 // escape analysis can go from the MemBarStoreStoreNode to the
1437 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1438 // based on the escape status of the AllocateNode.
1439 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1440 }
1441 if (compress) {
1442 set_result(_gvn.transform(count));
1443 }
1444 return true;
1445 }
1446
1447 #ifdef _LP64
1448 #define XTOP ,top() /*additional argument*/
1449 #else //_LP64
1450 #define XTOP /*no additional argument*/
1451 #endif //_LP64
1452
1453 //------------------------inline_string_toBytesU--------------------------
1454 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1455 bool LibraryCallKit::inline_string_toBytesU() {
1456 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1457 return false;
1458 }
1459 // Get the arguments.
1460 Node* value = argument(0);
1461 Node* offset = argument(1);
1462 Node* length = argument(2);
1463
1464 Node* newcopy = NULL;
1465
1466 // Set the original stack and the reexecute bit for the interpreter to reexecute
1467 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1468 { PreserveReexecuteState preexecs(this);
1469 jvms()->set_should_reexecute(true);
1470
1471 // Check if a null path was taken unconditionally.
1472 value = null_check(value);
1473
1474 RegionNode* bailout = new RegionNode(1);
1475 record_for_igvn(bailout);
1476
1477 // Range checks
1478 generate_negative_guard(offset, bailout);
1479 generate_negative_guard(length, bailout);
1480 generate_limit_guard(offset, length, load_array_length(value), bailout);
1481 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1482 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1483
1484 if (bailout->req() > 1) {
1485 PreserveJVMState pjvms(this);
1486 set_control(_gvn.transform(bailout));
1487 uncommon_trap(Deoptimization::Reason_intrinsic,
1488 Deoptimization::Action_maybe_recompile);
1489 }
1490 if (stopped()) {
1491 return true;
1492 }
1493
1494 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1495 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1496 newcopy = new_array(klass_node, size, 0); // no arguments to push
1497 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1498
1499 // Calculate starting addresses.
1500 Node* src_start = array_element_address(value, offset, T_CHAR);
1501 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1502
1503 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1504 const TypeInt* toffset = gvn().type(offset)->is_int();
1505 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1506
1507 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1508 const char* copyfunc_name = "arraycopy";
1509 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1510 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1511 OptoRuntime::fast_arraycopy_Type(),
1512 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1513 src_start, dst_start, ConvI2X(length) XTOP);
1514 // Do not let reads from the cloned object float above the arraycopy.
1515 if (alloc != NULL) {
1516 if (alloc->maybe_set_complete(&_gvn)) {
1517 // "You break it, you buy it."
1518 InitializeNode* init = alloc->initialization();
1519 assert(init->is_complete(), "we just did this");
1520 init->set_complete_with_arraycopy();
1521 assert(newcopy->is_CheckCastPP(), "sanity");
1522 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1523 }
1524 // Do not let stores that initialize this object be reordered with
1525 // a subsequent store that would make this object accessible by
1526 // other threads.
1527 // Record what AllocateNode this StoreStore protects so that
1528 // escape analysis can go from the MemBarStoreStoreNode to the
1529 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1530 // based on the escape status of the AllocateNode.
1531 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1532 } else {
1533 insert_mem_bar(Op_MemBarCPUOrder);
1534 }
1535 } // original reexecute is set back here
1536
1537 C->set_has_split_ifs(true); // Has chance for split-if optimization
1538 if (!stopped()) {
1539 set_result(newcopy);
1540 }
1541 return true;
1542 }
1543
1544 //------------------------inline_string_getCharsU--------------------------
1545 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1546 bool LibraryCallKit::inline_string_getCharsU() {
1547 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1548 return false;
1549 }
1550
1551 // Get the arguments.
1552 Node* src = argument(0);
1553 Node* src_begin = argument(1);
1554 Node* src_end = argument(2); // exclusive offset (i < src_end)
1555 Node* dst = argument(3);
1556 Node* dst_begin = argument(4);
1557
1558 // Check for allocation before we add nodes that would confuse
1559 // tightly_coupled_allocation()
1560 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1561
1562 // Check if a null path was taken unconditionally.
1563 src = null_check(src);
1564 dst = null_check(dst);
1565 if (stopped()) {
1566 return true;
1567 }
1568
1569 // Get length and convert char[] offset to byte[] offset
1570 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1571 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1572
1573 // Range checks
1574 generate_string_range_check(src, src_begin, length, true);
1575 generate_string_range_check(dst, dst_begin, length, false);
1576 if (stopped()) {
1577 return true;
1578 }
1579
1580 if (!stopped()) {
1581 // Calculate starting addresses.
1582 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1583 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1584
1585 // Check if array addresses are aligned to HeapWordSize
1586 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1587 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1588 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1589 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1590
1591 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1592 const char* copyfunc_name = "arraycopy";
1593 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1594 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1595 OptoRuntime::fast_arraycopy_Type(),
1596 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1597 src_start, dst_start, ConvI2X(length) XTOP);
1598 // Do not let reads from the cloned object float above the arraycopy.
1599 if (alloc != NULL) {
1600 if (alloc->maybe_set_complete(&_gvn)) {
1601 // "You break it, you buy it."
1602 InitializeNode* init = alloc->initialization();
1603 assert(init->is_complete(), "we just did this");
1604 init->set_complete_with_arraycopy();
1605 assert(dst->is_CheckCastPP(), "sanity");
1606 assert(dst->in(0)->in(0) == init, "dest pinned");
1607 }
1608 // Do not let stores that initialize this object be reordered with
1609 // a subsequent store that would make this object accessible by
1610 // other threads.
1611 // Record what AllocateNode this StoreStore protects so that
1612 // escape analysis can go from the MemBarStoreStoreNode to the
1613 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1614 // based on the escape status of the AllocateNode.
1615 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1616 } else {
1617 insert_mem_bar(Op_MemBarCPUOrder);
1618 }
1619 }
1620
1621 C->set_has_split_ifs(true); // Has chance for split-if optimization
1622 return true;
1623 }
1624
1625 //----------------------inline_string_char_access----------------------------
1626 // Store/Load char to/from byte[] array.
1627 // static void StringUTF16.putChar(byte[] val, int index, int c)
1628 // static char StringUTF16.getChar(byte[] val, int index)
1629 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1630 Node* value = argument(0);
1631 Node* index = argument(1);
1632 Node* ch = is_store ? argument(2) : NULL;
1633
1634 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1635 // correctly requires matched array shapes.
1636 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1637 "sanity: byte[] and char[] bases agree");
1638 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1639 "sanity: byte[] and char[] scales agree");
1640
1641 // Bail when getChar over constants is requested: constant folding would
1642 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1643 // Java method would constant fold nicely instead.
1644 if (!is_store && value->is_Con() && index->is_Con()) {
1645 return false;
1646 }
1647
1648 Node* adr = array_element_address(value, index, T_CHAR);
1649 if (is_store) {
1650 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1651 false, false, true /* mismatched */);
1652 } else {
1653 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1654 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1655 set_result(ch);
1656 }
1657 return true;
1658 }
1659
1660 //--------------------------round_double_node--------------------------------
1661 // Round a double node if necessary.
1662 Node* LibraryCallKit::round_double_node(Node* n) {
1663 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1664 n = _gvn.transform(new RoundDoubleNode(0, n));
1665 return n;
1666 }
1667
1668 //------------------------------inline_math-----------------------------------
1669 // public static double Math.abs(double)
1670 // public static double Math.sqrt(double)
1671 // public static double Math.log(double)
1672 // public static double Math.log10(double)
1673 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1674 Node* arg = round_double_node(argument(0));
1675 Node* n = NULL;
1676 switch (id) {
1677 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1678 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1679 case vmIntrinsics::_dlog10: n = new Log10DNode(C, control(), arg); break;
1680 default: fatal_unexpected_iid(id); break;
1681 }
1682 set_result(_gvn.transform(n));
1683 return true;
1684 }
1685
1686 //------------------------------inline_trig----------------------------------
1687 // Inline sin/cos/tan instructions, if possible. If rounding is required, do
1688 // argument reduction which will turn into a fast/slow diamond.
1689 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1690 Node* arg = round_double_node(argument(0));
1691 Node* n = NULL;
1692
1693 switch (id) {
1694 case vmIntrinsics::_dtan: n = new TanDNode(C, control(), arg); break;
1695 default: fatal_unexpected_iid(id); break;
1696 }
1697 n = _gvn.transform(n);
1698
1699 // Rounding required? Check for argument reduction!
1700 if (Matcher::strict_fp_requires_explicit_rounding) {
1701 static const double pi_4 = 0.7853981633974483;
1702 static const double neg_pi_4 = -0.7853981633974483;
1703 // pi/2 in 80-bit extended precision
1704 // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
1705 // -pi/2 in 80-bit extended precision
1706 // static const unsigned char neg_pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0xbf,0x00,0x00,0x00,0x00,0x00,0x00};
1707 // Cutoff value for using this argument reduction technique
1708 //static const double pi_2_minus_epsilon = 1.564660403643354;
1709 //static const double neg_pi_2_plus_epsilon = -1.564660403643354;
1710
1711 // Pseudocode for sin:
1712 // if (x <= Math.PI / 4.0) {
1713 // if (x >= -Math.PI / 4.0) return fsin(x);
1714 // if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
1715 // } else {
1716 // if (x <= Math.PI / 2.0) return fcos(x - Math.PI / 2.0);
1717 // }
1718 // return StrictMath.sin(x);
1719
1720 // Pseudocode for cos:
1721 // if (x <= Math.PI / 4.0) {
1722 // if (x >= -Math.PI / 4.0) return fcos(x);
1723 // if (x >= -Math.PI / 2.0) return fsin(x + Math.PI / 2.0);
1724 // } else {
1725 // if (x <= Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
1726 // }
1727 // return StrictMath.cos(x);
1728
1729 // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
1730 // requires a special machine instruction to load it. Instead we'll try
1731 // the 'easy' case. If we really need the extra range +/- PI/2 we'll
1732 // probably do the math inside the SIN encoding.
1733
1734 // Make the merge point
1735 RegionNode* r = new RegionNode(3);
1736 Node* phi = new PhiNode(r, Type::DOUBLE);
1737
1738 // Flatten arg so we need only 1 test
1739 Node *abs = _gvn.transform(new AbsDNode(arg));
1740 // Node for PI/4 constant
1741 Node *pi4 = makecon(TypeD::make(pi_4));
1742 // Check PI/4 : abs(arg)
1743 Node *cmp = _gvn.transform(new CmpDNode(pi4,abs));
1744 // Check: If PI/4 < abs(arg) then go slow
1745 Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt ));
1746 // Branch either way
1747 IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1748 set_control(opt_iff(r,iff));
1749
1750 // Set fast path result
1751 phi->init_req(2, n);
1752
1753 // Slow path - non-blocking leaf call
1754 Node* call = NULL;
1755 switch (id) {
1756 case vmIntrinsics::_dtan:
1757 call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1758 CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
1759 "Tan", NULL, arg, top());
1760 break;
1761 }
1762 assert(control()->in(0) == call, "");
1763 Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1764 r->init_req(1, control());
1765 phi->init_req(1, slow_result);
1766
1767 // Post-merge
1768 set_control(_gvn.transform(r));
1769 record_for_igvn(r);
1770 n = _gvn.transform(phi);
1771
1772 C->set_has_split_ifs(true); // Has chance for split-if optimization
1773 }
1774 set_result(n);
1775 return true;
1776 }
1777
1778 //------------------------------runtime_math-----------------------------
1779 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1780 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1781 "must be (DD)D or (D)D type");
1782
1783 // Inputs
1784 Node* a = round_double_node(argument(0));
1785 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1786
1787 const TypePtr* no_memory_effects = NULL;
1788 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1789 no_memory_effects,
1790 a, top(), b, b ? top() : NULL);
1791 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1792 #ifdef ASSERT
1793 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1794 assert(value_top == top(), "second value must be top");
1795 #endif
1796
1797 set_result(value);
1798 return true;
1799 }
1800
1801 //------------------------------inline_math_native-----------------------------
1802 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1803 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1804 switch (id) {
1805 // These intrinsics are not properly supported on all hardware
1806 case vmIntrinsics::_dsin:
1807 return StubRoutines::dsin() != NULL ?
1808 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1809 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1810 case vmIntrinsics::_dcos:
1811 return StubRoutines::dcos() != NULL ?
1812 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1813 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1814 case vmIntrinsics::_dtan: return Matcher::has_match_rule(Op_TanD) ? inline_trig(id) :
1815 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1816
1817 case vmIntrinsics::_dlog:
1818 return StubRoutines::dlog() != NULL ?
1819 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1820 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1821 case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_math(id) :
1822 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1823
1824 // These intrinsics are supported on all hardware
1825 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1826 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1827
1828 case vmIntrinsics::_dexp:
1829 return StubRoutines::dexp() != NULL ?
1830 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1831 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1832 case vmIntrinsics::_dpow:
1833 return StubRoutines::dpow() != NULL ?
1834 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1835 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1836 #undef FN_PTR
1837
1838 // These intrinsics are not yet correctly implemented
1839 case vmIntrinsics::_datan2:
1840 return false;
1841
1842 default:
1843 fatal_unexpected_iid(id);
1844 return false;
1845 }
1846 }
1847
1848 static bool is_simple_name(Node* n) {
1849 return (n->req() == 1 // constant
1850 || (n->is_Type() && n->as_Type()->type()->singleton())
1851 || n->is_Proj() // parameter or return value
1852 || n->is_Phi() // local of some sort
1853 );
1854 }
1855
1856 //----------------------------inline_notify-----------------------------------*
1857 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1858 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1859 address func;
1860 if (id == vmIntrinsics::_notify) {
1861 func = OptoRuntime::monitor_notify_Java();
1862 } else {
1863 func = OptoRuntime::monitor_notifyAll_Java();
1864 }
1865 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1866 make_slow_call_ex(call, env()->Throwable_klass(), false);
1867 return true;
1868 }
1869
1870
1871 //----------------------------inline_min_max-----------------------------------
1872 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1873 set_result(generate_min_max(id, argument(0), argument(1)));
1874 return true;
1875 }
1876
1877 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1878 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1879 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1880 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1881 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1882
1883 {
1884 PreserveJVMState pjvms(this);
1885 PreserveReexecuteState preexecs(this);
1886 jvms()->set_should_reexecute(true);
1887
1888 set_control(slow_path);
1889 set_i_o(i_o());
1890
1891 uncommon_trap(Deoptimization::Reason_intrinsic,
1892 Deoptimization::Action_none);
1893 }
1894
1895 set_control(fast_path);
1896 set_result(math);
1897 }
1898
1899 template <typename OverflowOp>
1900 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1901 typedef typename OverflowOp::MathOp MathOp;
1902
1903 MathOp* mathOp = new MathOp(arg1, arg2);
1904 Node* operation = _gvn.transform( mathOp );
1905 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1906 inline_math_mathExact(operation, ofcheck);
1907 return true;
1908 }
1909
1910 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1911 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1912 }
1913
1914 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1915 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1916 }
1917
1918 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1919 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1920 }
1921
1922 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1923 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1924 }
1925
1926 bool LibraryCallKit::inline_math_negateExactI() {
1927 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1928 }
1929
1930 bool LibraryCallKit::inline_math_negateExactL() {
1931 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1932 }
1933
1934 bool LibraryCallKit::inline_math_multiplyExactI() {
1935 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1936 }
1937
1938 bool LibraryCallKit::inline_math_multiplyExactL() {
1939 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1940 }
1941
1942 Node*
1943 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1944 // These are the candidate return value:
1945 Node* xvalue = x0;
1946 Node* yvalue = y0;
1947
1948 if (xvalue == yvalue) {
1949 return xvalue;
1950 }
1951
1952 bool want_max = (id == vmIntrinsics::_max);
1953
1954 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1955 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1956 if (txvalue == NULL || tyvalue == NULL) return top();
1957 // This is not really necessary, but it is consistent with a
1958 // hypothetical MaxINode::Value method:
1959 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1960
1961 // %%% This folding logic should (ideally) be in a different place.
1962 // Some should be inside IfNode, and there to be a more reliable
1963 // transformation of ?: style patterns into cmoves. We also want
1964 // more powerful optimizations around cmove and min/max.
1965
1966 // Try to find a dominating comparison of these guys.
1967 // It can simplify the index computation for Arrays.copyOf
1968 // and similar uses of System.arraycopy.
1969 // First, compute the normalized version of CmpI(x, y).
1970 int cmp_op = Op_CmpI;
1971 Node* xkey = xvalue;
1972 Node* ykey = yvalue;
1973 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1974 if (ideal_cmpxy->is_Cmp()) {
1975 // E.g., if we have CmpI(length - offset, count),
1976 // it might idealize to CmpI(length, count + offset)
1977 cmp_op = ideal_cmpxy->Opcode();
1978 xkey = ideal_cmpxy->in(1);
1979 ykey = ideal_cmpxy->in(2);
1980 }
1981
1982 // Start by locating any relevant comparisons.
1983 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1984 Node* cmpxy = NULL;
1985 Node* cmpyx = NULL;
1986 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1987 Node* cmp = start_from->fast_out(k);
1988 if (cmp->outcnt() > 0 && // must have prior uses
1989 cmp->in(0) == NULL && // must be context-independent
1990 cmp->Opcode() == cmp_op) { // right kind of compare
1991 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1992 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1993 }
1994 }
1995
1996 const int NCMPS = 2;
1997 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1998 int cmpn;
1999 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2000 if (cmps[cmpn] != NULL) break; // find a result
2001 }
2002 if (cmpn < NCMPS) {
2003 // Look for a dominating test that tells us the min and max.
2004 int depth = 0; // Limit search depth for speed
2005 Node* dom = control();
2006 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2007 if (++depth >= 100) break;
2008 Node* ifproj = dom;
2009 if (!ifproj->is_Proj()) continue;
2010 Node* iff = ifproj->in(0);
2011 if (!iff->is_If()) continue;
2012 Node* bol = iff->in(1);
2013 if (!bol->is_Bool()) continue;
2014 Node* cmp = bol->in(1);
2015 if (cmp == NULL) continue;
2016 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2017 if (cmps[cmpn] == cmp) break;
2018 if (cmpn == NCMPS) continue;
2019 BoolTest::mask btest = bol->as_Bool()->_test._test;
2020 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2021 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2022 // At this point, we know that 'x btest y' is true.
2023 switch (btest) {
2024 case BoolTest::eq:
2025 // They are proven equal, so we can collapse the min/max.
2026 // Either value is the answer. Choose the simpler.
2027 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2028 return yvalue;
2029 return xvalue;
2030 case BoolTest::lt: // x < y
2031 case BoolTest::le: // x <= y
2032 return (want_max ? yvalue : xvalue);
2033 case BoolTest::gt: // x > y
2034 case BoolTest::ge: // x >= y
2035 return (want_max ? xvalue : yvalue);
2036 }
2037 }
2038 }
2039
2040 // We failed to find a dominating test.
2041 // Let's pick a test that might GVN with prior tests.
2042 Node* best_bol = NULL;
2043 BoolTest::mask best_btest = BoolTest::illegal;
2044 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2045 Node* cmp = cmps[cmpn];
2046 if (cmp == NULL) continue;
2047 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2048 Node* bol = cmp->fast_out(j);
2049 if (!bol->is_Bool()) continue;
2050 BoolTest::mask btest = bol->as_Bool()->_test._test;
2051 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2052 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2053 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2054 best_bol = bol->as_Bool();
2055 best_btest = btest;
2056 }
2057 }
2058 }
2059
2060 Node* answer_if_true = NULL;
2061 Node* answer_if_false = NULL;
2062 switch (best_btest) {
2063 default:
2064 if (cmpxy == NULL)
2065 cmpxy = ideal_cmpxy;
2066 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2067 // and fall through:
2068 case BoolTest::lt: // x < y
2069 case BoolTest::le: // x <= y
2070 answer_if_true = (want_max ? yvalue : xvalue);
2071 answer_if_false = (want_max ? xvalue : yvalue);
2072 break;
2073 case BoolTest::gt: // x > y
2074 case BoolTest::ge: // x >= y
2075 answer_if_true = (want_max ? xvalue : yvalue);
2076 answer_if_false = (want_max ? yvalue : xvalue);
2077 break;
2078 }
2079
2080 jint hi, lo;
2081 if (want_max) {
2082 // We can sharpen the minimum.
2083 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2084 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2085 } else {
2086 // We can sharpen the maximum.
2087 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2088 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2089 }
2090
2091 // Use a flow-free graph structure, to avoid creating excess control edges
2092 // which could hinder other optimizations.
2093 // Since Math.min/max is often used with arraycopy, we want
2094 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2095 Node* cmov = CMoveNode::make(NULL, best_bol,
2096 answer_if_false, answer_if_true,
2097 TypeInt::make(lo, hi, widen));
2098
2099 return _gvn.transform(cmov);
2100
2101 /*
2102 // This is not as desirable as it may seem, since Min and Max
2103 // nodes do not have a full set of optimizations.
