1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "opto/addnode.hpp"
34 #include "opto/arraycopynode.hpp"
35 #include "opto/c2compiler.hpp"
36 #include "opto/callGenerator.hpp"
37 #include "opto/castnode.hpp"
38 #include "opto/cfgnode.hpp"
39 #include "opto/convertnode.hpp"
40 #include "opto/countbitsnode.hpp"
41 #include "opto/intrinsicnode.hpp"
42 #include "opto/idealKit.hpp"
43 #include "opto/mathexactnode.hpp"
44 #include "opto/movenode.hpp"
45 #include "opto/mulnode.hpp"
46 #include "opto/narrowptrnode.hpp"
47 #include "opto/opaquenode.hpp"
48 #include "opto/parse.hpp"
49 #include "opto/runtime.hpp"
50 #include "opto/subnode.hpp"
51 #include "prims/nativeLookup.hpp"
52 #include "prims/unsafe.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #ifdef TRACE_HAVE_INTRINSICS
55 #include "trace/traceMacros.hpp"
56 #endif
57
58 class LibraryIntrinsic : public InlineCallGenerator {
59 // Extend the set of intrinsics known to the runtime:
60 public:
61 private:
62 bool _is_virtual;
63 bool _does_virtual_dispatch;
64 int8_t _predicates_count; // Intrinsic is predicated by several conditions
65 int8_t _last_predicate; // Last generated predicate
66 vmIntrinsics::ID _intrinsic_id;
67
68 public:
69 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
70 : InlineCallGenerator(m),
71 _is_virtual(is_virtual),
72 _does_virtual_dispatch(does_virtual_dispatch),
73 _predicates_count((int8_t)predicates_count),
74 _last_predicate((int8_t)-1),
75 _intrinsic_id(id)
76 {
77 }
78 virtual bool is_intrinsic() const { return true; }
79 virtual bool is_virtual() const { return _is_virtual; }
80 virtual bool is_predicated() const { return _predicates_count > 0; }
81 virtual int predicates_count() const { return _predicates_count; }
82 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
83 virtual JVMState* generate(JVMState* jvms);
84 virtual Node* generate_predicate(JVMState* jvms, int predicate);
85 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
86 };
87
88
89 // Local helper class for LibraryIntrinsic:
90 class LibraryCallKit : public GraphKit {
91 private:
92 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
93 Node* _result; // the result node, if any
94 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
95
96 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
97
98 public:
99 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
100 : GraphKit(jvms),
101 _intrinsic(intrinsic),
102 _result(NULL)
103 {
104 // Check if this is a root compile. In that case we don't have a caller.
105 if (!jvms->has_method()) {
106 _reexecute_sp = sp();
107 } else {
108 // Find out how many arguments the interpreter needs when deoptimizing
109 // and save the stack pointer value so it can used by uncommon_trap.
110 // We find the argument count by looking at the declared signature.
111 bool ignored_will_link;
112 ciSignature* declared_signature = NULL;
113 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
114 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
115 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
116 }
117 }
118
119 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
120
121 ciMethod* caller() const { return jvms()->method(); }
122 int bci() const { return jvms()->bci(); }
123 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
124 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
125 ciMethod* callee() const { return _intrinsic->method(); }
126
127 bool try_to_inline(int predicate);
128 Node* try_to_predicate(int predicate);
129
130 void push_result() {
131 // Push the result onto the stack.
132 if (!stopped() && result() != NULL) {
133 BasicType bt = result()->bottom_type()->basic_type();
134 push_node(bt, result());
135 }
136 }
137
138 private:
139 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
140 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
141 }
142
143 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
144 void set_result(RegionNode* region, PhiNode* value);
145 Node* result() { return _result; }
146
147 virtual int reexecute_sp() { return _reexecute_sp; }
148
149 // Helper functions to inline natives
150 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
151 Node* generate_slow_guard(Node* test, RegionNode* region);
152 Node* generate_fair_guard(Node* test, RegionNode* region);
153 Node* generate_negative_guard(Node* index, RegionNode* region,
154 // resulting CastII of index:
155 Node* *pos_index = NULL);
156 Node* generate_limit_guard(Node* offset, Node* subseq_length,
157 Node* array_length,
158 RegionNode* region);
159 void generate_string_range_check(Node* array, Node* offset,
160 Node* length, bool char_count);
161 Node* generate_current_thread(Node* &tls_output);
162 Node* load_mirror_from_klass(Node* klass);
163 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
164 RegionNode* region, int null_path,
165 int offset);
166 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
167 RegionNode* region, int null_path) {
168 int offset = java_lang_Class::klass_offset_in_bytes();
169 return load_klass_from_mirror_common(mirror, never_see_null,
170 region, null_path,
171 offset);
172 }
173 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
174 RegionNode* region, int null_path) {
175 int offset = java_lang_Class::array_klass_offset_in_bytes();
176 return load_klass_from_mirror_common(mirror, never_see_null,
177 region, null_path,
178 offset);
179 }
180 Node* generate_access_flags_guard(Node* kls,
181 int modifier_mask, int modifier_bits,
182 RegionNode* region);
183 Node* generate_interface_guard(Node* kls, RegionNode* region);
184 Node* generate_array_guard(Node* kls, RegionNode* region) {
185 return generate_array_guard_common(kls, region, false, false);
186 }
187 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
188 return generate_array_guard_common(kls, region, false, true);
189 }
190 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
191 return generate_array_guard_common(kls, region, true, false);
192 }
193 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
194 return generate_array_guard_common(kls, region, true, true);
195 }
196 Node* generate_array_guard_common(Node* kls, RegionNode* region,
197 bool obj_array, bool not_array);
198 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
199 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
200 bool is_virtual = false, bool is_static = false);
201 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
202 return generate_method_call(method_id, false, true);
203 }
204 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
205 return generate_method_call(method_id, true, false);
206 }
207 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
208 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
209
210 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
211 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
213 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
214 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
215 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOfChar();
217 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_toBytesU();
219 bool inline_string_getCharsU();
220 bool inline_string_copy(bool compress);
221 bool inline_string_char_access(bool is_store);
222 Node* round_double_node(Node* n);
223 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
224 bool inline_math_native(vmIntrinsics::ID id);
225 bool inline_trig(vmIntrinsics::ID id);
226 bool inline_math(vmIntrinsics::ID id);
227 template <typename OverflowOp>
228 bool inline_math_overflow(Node* arg1, Node* arg2);
229 void inline_math_mathExact(Node* math, Node* test);
230 bool inline_math_addExactI(bool is_increment);
231 bool inline_math_addExactL(bool is_increment);
232 bool inline_math_multiplyExactI();
233 bool inline_math_multiplyExactL();
234 bool inline_math_negateExactI();
235 bool inline_math_negateExactL();
236 bool inline_math_subtractExactI(bool is_decrement);
237 bool inline_math_subtractExactL(bool is_decrement);
238 bool inline_min_max(vmIntrinsics::ID id);
239 bool inline_notify(vmIntrinsics::ID id);
240 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
241 // This returns Type::AnyPtr, RawPtr, or OopPtr.
242 int classify_unsafe_addr(Node* &base, Node* &offset);
243 Node* make_unsafe_address(Node* base, Node* offset);
244 // Helper for inline_unsafe_access.
245 // Generates the guards that check whether the result of
246 // Unsafe.getObject should be recorded in an SATB log buffer.
247 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
248
249 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
250 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
251 static bool klass_needs_init_guard(Node* kls);
252 bool inline_unsafe_allocate();
253 bool inline_unsafe_newArray(bool uninitialized);
254 bool inline_unsafe_copyMemory();
255 bool inline_native_currentThread();
256
257 bool inline_native_time_funcs(address method, const char* funcName);
258 bool inline_native_isInterrupted();
259 bool inline_native_Class_query(vmIntrinsics::ID id);
260 bool inline_native_subtype_check();
261 bool inline_native_getLength();
262 bool inline_array_copyOf(bool is_copyOfRange);
263 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
264 bool inline_preconditions_checkIndex();
265 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
266 bool inline_native_clone(bool is_virtual);
267 bool inline_native_Reflection_getCallerClass();
268 // Helper function for inlining native object hash method