2104 // And they would interfere, anyway, with 'if' optimizations
2105 // and with CMoveI canonical forms.
2106 switch (id) {
2107 case vmIntrinsics::_min:
2108 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2109 case vmIntrinsics::_max:
2110 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2111 default:
2112 ShouldNotReachHere();
2113 }
2114 */
2115 }
2116
2117 inline int
2118 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2119 const TypePtr* base_type = TypePtr::NULL_PTR;
2120 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2121 if (base_type == NULL) {
2122 // Unknown type.
2123 return Type::AnyPtr;
2124 } else if (base_type == TypePtr::NULL_PTR) {
2125 // Since this is a NULL+long form, we have to switch to a rawptr.
2126 base = _gvn.transform(new CastX2PNode(offset));
2127 offset = MakeConX(0);
2128 return Type::RawPtr;
2129 } else if (base_type->base() == Type::RawPtr) {
2130 return Type::RawPtr;
2131 } else if (base_type->isa_oopptr()) {
2132 // Base is never null => always a heap address.
2133 if (base_type->ptr() == TypePtr::NotNull) {
2134 return Type::OopPtr;
2135 }
2136 // Offset is small => always a heap address.
2137 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2138 if (offset_type != NULL &&
2139 base_type->offset() == 0 && // (should always be?)
2140 offset_type->_lo >= 0 &&
2141 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2142 return Type::OopPtr;
2143 }
2144 // Otherwise, it might either be oop+off or NULL+addr.
2145 return Type::AnyPtr;
2146 } else {
2147 // No information:
2148 return Type::AnyPtr;
2149 }
2150 }
2151
2152 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2153 int kind = classify_unsafe_addr(base, offset);
2154 if (kind == Type::RawPtr) {
2155 return basic_plus_adr(top(), base, offset);
2156 } else {
2157 return basic_plus_adr(base, offset);
2158 }
2159 }
2160
2161 //--------------------------inline_number_methods-----------------------------
2162 // inline int Integer.numberOfLeadingZeros(int)
2163 // inline int Long.numberOfLeadingZeros(long)
2164 //
2165 // inline int Integer.numberOfTrailingZeros(int)
2166 // inline int Long.numberOfTrailingZeros(long)
2167 //
2168 // inline int Integer.bitCount(int)
2169 // inline int Long.bitCount(long)
2170 //
2171 // inline char Character.reverseBytes(char)
2172 // inline short Short.reverseBytes(short)
2173 // inline int Integer.reverseBytes(int)
2174 // inline long Long.reverseBytes(long)
2175 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2176 Node* arg = argument(0);
2177 Node* n = NULL;
2178 switch (id) {
2179 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2180 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2181 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2182 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2183 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2184 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2185 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2186 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2187 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2188 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2189 default: fatal_unexpected_iid(id); break;
2190 }
2191 set_result(_gvn.transform(n));
2192 return true;
2193 }
2194
2195 //----------------------------inline_unsafe_access----------------------------
2196
2197 const static BasicType T_ADDRESS_HOLDER = T_LONG;
2198
2199 // Helper that guards and inserts a pre-barrier.
2200 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2201 Node* pre_val, bool need_mem_bar) {
2202 // We could be accessing the referent field of a reference object. If so, when G1
2203 // is enabled, we need to log the value in the referent field in an SATB buffer.
2204 // This routine performs some compile time filters and generates suitable
2205 // runtime filters that guard the pre-barrier code.
2206 // Also add memory barrier for non volatile load from the referent field
2207 // to prevent commoning of loads across safepoint.
2208 if (!UseG1GC && !need_mem_bar)
2209 return;
2210
2211 // Some compile time checks.
2212
2213 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2214 const TypeX* otype = offset->find_intptr_t_type();
2215 if (otype != NULL && otype->is_con() &&
2216 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2217 // Constant offset but not the reference_offset so just return
2218 return;
2219 }
2220
2221 // We only need to generate the runtime guards for instances.
2222 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2223 if (btype != NULL) {
2224 if (btype->isa_aryptr()) {
2225 // Array type so nothing to do
2226 return;
2227 }
2228
2229 const TypeInstPtr* itype = btype->isa_instptr();
2230 if (itype != NULL) {
2231 // Can the klass of base_oop be statically determined to be
2232 // _not_ a sub-class of Reference and _not_ Object?
2233 ciKlass* klass = itype->klass();
2234 if ( klass->is_loaded() &&
2235 !klass->is_subtype_of(env()->Reference_klass()) &&
2236 !env()->Object_klass()->is_subtype_of(klass)) {
2237 return;
2238 }
2239 }
2240 }
2241
2242 // The compile time filters did not reject base_oop/offset so
2243 // we need to generate the following runtime filters
2244 //
2245 // if (offset == java_lang_ref_Reference::_reference_offset) {
2246 // if (instance_of(base, java.lang.ref.Reference)) {
2247 // pre_barrier(_, pre_val, ...);
2248 // }
2249 // }
2250
2251 float likely = PROB_LIKELY( 0.999);
2252 float unlikely = PROB_UNLIKELY(0.999);
2253
2254 IdealKit ideal(this);
2255 #define __ ideal.
2256
2257 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2258
2259 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2260 // Update graphKit memory and control from IdealKit.
2261 sync_kit(ideal);
2262
2263 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2264 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2265
2266 // Update IdealKit memory and control from graphKit.
2267 __ sync_kit(this);
2268
2269 Node* one = __ ConI(1);
2270 // is_instof == 0 if base_oop == NULL
2271 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2272
2273 // Update graphKit from IdeakKit.
2274 sync_kit(ideal);
2275
2276 // Use the pre-barrier to record the value in the referent field
2277 pre_barrier(false /* do_load */,
2278 __ ctrl(),
2279 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2280 pre_val /* pre_val */,
2281 T_OBJECT);
2282 if (need_mem_bar) {
2283 // Add memory barrier to prevent commoning reads from this field
2284 // across safepoint since GC can change its value.
2285 insert_mem_bar(Op_MemBarCPUOrder);
2286 }
2287 // Update IdealKit from graphKit.
2288 __ sync_kit(this);
2289
2290 } __ end_if(); // _ref_type != ref_none
2291 } __ end_if(); // offset == referent_offset
2292
2293 // Final sync IdealKit and GraphKit.
2294 final_sync(ideal);
2295 #undef __
2296 }
2297
2298
2299 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
2300 // Attempt to infer a sharper value type from the offset and base type.
2301 ciKlass* sharpened_klass = NULL;
2302
2303 // See if it is an instance field, with an object type.
2304 if (alias_type->field() != NULL) {
2305 assert(!is_native_ptr, "native pointer op cannot use a java address");
2306 if (alias_type->field()->type()->is_klass()) {
2307 sharpened_klass = alias_type->field()->type()->as_klass();
2308 }
2309 }
2310
2311 // See if it is a narrow oop array.
2312 if (adr_type->isa_aryptr()) {
2313 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2314 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2315 if (elem_type != NULL) {
2316 sharpened_klass = elem_type->klass();
2317 }
2318 }
2319 }
2320
2321 // The sharpened class might be unloaded if there is no class loader
2322 // contraint in place.
2323 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2324 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2325
2326 #ifndef PRODUCT
2327 if (C->print_intrinsics() || C->print_inlining()) {
2328 tty->print(" from base type: "); adr_type->dump();
2329 tty->print(" sharpened value: "); tjp->dump();
2330 }
2331 #endif
2332 // Sharpen the value type.
2333 return tjp;
2334 }
2335 return NULL;
2336 }
2337
2338 bool LibraryCallKit::inline_unsafe_access(const bool is_native_ptr, bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2339 if (callee()->is_static()) return false; // caller must have the capability!
2340 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2341 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2342
2343 #ifndef PRODUCT
2344 {
2345 ResourceMark rm;
2346 // Check the signatures.
2347 ciSignature* sig = callee()->signature();
2348 #ifdef ASSERT
2349 if (!is_store) {
2350 // Object getObject(Object base, int/long offset), etc.
2351 BasicType rtype = sig->return_type()->basic_type();
2352 if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
2353 rtype = T_ADDRESS; // it is really a C void*
2354 assert(rtype == type, "getter must return the expected value");
2355 if (!is_native_ptr) {
2356 assert(sig->count() == 2, "oop getter has 2 arguments");
2357 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2358 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2359 } else {
2360 assert(sig->count() == 1, "native getter has 1 argument");
2361 assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
2362 }
2363 } else {
2364 // void putObject(Object base, int/long offset, Object x), etc.
2365 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2366 if (!is_native_ptr) {
2367 assert(sig->count() == 3, "oop putter has 3 arguments");
2368 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2369 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2370 } else {
2371 assert(sig->count() == 2, "native putter has 2 arguments");
2372 assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
2373 }
2374 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2375 if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
2376 vtype = T_ADDRESS; // it is really a C void*
2377 assert(vtype == type, "putter must accept the expected value");
2378 }
2379 #endif // ASSERT
2380 }
2381 #endif //PRODUCT
2382
2383 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2384
2385 Node* receiver = argument(0); // type: oop
2386
2387 // Build address expression.
2388 Node* adr;
2389 Node* heap_base_oop = top();
2390 Node* offset = top();
2391 Node* val;
2392
2393 if (!is_native_ptr) {
2394 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2395 Node* base = argument(1); // type: oop
2396 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2397 offset = argument(2); // type: long
2398 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2399 // to be plain byte offsets, which are also the same as those accepted
2400 // by oopDesc::field_base.
2401 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2402 "fieldOffset must be byte-scaled");
2403 // 32-bit machines ignore the high half!
2404 offset = ConvL2X(offset);
2405 adr = make_unsafe_address(base, offset);
2406 heap_base_oop = base;
2407 val = is_store ? argument(4) : NULL;
2408 } else {
2409 Node* ptr = argument(1); // type: long
2410 ptr = ConvL2X(ptr); // adjust Java long to machine word
2411 adr = make_unsafe_address(NULL, ptr);
2412 val = is_store ? argument(3) : NULL;
2413 }
2414
2415 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2416
2417 // First guess at the value type.
2418 const Type *value_type = Type::get_const_basic_type(type);
2419
2420 // Try to categorize the address. If it comes up as TypeJavaPtr::BOTTOM,
2421 // there was not enough information to nail it down.
2422 Compile::AliasType* alias_type = C->alias_type(adr_type);
2423 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2424
2425 // We will need memory barriers unless we can determine a unique
2426 // alias category for this reference. (Note: If for some reason
2427 // the barriers get omitted and the unsafe reference begins to "pollute"
2428 // the alias analysis of the rest of the graph, either Compile::can_alias
2429 // or Compile::must_alias will throw a diagnostic assert.)
2430 bool need_mem_bar;
2431 switch (kind) {
2432 case Relaxed:
2433 need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2434 break;
2435 case Opaque:
2436 // Opaque uses CPUOrder membars for protection against code movement.
2437 case Acquire:
2438 case Release:
2439 case Volatile:
2440 need_mem_bar = true;
2441 break;
2442 default:
2443 ShouldNotReachHere();
2444 }
2445
2446 // Some accesses require access atomicity for all types, notably longs and doubles.
2447 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2448 bool requires_atomic_access = false;
2449 switch (kind) {
2450 case Relaxed:
2451 case Opaque:
2452 requires_atomic_access = AlwaysAtomicAccesses;
2453 break;
2454 case Acquire:
2455 case Release:
2456 case Volatile:
2457 requires_atomic_access = true;
2458 break;
2459 default:
2460 ShouldNotReachHere();
2461 }
2462
2463 // Figure out the memory ordering.
2464 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2465 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2466
2467 // If we are reading the value of the referent field of a Reference
2468 // object (either by using Unsafe directly or through reflection)
2469 // then, if G1 is enabled, we need to record the referent in an
2470 // SATB log buffer using the pre-barrier mechanism.
2471 // Also we need to add memory barrier to prevent commoning reads
2472 // from this field across safepoint since GC can change its value.
2473 bool need_read_barrier = !is_native_ptr && !is_store &&
2474 offset != top() && heap_base_oop != top();
2475
2476 if (!is_store && type == T_OBJECT) {
2477 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
2478 if (tjp != NULL) {
2479 value_type = tjp;
2480 }
2481 }
2482
2483 receiver = null_check(receiver);
2484 if (stopped()) {
2485 return true;
2486 }
2487 // Heap pointers get a null-check from the interpreter,
2488 // as a courtesy. However, this is not guaranteed by Unsafe,
2489 // and it is not possible to fully distinguish unintended nulls
2490 // from intended ones in this API.
2491
2492 // We need to emit leading and trailing CPU membars (see below) in
2493 // addition to memory membars for special access modes. This is a little
2494 // too strong, but avoids the need to insert per-alias-type
2495 // volatile membars (for stores; compare Parse::do_put_xxx), which
2496 // we cannot do effectively here because we probably only have a
2497 // rough approximation of type.
2498
2499 switch(kind) {
2500 case Relaxed:
2501 case Opaque:
2502 case Acquire:
2503 break;
2504 case Release:
2505 case Volatile:
2506 if (is_store) {
2507 insert_mem_bar(Op_MemBarRelease);
2508 } else {
2509 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2510 insert_mem_bar(Op_MemBarVolatile);
2511 }
2512 }
2513 break;
2514 default:
2515 ShouldNotReachHere();
2516 }
2517
2518 // Memory barrier to prevent normal and 'unsafe' accesses from
2519 // bypassing each other. Happens after null checks, so the
2520 // exception paths do not take memory state from the memory barrier,
2521 // so there's no problems making a strong assert about mixing users
2522 // of safe & unsafe memory.
2523 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2524
2525 assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2526 alias_type->field() != NULL || alias_type->element() != NULL, "field, array element or unknown");
2527 bool mismatched = false;
2528 if (alias_type->element() != NULL || alias_type->field() != NULL) {
2529 BasicType bt;
2530 if (alias_type->element() != NULL) {
2531 // Use address type to get the element type. Alias type doesn't provide
2532 // enough information (e.g., doesn't differentiate between byte[] and boolean[]).
2533 const Type* element = adr_type->is_aryptr()->elem();
2534 bt = element->isa_narrowoop() ? T_OBJECT : element->array_element_basic_type();
2535 } else {
2536 bt = alias_type->field()->layout_type();
2537 }
2538 if (bt == T_ARRAY) {
2539 // accessing an array field with getObject is not a mismatch
2540 bt = T_OBJECT;
2541 }
2542 if (bt != type) {
2543 mismatched = true;
2544 }
2545 }
2546 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2547
2548 if (!is_store) {
2549 Node* p = NULL;
2550 // Try to constant fold a load from a constant field
2551 ciField* field = alias_type->field();
2552 if (heap_base_oop != top() &&
2553 field != NULL && field->is_constant() && !mismatched) {
2554 // final or stable field
2555 const Type* con_type = Type::make_constant(alias_type->field(), heap_base_oop);
2556 if (con_type != NULL) {
2557 p = makecon(con_type);
2558 }
2559 }
2560 if (p == NULL) {
2561 // To be valid, unsafe loads may depend on other conditions than
2562 // the one that guards them: pin the Load node
2563 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2564 // load value
2565 switch (type) {
2566 case T_BOOLEAN:
2567 case T_CHAR:
2568 case T_BYTE:
2569 case T_SHORT:
2570 case T_INT:
2571 case T_LONG:
2572 case T_FLOAT:
2573 case T_DOUBLE:
2574 break;
2575 case T_OBJECT:
2576 if (need_read_barrier) {
2577 // We do not require a mem bar inside pre_barrier if need_mem_bar
2578 // is set: the barriers would be emitted by us.
2579 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2580 }
2581 break;
2582 case T_ADDRESS:
2583 // Cast to an int type.
2584 p = _gvn.transform(new CastP2XNode(NULL, p));
2585 p = ConvX2UL(p);
2586 break;
2587 default:
2588 fatal("unexpected type %d: %s", type, type2name(type));
2589 break;
2590 }
2591 }
2592 // The load node has the control of the preceding MemBarCPUOrder. All
2593 // following nodes will have the control of the MemBarCPUOrder inserted at
2594 // the end of this method. So, pushing the load onto the stack at a later
2595 // point is fine.
2596 set_result(p);
2597 } else {
2598 // place effect of store into memory
2599 switch (type) {
2600 case T_DOUBLE:
2601 val = dstore_rounding(val);
2602 break;
2603 case T_ADDRESS:
2604 // Repackage the long as a pointer.
2605 val = ConvL2X(val);
2606 val = _gvn.transform(new CastX2PNode(val));
2607 break;
2608 }
2609
2610 if (type != T_OBJECT) {
2611 (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2612 } else {
2613 // Possibly an oop being stored to Java heap or native memory
2614 if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2615 // oop to Java heap.
2616 (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2617 } else {
2618 // We can't tell at compile time if we are storing in the Java heap or outside
2619 // of it. So we need to emit code to conditionally do the proper type of
2620 // store.
2621
2622 IdealKit ideal(this);
2623 #define __ ideal.
2624 // QQQ who knows what probability is here??
2625 __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2626 // Sync IdealKit and graphKit.
2627 sync_kit(ideal);
2628 Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2629 // Update IdealKit memory.
2630 __ sync_kit(this);
2631 } __ else_(); {
2632 __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched);
2633 } __ end_if();
2634 // Final sync IdealKit and GraphKit.
2635 final_sync(ideal);
2636 #undef __
2637 }
2638 }
2639 }
2640
2641 switch(kind) {
2642 case Relaxed:
2643 case Opaque:
2644 case Release:
2645 break;
2646 case Acquire:
2647 case Volatile:
2648 if (!is_store) {
2649 insert_mem_bar(Op_MemBarAcquire);
2650 } else {
2651 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2652 insert_mem_bar(Op_MemBarVolatile);
2653 }
2654 }
2655 break;
2656 default:
2657 ShouldNotReachHere();
2658 }
2659
2660 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2661
2662 return true;
2663 }
2664
2665 //----------------------------inline_unsafe_load_store----------------------------
2666 // This method serves a couple of different customers (depending on LoadStoreKind):
2667 //
2668 // LS_cmp_swap:
2669 //
2670 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2671 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2672 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2673 //
2674 // LS_cmp_swap_weak:
2675 //
2676 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2677 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2678 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2679 //
2680 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2681 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2682 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2683 //
2684 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2685 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2686 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2687 //
2688 // LS_cmp_exchange:
2689 //
2690 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2691 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2692 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2693 //
2694 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2695 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2696 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2697 //
2698 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2699 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2700 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2701 //
2702 // LS_get_add:
2703 //
2704 // int getAndAddInt( Object o, long offset, int delta)
2705 // long getAndAddLong(Object o, long offset, long delta)
2706 //
2707 // LS_get_set:
2708 //
2709 // int getAndSet(Object o, long offset, int newValue)
2710 // long getAndSet(Object o, long offset, long newValue)
2711 // Object getAndSet(Object o, long offset, Object newValue)
2712 //
2713 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2714 // This basic scheme here is the same as inline_unsafe_access, but
2715 // differs in enough details that combining them would make the code
2716 // overly confusing. (This is a true fact! I originally combined
2717 // them, but even I was confused by it!) As much code/comments as
2718 // possible are retained from inline_unsafe_access though to make
2719 // the correspondences clearer. - dl
2720
2721 if (callee()->is_static()) return false; // caller must have the capability!
2722
2723 #ifndef PRODUCT
2724 BasicType rtype;
2725 {
2726 ResourceMark rm;
2727 // Check the signatures.
2728 ciSignature* sig = callee()->signature();
2729 rtype = sig->return_type()->basic_type();
2730 switch(kind) {
2731 case LS_get_add:
2732 case LS_get_set: {
2733 // Check the signatures.
2734 #ifdef ASSERT
2735 assert(rtype == type, "get and set must return the expected type");
2736 assert(sig->count() == 3, "get and set has 3 arguments");
2737 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2738 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2739 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2740 #endif // ASSERT
2741 break;
2742 }
2743 case LS_cmp_swap:
2744 case LS_cmp_swap_weak: {
2745 // Check the signatures.
2746 #ifdef ASSERT
2747 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2748 assert(sig->count() == 4, "CAS has 4 arguments");
2749 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2750 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2751 #endif // ASSERT
2752 break;
2753 }
2754 case LS_cmp_exchange: {
2755 // Check the signatures.
2756 #ifdef ASSERT
2757 assert(rtype == type, "CAS must return the expected type");
2758 assert(sig->count() == 4, "CAS has 4 arguments");
2759 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2760 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2761 #endif // ASSERT
2762 break;
2763 }
2764 default:
2765 ShouldNotReachHere();
2766 }
2767 }
2768 #endif //PRODUCT
2769
2770 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2771
2772 // Get arguments:
2773 Node* receiver = NULL;
2774 Node* base = NULL;
2775 Node* offset = NULL;
2776 Node* oldval = NULL;
2777 Node* newval = NULL;
2778 switch(kind) {
2779 case LS_cmp_swap:
2780 case LS_cmp_swap_weak:
2781 case LS_cmp_exchange: {
2782 const bool two_slot_type = type2size[type] == 2;
2783 receiver = argument(0); // type: oop
2784 base = argument(1); // type: oop
2785 offset = argument(2); // type: long
2786 oldval = argument(4); // type: oop, int, or long
2787 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2788 break;
2789 }
2790 case LS_get_add:
2791 case LS_get_set: {
2792 receiver = argument(0); // type: oop
2793 base = argument(1); // type: oop
2794 offset = argument(2); // type: long
2795 oldval = NULL;
2796 newval = argument(4); // type: oop, int, or long
2797 break;
2798 }
2799 default:
2800 ShouldNotReachHere();
2801 }
2802
2803 // Null check receiver.
2804 receiver = null_check(receiver);
2805 if (stopped()) {
2806 return true;
2807 }
2808
2809 // Build field offset expression.
2810 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2811 // to be plain byte offsets, which are also the same as those accepted
2812 // by oopDesc::field_base.
2813 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2814 // 32-bit machines ignore the high half of long offsets
2815 offset = ConvL2X(offset);
2816 Node* adr = make_unsafe_address(base, offset);
2817 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2818
2819 // For CAS, unlike inline_unsafe_access, there seems no point in
2820 // trying to refine types. Just use the coarse types here.
2821 const Type *value_type = Type::get_const_basic_type(type);
2822 Compile::AliasType* alias_type = C->alias_type(adr_type);
2823 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2824
2825 switch (kind) {
2826 case LS_get_set:
2827 case LS_cmp_exchange: {
2828 if (type == T_OBJECT) {
2829 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2830 if (tjp != NULL) {
2831 value_type = tjp;
2832 }
2833 }
2834 break;
2835 }
2836 case LS_cmp_swap:
2837 case LS_cmp_swap_weak:
2838 case LS_get_add:
2839 break;
2840 default:
2841 ShouldNotReachHere();
2842 }
2843
2844 int alias_idx = C->get_alias_index(adr_type);
2845
2846 // Memory-model-wise, a LoadStore acts like a little synchronized
2847 // block, so needs barriers on each side. These don't translate
2848 // into actual barriers on most machines, but we still need rest of
2849 // compiler to respect ordering.
2850
2851 switch (access_kind) {
2852 case Relaxed:
2853 case Acquire:
2854 break;
2855 case Release:
2856 case Volatile:
2857 insert_mem_bar(Op_MemBarRelease);
2858 break;
2859 default:
2860 ShouldNotReachHere();
2861 }
2862 insert_mem_bar(Op_MemBarCPUOrder);
2863
2864 // Figure out the memory ordering.
2865 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2866
2867 // 4984716: MemBars must be inserted before this
2868 // memory node in order to avoid a false
2869 // dependency which will confuse the scheduler.
2870 Node *mem = memory(alias_idx);
2871
2872 // For now, we handle only those cases that actually exist: ints,
2873 // longs, and Object. Adding others should be straightforward.
2874 Node* load_store = NULL;
2875 switch(type) {
2876 case T_INT:
2877 switch(kind) {
2878 case LS_get_add:
2879 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2880 break;
2881 case LS_get_set:
2882 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2883 break;
2884 case LS_cmp_swap_weak:
2885 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2886 break;
2887 case LS_cmp_swap:
2888 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2889 break;
2890 case LS_cmp_exchange:
2891 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2892 break;
2893 default:
2894 ShouldNotReachHere();
2895 }
2896 break;
2897 case T_LONG:
2898 switch(kind) {
2899 case LS_get_add:
2900 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2901 break;
2902 case LS_get_set:
2903 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2904 break;
2905 case LS_cmp_swap_weak:
2906 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2907 break;
2908 case LS_cmp_swap:
2909 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2910 break;
2911 case LS_cmp_exchange:
2912 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2913 break;
2914 default:
2915 ShouldNotReachHere();
2916 }
2917 break;
2918 case T_OBJECT:
2919 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2920 // could be delayed during Parse (for example, in adjust_map_after_if()).
2921 // Execute transformation here to avoid barrier generation in such case.
2922 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2923 newval = _gvn.makecon(TypePtr::NULL_PTR);
2924
2925 // Reference stores need a store barrier.
2926 switch(kind) {
2927 case LS_get_set: {
2928 // If pre-barrier must execute before the oop store, old value will require do_load here.
2929 if (!can_move_pre_barrier()) {
2930 pre_barrier(true /* do_load*/,
2931 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2932 NULL /* pre_val*/,
2933 T_OBJECT);
2934 } // Else move pre_barrier to use load_store value, see below.
2935 break;
2936 }
2937 case LS_cmp_swap_weak:
2938 case LS_cmp_swap:
2939 case LS_cmp_exchange: {
2940 // Same as for newval above:
2941 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2942 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2943 }
2944 // The only known value which might get overwritten is oldval.
2945 pre_barrier(false /* do_load */,
2946 control(), NULL, NULL, max_juint, NULL, NULL,
2947 oldval /* pre_val */,
2948 T_OBJECT);
2949 break;
2950 }
2951 default:
2952 ShouldNotReachHere();
2953 }
2954
2955 #ifdef _LP64
2956 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2957 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2958
2959 switch(kind) {
2960 case LS_get_set:
2961 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2962 break;
2963 case LS_cmp_swap_weak: {
2964 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2965 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2966 break;
2967 }
2968 case LS_cmp_swap: {
2969 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2970 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2971 break;
2972 }
2973 case LS_cmp_exchange: {
2974 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2975 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2976 break;
2977 }
2978 default:
2979 ShouldNotReachHere();
2980 }
2981 } else
2982 #endif
2983 switch (kind) {
2984 case LS_get_set:
2985 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2986 break;
2987 case LS_cmp_swap_weak:
2988 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2989 break;
2990 case LS_cmp_swap:
2991 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2992 break;
2993 case LS_cmp_exchange:
2994 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
2995 break;
2996 default:
2997 ShouldNotReachHere();
2998 }
2999
3000 // Emit the post barrier only when the actual store happened. This makes sense
3001 // to check only for LS_cmp_* that can fail to set the value.
3002 // LS_cmp_exchange does not produce any branches by default, so there is no
3003 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
3004 // CompareAndExchange and move branches here, it would make sense to conditionalize
3005 // post_barriers for LS_cmp_exchange as well.
3006 //
3007 // CAS success path is marked more likely since we anticipate this is a performance
3008 // critical path, while CAS failure path can use the penalty for going through unlikely
3009 // path as backoff. Which is still better than doing a store barrier there.
3010 switch (kind) {
3011 case LS_get_set:
3012 case LS_cmp_exchange: {
3013 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3014 break;
3015 }
3016 case LS_cmp_swap_weak:
3017 case LS_cmp_swap: {
3018 IdealKit ideal(this);
3019 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3020 sync_kit(ideal);
3021 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3022 ideal.sync_kit(this);
3023 } ideal.end_if();
3024 final_sync(ideal);
3025 break;
3026 }
3027 default:
3028 ShouldNotReachHere();
3029 }
3030 break;
3031 default:
3032 fatal("unexpected type %d: %s", type, type2name(type));
3033 break;
3034 }
3035
3036 // SCMemProjNodes represent the memory state of a LoadStore. Their
3037 // main role is to prevent LoadStore nodes from being optimized away
3038 // when their results aren't used.
3039 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3040 set_memory(proj, alias_idx);
3041
3042 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3043 #ifdef _LP64
3044 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3045 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3046 }
3047 #endif
3048 if (can_move_pre_barrier()) {
3049 // Don't need to load pre_val. The old value is returned by load_store.
3050 // The pre_barrier can execute after the xchg as long as no safepoint
3051 // gets inserted between them.
3052 pre_barrier(false /* do_load */,
3053 control(), NULL, NULL, max_juint, NULL, NULL,
3054 load_store /* pre_val */,
3055 T_OBJECT);
3056 }
3057 }
3058
3059 // Add the trailing membar surrounding the access
3060 insert_mem_bar(Op_MemBarCPUOrder);
3061
3062 switch (access_kind) {
3063 case Relaxed:
3064 case Release:
3065 break; // do nothing
3066 case Acquire:
3067 case Volatile:
3068 insert_mem_bar(Op_MemBarAcquire);
3069 break;
3070 default:
3071 ShouldNotReachHere();
3072 }
3073
3074 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3075 set_result(load_store);
3076 return true;
3077 }
3078
3079 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
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 = is_store ? MemNode::release : MemNode::acquire; break;
3087 default:
3088 ShouldNotReachHere();
3089 }
3090 guarantee(mo != MemNode::unset, "Should select memory ordering");
3091 return mo;
3092 }
3093
3094 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3095 MemNode::MemOrd mo = MemNode::unset;
3096 switch(kind) {
3097 case Opaque:
3098 case Relaxed: mo = MemNode::unordered; break;
3099 case Acquire: mo = MemNode::acquire; break;
3100 case Release: mo = MemNode::release; break;
3101 case Volatile: mo = MemNode::seqcst; break;
3102 default:
3103 ShouldNotReachHere();
3104 }
3105 guarantee(mo != MemNode::unset, "Should select memory ordering");
3106 return mo;
3107 }
3108
3109 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3110 // Regardless of form, don't allow previous ld/st to move down,
3111 // then issue acquire, release, or volatile mem_bar.
3112 insert_mem_bar(Op_MemBarCPUOrder);
3113 switch(id) {
3114 case vmIntrinsics::_loadFence:
3115 insert_mem_bar(Op_LoadFence);
3116 return true;
3117 case vmIntrinsics::_storeFence:
3118 insert_mem_bar(Op_StoreFence);
3119 return true;
3120 case vmIntrinsics::_fullFence:
3121 insert_mem_bar(Op_MemBarVolatile);
3122 return true;
3123 default:
3124 fatal_unexpected_iid(id);
3125 return false;
3126 }
3127 }
3128
3129 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3130 if (!kls->is_Con()) {
3131 return true;
3132 }
3133 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3134 if (klsptr == NULL) {
3135 return true;
3136 }
3137 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3138 // don't need a guard for a klass that is already initialized
3139 return !ik->is_initialized();
3140 }
3141
3142 //----------------------------inline_unsafe_allocate---------------------------
3143 // public native Object Unsafe.allocateInstance(Class<?> cls);
3144 bool LibraryCallKit::inline_unsafe_allocate() {
3145 if (callee()->is_static()) return false; // caller must have the capability!
3146
3147 null_check_receiver(); // null-check, then ignore
3148 Node* cls = null_check(argument(1));
3149 if (stopped()) return true;
3150
3151 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3152 kls = null_check(kls);
3153 if (stopped()) return true; // argument was like int.class
3154
3155 Node* test = NULL;
3156 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3157 // Note: The argument might still be an illegal value like
3158 // Serializable.class or Object[].class. The runtime will handle it.
3159 // But we must make an explicit check for initialization.
3160 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3161 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3162 // can generate code to load it as unsigned byte.
3163 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3164 Node* bits = intcon(InstanceKlass::fully_initialized);
3165 test = _gvn.transform(new SubINode(inst, bits));
3166 // The 'test' is non-zero if we need to take a slow path.
3167 }
3168
3169 Node* obj = new_instance(kls, test);
3170 set_result(obj);
3171 return true;
3172 }
3173
3174 //------------------------inline_native_time_funcs--------------
3175 // inline code for System.currentTimeMillis() and System.nanoTime()
3176 // these have the same type and signature
3177 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3178 const TypeFunc* tf = OptoRuntime::void_long_Type();
3179 const TypePtr* no_memory_effects = NULL;
3180 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3181 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3182 #ifdef ASSERT
3183 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3184 assert(value_top == top(), "second value must be top");
3185 #endif
3186 set_result(value);
3187 return true;
3188 }
3189
3190 //------------------------inline_native_currentThread------------------
3191 bool LibraryCallKit::inline_native_currentThread() {
3192 Node* junk = NULL;
3193 set_result(generate_current_thread(junk));
3194 return true;
3195 }
3196
3197 //------------------------inline_native_isInterrupted------------------
3198 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3199 bool LibraryCallKit::inline_native_isInterrupted() {
3200 // Add a fast path to t.isInterrupted(clear_int):
3201 // (t == Thread.current() &&
3202 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3203 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3204 // So, in the common case that the interrupt bit is false,
3205 // we avoid making a call into the VM. Even if the interrupt bit
3206 // is true, if the clear_int argument is false, we avoid the VM call.
3207 // However, if the receiver is not currentThread, we must call the VM,
3208 // because there must be some locking done around the operation.
3209
3210 // We only go to the fast case code if we pass two guards.
3211 // Paths which do not pass are accumulated in the slow_region.
3212
3213 enum {
3214 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3215 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3216 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3217 PATH_LIMIT
3218 };
3219
3220 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3221 // out of the function.
3222 insert_mem_bar(Op_MemBarCPUOrder);
3223
3224 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3225 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3226
3227 RegionNode* slow_region = new RegionNode(1);
3228 record_for_igvn(slow_region);
3229
3230 // (a) Receiving thread must be the current thread.
3231 Node* rec_thr = argument(0);
3232 Node* tls_ptr = NULL;
3233 Node* cur_thr = generate_current_thread(tls_ptr);
3234 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3235 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3236
3237 generate_slow_guard(bol_thr, slow_region);
3238
3239 // (b) Interrupt bit on TLS must be false.
3240 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3241 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3242 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3243
3244 // Set the control input on the field _interrupted read to prevent it floating up.
3245 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3246 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3247 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3248
3249 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3250
3251 // First fast path: if (!TLS._interrupted) return false;
3252 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3253 result_rgn->init_req(no_int_result_path, false_bit);
3254 result_val->init_req(no_int_result_path, intcon(0));
3255
3256 // drop through to next case
3257 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3258
3259 #ifndef TARGET_OS_FAMILY_windows
3260 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3261 Node* clr_arg = argument(1);
3262 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3263 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3264 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3265
3266 // Second fast path: ... else if (!clear_int) return true;
3267 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3268 result_rgn->init_req(no_clear_result_path, false_arg);
3269 result_val->init_req(no_clear_result_path, intcon(1));
3270
3271 // drop through to next case
3272 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3273 #else
3274 // To return true on Windows you must read the _interrupted field
3275 // and check the event state i.e. take the slow path.
3276 #endif // TARGET_OS_FAMILY_windows
3277
3278 // (d) Otherwise, go to the slow path.
3279 slow_region->add_req(control());
3280 set_control( _gvn.transform(slow_region));
3281
3282 if (stopped()) {
3283 // There is no slow path.
3284 result_rgn->init_req(slow_result_path, top());
3285 result_val->init_req(slow_result_path, top());
3286 } else {
3287 // non-virtual because it is a private non-static
3288 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3289
3290 Node* slow_val = set_results_for_java_call(slow_call);
3291 // this->control() comes from set_results_for_java_call
3292
3293 Node* fast_io = slow_call->in(TypeFunc::I_O);
3294 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3295
3296 // These two phis are pre-filled with copies of of the fast IO and Memory
3297 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3298 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3299
3300 result_rgn->init_req(slow_result_path, control());
3301 result_io ->init_req(slow_result_path, i_o());
3302 result_mem->init_req(slow_result_path, reset_memory());
3303 result_val->init_req(slow_result_path, slow_val);
3304
3305 set_all_memory(_gvn.transform(result_mem));
3306 set_i_o( _gvn.transform(result_io));
3307 }
3308
3309 C->set_has_split_ifs(true); // Has chance for split-if optimization
3310 set_result(result_rgn, result_val);
3311 return true;
3312 }
3313
3314 //---------------------------load_mirror_from_klass----------------------------
3315 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3316 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3317 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3318 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3319 }
3320
3321 //-----------------------load_klass_from_mirror_common-------------------------
3322 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3323 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3324 // and branch to the given path on the region.
3325 // If never_see_null, take an uncommon trap on null, so we can optimistically
3326 // compile for the non-null case.
3327 // If the region is NULL, force never_see_null = true.
3328 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3329 bool never_see_null,
3330 RegionNode* region,
3331 int null_path,
3332 int offset) {
3333 if (region == NULL) never_see_null = true;
3334 Node* p = basic_plus_adr(mirror, offset);
3335 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3336 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3337 Node* null_ctl = top();
3338 kls = null_check_oop(kls, &null_ctl, never_see_null);
3339 if (region != NULL) {
3340 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3341 region->init_req(null_path, null_ctl);
3342 } else {
3343 assert(null_ctl == top(), "no loose ends");
3344 }
3345 return kls;
3346 }
3347
3348 //--------------------(inline_native_Class_query helpers)---------------------
3349 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3350 // Fall through if (mods & mask) == bits, take the guard otherwise.
3351 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3352 // Branch around if the given klass has the given modifier bit set.
3353 // Like generate_guard, adds a new path onto the region.
3354 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3355 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3356 Node* mask = intcon(modifier_mask);
3357 Node* bits = intcon(modifier_bits);
3358 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3359 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3360 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3361 return generate_fair_guard(bol, region);
3362 }
3363 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3364 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3365 }
3366
3367 //-------------------------inline_native_Class_query-------------------
3368 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3369 const Type* return_type = TypeInt::BOOL;
3370 Node* prim_return_value = top(); // what happens if it's a primitive class?
3371 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3372 bool expect_prim = false; // most of these guys expect to work on refs
3373
3374 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3375
3376 Node* mirror = argument(0);
3377 Node* obj = top();
3378
3379 switch (id) {
3380 case vmIntrinsics::_isInstance:
3381 // nothing is an instance of a primitive type
3382 prim_return_value = intcon(0);
3383 obj = argument(1);
3384 break;
3385 case vmIntrinsics::_getModifiers:
3386 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3387 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3388 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3389 break;
3390 case vmIntrinsics::_isInterface:
3391 prim_return_value = intcon(0);
3392 break;
3393 case vmIntrinsics::_isArray:
3394 prim_return_value = intcon(0);
3395 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3396 break;
3397 case vmIntrinsics::_isPrimitive:
3398 prim_return_value = intcon(1);
3399 expect_prim = true; // obviously
3400 break;
3401 case vmIntrinsics::_getSuperclass:
3402 prim_return_value = null();
3403 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3404 break;
3405 case vmIntrinsics::_getClassAccessFlags:
3406 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3407 return_type = TypeInt::INT; // not bool! 6297094
3408 break;
3409 default:
3410 fatal_unexpected_iid(id);
3411 break;
3412 }
3413
3414 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3415 if (mirror_con == NULL) return false; // cannot happen?
3416
3417 #ifndef PRODUCT
3418 if (C->print_intrinsics() || C->print_inlining()) {
3419 ciType* k = mirror_con->java_mirror_type();
3420 if (k) {
3421 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3422 k->print_name();
3423 tty->cr();
3424 }
3425 }
3426 #endif
3427
3428 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3429 RegionNode* region = new RegionNode(PATH_LIMIT);
3430 record_for_igvn(region);
3431 PhiNode* phi = new PhiNode(region, return_type);
3432
3433 // The mirror will never be null of Reflection.getClassAccessFlags, however
3434 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3435 // if it is. See bug 4774291.
3436
3437 // For Reflection.getClassAccessFlags(), the null check occurs in
3438 // the wrong place; see inline_unsafe_access(), above, for a similar
3439 // situation.
3440 mirror = null_check(mirror);
3441 // If mirror or obj is dead, only null-path is taken.
3442 if (stopped()) return true;
3443
3444 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3445
3446 // Now load the mirror's klass metaobject, and null-check it.
3447 // Side-effects region with the control path if the klass is null.
3448 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3449 // If kls is null, we have a primitive mirror.
3450 phi->init_req(_prim_path, prim_return_value);
3451 if (stopped()) { set_result(region, phi); return true; }
3452 bool safe_for_replace = (region->in(_prim_path) == top());
3453
3454 Node* p; // handy temp
3455 Node* null_ctl;
3456
3457 // Now that we have the non-null klass, we can perform the real query.
3458 // For constant classes, the query will constant-fold in LoadNode::Value.
3459 Node* query_value = top();
3460 switch (id) {
3461 case vmIntrinsics::_isInstance:
3462 // nothing is an instance of a primitive type
3463 query_value = gen_instanceof(obj, kls, safe_for_replace);
3464 break;
3465
3466 case vmIntrinsics::_getModifiers:
3467 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3468 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3469 break;
3470
3471 case vmIntrinsics::_isInterface:
3472 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3473 if (generate_interface_guard(kls, region) != NULL)
3474 // A guard was added. If the guard is taken, it was an interface.
3475 phi->add_req(intcon(1));
3476 // If we fall through, it's a plain class.
3477 query_value = intcon(0);
3478 break;
3479
3480 case vmIntrinsics::_isArray:
3481 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3482 if (generate_array_guard(kls, region) != NULL)
3483 // A guard was added. If the guard is taken, it was an array.
3484 phi->add_req(intcon(1));
3485 // If we fall through, it's a plain class.
3486 query_value = intcon(0);
3487 break;
3488
3489 case vmIntrinsics::_isPrimitive:
3490 query_value = intcon(0); // "normal" path produces false
3491 break;
3492
3493 case vmIntrinsics::_getSuperclass:
3494 // The rules here are somewhat unfortunate, but we can still do better
3495 // with random logic than with a JNI call.
3496 // Interfaces store null or Object as _super, but must report null.
3497 // Arrays store an intermediate super as _super, but must report Object.
3498 // Other types can report the actual _super.
3499 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3500 if (generate_interface_guard(kls, region) != NULL)
3501 // A guard was added. If the guard is taken, it was an interface.
3502 phi->add_req(null());
3503 if (generate_array_guard(kls, region) != NULL)
3504 // A guard was added. If the guard is taken, it was an array.
3505 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3506 // If we fall through, it's a plain class. Get its _super.
3507 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3508 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3509 null_ctl = top();
3510 kls = null_check_oop(kls, &null_ctl);
3511 if (null_ctl != top()) {
3512 // If the guard is taken, Object.superClass is null (both klass and mirror).
3513 region->add_req(null_ctl);
3514 phi ->add_req(null());
3515 }
3516 if (!stopped()) {
3517 query_value = load_mirror_from_klass(kls);
3518 }
3519 break;
3520
3521 case vmIntrinsics::_getClassAccessFlags:
3522 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3523 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3524 break;
3525
3526 default:
3527 fatal_unexpected_iid(id);
3528 break;
3529 }
3530
3531 // Fall-through is the normal case of a query to a real class.
3532 phi->init_req(1, query_value);
3533 region->init_req(1, control());
3534
3535 C->set_has_split_ifs(true); // Has chance for split-if optimization
3536 set_result(region, phi);
3537 return true;
3538 }
3539
3540 //-------------------------inline_Class_cast-------------------
3541 bool LibraryCallKit::inline_Class_cast() {
3542 Node* mirror = argument(0); // Class
3543 Node* obj = argument(1);
3544 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3545 if (mirror_con == NULL) {
3546 return false; // dead path (mirror->is_top()).
3547 }
3548 if (obj == NULL || obj->is_top()) {
3549 return false; // dead path
3550 }
3551 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3552
3553 // First, see if Class.cast() can be folded statically.
3554 // java_mirror_type() returns non-null for compile-time Class constants.
3555 ciType* tm = mirror_con->java_mirror_type();
3556 if (tm != NULL && tm->is_klass() &&
3557 tp != NULL && tp->klass() != NULL) {
3558 if (!tp->klass()->is_loaded()) {
3559 // Don't use intrinsic when class is not loaded.
3560 return false;
3561 } else {
3562 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3563 if (static_res == Compile::SSC_always_true) {
3564 // isInstance() is true - fold the code.
3565 set_result(obj);
3566 return true;
3567 } else if (static_res == Compile::SSC_always_false) {
3568 // Don't use intrinsic, have to throw ClassCastException.
3569 // If the reference is null, the non-intrinsic bytecode will
3570 // be optimized appropriately.
3571 return false;
3572 }
3573 }
3574 }
3575
3576 // Bailout intrinsic and do normal inlining if exception path is frequent.
3577 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3578 return false;
3579 }
3580
3581 // Generate dynamic checks.
3582 // Class.cast() is java implementation of _checkcast bytecode.
3583 // Do checkcast (Parse::do_checkcast()) optimizations here.
3584
3585 mirror = null_check(mirror);
3586 // If mirror is dead, only null-path is taken.
3587 if (stopped()) {
3588 return true;
3589 }
3590
3591 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3592 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3593 RegionNode* region = new RegionNode(PATH_LIMIT);
3594 record_for_igvn(region);
3595
3596 // Now load the mirror's klass metaobject, and null-check it.
3597 // If kls is null, we have a primitive mirror and
3598 // nothing is an instance of a primitive type.
3599 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3600
3601 Node* res = top();
3602 if (!stopped()) {
3603 Node* bad_type_ctrl = top();
3604 // Do checkcast optimizations.
3605 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3606 region->init_req(_bad_type_path, bad_type_ctrl);
3607 }
3608 if (region->in(_prim_path) != top() ||
3609 region->in(_bad_type_path) != top()) {
3610 // Let Interpreter throw ClassCastException.
3611 PreserveJVMState pjvms(this);
3612 set_control(_gvn.transform(region));
3613 uncommon_trap(Deoptimization::Reason_intrinsic,
3614 Deoptimization::Action_maybe_recompile);
3615 }
3616 if (!stopped()) {
3617 set_result(res);
3618 }
3619 return true;
3620 }
3621
3622
3623 //--------------------------inline_native_subtype_check------------------------
3624 // This intrinsic takes the JNI calls out of the heart of
3625 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3626 bool LibraryCallKit::inline_native_subtype_check() {
3627 // Pull both arguments off the stack.
3628 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3629 args[0] = argument(0);
3630 args[1] = argument(1);
3631 Node* klasses[2]; // corresponding Klasses: superk, subk
3632 klasses[0] = klasses[1] = top();
3633
3634 enum {
3635 // A full decision tree on {superc is prim, subc is prim}:
3636 _prim_0_path = 1, // {P,N} => false
3637 // {P,P} & superc!=subc => false
3638 _prim_same_path, // {P,P} & superc==subc => true
3639 _prim_1_path, // {N,P} => false
3640 _ref_subtype_path, // {N,N} & subtype check wins => true
3641 _both_ref_path, // {N,N} & subtype check loses => false
3642 PATH_LIMIT
3643 };
3644
3645 RegionNode* region = new RegionNode(PATH_LIMIT);
3646 Node* phi = new PhiNode(region, TypeInt::BOOL);
3647 record_for_igvn(region);
3648
3649 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3650 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3651 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3652
3653 // First null-check both mirrors and load each mirror's klass metaobject.
3654 int which_arg;
3655 for (which_arg = 0; which_arg <= 1; which_arg++) {
3656 Node* arg = args[which_arg];
3657 arg = null_check(arg);
3658 if (stopped()) break;
3659 args[which_arg] = arg;
3660
3661 Node* p = basic_plus_adr(arg, class_klass_offset);
3662 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3663 klasses[which_arg] = _gvn.transform(kls);
3664 }
3665
3666 // Having loaded both klasses, test each for null.
3667 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3668 for (which_arg = 0; which_arg <= 1; which_arg++) {
3669 Node* kls = klasses[which_arg];
3670 Node* null_ctl = top();
3671 kls = null_check_oop(kls, &null_ctl, never_see_null);
3672 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3673 region->init_req(prim_path, null_ctl);
3674 if (stopped()) break;
3675 klasses[which_arg] = kls;
3676 }
3677
3678 if (!stopped()) {
3679 // now we have two reference types, in klasses[0..1]
3680 Node* subk = klasses[1]; // the argument to isAssignableFrom
3681 Node* superk = klasses[0]; // the receiver
3682 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3683 // now we have a successful reference subtype check
3684 region->set_req(_ref_subtype_path, control());
3685 }
3686
3687 // If both operands are primitive (both klasses null), then
3688 // we must return true when they are identical primitives.
3689 // It is convenient to test this after the first null klass check.
3690 set_control(region->in(_prim_0_path)); // go back to first null check
3691 if (!stopped()) {
3692 // Since superc is primitive, make a guard for the superc==subc case.
3693 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3694 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3695 generate_guard(bol_eq, region, PROB_FAIR);
3696 if (region->req() == PATH_LIMIT+1) {
3697 // A guard was added. If the added guard is taken, superc==subc.
3698 region->swap_edges(PATH_LIMIT, _prim_same_path);
3699 region->del_req(PATH_LIMIT);
3700 }
3701 region->set_req(_prim_0_path, control()); // Not equal after all.
3702 }
3703
3704 // these are the only paths that produce 'true':
3705 phi->set_req(_prim_same_path, intcon(1));
3706 phi->set_req(_ref_subtype_path, intcon(1));
3707
3708 // pull together the cases:
3709 assert(region->req() == PATH_LIMIT, "sane region");
3710 for (uint i = 1; i < region->req(); i++) {
3711 Node* ctl = region->in(i);
3712 if (ctl == NULL || ctl == top()) {
3713 region->set_req(i, top());
3714 phi ->set_req(i, top());
3715 } else if (phi->in(i) == NULL) {
3716 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3717 }
3718 }
3719
3720 set_control(_gvn.transform(region));
3721 set_result(_gvn.transform(phi));
3722 return true;
3723 }
3724
3725 //---------------------generate_array_guard_common------------------------
3726 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3727 bool obj_array, bool not_array) {
3728
3729 if (stopped()) {
3730 return NULL;
3731 }
3732
3733 // If obj_array/non_array==false/false:
3734 // Branch around if the given klass is in fact an array (either obj or prim).
3735 // If obj_array/non_array==false/true:
3736 // Branch around if the given klass is not an array klass of any kind.
3737 // If obj_array/non_array==true/true:
3738 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3739 // If obj_array/non_array==true/false:
3740 // Branch around if the kls is an oop array (Object[] or subtype)
3741 //
3742 // Like generate_guard, adds a new path onto the region.
3743 jint layout_con = 0;
3744 Node* layout_val = get_layout_helper(kls, layout_con);
3745 if (layout_val == NULL) {
3746 bool query = (obj_array
3747 ? Klass::layout_helper_is_objArray(layout_con)
3748 : Klass::layout_helper_is_array(layout_con));
3749 if (query == not_array) {
3750 return NULL; // never a branch
3751 } else { // always a branch
3752 Node* always_branch = control();
3753 if (region != NULL)
3754 region->add_req(always_branch);
3755 set_control(top());
3756 return always_branch;
3757 }
3758 }
3759 // Now test the correct condition.
3760 jint nval = (obj_array
3761 ? ((jint)Klass::_lh_array_tag_type_value
3762 << Klass::_lh_array_tag_shift)
3763 : Klass::_lh_neutral_value);
3764 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3765 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3766 // invert the test if we are looking for a non-array
3767 if (not_array) btest = BoolTest(btest).negate();
3768 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3769 return generate_fair_guard(bol, region);
3770 }
3771
3772
3773 //-----------------------inline_native_newArray--------------------------
3774 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3775 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3776 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3777 Node* mirror;
3778 Node* count_val;
3779 if (uninitialized) {
3780 mirror = argument(1);
3781 count_val = argument(2);
3782 } else {
3783 mirror = argument(0);
3784 count_val = argument(1);
3785 }
3786
3787 mirror = null_check(mirror);
3788 // If mirror or obj is dead, only null-path is taken.
3789 if (stopped()) return true;
3790
3791 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3792 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3793 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3794 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3795 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3796
3797 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3798 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3799 result_reg, _slow_path);
3800 Node* normal_ctl = control();
3801 Node* no_array_ctl = result_reg->in(_slow_path);
3802
3803 // Generate code for the slow case. We make a call to newArray().
3804 set_control(no_array_ctl);
3805 if (!stopped()) {
3806 // Either the input type is void.class, or else the
3807 // array klass has not yet been cached. Either the
3808 // ensuing call will throw an exception, or else it
3809 // will cache the array klass for next time.
3810 PreserveJVMState pjvms(this);
3811 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3812 Node* slow_result = set_results_for_java_call(slow_call);
3813 // this->control() comes from set_results_for_java_call
3814 result_reg->set_req(_slow_path, control());
3815 result_val->set_req(_slow_path, slow_result);
3816 result_io ->set_req(_slow_path, i_o());
3817 result_mem->set_req(_slow_path, reset_memory());
3818 }
3819
3820 set_control(normal_ctl);
3821 if (!stopped()) {
3822 // Normal case: The array type has been cached in the java.lang.Class.
3823 // The following call works fine even if the array type is polymorphic.
3824 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3825 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3826 result_reg->init_req(_normal_path, control());
3827 result_val->init_req(_normal_path, obj);
3828 result_io ->init_req(_normal_path, i_o());
3829 result_mem->init_req(_normal_path, reset_memory());
3830
3831 if (uninitialized) {
3832 // Mark the allocation so that zeroing is skipped
3833 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3834 alloc->maybe_set_complete(&_gvn);
3835 }
3836 }
3837
3838 // Return the combined state.
3839 set_i_o( _gvn.transform(result_io) );
3840 set_all_memory( _gvn.transform(result_mem));
3841
3842 C->set_has_split_ifs(true); // Has chance for split-if optimization
3843 set_result(result_reg, result_val);
3844 return true;
3845 }
3846
3847 //----------------------inline_native_getLength--------------------------
3848 // public static native int java.lang.reflect.Array.getLength(Object array);
3849 bool LibraryCallKit::inline_native_getLength() {
3850 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3851
3852 Node* array = null_check(argument(0));
3853 // If array is dead, only null-path is taken.
3854 if (stopped()) return true;
3855
3856 // Deoptimize if it is a non-array.
3857 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3858
3859 if (non_array != NULL) {
3860 PreserveJVMState pjvms(this);
3861 set_control(non_array);
3862 uncommon_trap(Deoptimization::Reason_intrinsic,
3863 Deoptimization::Action_maybe_recompile);
3864 }
3865
3866 // If control is dead, only non-array-path is taken.
3867 if (stopped()) return true;
3868
3869 // The works fine even if the array type is polymorphic.
3870 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3871 Node* result = load_array_length(array);
3872
3873 C->set_has_split_ifs(true); // Has chance for split-if optimization
3874 set_result(result);
3875 return true;
3876 }
3877
3878 //------------------------inline_array_copyOf----------------------------
3879 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3880 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3881 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3882 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3883
3884 // Get the arguments.
3885 Node* original = argument(0);
3886 Node* start = is_copyOfRange? argument(1): intcon(0);
3887 Node* end = is_copyOfRange? argument(2): argument(1);
3888 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3889
3890 Node* newcopy = NULL;
3891
3892 // Set the original stack and the reexecute bit for the interpreter to reexecute
3893 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3894 { PreserveReexecuteState preexecs(this);
3895 jvms()->set_should_reexecute(true);
3896
3897 array_type_mirror = null_check(array_type_mirror);
3898 original = null_check(original);
3899
3900 // Check if a null path was taken unconditionally.
3901 if (stopped()) return true;
3902
3903 Node* orig_length = load_array_length(original);
3904
3905 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3906 klass_node = null_check(klass_node);
3907
3908 RegionNode* bailout = new RegionNode(1);
3909 record_for_igvn(bailout);
3910
3911 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3912 // Bail out if that is so.
3913 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3914 if (not_objArray != NULL) {
3915 // Improve the klass node's type from the new optimistic assumption:
3916 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3917 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3918 Node* cast = new CastPPNode(klass_node, akls);
3919 cast->init_req(0, control());
3920 klass_node = _gvn.transform(cast);
3921 }
3922
3923 // Bail out if either start or end is negative.
3924 generate_negative_guard(start, bailout, &start);
3925 generate_negative_guard(end, bailout, &end);
3926
3927 Node* length = end;
3928 if (_gvn.type(start) != TypeInt::ZERO) {
3929 length = _gvn.transform(new SubINode(end, start));
3930 }
3931
3932 // Bail out if length is negative.
3933 // Without this the new_array would throw
3934 // NegativeArraySizeException but IllegalArgumentException is what
3935 // should be thrown
3936 generate_negative_guard(length, bailout, &length);
3937
3938 if (bailout->req() > 1) {
3939 PreserveJVMState pjvms(this);
3940 set_control(_gvn.transform(bailout));
3941 uncommon_trap(Deoptimization::Reason_intrinsic,
3942 Deoptimization::Action_maybe_recompile);
3943 }
3944
3945 if (!stopped()) {
3946 // How many elements will we copy from the original?
3947 // The answer is MinI(orig_length - start, length).
3948 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3949 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3950
3951 // Generate a direct call to the right arraycopy function(s).
3952 // We know the copy is disjoint but we might not know if the
3953 // oop stores need checking.
3954 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3955 // This will fail a store-check if x contains any non-nulls.
3956
3957 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3958 // loads/stores but it is legal only if we're sure the
3959 // Arrays.copyOf would succeed. So we need all input arguments
3960 // to the copyOf to be validated, including that the copy to the
3961 // new array won't trigger an ArrayStoreException. That subtype
3962 // check can be optimized if we know something on the type of
3963 // the input array from type speculation.
3964 if (_gvn.type(klass_node)->singleton()) {
3965 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3966 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3967
3968 int test = C->static_subtype_check(superk, subk);
3969 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3970 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3971 if (t_original->speculative_type() != NULL) {
3972 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3973 }
3974 }
3975 }
3976
3977 bool validated = false;
3978 // Reason_class_check rather than Reason_intrinsic because we
3979 // want to intrinsify even if this traps.
3980 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3981 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3982 klass_node);
3983
3984 if (not_subtype_ctrl != top()) {
3985 PreserveJVMState pjvms(this);
3986 set_control(not_subtype_ctrl);
3987 uncommon_trap(Deoptimization::Reason_class_check,
3988 Deoptimization::Action_make_not_entrant);
3989 assert(stopped(), "Should be stopped");
3990 }
3991 validated = true;
3992 }
3993
3994 if (!stopped()) {
3995 newcopy = new_array(klass_node, length, 0); // no arguments to push
3996
3997 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
3998 load_object_klass(original), klass_node);
3999 if (!is_copyOfRange) {
4000 ac->set_copyof(validated);
4001 } else {
4002 ac->set_copyofrange(validated);
4003 }
4004 Node* n = _gvn.transform(ac);
4005 if (n == ac) {
4006 ac->connect_outputs(this);
4007 } else {
4008 assert(validated, "shouldn't transform if all arguments not validated");
4009 set_all_memory(n);
4010 }
4011 }
4012 }
4013 } // original reexecute is set back here
4014
4015 C->set_has_split_ifs(true); // Has chance for split-if optimization
4016 if (!stopped()) {
4017 set_result(newcopy);
4018 }
4019 return true;
4020 }
4021
4022
4023 //----------------------generate_virtual_guard---------------------------
4024 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4025 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4026 RegionNode* slow_region) {
4027 ciMethod* method = callee();
4028 int vtable_index = method->vtable_index();
4029 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4030 "bad index %d", vtable_index);
4031 // Get the Method* out of the appropriate vtable entry.
4032 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4033 vtable_index*vtableEntry::size_in_bytes() +
4034 vtableEntry::method_offset_in_bytes();
4035 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4036 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4037
4038 // Compare the target method with the expected method (e.g., Object.hashCode).
4039 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4040
4041 Node* native_call = makecon(native_call_addr);
4042 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4043 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4044
4045 return generate_slow_guard(test_native, slow_region);
4046 }
4047
4048 //-----------------------generate_method_call----------------------------
4049 // Use generate_method_call to make a slow-call to the real
4050 // method if the fast path fails. An alternative would be to
4051 // use a stub like OptoRuntime::slow_arraycopy_Java.
4052 // This only works for expanding the current library call,
4053 // not another intrinsic. (E.g., don't use this for making an
4054 // arraycopy call inside of the copyOf intrinsic.)
4055 CallJavaNode*
4056 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4057 // When compiling the intrinsic method itself, do not use this technique.
4058 guarantee(callee() != C->method(), "cannot make slow-call to self");
4059
4060 ciMethod* method = callee();
4061 // ensure the JVMS we have will be correct for this call
4062 guarantee(method_id == method->intrinsic_id(), "must match");
4063
4064 const TypeFunc* tf = TypeFunc::make(method);
4065 CallJavaNode* slow_call;
4066 if (is_static) {
4067 assert(!is_virtual, "");
4068 slow_call = new CallStaticJavaNode(C, tf,
4069 SharedRuntime::get_resolve_static_call_stub(),
4070 method, bci());
4071 } else if (is_virtual) {
4072 null_check_receiver();
4073 int vtable_index = Method::invalid_vtable_index;
4074 if (UseInlineCaches) {
4075 // Suppress the vtable call
4076 } else {
4077 // hashCode and clone are not a miranda methods,
4078 // so the vtable index is fixed.
4079 // No need to use the linkResolver to get it.
4080 vtable_index = method->vtable_index();
4081 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4082 "bad index %d", vtable_index);
4083 }
4084 slow_call = new CallDynamicJavaNode(tf,
4085 SharedRuntime::get_resolve_virtual_call_stub(),
4086 method, vtable_index, bci());
4087 } else { // neither virtual nor static: opt_virtual
4088 null_check_receiver();
4089 slow_call = new CallStaticJavaNode(C, tf,
4090 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4091 method, bci());
4092 slow_call->set_optimized_virtual(true);
4093 }
4094 set_arguments_for_java_call(slow_call);
4095 set_edges_for_java_call(slow_call);
4096 return slow_call;
4097 }
4098
4099
4100 /**
4101 * Build special case code for calls to hashCode on an object. This call may
4102 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4103 * slightly different code.
4104 */
4105 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4106 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4107 assert(!(is_virtual && is_static), "either virtual, special, or static");
4108
4109 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4110
4111 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4112 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4113 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4114 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4115 Node* obj = NULL;
4116 if (!is_static) {
4117 // Check for hashing null object
4118 obj = null_check_receiver();
4119 if (stopped()) return true; // unconditionally null
4120 result_reg->init_req(_null_path, top());
4121 result_val->init_req(_null_path, top());
4122 } else {
4123 // Do a null check, and return zero if null.
4124 // System.identityHashCode(null) == 0
4125 obj = argument(0);
4126 Node* null_ctl = top();
4127 obj = null_check_oop(obj, &null_ctl);
4128 result_reg->init_req(_null_path, null_ctl);
4129 result_val->init_req(_null_path, _gvn.intcon(0));
4130 }
4131
4132 // Unconditionally null? Then return right away.
4133 if (stopped()) {
4134 set_control( result_reg->in(_null_path));
4135 if (!stopped())
4136 set_result(result_val->in(_null_path));
4137 return true;
4138 }
4139
4140 // We only go to the fast case code if we pass a number of guards. The
4141 // paths which do not pass are accumulated in the slow_region.
4142 RegionNode* slow_region = new RegionNode(1);
4143 record_for_igvn(slow_region);
4144
4145 // If this is a virtual call, we generate a funny guard. We pull out
4146 // the vtable entry corresponding to hashCode() from the target object.
4147 // If the target method which we are calling happens to be the native
4148 // Object hashCode() method, we pass the guard. We do not need this
4149 // guard for non-virtual calls -- the caller is known to be the native
4150 // Object hashCode().
4151 if (is_virtual) {
4152 // After null check, get the object's klass.
4153 Node* obj_klass = load_object_klass(obj);
4154 generate_virtual_guard(obj_klass, slow_region);
4155 }
4156
4157 // Get the header out of the object, use LoadMarkNode when available
4158 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4159 // The control of the load must be NULL. Otherwise, the load can move before
4160 // the null check after castPP removal.
4161 Node* no_ctrl = NULL;
4162 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4163
4164 // Test the header to see if it is unlocked.
4165 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4166 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4167 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4168 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4169 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4170
4171 generate_slow_guard(test_unlocked, slow_region);
4172
4173 // Get the hash value and check to see that it has been properly assigned.
4174 // We depend on hash_mask being at most 32 bits and avoid the use of
4175 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4176 // vm: see markOop.hpp.
4177 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4178 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4179 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4180 // This hack lets the hash bits live anywhere in the mark object now, as long
4181 // as the shift drops the relevant bits into the low 32 bits. Note that
4182 // Java spec says that HashCode is an int so there's no point in capturing
4183 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4184 hshifted_header = ConvX2I(hshifted_header);
4185 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4186
4187 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4188 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4189 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4190
4191 generate_slow_guard(test_assigned, slow_region);
4192
4193 Node* init_mem = reset_memory();
4194 // fill in the rest of the null path:
4195 result_io ->init_req(_null_path, i_o());
4196 result_mem->init_req(_null_path, init_mem);
4197
4198 result_val->init_req(_fast_path, hash_val);
4199 result_reg->init_req(_fast_path, control());
4200 result_io ->init_req(_fast_path, i_o());
4201 result_mem->init_req(_fast_path, init_mem);
4202
4203 // Generate code for the slow case. We make a call to hashCode().
4204 set_control(_gvn.transform(slow_region));
4205 if (!stopped()) {
4206 // No need for PreserveJVMState, because we're using up the present state.
4207 set_all_memory(init_mem);
4208 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4209 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4210 Node* slow_result = set_results_for_java_call(slow_call);
4211 // this->control() comes from set_results_for_java_call
4212 result_reg->init_req(_slow_path, control());
4213 result_val->init_req(_slow_path, slow_result);
4214 result_io ->set_req(_slow_path, i_o());
4215 result_mem ->set_req(_slow_path, reset_memory());
4216 }
4217
4218 // Return the combined state.
4219 set_i_o( _gvn.transform(result_io) );
4220 set_all_memory( _gvn.transform(result_mem));
4221
4222 set_result(result_reg, result_val);
4223 return true;
4224 }
4225
4226 //---------------------------inline_native_getClass----------------------------
4227 // public final native Class<?> java.lang.Object.getClass();
4228 //
4229 // Build special case code for calls to getClass on an object.
4230 bool LibraryCallKit::inline_native_getClass() {
4231 Node* obj = null_check_receiver();
4232 if (stopped()) return true;
4233 set_result(load_mirror_from_klass(load_object_klass(obj)));
4234 return true;
4235 }
4236
4237 //-----------------inline_native_Reflection_getCallerClass---------------------
4238 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4239 //
4240 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4241 //
4242 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4243 // in that it must skip particular security frames and checks for
4244 // caller sensitive methods.
4245 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4246 #ifndef PRODUCT
4247 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4248 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4249 }
4250 #endif
4251
4252 if (!jvms()->has_method()) {
4253 #ifndef PRODUCT
4254 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4255 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4256 }
4257 #endif
4258 return false;
4259 }
4260
4261 // Walk back up the JVM state to find the caller at the required
4262 // depth.
4263 JVMState* caller_jvms = jvms();
4264
4265 // Cf. JVM_GetCallerClass
4266 // NOTE: Start the loop at depth 1 because the current JVM state does
4267 // not include the Reflection.getCallerClass() frame.
4268 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4269 ciMethod* m = caller_jvms->method();
4270 switch (n) {
4271 case 0:
4272 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4273 break;
4274 case 1:
4275 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4276 if (!m->caller_sensitive()) {
4277 #ifndef PRODUCT
4278 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4279 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4280 }
4281 #endif
4282 return false; // bail-out; let JVM_GetCallerClass do the work
4283 }
4284 break;
4285 default:
4286 if (!m->is_ignored_by_security_stack_walk()) {
4287 // We have reached the desired frame; return the holder class.
4288 // Acquire method holder as java.lang.Class and push as constant.
4289 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4290 ciInstance* caller_mirror = caller_klass->java_mirror();
4291 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4292
4293 #ifndef PRODUCT
4294 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4295 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());
4296 tty->print_cr(" JVM state at this point:");
4297 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4298 ciMethod* m = jvms()->of_depth(i)->method();
4299 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4300 }
4301 }
4302 #endif
4303 return true;
4304 }
4305 break;
4306 }
4307 }
4308
4309 #ifndef PRODUCT
4310 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4311 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4312 tty->print_cr(" JVM state at this point:");
4313 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4314 ciMethod* m = jvms()->of_depth(i)->method();
4315 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4316 }
4317 }
4318 #endif
4319
4320 return false; // bail-out; let JVM_GetCallerClass do the work
4321 }
4322
4323 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4324 Node* arg = argument(0);
4325 Node* result = NULL;
4326
4327 switch (id) {
4328 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4329 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4330 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4331 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4332
4333 case vmIntrinsics::_doubleToLongBits: {
4334 // two paths (plus control) merge in a wood
4335 RegionNode *r = new RegionNode(3);
4336 Node *phi = new PhiNode(r, TypeLong::LONG);
4337
4338 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4339 // Build the boolean node
4340 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4341
4342 // Branch either way.
4343 // NaN case is less traveled, which makes all the difference.
4344 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4345 Node *opt_isnan = _gvn.transform(ifisnan);
4346 assert( opt_isnan->is_If(), "Expect an IfNode");
4347 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4348 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4349
4350 set_control(iftrue);
4351
4352 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4353 Node *slow_result = longcon(nan_bits); // return NaN
4354 phi->init_req(1, _gvn.transform( slow_result ));
4355 r->init_req(1, iftrue);
4356
4357 // Else fall through
4358 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4359 set_control(iffalse);
4360
4361 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4362 r->init_req(2, iffalse);
4363
4364 // Post merge
4365 set_control(_gvn.transform(r));
4366 record_for_igvn(r);
4367
4368 C->set_has_split_ifs(true); // Has chance for split-if optimization
4369 result = phi;
4370 assert(result->bottom_type()->isa_long(), "must be");
4371 break;
4372 }
4373
4374 case vmIntrinsics::_floatToIntBits: {
4375 // two paths (plus control) merge in a wood
4376 RegionNode *r = new RegionNode(3);
4377 Node *phi = new PhiNode(r, TypeInt::INT);
4378
4379 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4380 // Build the boolean node
4381 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4382
4383 // Branch either way.
4384 // NaN case is less traveled, which makes all the difference.
4385 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4386 Node *opt_isnan = _gvn.transform(ifisnan);
4387 assert( opt_isnan->is_If(), "Expect an IfNode");
4388 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4389 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4390
4391 set_control(iftrue);
4392
4393 static const jint nan_bits = 0x7fc00000;
4394 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4395 phi->init_req(1, _gvn.transform( slow_result ));
4396 r->init_req(1, iftrue);
4397
4398 // Else fall through
4399 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4400 set_control(iffalse);
4401
4402 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4403 r->init_req(2, iffalse);
4404
4405 // Post merge
4406 set_control(_gvn.transform(r));
4407 record_for_igvn(r);
4408
4409 C->set_has_split_ifs(true); // Has chance for split-if optimization
4410 result = phi;
4411 assert(result->bottom_type()->isa_int(), "must be");
4412 break;
4413 }
4414
4415 default:
4416 fatal_unexpected_iid(id);
4417 break;
4418 }
4419 set_result(_gvn.transform(result));
4420 return true;
4421 }
4422
4423 //----------------------inline_unsafe_copyMemory-------------------------
4424 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4425 bool LibraryCallKit::inline_unsafe_copyMemory() {
4426 if (callee()->is_static()) return false; // caller must have the capability!
4427 null_check_receiver(); // null-check receiver
4428 if (stopped()) return true;
4429
4430 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4431
4432 Node* src_ptr = argument(1); // type: oop
4433 Node* src_off = ConvL2X(argument(2)); // type: long
4434 Node* dst_ptr = argument(4); // type: oop
4435 Node* dst_off = ConvL2X(argument(5)); // type: long
4436 Node* size = ConvL2X(argument(7)); // type: long
4437
4438 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4439 "fieldOffset must be byte-scaled");
4440
4441 Node* src = make_unsafe_address(src_ptr, src_off);
4442 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4443
4444 // Conservatively insert a memory barrier on all memory slices.
4445 // Do not let writes of the copy source or destination float below the copy.
4446 insert_mem_bar(Op_MemBarCPUOrder);
4447
4448 // Call it. Note that the length argument is not scaled.
4449 make_runtime_call(RC_LEAF|RC_NO_FP,
4450 OptoRuntime::fast_arraycopy_Type(),
4451 StubRoutines::unsafe_arraycopy(),
4452 "unsafe_arraycopy",
4453 TypeRawPtr::BOTTOM,
4454 src, dst, size XTOP);
4455
4456 // Do not let reads of the copy destination float above the copy.
4457 insert_mem_bar(Op_MemBarCPUOrder);
4458
4459 return true;
4460 }
4461
4462 //------------------------clone_coping-----------------------------------
4463 // Helper function for inline_native_clone.
4464 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4465 assert(obj_size != NULL, "");
4466 Node* raw_obj = alloc_obj->in(1);
4467 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4468
4469 AllocateNode* alloc = NULL;
4470 if (ReduceBulkZeroing) {
4471 // We will be completely responsible for initializing this object -
4472 // mark Initialize node as complete.
4473 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4474 // The object was just allocated - there should be no any stores!
4475 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4476 // Mark as complete_with_arraycopy so that on AllocateNode
4477 // expansion, we know this AllocateNode is initialized by an array
4478 // copy and a StoreStore barrier exists after the array copy.
4479 alloc->initialization()->set_complete_with_arraycopy();
4480 }
4481
4482 // Copy the fastest available way.
4483 // TODO: generate fields copies for small objects instead.
4484 Node* src = obj;
4485 Node* dest = alloc_obj;
4486 Node* size = _gvn.transform(obj_size);
4487
4488 // Exclude the header but include array length to copy by 8 bytes words.
4489 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4490 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4491 instanceOopDesc::base_offset_in_bytes();
4492 // base_off:
4493 // 8 - 32-bit VM
4494 // 12 - 64-bit VM, compressed klass
4495 // 16 - 64-bit VM, normal klass
4496 if (base_off % BytesPerLong != 0) {
4497 assert(UseCompressedClassPointers, "");
4498 if (is_array) {
4499 // Exclude length to copy by 8 bytes words.
4500 base_off += sizeof(int);
4501 } else {
4502 // Include klass to copy by 8 bytes words.
4503 base_off = instanceOopDesc::klass_offset_in_bytes();
4504 }
4505 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4506 }
4507 src = basic_plus_adr(src, base_off);
4508 dest = basic_plus_adr(dest, base_off);
4509
4510 // Compute the length also, if needed:
4511 Node* countx = size;
4512 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4513 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4514
4515 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4516
4517 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4518 ac->set_clonebasic();
4519 Node* n = _gvn.transform(ac);
4520 if (n == ac) {
4521 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4522 } else {
4523 set_all_memory(n);
4524 }
4525
4526 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4527 if (card_mark) {
4528 assert(!is_array, "");
4529 // Put in store barrier for any and all oops we are sticking
4530 // into this object. (We could avoid this if we could prove
4531 // that the object type contains no oop fields at all.)
4532 Node* no_particular_value = NULL;
4533 Node* no_particular_field = NULL;
4534 int raw_adr_idx = Compile::AliasIdxRaw;
4535 post_barrier(control(),
4536 memory(raw_adr_type),
4537 alloc_obj,
4538 no_particular_field,
4539 raw_adr_idx,
4540 no_particular_value,
4541 T_OBJECT,
4542 false);
4543 }
4544
4545 // Do not let reads from the cloned object float above the arraycopy.
4546 if (alloc != NULL) {
4547 // Do not let stores that initialize this object be reordered with
4548 // a subsequent store that would make this object accessible by
4549 // other threads.
4550 // Record what AllocateNode this StoreStore protects so that
4551 // escape analysis can go from the MemBarStoreStoreNode to the
4552 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4553 // based on the escape status of the AllocateNode.
4554 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4555 } else {
4556 insert_mem_bar(Op_MemBarCPUOrder);
4557 }
4558 }
4559
4560 //------------------------inline_native_clone----------------------------
4561 // protected native Object java.lang.Object.clone();
4562 //
4563 // Here are the simple edge cases:
4564 // null receiver => normal trap
4565 // virtual and clone was overridden => slow path to out-of-line clone
4566 // not cloneable or finalizer => slow path to out-of-line Object.clone
4567 //
4568 // The general case has two steps, allocation and copying.
4569 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4570 //
4571 // Copying also has two cases, oop arrays and everything else.
4572 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4573 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4574 //
4575 // These steps fold up nicely if and when the cloned object's klass
4576 // can be sharply typed as an object array, a type array, or an instance.
4577 //
4578 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4579 PhiNode* result_val;
4580
4581 // Set the reexecute bit for the interpreter to reexecute
4582 // the bytecode that invokes Object.clone if deoptimization happens.
4583 { PreserveReexecuteState preexecs(this);
4584 jvms()->set_should_reexecute(true);
4585
4586 Node* obj = null_check_receiver();
4587 if (stopped()) return true;
4588
4589 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4590
4591 // If we are going to clone an instance, we need its exact type to
4592 // know the number and types of fields to convert the clone to
4593 // loads/stores. Maybe a speculative type can help us.
4594 if (!obj_type->klass_is_exact() &&
4595 obj_type->speculative_type() != NULL &&
4596 obj_type->speculative_type()->is_instance_klass()) {
4597 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4598 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4599 !spec_ik->has_injected_fields()) {
4600 ciKlass* k = obj_type->klass();
4601 if (!k->is_instance_klass() ||
4602 k->as_instance_klass()->is_interface() ||
4603 k->as_instance_klass()->has_subklass()) {
4604 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4605 }
4606 }
4607 }
4608
4609 Node* obj_klass = load_object_klass(obj);
4610 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4611 const TypeOopPtr* toop = ((tklass != NULL)
4612 ? tklass->as_instance_type()
4613 : TypeInstPtr::NOTNULL);
4614
4615 // Conservatively insert a memory barrier on all memory slices.
4616 // Do not let writes into the original float below the clone.
4617 insert_mem_bar(Op_MemBarCPUOrder);
4618
4619 // paths into result_reg:
4620 enum {
4621 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4622 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4623 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4624 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4625 PATH_LIMIT
4626 };
4627 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4628 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4629 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4630 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4631 record_for_igvn(result_reg);
4632
4633 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4634 int raw_adr_idx = Compile::AliasIdxRaw;
4635
4636 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4637 if (array_ctl != NULL) {
4638 // It's an array.
4639 PreserveJVMState pjvms(this);
4640 set_control(array_ctl);
4641 Node* obj_length = load_array_length(obj);
4642 Node* obj_size = NULL;
4643 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4644
4645 if (!use_ReduceInitialCardMarks()) {
4646 // If it is an oop array, it requires very special treatment,
4647 // because card marking is required on each card of the array.
4648 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4649 if (is_obja != NULL) {
4650 PreserveJVMState pjvms2(this);
4651 set_control(is_obja);
4652 // Generate a direct call to the right arraycopy function(s).
4653 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4654 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4655 ac->set_cloneoop();
4656 Node* n = _gvn.transform(ac);
4657 assert(n == ac, "cannot disappear");
4658 ac->connect_outputs(this);
4659
4660 result_reg->init_req(_objArray_path, control());
4661 result_val->init_req(_objArray_path, alloc_obj);
4662 result_i_o ->set_req(_objArray_path, i_o());
4663 result_mem ->set_req(_objArray_path, reset_memory());
4664 }
4665 }
4666 // Otherwise, there are no card marks to worry about.
4667 // (We can dispense with card marks if we know the allocation
4668 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4669 // causes the non-eden paths to take compensating steps to
4670 // simulate a fresh allocation, so that no further
4671 // card marks are required in compiled code to initialize
4672 // the object.)
4673
4674 if (!stopped()) {
4675 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4676
4677 // Present the results of the copy.
4678 result_reg->init_req(_array_path, control());
4679 result_val->init_req(_array_path, alloc_obj);
4680 result_i_o ->set_req(_array_path, i_o());
4681 result_mem ->set_req(_array_path, reset_memory());
4682 }
4683 }
4684
4685 // We only go to the instance fast case code if we pass a number of guards.
4686 // The paths which do not pass are accumulated in the slow_region.
4687 RegionNode* slow_region = new RegionNode(1);
4688 record_for_igvn(slow_region);
4689 if (!stopped()) {
4690 // It's an instance (we did array above). Make the slow-path tests.
4691 // If this is a virtual call, we generate a funny guard. We grab
4692 // the vtable entry corresponding to clone() from the target object.
4693 // If the target method which we are calling happens to be the
4694 // Object clone() method, we pass the guard. We do not need this
4695 // guard for non-virtual calls; the caller is known to be the native
4696 // Object clone().
4697 if (is_virtual) {
4698 generate_virtual_guard(obj_klass, slow_region);
4699 }
4700
4701 // The object must be easily cloneable and must not have a finalizer.
4702 // Both of these conditions may be checked in a single test.
4703 // We could optimize the test further, but we don't care.
4704 generate_access_flags_guard(obj_klass,
4705 // Test both conditions:
4706 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4707 // Must be cloneable but not finalizer:
4708 JVM_ACC_IS_CLONEABLE_FAST,
4709 slow_region);
4710 }
4711
4712 if (!stopped()) {
4713 // It's an instance, and it passed the slow-path tests.
4714 PreserveJVMState pjvms(this);
4715 Node* obj_size = NULL;
4716 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4717 // is reexecuted if deoptimization occurs and there could be problems when merging
4718 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4719 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4720
4721 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4722
4723 // Present the results of the slow call.
4724 result_reg->init_req(_instance_path, control());
4725 result_val->init_req(_instance_path, alloc_obj);
4726 result_i_o ->set_req(_instance_path, i_o());
4727 result_mem ->set_req(_instance_path, reset_memory());
4728 }
4729
4730 // Generate code for the slow case. We make a call to clone().
4731 set_control(_gvn.transform(slow_region));
4732 if (!stopped()) {
4733 PreserveJVMState pjvms(this);
4734 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4735 Node* slow_result = set_results_for_java_call(slow_call);
4736 // this->control() comes from set_results_for_java_call
4737 result_reg->init_req(_slow_path, control());
4738 result_val->init_req(_slow_path, slow_result);
4739 result_i_o ->set_req(_slow_path, i_o());
4740 result_mem ->set_req(_slow_path, reset_memory());
4741 }
4742
4743 // Return the combined state.
4744 set_control( _gvn.transform(result_reg));
4745 set_i_o( _gvn.transform(result_i_o));
4746 set_all_memory( _gvn.transform(result_mem));
4747 } // original reexecute is set back here
4748
4749 set_result(_gvn.transform(result_val));
4750 return true;
4751 }
4752
4753 // If we have a tighly coupled allocation, the arraycopy may take care
4754 // of the array initialization. If one of the guards we insert between
4755 // the allocation and the arraycopy causes a deoptimization, an
4756 // unitialized array will escape the compiled method. To prevent that
4757 // we set the JVM state for uncommon traps between the allocation and
4758 // the arraycopy to the state before the allocation so, in case of
4759 // deoptimization, we'll reexecute the allocation and the
4760 // initialization.
4761 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4762 if (alloc != NULL) {
4763 ciMethod* trap_method = alloc->jvms()->method();
4764 int trap_bci = alloc->jvms()->bci();
4765
4766 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4767 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4768 // Make sure there's no store between the allocation and the
4769 // arraycopy otherwise visible side effects could be rexecuted
4770 // in case of deoptimization and cause incorrect execution.
4771 bool no_interfering_store = true;
4772 Node* mem = alloc->in(TypeFunc::Memory);
4773 if (mem->is_MergeMem()) {
4774 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4775 Node* n = mms.memory();
4776 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4777 assert(n->is_Store(), "what else?");
4778 no_interfering_store = false;
4779 break;
4780 }
4781 }
4782 } else {
4783 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4784 Node* n = mms.memory();
4785 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4786 assert(n->is_Store(), "what else?");
4787 no_interfering_store = false;
4788 break;
4789 }
4790 }
4791 }
4792
4793 if (no_interfering_store) {
4794 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4795 uint size = alloc->req();
4796 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4797 old_jvms->set_map(sfpt);
4798 for (uint i = 0; i < size; i++) {
4799 sfpt->init_req(i, alloc->in(i));
4800 }
4801 // re-push array length for deoptimization
4802 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4803 old_jvms->set_sp(old_jvms->sp()+1);
4804 old_jvms->set_monoff(old_jvms->monoff()+1);
4805 old_jvms->set_scloff(old_jvms->scloff()+1);
4806 old_jvms->set_endoff(old_jvms->endoff()+1);
4807 old_jvms->set_should_reexecute(true);
4808
4809 sfpt->set_i_o(map()->i_o());
4810 sfpt->set_memory(map()->memory());
4811 sfpt->set_control(map()->control());
4812
4813 JVMState* saved_jvms = jvms();
4814 saved_reexecute_sp = _reexecute_sp;
4815
4816 set_jvms(sfpt->jvms());
4817 _reexecute_sp = jvms()->sp();
4818
4819 return saved_jvms;
4820 }
4821 }
4822 }
4823 return NULL;
4824 }
4825
4826 // In case of a deoptimization, we restart execution at the
4827 // allocation, allocating a new array. We would leave an uninitialized
4828 // array in the heap that GCs wouldn't expect. Move the allocation
4829 // after the traps so we don't allocate the array if we
4830 // deoptimize. This is possible because tightly_coupled_allocation()
4831 // guarantees there's no observer of the allocated array at this point
4832 // and the control flow is simple enough.
4833 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4834 if (saved_jvms != NULL && !stopped()) {
4835 assert(alloc != NULL, "only with a tightly coupled allocation");
4836 // restore JVM state to the state at the arraycopy
4837 saved_jvms->map()->set_control(map()->control());
4838 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4839 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4840 // If we've improved the types of some nodes (null check) while
4841 // emitting the guards, propagate them to the current state
4842 map()->replaced_nodes().apply(saved_jvms->map());
4843 set_jvms(saved_jvms);
4844 _reexecute_sp = saved_reexecute_sp;
4845
4846 // Remove the allocation from above the guards
4847 CallProjections callprojs;
4848 alloc->extract_projections(&callprojs, true);
4849 InitializeNode* init = alloc->initialization();
4850 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4851 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4852 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4853 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4854
4855 // move the allocation here (after the guards)
4856 _gvn.hash_delete(alloc);
4857 alloc->set_req(TypeFunc::Control, control());
4858 alloc->set_req(TypeFunc::I_O, i_o());
4859 Node *mem = reset_memory();
4860 set_all_memory(mem);
4861 alloc->set_req(TypeFunc::Memory, mem);
4862 set_control(init->proj_out(TypeFunc::Control));
4863 set_i_o(callprojs.fallthrough_ioproj);
4864
4865 // Update memory as done in GraphKit::set_output_for_allocation()
4866 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4867 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4868 if (ary_type->isa_aryptr() && length_type != NULL) {
4869 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4870 }
4871 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4872 int elemidx = C->get_alias_index(telemref);
4873 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4874 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4875
4876 Node* allocx = _gvn.transform(alloc);
4877 assert(allocx == alloc, "where has the allocation gone?");
4878 assert(dest->is_CheckCastPP(), "not an allocation result?");
4879
4880 _gvn.hash_delete(dest);
4881 dest->set_req(0, control());
4882 Node* destx = _gvn.transform(dest);
4883 assert(destx == dest, "where has the allocation result gone?");
4884 }
4885 }
4886
4887
4888 //------------------------------inline_arraycopy-----------------------
4889 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4890 // Object dest, int destPos,
4891 // int length);
4892 bool LibraryCallKit::inline_arraycopy() {
4893 // Get the arguments.
4894 Node* src = argument(0); // type: oop
4895 Node* src_offset = argument(1); // type: int
4896 Node* dest = argument(2); // type: oop
4897 Node* dest_offset = argument(3); // type: int
4898 Node* length = argument(4); // type: int
4899
4900
4901 // Check for allocation before we add nodes that would confuse
4902 // tightly_coupled_allocation()
4903 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4904
4905 int saved_reexecute_sp = -1;
4906 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4907 // See arraycopy_restore_alloc_state() comment
4908 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4909 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4910 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4911 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4912
4913 // The following tests must be performed
4914 // (1) src and dest are arrays.
4915 // (2) src and dest arrays must have elements of the same BasicType
4916 // (3) src and dest must not be null.
4917 // (4) src_offset must not be negative.
4918 // (5) dest_offset must not be negative.
4919 // (6) length must not be negative.
4920 // (7) src_offset + length must not exceed length of src.
4921 // (8) dest_offset + length must not exceed length of dest.
4922 // (9) each element of an oop array must be assignable
4923
4924 // (3) src and dest must not be null.
4925 // always do this here because we need the JVM state for uncommon traps
4926 Node* null_ctl = top();
4927 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4928 assert(null_ctl->is_top(), "no null control here");
4929 dest = null_check(dest, T_ARRAY);
4930
4931 if (!can_emit_guards) {
4932 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4933 // guards but the arraycopy node could still take advantage of a
4934 // tightly allocated allocation. tightly_coupled_allocation() is
4935 // called again to make sure it takes the null check above into
4936 // account: the null check is mandatory and if it caused an
4937 // uncommon trap to be emitted then the allocation can't be
4938 // considered tightly coupled in this context.
4939 alloc = tightly_coupled_allocation(dest, NULL);
4940 }
4941
4942 bool validated = false;
4943
4944 const Type* src_type = _gvn.type(src);
4945 const Type* dest_type = _gvn.type(dest);
4946 const TypeAryPtr* top_src = src_type->isa_aryptr();
4947 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4948
4949 // Do we have the type of src?
4950 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4951 // Do we have the type of dest?
4952 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4953 // Is the type for src from speculation?
4954 bool src_spec = false;
4955 // Is the type for dest from speculation?
4956 bool dest_spec = false;
4957
4958 if ((!has_src || !has_dest) && can_emit_guards) {
4959 // We don't have sufficient type information, let's see if
4960 // speculative types can help. We need to have types for both src
4961 // and dest so that it pays off.
4962
4963 // Do we already have or could we have type information for src
4964 bool could_have_src = has_src;
4965 // Do we already have or could we have type information for dest
4966 bool could_have_dest = has_dest;
4967
4968 ciKlass* src_k = NULL;
4969 if (!has_src) {
4970 src_k = src_type->speculative_type_not_null();
4971 if (src_k != NULL && src_k->is_array_klass()) {
4972 could_have_src = true;
4973 }
4974 }
4975
4976 ciKlass* dest_k = NULL;
4977 if (!has_dest) {
4978 dest_k = dest_type->speculative_type_not_null();
4979 if (dest_k != NULL && dest_k->is_array_klass()) {
4980 could_have_dest = true;
4981 }
4982 }
4983
4984 if (could_have_src && could_have_dest) {
4985 // This is going to pay off so emit the required guards
4986 if (!has_src) {
4987 src = maybe_cast_profiled_obj(src, src_k, true);
4988 src_type = _gvn.type(src);
4989 top_src = src_type->isa_aryptr();
4990 has_src = (top_src != NULL && top_src->klass() != NULL);
4991 src_spec = true;
4992 }
4993 if (!has_dest) {
4994 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4995 dest_type = _gvn.type(dest);
4996 top_dest = dest_type->isa_aryptr();
4997 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4998 dest_spec = true;
4999 }
5000 }
5001 }
5002
5003 if (has_src && has_dest && can_emit_guards) {
5004 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5005 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5006 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5007 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5008
5009 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5010 // If both arrays are object arrays then having the exact types
5011 // for both will remove the need for a subtype check at runtime
5012 // before the call and may make it possible to pick a faster copy
5013 // routine (without a subtype check on every element)
5014 // Do we have the exact type of src?
5015 bool could_have_src = src_spec;
5016 // Do we have the exact type of dest?
5017 bool could_have_dest = dest_spec;
5018 ciKlass* src_k = top_src->klass();
5019 ciKlass* dest_k = top_dest->klass();
5020 if (!src_spec) {
5021 src_k = src_type->speculative_type_not_null();
5022 if (src_k != NULL && src_k->is_array_klass()) {
5023 could_have_src = true;
5024 }
5025 }
5026 if (!dest_spec) {
5027 dest_k = dest_type->speculative_type_not_null();
5028 if (dest_k != NULL && dest_k->is_array_klass()) {
5029 could_have_dest = true;
5030 }
5031 }
5032 if (could_have_src && could_have_dest) {
5033 // If we can have both exact types, emit the missing guards
5034 if (could_have_src && !src_spec) {
5035 src = maybe_cast_profiled_obj(src, src_k, true);
5036 }
5037 if (could_have_dest && !dest_spec) {
5038 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5039 }
5040 }
5041 }
5042 }
5043
5044 ciMethod* trap_method = method();
5045 int trap_bci = bci();
5046 if (saved_jvms != NULL) {
5047 trap_method = alloc->jvms()->method();
5048 trap_bci = alloc->jvms()->bci();
5049 }
5050
5051 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5052 can_emit_guards &&
5053 !src->is_top() && !dest->is_top()) {
5054 // validate arguments: enables transformation the ArrayCopyNode
5055 validated = true;
5056
5057 RegionNode* slow_region = new RegionNode(1);
5058 record_for_igvn(slow_region);
5059
5060 // (1) src and dest are arrays.
5061 generate_non_array_guard(load_object_klass(src), slow_region);
5062 generate_non_array_guard(load_object_klass(dest), slow_region);
5063
5064 // (2) src and dest arrays must have elements of the same BasicType
5065 // done at macro expansion or at Ideal transformation time
5066
5067 // (4) src_offset must not be negative.
5068 generate_negative_guard(src_offset, slow_region);
5069
5070 // (5) dest_offset must not be negative.
5071 generate_negative_guard(dest_offset, slow_region);
5072
5073 // (7) src_offset + length must not exceed length of src.
5074 generate_limit_guard(src_offset, length,
5075 load_array_length(src),
5076 slow_region);
5077
5078 // (8) dest_offset + length must not exceed length of dest.
5079 generate_limit_guard(dest_offset, length,
5080 load_array_length(dest),
5081 slow_region);
5082
5083 // (9) each element of an oop array must be assignable
5084 Node* src_klass = load_object_klass(src);
5085 Node* dest_klass = load_object_klass(dest);
5086 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5087
5088 if (not_subtype_ctrl != top()) {
5089 PreserveJVMState pjvms(this);
5090 set_control(not_subtype_ctrl);
5091 uncommon_trap(Deoptimization::Reason_intrinsic,
5092 Deoptimization::Action_make_not_entrant);
5093 assert(stopped(), "Should be stopped");
5094 }
5095 {
5096 PreserveJVMState pjvms(this);
5097 set_control(_gvn.transform(slow_region));
5098 uncommon_trap(Deoptimization::Reason_intrinsic,
5099 Deoptimization::Action_make_not_entrant);
5100 assert(stopped(), "Should be stopped");
5101 }
5102 }
5103
5104 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5105
5106 if (stopped()) {
5107 return true;
5108 }
5109
5110 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5111 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5112 // so the compiler has a chance to eliminate them: during macro expansion,
5113 // we have to set their control (CastPP nodes are eliminated).
5114 load_object_klass(src), load_object_klass(dest),
5115 load_array_length(src), load_array_length(dest));
5116
5117 ac->set_arraycopy(validated);
5118
5119 Node* n = _gvn.transform(ac);
5120 if (n == ac) {
5121 ac->connect_outputs(this);
5122 } else {
5123 assert(validated, "shouldn't transform if all arguments not validated");
5124 set_all_memory(n);
5125 }
5126
5127 return true;
5128 }
5129
5130
5131 // Helper function which determines if an arraycopy immediately follows
5132 // an allocation, with no intervening tests or other escapes for the object.
5133 AllocateArrayNode*
5134 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5135 RegionNode* slow_region) {
5136 if (stopped()) return NULL; // no fast path
5137 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5138
5139 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5140 if (alloc == NULL) return NULL;
5141
5142 Node* rawmem = memory(Compile::AliasIdxRaw);
5143 // Is the allocation's memory state untouched?
5144 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5145 // Bail out if there have been raw-memory effects since the allocation.
5146 // (Example: There might have been a call or safepoint.)
5147 return NULL;
5148 }
5149 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5150 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5151 return NULL;
5152 }
5153
5154 // There must be no unexpected observers of this allocation.
5155 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5156 Node* obs = ptr->fast_out(i);
5157 if (obs != this->map()) {
5158 return NULL;
5159 }
5160 }
5161
5162 // This arraycopy must unconditionally follow the allocation of the ptr.
5163 Node* alloc_ctl = ptr->in(0);
5164 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5165
5166 Node* ctl = control();
5167 while (ctl != alloc_ctl) {
5168 // There may be guards which feed into the slow_region.
5169 // Any other control flow means that we might not get a chance
5170 // to finish initializing the allocated object.
5171 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5172 IfNode* iff = ctl->in(0)->as_If();
5173 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5174 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5175 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5176 ctl = iff->in(0); // This test feeds the known slow_region.
5177 continue;
5178 }
5179 // One more try: Various low-level checks bottom out in
5180 // uncommon traps. If the debug-info of the trap omits
5181 // any reference to the allocation, as we've already
5182 // observed, then there can be no objection to the trap.
5183 bool found_trap = false;
5184 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5185 Node* obs = not_ctl->fast_out(j);
5186 if (obs->in(0) == not_ctl && obs->is_Call() &&
5187 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5188 found_trap = true; break;
5189 }
5190 }
5191 if (found_trap) {
5192 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5193 continue;
5194 }
5195 }
5196 return NULL;
5197 }
5198
5199 // If we get this far, we have an allocation which immediately
5200 // precedes the arraycopy, and we can take over zeroing the new object.
5201 // The arraycopy will finish the initialization, and provide
5202 // a new control state to which we will anchor the destination pointer.
5203
5204 return alloc;
5205 }
5206
5207 //-------------inline_encodeISOArray-----------------------------------
5208 // encode char[] to byte[] in ISO_8859_1
5209 bool LibraryCallKit::inline_encodeISOArray() {
5210 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5211 // no receiver since it is static method
5212 Node *src = argument(0);
5213 Node *src_offset = argument(1);
5214 Node *dst = argument(2);
5215 Node *dst_offset = argument(3);
5216 Node *length = argument(4);
5217
5218 const Type* src_type = src->Value(&_gvn);
5219 const Type* dst_type = dst->Value(&_gvn);
5220 const TypeAryPtr* top_src = src_type->isa_aryptr();
5221 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5222 if (top_src == NULL || top_src->klass() == NULL ||
5223 top_dest == NULL || top_dest->klass() == NULL) {
5224 // failed array check
5225 return false;
5226 }
5227
5228 // Figure out the size and type of the elements we will be copying.
5229 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5230 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5231 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5232 return false;
5233 }
5234
5235 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5236 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5237 // 'src_start' points to src array + scaled offset
5238 // 'dst_start' points to dst array + scaled offset
5239
5240 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5241 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5242 enc = _gvn.transform(enc);
5243 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5244 set_memory(res_mem, mtype);
5245 set_result(enc);
5246 return true;
5247 }
5248
5249 //-------------inline_multiplyToLen-----------------------------------
5250 bool LibraryCallKit::inline_multiplyToLen() {
5251 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5252
5253 address stubAddr = StubRoutines::multiplyToLen();
5254 if (stubAddr == NULL) {
5255 return false; // Intrinsic's stub is not implemented on this platform
5256 }
5257 const char* stubName = "multiplyToLen";
5258
5259 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5260
5261 // no receiver because it is a static method
5262 Node* x = argument(0);
5263 Node* xlen = argument(1);
5264 Node* y = argument(2);
5265 Node* ylen = argument(3);
5266 Node* z = argument(4);
5267
5268 const Type* x_type = x->Value(&_gvn);
5269 const Type* y_type = y->Value(&_gvn);
5270 const TypeAryPtr* top_x = x_type->isa_aryptr();
5271 const TypeAryPtr* top_y = y_type->isa_aryptr();
5272 if (top_x == NULL || top_x->klass() == NULL ||
5273 top_y == NULL || top_y->klass() == NULL) {
5274 // failed array check
5275 return false;
5276 }
5277
5278 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5279 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5280 if (x_elem != T_INT || y_elem != T_INT) {
5281 return false;
5282 }
5283
5284 // Set the original stack and the reexecute bit for the interpreter to reexecute
5285 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5286 // on the return from z array allocation in runtime.
5287 { PreserveReexecuteState preexecs(this);
5288 jvms()->set_should_reexecute(true);
5289
5290 Node* x_start = array_element_address(x, intcon(0), x_elem);
5291 Node* y_start = array_element_address(y, intcon(0), y_elem);
5292 // 'x_start' points to x array + scaled xlen
5293 // 'y_start' points to y array + scaled ylen
5294
5295 // Allocate the result array
5296 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5297 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5298 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5299
5300 IdealKit ideal(this);
5301
5302 #define __ ideal.
5303 Node* one = __ ConI(1);
5304 Node* zero = __ ConI(0);
5305 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5306 __ set(need_alloc, zero);
5307 __ set(z_alloc, z);
5308 __ if_then(z, BoolTest::eq, null()); {
5309 __ increment (need_alloc, one);
5310 } __ else_(); {
5311 // Update graphKit memory and control from IdealKit.
5312 sync_kit(ideal);
5313 Node* zlen_arg = load_array_length(z);
5314 // Update IdealKit memory and control from graphKit.
5315 __ sync_kit(this);
5316 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5317 __ increment (need_alloc, one);
5318 } __ end_if();
5319 } __ end_if();
5320
5321 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5322 // Update graphKit memory and control from IdealKit.
5323 sync_kit(ideal);
5324 Node * narr = new_array(klass_node, zlen, 1);
5325 // Update IdealKit memory and control from graphKit.
5326 __ sync_kit(this);
5327 __ set(z_alloc, narr);
5328 } __ end_if();
5329
5330 sync_kit(ideal);
5331 z = __ value(z_alloc);
5332 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5333 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5334 // Final sync IdealKit and GraphKit.
5335 final_sync(ideal);
5336 #undef __
5337
5338 Node* z_start = array_element_address(z, intcon(0), T_INT);
5339
5340 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5341 OptoRuntime::multiplyToLen_Type(),
5342 stubAddr, stubName, TypePtr::BOTTOM,
5343 x_start, xlen, y_start, ylen, z_start, zlen);
5344 } // original reexecute is set back here
5345
5346 C->set_has_split_ifs(true); // Has chance for split-if optimization
5347 set_result(z);
5348 return true;
5349 }
5350
5351 //-------------inline_squareToLen------------------------------------
5352 bool LibraryCallKit::inline_squareToLen() {
5353 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5354
5355 address stubAddr = StubRoutines::squareToLen();
5356 if (stubAddr == NULL) {
5357 return false; // Intrinsic's stub is not implemented on this platform
5358 }
5359 const char* stubName = "squareToLen";
5360
5361 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5362
5363 Node* x = argument(0);
5364 Node* len = argument(1);
5365 Node* z = argument(2);
5366 Node* zlen = argument(3);
5367
5368 const Type* x_type = x->Value(&_gvn);
5369 const Type* z_type = z->Value(&_gvn);
5370 const TypeAryPtr* top_x = x_type->isa_aryptr();
5371 const TypeAryPtr* top_z = z_type->isa_aryptr();
5372 if (top_x == NULL || top_x->klass() == NULL ||
5373 top_z == NULL || top_z->klass() == NULL) {
5374 // failed array check
5375 return false;
5376 }
5377
5378 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5379 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5380 if (x_elem != T_INT || z_elem != T_INT) {
5381 return false;
5382 }
5383
5384
5385 Node* x_start = array_element_address(x, intcon(0), x_elem);
5386 Node* z_start = array_element_address(z, intcon(0), z_elem);
5387
5388 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5389 OptoRuntime::squareToLen_Type(),
5390 stubAddr, stubName, TypePtr::BOTTOM,
5391 x_start, len, z_start, zlen);
5392
5393 set_result(z);
5394 return true;
5395 }
5396
5397 //-------------inline_mulAdd------------------------------------------
5398 bool LibraryCallKit::inline_mulAdd() {
5399 assert(UseMulAddIntrinsic, "not implemented on this platform");
5400
5401 address stubAddr = StubRoutines::mulAdd();
5402 if (stubAddr == NULL) {
5403 return false; // Intrinsic's stub is not implemented on this platform
5404 }
5405 const char* stubName = "mulAdd";
5406
5407 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5408
5409 Node* out = argument(0);
5410 Node* in = argument(1);
5411 Node* offset = argument(2);
5412 Node* len = argument(3);
5413 Node* k = argument(4);
5414
5415 const Type* out_type = out->Value(&_gvn);
5416 const Type* in_type = in->Value(&_gvn);
5417 const TypeAryPtr* top_out = out_type->isa_aryptr();
5418 const TypeAryPtr* top_in = in_type->isa_aryptr();
5419 if (top_out == NULL || top_out->klass() == NULL ||
5420 top_in == NULL || top_in->klass() == NULL) {
5421 // failed array check
5422 return false;
5423 }
5424
5425 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5426 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5427 if (out_elem != T_INT || in_elem != T_INT) {
5428 return false;
5429 }
5430
5431 Node* outlen = load_array_length(out);
5432 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5433 Node* out_start = array_element_address(out, intcon(0), out_elem);
5434 Node* in_start = array_element_address(in, intcon(0), in_elem);
5435
5436 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5437 OptoRuntime::mulAdd_Type(),
5438 stubAddr, stubName, TypePtr::BOTTOM,
5439 out_start,in_start, new_offset, len, k);
5440 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5441 set_result(result);
5442 return true;
5443 }
5444
5445 //-------------inline_montgomeryMultiply-----------------------------------
5446 bool LibraryCallKit::inline_montgomeryMultiply() {
5447 address stubAddr = StubRoutines::montgomeryMultiply();
5448 if (stubAddr == NULL) {
5449 return false; // Intrinsic's stub is not implemented on this platform
5450 }
5451
5452 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5453 const char* stubName = "montgomery_square";
5454
5455 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5456
5457 Node* a = argument(0);
5458 Node* b = argument(1);
5459 Node* n = argument(2);
5460 Node* len = argument(3);
5461 Node* inv = argument(4);
5462 Node* m = argument(6);
5463
5464 const Type* a_type = a->Value(&_gvn);
5465 const TypeAryPtr* top_a = a_type->isa_aryptr();
5466 const Type* b_type = b->Value(&_gvn);
5467 const TypeAryPtr* top_b = b_type->isa_aryptr();
5468 const Type* n_type = a->Value(&_gvn);
5469 const TypeAryPtr* top_n = n_type->isa_aryptr();
5470 const Type* m_type = a->Value(&_gvn);
5471 const TypeAryPtr* top_m = m_type->isa_aryptr();
5472 if (top_a == NULL || top_a->klass() == NULL ||
5473 top_b == NULL || top_b->klass() == NULL ||
5474 top_n == NULL || top_n->klass() == NULL ||
5475 top_m == NULL || top_m->klass() == NULL) {
5476 // failed array check
5477 return false;
5478 }
5479
5480 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5481 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5482 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5483 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5484 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5485 return false;
5486 }
5487
5488 // Make the call
5489 {
5490 Node* a_start = array_element_address(a, intcon(0), a_elem);
5491 Node* b_start = array_element_address(b, intcon(0), b_elem);
5492 Node* n_start = array_element_address(n, intcon(0), n_elem);
5493 Node* m_start = array_element_address(m, intcon(0), m_elem);
5494
5495 Node* call = make_runtime_call(RC_LEAF,
5496 OptoRuntime::montgomeryMultiply_Type(),
5497 stubAddr, stubName, TypePtr::BOTTOM,
5498 a_start, b_start, n_start, len, inv, top(),
5499 m_start);
5500 set_result(m);
5501 }
5502
5503 return true;
5504 }
5505
5506 bool LibraryCallKit::inline_montgomerySquare() {
5507 address stubAddr = StubRoutines::montgomerySquare();
5508 if (stubAddr == NULL) {
5509 return false; // Intrinsic's stub is not implemented on this platform
5510 }
5511
5512 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5513 const char* stubName = "montgomery_square";
5514
5515 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5516
5517 Node* a = argument(0);
5518 Node* n = argument(1);
5519 Node* len = argument(2);
5520 Node* inv = argument(3);
5521 Node* m = argument(5);
5522
5523 const Type* a_type = a->Value(&_gvn);
5524 const TypeAryPtr* top_a = a_type->isa_aryptr();
5525 const Type* n_type = a->Value(&_gvn);
5526 const TypeAryPtr* top_n = n_type->isa_aryptr();
5527 const Type* m_type = a->Value(&_gvn);
5528 const TypeAryPtr* top_m = m_type->isa_aryptr();
5529 if (top_a == NULL || top_a->klass() == NULL ||
5530 top_n == NULL || top_n->klass() == NULL ||
5531 top_m == NULL || top_m->klass() == NULL) {
5532 // failed array check
5533 return false;
5534 }
5535
5536 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5537 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5538 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5539 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5540 return false;
5541 }
5542
5543 // Make the call
5544 {
5545 Node* a_start = array_element_address(a, intcon(0), a_elem);
5546 Node* n_start = array_element_address(n, intcon(0), n_elem);
5547 Node* m_start = array_element_address(m, intcon(0), m_elem);
5548
5549 Node* call = make_runtime_call(RC_LEAF,
5550 OptoRuntime::montgomerySquare_Type(),
5551 stubAddr, stubName, TypePtr::BOTTOM,
5552 a_start, n_start, len, inv, top(),
5553 m_start);
5554 set_result(m);
5555 }
5556
5557 return true;
5558 }
5559
5560 //-------------inline_vectorizedMismatch------------------------------
5561 bool LibraryCallKit::inline_vectorizedMismatch() {
5562 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5563
5564 address stubAddr = StubRoutines::vectorizedMismatch();
5565 if (stubAddr == NULL) {
5566 return false; // Intrinsic's stub is not implemented on this platform
5567 }
5568 const char* stubName = "vectorizedMismatch";
5569 int size_l = callee()->signature()->size();
5570 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5571
5572 Node* obja = argument(0);
5573 Node* aoffset = argument(1);
5574 Node* objb = argument(3);
5575 Node* boffset = argument(4);
5576 Node* length = argument(6);
5577 Node* scale = argument(7);
5578
5579 const Type* a_type = obja->Value(&_gvn);
5580 const Type* b_type = objb->Value(&_gvn);
5581 const TypeAryPtr* top_a = a_type->isa_aryptr();
5582 const TypeAryPtr* top_b = b_type->isa_aryptr();
5583 if (top_a == NULL || top_a->klass() == NULL ||
5584 top_b == NULL || top_b->klass() == NULL) {
5585 // failed array check
5586 return false;
5587 }
5588
5589 Node* call;
5590 jvms()->set_should_reexecute(true);
5591
5592 Node* obja_adr = make_unsafe_address(obja, aoffset);
5593 Node* objb_adr = make_unsafe_address(objb, boffset);
5594
5595 call = make_runtime_call(RC_LEAF,
5596 OptoRuntime::vectorizedMismatch_Type(),
5597 stubAddr, stubName, TypePtr::BOTTOM,
5598 obja_adr, objb_adr, length, scale);
5599
5600 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5601 set_result(result);
5602 return true;
5603 }
5604
5605 /**
5606 * Calculate CRC32 for byte.
5607 * int java.util.zip.CRC32.update(int crc, int b)
5608 */
5609 bool LibraryCallKit::inline_updateCRC32() {
5610 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5611 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5612 // no receiver since it is static method
5613 Node* crc = argument(0); // type: int
5614 Node* b = argument(1); // type: int
5615
5616 /*
5617 * int c = ~ crc;
5618 * b = timesXtoThe32[(b ^ c) & 0xFF];
5619 * b = b ^ (c >>> 8);
5620 * crc = ~b;
5621 */
5622
5623 Node* M1 = intcon(-1);
5624 crc = _gvn.transform(new XorINode(crc, M1));
5625 Node* result = _gvn.transform(new XorINode(crc, b));
5626 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5627
5628 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5629 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5630 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5631 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5632
5633 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5634 result = _gvn.transform(new XorINode(crc, result));
5635 result = _gvn.transform(new XorINode(result, M1));
5636 set_result(result);
5637 return true;
5638 }
5639
5640 /**
5641 * Calculate CRC32 for byte[] array.
5642 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5643 */
5644 bool LibraryCallKit::inline_updateBytesCRC32() {
5645 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5646 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5647 // no receiver since it is static method
5648 Node* crc = argument(0); // type: int
5649 Node* src = argument(1); // type: oop
5650 Node* offset = argument(2); // type: int
5651 Node* length = argument(3); // type: int
5652
5653 const Type* src_type = src->Value(&_gvn);
5654 const TypeAryPtr* top_src = src_type->isa_aryptr();
5655 if (top_src == NULL || top_src->klass() == NULL) {
5656 // failed array check
5657 return false;
5658 }
5659
5660 // Figure out the size and type of the elements we will be copying.
5661 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5662 if (src_elem != T_BYTE) {
5663 return false;
5664 }
5665
5666 // 'src_start' points to src array + scaled offset
5667 Node* src_start = array_element_address(src, offset, src_elem);
5668
5669 // We assume that range check is done by caller.
5670 // TODO: generate range check (offset+length < src.length) in debug VM.
5671
5672 // Call the stub.
5673 address stubAddr = StubRoutines::updateBytesCRC32();
5674 const char *stubName = "updateBytesCRC32";
5675
5676 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5677 stubAddr, stubName, TypePtr::BOTTOM,
5678 crc, src_start, length);
5679 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5680 set_result(result);
5681 return true;
5682 }
5683
5684 /**
5685 * Calculate CRC32 for ByteBuffer.
5686 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5687 */
5688 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5689 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5690 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5691 // no receiver since it is static method
5692 Node* crc = argument(0); // type: int
5693 Node* src = argument(1); // type: long
5694 Node* offset = argument(3); // type: int
5695 Node* length = argument(4); // type: int
5696
5697 src = ConvL2X(src); // adjust Java long to machine word
5698 Node* base = _gvn.transform(new CastX2PNode(src));
5699 offset = ConvI2X(offset);
5700
5701 // 'src_start' points to src array + scaled offset
5702 Node* src_start = basic_plus_adr(top(), base, offset);
5703
5704 // Call the stub.
5705 address stubAddr = StubRoutines::updateBytesCRC32();
5706 const char *stubName = "updateBytesCRC32";
5707
5708 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5709 stubAddr, stubName, TypePtr::BOTTOM,
5710 crc, src_start, length);
5711 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5712 set_result(result);
5713 return true;
5714 }
5715
5716 //------------------------------get_table_from_crc32c_class-----------------------
5717 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5718 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5719 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5720
5721 return table;
5722 }
5723
5724 //------------------------------inline_updateBytesCRC32C-----------------------
5725 //
5726 // Calculate CRC32C for byte[] array.
5727 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5728 //
5729 bool LibraryCallKit::inline_updateBytesCRC32C() {
5730 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5731 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5732 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5733 // no receiver since it is a static method
5734 Node* crc = argument(0); // type: int
5735 Node* src = argument(1); // type: oop
5736 Node* offset = argument(2); // type: int
5737 Node* end = argument(3); // type: int
5738
5739 Node* length = _gvn.transform(new SubINode(end, offset));
5740
5741 const Type* src_type = src->Value(&_gvn);
5742 const TypeAryPtr* top_src = src_type->isa_aryptr();
5743 if (top_src == NULL || top_src->klass() == NULL) {
5744 // failed array check
5745 return false;
5746 }
5747
5748 // Figure out the size and type of the elements we will be copying.
5749 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5750 if (src_elem != T_BYTE) {
5751 return false;
5752 }
5753
5754 // 'src_start' points to src array + scaled offset
5755 Node* src_start = array_element_address(src, offset, src_elem);
5756
5757 // static final int[] byteTable in class CRC32C
5758 Node* table = get_table_from_crc32c_class(callee()->holder());
5759 Node* table_start = array_element_address(table, intcon(0), T_INT);
5760
5761 // We assume that range check is done by caller.
5762 // TODO: generate range check (offset+length < src.length) in debug VM.
5763
5764 // Call the stub.
5765 address stubAddr = StubRoutines::updateBytesCRC32C();
5766 const char *stubName = "updateBytesCRC32C";
5767
5768 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5769 stubAddr, stubName, TypePtr::BOTTOM,
5770 crc, src_start, length, table_start);
5771 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5772 set_result(result);
5773 return true;
5774 }
5775
5776 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5777 //
5778 // Calculate CRC32C for DirectByteBuffer.
5779 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5780 //
5781 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5782 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5783 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5784 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5785 // no receiver since it is a static method
5786 Node* crc = argument(0); // type: int
5787 Node* src = argument(1); // type: long
5788 Node* offset = argument(3); // type: int
5789 Node* end = argument(4); // type: int
5790
5791 Node* length = _gvn.transform(new SubINode(end, offset));
5792
5793 src = ConvL2X(src); // adjust Java long to machine word
5794 Node* base = _gvn.transform(new CastX2PNode(src));
5795 offset = ConvI2X(offset);
5796
5797 // 'src_start' points to src array + scaled offset
5798 Node* src_start = basic_plus_adr(top(), base, offset);
5799
5800 // static final int[] byteTable in class CRC32C
5801 Node* table = get_table_from_crc32c_class(callee()->holder());
5802 Node* table_start = array_element_address(table, intcon(0), T_INT);
5803
5804 // Call the stub.
5805 address stubAddr = StubRoutines::updateBytesCRC32C();
5806 const char *stubName = "updateBytesCRC32C";
5807
5808 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5809 stubAddr, stubName, TypePtr::BOTTOM,
5810 crc, src_start, length, table_start);
5811 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5812 set_result(result);
5813 return true;
5814 }
5815
5816 //------------------------------inline_updateBytesAdler32----------------------
5817 //
5818 // Calculate Adler32 checksum for byte[] array.
5819 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5820 //
5821 bool LibraryCallKit::inline_updateBytesAdler32() {
5822 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5823 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5824 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5825 // no receiver since it is static method
5826 Node* crc = argument(0); // type: int
5827 Node* src = argument(1); // type: oop
5828 Node* offset = argument(2); // type: int
5829 Node* length = argument(3); // type: int
5830
5831 const Type* src_type = src->Value(&_gvn);
5832 const TypeAryPtr* top_src = src_type->isa_aryptr();
5833 if (top_src == NULL || top_src->klass() == NULL) {
5834 // failed array check
5835 return false;
5836 }
5837
5838 // Figure out the size and type of the elements we will be copying.
5839 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5840 if (src_elem != T_BYTE) {
5841 return false;
5842 }
5843
5844 // 'src_start' points to src array + scaled offset
5845 Node* src_start = array_element_address(src, offset, src_elem);
5846
5847 // We assume that range check is done by caller.
5848 // TODO: generate range check (offset+length < src.length) in debug VM.
5849
5850 // Call the stub.
5851 address stubAddr = StubRoutines::updateBytesAdler32();
5852 const char *stubName = "updateBytesAdler32";
5853
5854 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5855 stubAddr, stubName, TypePtr::BOTTOM,
5856 crc, src_start, length);
5857 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5858 set_result(result);
5859 return true;
5860 }
5861
5862 //------------------------------inline_updateByteBufferAdler32---------------
5863 //
5864 // Calculate Adler32 checksum for DirectByteBuffer.
5865 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5866 //
5867 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5868 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5869 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5870 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5871 // no receiver since it is static method
5872 Node* crc = argument(0); // type: int
5873 Node* src = argument(1); // type: long
5874 Node* offset = argument(3); // type: int
5875 Node* length = argument(4); // type: int
5876
5877 src = ConvL2X(src); // adjust Java long to machine word
5878 Node* base = _gvn.transform(new CastX2PNode(src));
5879 offset = ConvI2X(offset);
5880
5881 // 'src_start' points to src array + scaled offset
5882 Node* src_start = basic_plus_adr(top(), base, offset);
5883
5884 // Call the stub.
5885 address stubAddr = StubRoutines::updateBytesAdler32();
5886 const char *stubName = "updateBytesAdler32";
5887
5888 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5889 stubAddr, stubName, TypePtr::BOTTOM,
5890 crc, src_start, length);
5891
5892 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5893 set_result(result);
5894 return true;
5895 }
5896
5897 //----------------------------inline_reference_get----------------------------
5898 // public T java.lang.ref.Reference.get();
5899 bool LibraryCallKit::inline_reference_get() {
5900 const int referent_offset = java_lang_ref_Reference::referent_offset;
5901 guarantee(referent_offset > 0, "should have already been set");
5902
5903 // Get the argument:
5904 Node* reference_obj = null_check_receiver();
5905 if (stopped()) return true;
5906
5907 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5908
5909 ciInstanceKlass* klass = env()->Object_klass();
5910 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5911
5912 Node* no_ctrl = NULL;
5913 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5914
5915 // Use the pre-barrier to record the value in the referent field
5916 pre_barrier(false /* do_load */,
5917 control(),
5918 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5919 result /* pre_val */,
5920 T_OBJECT);
5921
5922 // Add memory barrier to prevent commoning reads from this field
5923 // across safepoint since GC can change its value.
5924 insert_mem_bar(Op_MemBarCPUOrder);
5925
5926 set_result(result);
5927 return true;
5928 }
5929
5930
5931 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5932 bool is_exact=true, bool is_static=false,
5933 ciInstanceKlass * fromKls=NULL) {
5934 if (fromKls == NULL) {
5935 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5936 assert(tinst != NULL, "obj is null");
5937 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5938 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5939 fromKls = tinst->klass()->as_instance_klass();
5940 } else {
5941 assert(is_static, "only for static field access");
5942 }
5943 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5944 ciSymbol::make(fieldTypeString),
5945 is_static);
5946
5947 assert (field != NULL, "undefined field");
5948 if (field == NULL) return (Node *) NULL;
5949
5950 if (is_static) {
5951 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5952 fromObj = makecon(tip);
5953 }
5954
5955 // Next code copied from Parse::do_get_xxx():
5956
5957 // Compute address and memory type.
5958 int offset = field->offset_in_bytes();
5959 bool is_vol = field->is_volatile();
5960 ciType* field_klass = field->type();
5961 assert(field_klass->is_loaded(), "should be loaded");
5962 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5963 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5964 BasicType bt = field->layout_type();
5965
5966 // Build the resultant type of the load
5967 const Type *type;
5968 if (bt == T_OBJECT) {
5969 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5970 } else {
5971 type = Type::get_const_basic_type(bt);
5972 }
5973
5974 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5975 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
5976 }
5977 // Build the load.
5978 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5979 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5980 // If reference is volatile, prevent following memory ops from
5981 // floating up past the volatile read. Also prevents commoning
5982 // another volatile read.
5983 if (is_vol) {
5984 // Memory barrier includes bogus read of value to force load BEFORE membar
5985 insert_mem_bar(Op_MemBarAcquire, loadedField);
5986 }
5987 return loadedField;
5988 }
5989
5990 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5991 bool is_exact = true, bool is_static = false,
5992 ciInstanceKlass * fromKls = NULL) {
5993 if (fromKls == NULL) {
5994 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5995 assert(tinst != NULL, "obj is null");
5996 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5997 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5998 fromKls = tinst->klass()->as_instance_klass();
5999 }
6000 else {
6001 assert(is_static, "only for static field access");
6002 }
6003 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6004 ciSymbol::make(fieldTypeString),
6005 is_static);
6006
6007 assert(field != NULL, "undefined field");
6008 assert(!field->is_volatile(), "not defined for volatile fields");
6009
6010 if (is_static) {
6011 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6012 fromObj = makecon(tip);
6013 }
6014
6015 // Next code copied from Parse::do_get_xxx():
6016
6017 // Compute address and memory type.
6018 int offset = field->offset_in_bytes();
6019 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6020
6021 return adr;
6022 }
6023
6024 //------------------------------inline_aescrypt_Block-----------------------
6025 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6026 address stubAddr = NULL;
6027 const char *stubName;
6028 assert(UseAES, "need AES instruction support");
6029
6030 switch(id) {
6031 case vmIntrinsics::_aescrypt_encryptBlock:
6032 stubAddr = StubRoutines::aescrypt_encryptBlock();
6033 stubName = "aescrypt_encryptBlock";
6034 break;
6035 case vmIntrinsics::_aescrypt_decryptBlock:
6036 stubAddr = StubRoutines::aescrypt_decryptBlock();
6037 stubName = "aescrypt_decryptBlock";
6038 break;
6039 }
6040 if (stubAddr == NULL) return false;
6041
6042 Node* aescrypt_object = argument(0);
6043 Node* src = argument(1);
6044 Node* src_offset = argument(2);
6045 Node* dest = argument(3);
6046 Node* dest_offset = argument(4);
6047
6048 // (1) src and dest are arrays.
6049 const Type* src_type = src->Value(&_gvn);
6050 const Type* dest_type = dest->Value(&_gvn);
6051 const TypeAryPtr* top_src = src_type->isa_aryptr();
6052 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6053 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6054
6055 // for the quick and dirty code we will skip all the checks.
6056 // we are just trying to get the call to be generated.
6057 Node* src_start = src;
6058 Node* dest_start = dest;
6059 if (src_offset != NULL || dest_offset != NULL) {
6060 assert(src_offset != NULL && dest_offset != NULL, "");
6061 src_start = array_element_address(src, src_offset, T_BYTE);
6062 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6063 }
6064
6065 // now need to get the start of its expanded key array
6066 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6067 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6068 if (k_start == NULL) return false;
6069
6070 if (Matcher::pass_original_key_for_aes()) {
6071 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6072 // compatibility issues between Java key expansion and SPARC crypto instructions
6073 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6074 if (original_k_start == NULL) return false;
6075
6076 // Call the stub.
6077 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6078 stubAddr, stubName, TypePtr::BOTTOM,
6079 src_start, dest_start, k_start, original_k_start);
6080 } else {
6081 // Call the stub.
6082 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6083 stubAddr, stubName, TypePtr::BOTTOM,
6084 src_start, dest_start, k_start);
6085 }
6086
6087 return true;
6088 }
6089
6090 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6091 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6092 address stubAddr = NULL;
6093 const char *stubName = NULL;
6094
6095 assert(UseAES, "need AES instruction support");
6096
6097 switch(id) {
6098 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6099 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6100 stubName = "cipherBlockChaining_encryptAESCrypt";
6101 break;
6102 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6103 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6104 stubName = "cipherBlockChaining_decryptAESCrypt";
6105 break;
6106 }
6107 if (stubAddr == NULL) return false;
6108
6109 Node* cipherBlockChaining_object = argument(0);
6110 Node* src = argument(1);
6111 Node* src_offset = argument(2);
6112 Node* len = argument(3);
6113 Node* dest = argument(4);
6114 Node* dest_offset = argument(5);
6115
6116 // (1) src and dest are arrays.
6117 const Type* src_type = src->Value(&_gvn);
6118 const Type* dest_type = dest->Value(&_gvn);
6119 const TypeAryPtr* top_src = src_type->isa_aryptr();
6120 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6121 assert (top_src != NULL && top_src->klass() != NULL
6122 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6123
6124 // checks are the responsibility of the caller
6125 Node* src_start = src;
6126 Node* dest_start = dest;
6127 if (src_offset != NULL || dest_offset != NULL) {
6128 assert(src_offset != NULL && dest_offset != NULL, "");
6129 src_start = array_element_address(src, src_offset, T_BYTE);
6130 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6131 }
6132
6133 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6134 // (because of the predicated logic executed earlier).
6135 // so we cast it here safely.
6136 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6137
6138 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6139 if (embeddedCipherObj == NULL) return false;
6140
6141 // cast it to what we know it will be at runtime
6142 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6143 assert(tinst != NULL, "CBC obj is null");
6144 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6145 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6146 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6147
6148 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6149 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6150 const TypeOopPtr* xtype = aklass->as_instance_type();
6151 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6152 aescrypt_object = _gvn.transform(aescrypt_object);
6153
6154 // we need to get the start of the aescrypt_object's expanded key array
6155 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6156 if (k_start == NULL) return false;
6157
6158 // similarly, get the start address of the r vector
6159 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6160 if (objRvec == NULL) return false;
6161 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6162
6163 Node* cbcCrypt;
6164 if (Matcher::pass_original_key_for_aes()) {
6165 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6166 // compatibility issues between Java key expansion and SPARC crypto instructions
6167 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6168 if (original_k_start == NULL) return false;
6169
6170 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6171 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6172 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6173 stubAddr, stubName, TypePtr::BOTTOM,
6174 src_start, dest_start, k_start, r_start, len, original_k_start);
6175 } else {
6176 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6177 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6178 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6179 stubAddr, stubName, TypePtr::BOTTOM,
6180 src_start, dest_start, k_start, r_start, len);
6181 }
6182
6183 // return cipher length (int)
6184 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6185 set_result(retvalue);
6186 return true;
6187 }
6188
6189 //------------------------------inline_counterMode_AESCrypt-----------------------
6190 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6191 assert(UseAES, "need AES instruction support");
6192 if (!UseAESCTRIntrinsics) return false;
6193
6194 address stubAddr = NULL;
6195 const char *stubName = NULL;
6196 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6197 stubAddr = StubRoutines::counterMode_AESCrypt();
6198 stubName = "counterMode_AESCrypt";
6199 }
6200 if (stubAddr == NULL) return false;
6201
6202 Node* counterMode_object = argument(0);
6203 Node* src = argument(1);
6204 Node* src_offset = argument(2);
6205 Node* len = argument(3);
6206 Node* dest = argument(4);
6207 Node* dest_offset = argument(5);
6208
6209 // (1) src and dest are arrays.
6210 const Type* src_type = src->Value(&_gvn);
6211 const Type* dest_type = dest->Value(&_gvn);
6212 const TypeAryPtr* top_src = src_type->isa_aryptr();
6213 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6214 assert(top_src != NULL && top_src->klass() != NULL &&
6215 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6216
6217 // checks are the responsibility of the caller
6218 Node* src_start = src;
6219 Node* dest_start = dest;
6220 if (src_offset != NULL || dest_offset != NULL) {
6221 assert(src_offset != NULL && dest_offset != NULL, "");
6222 src_start = array_element_address(src, src_offset, T_BYTE);
6223 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6224 }
6225
6226 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6227 // (because of the predicated logic executed earlier).
6228 // so we cast it here safely.
6229 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6230 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6231 if (embeddedCipherObj == NULL) return false;
6232 // cast it to what we know it will be at runtime
6233 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6234 assert(tinst != NULL, "CTR obj is null");
6235 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6236 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6237 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6238 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6239 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6240 const TypeOopPtr* xtype = aklass->as_instance_type();
6241 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6242 aescrypt_object = _gvn.transform(aescrypt_object);
6243 // we need to get the start of the aescrypt_object's expanded key array
6244 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6245 if (k_start == NULL) return false;
6246 // similarly, get the start address of the r vector
6247 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6248 if (obj_counter == NULL) return false;
6249 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6250
6251 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6252 if (saved_encCounter == NULL) return false;
6253 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6254 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6255
6256 Node* ctrCrypt;
6257 if (Matcher::pass_original_key_for_aes()) {
6258 // no SPARC version for AES/CTR intrinsics now.
6259 return false;
6260 }
6261 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6262 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6263 OptoRuntime::counterMode_aescrypt_Type(),
6264 stubAddr, stubName, TypePtr::BOTTOM,
6265 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6266
6267 // return cipher length (int)
6268 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6269 set_result(retvalue);
6270 return true;
6271 }
6272
6273 //------------------------------get_key_start_from_aescrypt_object-----------------------
6274 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6275 #ifdef PPC64
6276 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6277 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6278 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6279 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6280 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6281 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6282 if (objSessionK == NULL) {
6283 return (Node *) NULL;
6284 }
6285 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6286 #else
6287 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6288 #endif // PPC64
6289 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6290 if (objAESCryptKey == NULL) return (Node *) NULL;
6291
6292 // now have the array, need to get the start address of the K array
6293 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6294 return k_start;
6295 }
6296
6297 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6298 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6299 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6300 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6301 if (objAESCryptKey == NULL) return (Node *) NULL;
6302
6303 // now have the array, need to get the start address of the lastKey array
6304 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6305 return original_k_start;
6306 }
6307
6308 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6309 // Return node representing slow path of predicate check.
6310 // the pseudo code we want to emulate with this predicate is:
6311 // for encryption:
6312 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6313 // for decryption:
6314 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6315 // note cipher==plain is more conservative than the original java code but that's OK
6316 //
6317 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6318 // The receiver was checked for NULL already.
6319 Node* objCBC = argument(0);
6320
6321 // Load embeddedCipher field of CipherBlockChaining object.
6322 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6323
6324 // get AESCrypt klass for instanceOf check
6325 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6326 // will have same classloader as CipherBlockChaining object
6327 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6328 assert(tinst != NULL, "CBCobj is null");
6329 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6330
6331 // we want to do an instanceof comparison against the AESCrypt class
6332 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6333 if (!klass_AESCrypt->is_loaded()) {
6334 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6335 Node* ctrl = control();
6336 set_control(top()); // no regular fast path
6337 return ctrl;
6338 }
6339 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6340
6341 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6342 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6343 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6344
6345 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6346
6347 // for encryption, we are done
6348 if (!decrypting)
6349 return instof_false; // even if it is NULL
6350
6351 // for decryption, we need to add a further check to avoid
6352 // taking the intrinsic path when cipher and plain are the same
6353 // see the original java code for why.
6354 RegionNode* region = new RegionNode(3);
6355 region->init_req(1, instof_false);
6356 Node* src = argument(1);
6357 Node* dest = argument(4);
6358 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6359 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6360 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6361 region->init_req(2, src_dest_conjoint);
6362
6363 record_for_igvn(region);
6364 return _gvn.transform(region);
6365 }
6366
6367 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6368 // Return node representing slow path of predicate check.
6369 // the pseudo code we want to emulate with this predicate is:
6370 // for encryption:
6371 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6372 // for decryption:
6373 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6374 // note cipher==plain is more conservative than the original java code but that's OK
6375 //
6376
6377 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6378 // The receiver was checked for NULL already.
6379 Node* objCTR = argument(0);
6380
6381 // Load embeddedCipher field of CipherBlockChaining object.
6382 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6383
6384 // get AESCrypt klass for instanceOf check
6385 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6386 // will have same classloader as CipherBlockChaining object
6387 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6388 assert(tinst != NULL, "CTRobj is null");
6389 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6390
6391 // we want to do an instanceof comparison against the AESCrypt class
6392 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6393 if (!klass_AESCrypt->is_loaded()) {
6394 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6395 Node* ctrl = control();
6396 set_control(top()); // no regular fast path
6397 return ctrl;
6398 }
6399
6400 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6401 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6402 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6403 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6404 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6405
6406 return instof_false; // even if it is NULL
6407 }
6408
6409 //------------------------------inline_ghash_processBlocks
6410 bool LibraryCallKit::inline_ghash_processBlocks() {
6411 address stubAddr;
6412 const char *stubName;
6413 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6414
6415 stubAddr = StubRoutines::ghash_processBlocks();
6416 stubName = "ghash_processBlocks";
6417
6418 Node* data = argument(0);
6419 Node* offset = argument(1);
6420 Node* len = argument(2);
6421 Node* state = argument(3);
6422 Node* subkeyH = argument(4);
6423
6424 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6425 assert(state_start, "state is NULL");
6426 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6427 assert(subkeyH_start, "subkeyH is NULL");
6428 Node* data_start = array_element_address(data, offset, T_BYTE);
6429 assert(data_start, "data is NULL");
6430
6431 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6432 OptoRuntime::ghash_processBlocks_Type(),
6433 stubAddr, stubName, TypePtr::BOTTOM,
6434 state_start, subkeyH_start, data_start, len);
6435 return true;
6436 }
6437
6438 //------------------------------inline_sha_implCompress-----------------------
6439 //
6440 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6441 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6442 //
6443 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6444 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6445 //
6446 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6447 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6448 //
6449 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6450 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6451
6452 Node* sha_obj = argument(0);
6453 Node* src = argument(1); // type oop
6454 Node* ofs = argument(2); // type int
6455
6456 const Type* src_type = src->Value(&_gvn);
6457 const TypeAryPtr* top_src = src_type->isa_aryptr();
6458 if (top_src == NULL || top_src->klass() == NULL) {
6459 // failed array check
6460 return false;
6461 }
6462 // Figure out the size and type of the elements we will be copying.
6463 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6464 if (src_elem != T_BYTE) {
6465 return false;
6466 }
6467 // 'src_start' points to src array + offset
6468 Node* src_start = array_element_address(src, ofs, src_elem);
6469 Node* state = NULL;
6470 address stubAddr;
6471 const char *stubName;
6472
6473 switch(id) {
6474 case vmIntrinsics::_sha_implCompress:
6475 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6476 state = get_state_from_sha_object(sha_obj);
6477 stubAddr = StubRoutines::sha1_implCompress();
6478 stubName = "sha1_implCompress";
6479 break;
6480 case vmIntrinsics::_sha2_implCompress:
6481 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6482 state = get_state_from_sha_object(sha_obj);
6483 stubAddr = StubRoutines::sha256_implCompress();
6484 stubName = "sha256_implCompress";
6485 break;
6486 case vmIntrinsics::_sha5_implCompress:
6487 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6488 state = get_state_from_sha5_object(sha_obj);
6489 stubAddr = StubRoutines::sha512_implCompress();
6490 stubName = "sha512_implCompress";
6491 break;
6492 default:
6493 fatal_unexpected_iid(id);
6494 return false;
6495 }
6496 if (state == NULL) return false;
6497
6498 // Call the stub.
6499 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6500 stubAddr, stubName, TypePtr::BOTTOM,
6501 src_start, state);
6502
6503 return true;
6504 }
6505
6506 //------------------------------inline_digestBase_implCompressMB-----------------------
6507 //
6508 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6509 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6510 //
6511 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6512 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6513 "need SHA1/SHA256/SHA512 instruction support");
6514 assert((uint)predicate < 3, "sanity");
6515 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6516
6517 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6518 Node* src = argument(1); // byte[] array
6519 Node* ofs = argument(2); // type int
6520 Node* limit = argument(3); // type int
6521
6522 const Type* src_type = src->Value(&_gvn);
6523 const TypeAryPtr* top_src = src_type->isa_aryptr();
6524 if (top_src == NULL || top_src->klass() == NULL) {
6525 // failed array check
6526 return false;
6527 }
6528 // Figure out the size and type of the elements we will be copying.
6529 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6530 if (src_elem != T_BYTE) {
6531 return false;
6532 }
6533 // 'src_start' points to src array + offset
6534 Node* src_start = array_element_address(src, ofs, src_elem);
6535
6536 const char* klass_SHA_name = NULL;
6537 const char* stub_name = NULL;
6538 address stub_addr = NULL;
6539 bool long_state = false;
6540
6541 switch (predicate) {
6542 case 0:
6543 if (UseSHA1Intrinsics) {
6544 klass_SHA_name = "sun/security/provider/SHA";
6545 stub_name = "sha1_implCompressMB";
6546 stub_addr = StubRoutines::sha1_implCompressMB();
6547 }
6548 break;
6549 case 1:
6550 if (UseSHA256Intrinsics) {
6551 klass_SHA_name = "sun/security/provider/SHA2";
6552 stub_name = "sha256_implCompressMB";
6553 stub_addr = StubRoutines::sha256_implCompressMB();
6554 }
6555 break;
6556 case 2:
6557 if (UseSHA512Intrinsics) {
6558 klass_SHA_name = "sun/security/provider/SHA5";
6559 stub_name = "sha512_implCompressMB";
6560 stub_addr = StubRoutines::sha512_implCompressMB();
6561 long_state = true;
6562 }
6563 break;
6564 default:
6565 fatal("unknown SHA intrinsic predicate: %d", predicate);
6566 }
6567 if (klass_SHA_name != NULL) {
6568 // get DigestBase klass to lookup for SHA klass
6569 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6570 assert(tinst != NULL, "digestBase_obj is not instance???");
6571 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6572
6573 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6574 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6575 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6576 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6577 }
6578 return false;
6579 }
6580 //------------------------------inline_sha_implCompressMB-----------------------
6581 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6582 bool long_state, address stubAddr, const char *stubName,
6583 Node* src_start, Node* ofs, Node* limit) {
6584 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6585 const TypeOopPtr* xtype = aklass->as_instance_type();
6586 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6587 sha_obj = _gvn.transform(sha_obj);
6588
6589 Node* state;
6590 if (long_state) {
6591 state = get_state_from_sha5_object(sha_obj);
6592 } else {
6593 state = get_state_from_sha_object(sha_obj);
6594 }
6595 if (state == NULL) return false;
6596
6597 // Call the stub.
6598 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6599 OptoRuntime::digestBase_implCompressMB_Type(),
6600 stubAddr, stubName, TypePtr::BOTTOM,
6601 src_start, state, ofs, limit);
6602 // return ofs (int)
6603 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6604 set_result(result);
6605
6606 return true;
6607 }
6608
6609 //------------------------------get_state_from_sha_object-----------------------
6610 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6611 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6612 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6613 if (sha_state == NULL) return (Node *) NULL;
6614
6615 // now have the array, need to get the start address of the state array
6616 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6617 return state;
6618 }
6619
6620 //------------------------------get_state_from_sha5_object-----------------------
6621 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6622 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6623 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6624 if (sha_state == NULL) return (Node *) NULL;
6625
6626 // now have the array, need to get the start address of the state array
6627 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6628 return state;
6629 }
6630
6631 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6632 // Return node representing slow path of predicate check.
6633 // the pseudo code we want to emulate with this predicate is:
6634 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6635 //
6636 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6637 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6638 "need SHA1/SHA256/SHA512 instruction support");
6639 assert((uint)predicate < 3, "sanity");
6640
6641 // The receiver was checked for NULL already.
6642 Node* digestBaseObj = argument(0);
6643
6644 // get DigestBase klass for instanceOf check
6645 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6646 assert(tinst != NULL, "digestBaseObj is null");
6647 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6648
6649 const char* klass_SHA_name = NULL;
6650 switch (predicate) {
6651 case 0:
6652 if (UseSHA1Intrinsics) {
6653 // we want to do an instanceof comparison against the SHA class
6654 klass_SHA_name = "sun/security/provider/SHA";
6655 }
6656 break;
6657 case 1:
6658 if (UseSHA256Intrinsics) {
6659 // we want to do an instanceof comparison against the SHA2 class
6660 klass_SHA_name = "sun/security/provider/SHA2";
6661 }
6662 break;
6663 case 2:
6664 if (UseSHA512Intrinsics) {
6665 // we want to do an instanceof comparison against the SHA5 class
6666 klass_SHA_name = "sun/security/provider/SHA5";
6667 }
6668 break;
6669 default:
6670 fatal("unknown SHA intrinsic predicate: %d", predicate);
6671 }
6672
6673 ciKlass* klass_SHA = NULL;
6674 if (klass_SHA_name != NULL) {
6675 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6676 }
6677 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6678 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6679 Node* ctrl = control();
6680 set_control(top()); // no intrinsic path
6681 return ctrl;
6682 }
6683 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6684
6685 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6686 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6687 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6688 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6689
6690 return instof_false; // even if it is NULL
6691 }
6692
6693 bool LibraryCallKit::inline_profileBoolean() {
6694 Node* counts = argument(1);
6695 const TypeAryPtr* ary = NULL;
6696 ciArray* aobj = NULL;
6697 if (counts->is_Con()
6698 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6699 && (aobj = ary->const_oop()->as_array()) != NULL
6700 && (aobj->length() == 2)) {
6701 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6702 jint false_cnt = aobj->element_value(0).as_int();
6703 jint true_cnt = aobj->element_value(1).as_int();
6704
6705 if (C->log() != NULL) {
6706 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6707 false_cnt, true_cnt);
6708 }
6709
6710 if (false_cnt + true_cnt == 0) {
6711 // According to profile, never executed.
6712 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6713 Deoptimization::Action_reinterpret);
6714 return true;
6715 }
6716
6717 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6718 // is a number of each value occurrences.
6719 Node* result = argument(0);
6720 if (false_cnt == 0 || true_cnt == 0) {
6721 // According to profile, one value has been never seen.
6722 int expected_val = (false_cnt == 0) ? 1 : 0;
6723
6724 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6725 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6726
6727 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6728 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6729 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6730
6731 { // Slow path: uncommon trap for never seen value and then reexecute
6732 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6733 // the value has been seen at least once.
6734 PreserveJVMState pjvms(this);
6735 PreserveReexecuteState preexecs(this);
6736 jvms()->set_should_reexecute(true);
6737
6738 set_control(slow_path);
6739 set_i_o(i_o());
6740
6741 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6742 Deoptimization::Action_reinterpret);
6743 }
6744 // The guard for never seen value enables sharpening of the result and
6745 // returning a constant. It allows to eliminate branches on the same value
6746 // later on.
6747 set_control(fast_path);
6748 result = intcon(expected_val);
6749 }
6750 // Stop profiling.
6751 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6752 // By replacing method body with profile data (represented as ProfileBooleanNode
6753 // on IR level) we effectively disable profiling.
6754 // It enables full speed execution once optimized code is generated.
6755 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6756 C->record_for_igvn(profile);
6757 set_result(profile);
6758 return true;
6759 } else {
6760 // Continue profiling.
6761 // Profile data isn't available at the moment. So, execute method's bytecode version.
6762 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6763 // is compiled and counters aren't available since corresponding MethodHandle
6764 // isn't a compile-time constant.
6765 return false;
6766 }
6767 }
6768
6769 bool LibraryCallKit::inline_isCompileConstant() {
6770 Node* n = argument(0);
6771 set_result(n->is_Con() ? intcon(1) : intcon(0));
6772 return true;
6773 }