269 bool inline_native_hashcode(bool is_virtual, bool is_static);
270 bool inline_native_getClass();
271
272 // Helper functions for inlining arraycopy
273 bool inline_arraycopy();
274 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
275 RegionNode* slow_region);
276 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
277 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
278
279 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
280 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
281 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
282 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
283 bool inline_unsafe_fence(vmIntrinsics::ID id);
284 bool inline_onspinwait();
285 bool inline_fp_conversions(vmIntrinsics::ID id);
286 bool inline_number_methods(vmIntrinsics::ID id);
287 bool inline_reference_get();
288 bool inline_Class_cast();
289 bool inline_aescrypt_Block(vmIntrinsics::ID id);
290 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
291 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
292 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
293 Node* inline_counterMode_AESCrypt_predicate();
294 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
295 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
296 bool inline_ghash_processBlocks();
297 bool inline_sha_implCompress(vmIntrinsics::ID id);
298 bool inline_digestBase_implCompressMB(int predicate);
299 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
300 bool long_state, address stubAddr, const char *stubName,
301 Node* src_start, Node* ofs, Node* limit);
302 Node* get_state_from_sha_object(Node *sha_object);
303 Node* get_state_from_sha5_object(Node *sha_object);
304 Node* inline_digestBase_implCompressMB_predicate(int predicate);
305 bool inline_encodeISOArray();
306 bool inline_updateCRC32();
307 bool inline_updateBytesCRC32();
308 bool inline_updateByteBufferCRC32();
309 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
310 bool inline_updateBytesCRC32C();
311 bool inline_updateDirectByteBufferCRC32C();
312 bool inline_updateBytesAdler32();
313 bool inline_updateByteBufferAdler32();
314 bool inline_multiplyToLen();
315 bool inline_hasNegatives();
316 bool inline_squareToLen();
317 bool inline_mulAdd();
318 bool inline_montgomeryMultiply();
319 bool inline_montgomerySquare();
320 bool inline_vectorizedMismatch();
321
322 bool inline_profileBoolean();
323 bool inline_isCompileConstant();
324 };
325
326 //---------------------------make_vm_intrinsic----------------------------
327 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
328 vmIntrinsics::ID id = m->intrinsic_id();
329 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
330
331 if (!m->is_loaded()) {
332 // Do not attempt to inline unloaded methods.
333 return NULL;
334 }
335
336 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
337 bool is_available = false;
338
339 {
340 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
341 // the compiler must transition to '_thread_in_vm' state because both
342 // methods access VM-internal data.
343 VM_ENTRY_MARK;
344 methodHandle mh(THREAD, m->get_Method());
345 is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
346 !C->directive()->is_intrinsic_disabled(mh) &&
347 !vmIntrinsics::is_disabled_by_flags(mh);
348
349 }
350
351 if (is_available) {
352 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
353 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
354 return new LibraryIntrinsic(m, is_virtual,
355 vmIntrinsics::predicates_needed(id),
356 vmIntrinsics::does_virtual_dispatch(id),
357 (vmIntrinsics::ID) id);
358 } else {
359 return NULL;
360 }
361 }
362
363 //----------------------register_library_intrinsics-----------------------
364 // Initialize this file's data structures, for each Compile instance.
365 void Compile::register_library_intrinsics() {
366 // Nothing to do here.
367 }
368
369 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
370 LibraryCallKit kit(jvms, this);
371 Compile* C = kit.C;
372 int nodes = C->unique();
373 #ifndef PRODUCT
374 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
375 char buf[1000];
376 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
377 tty->print_cr("Intrinsic %s", str);
378 }
379 #endif
380 ciMethod* callee = kit.callee();
381 const int bci = kit.bci();
382
383 // Try to inline the intrinsic.
384 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
385 kit.try_to_inline(_last_predicate)) {
386 if (C->print_intrinsics() || C->print_inlining()) {
387 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
388 }
389 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
390 if (C->log()) {
391 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
392 vmIntrinsics::name_at(intrinsic_id()),
393 (is_virtual() ? " virtual='1'" : ""),
394 C->unique() - nodes);
395 }
396 // Push the result from the inlined method onto the stack.
397 kit.push_result();
398 C->print_inlining_update(this);
399 return kit.transfer_exceptions_into_jvms();
400 }
401
402 // The intrinsic bailed out
403 if (C->print_intrinsics() || C->print_inlining()) {
404 if (jvms->has_method()) {
405 // Not a root compile.
406 const char* msg;
407 if (callee->intrinsic_candidate()) {
408 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
409 } else {
410 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
411 : "failed to inline (intrinsic), method not annotated";
412 }
413 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
414 } else {
415 // Root compile
416 tty->print("Did not generate intrinsic %s%s at bci:%d in",
417 vmIntrinsics::name_at(intrinsic_id()),
418 (is_virtual() ? " (virtual)" : ""), bci);
419 }
420 }
421 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
422 C->print_inlining_update(this);
423 return NULL;
424 }
425
426 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
427 LibraryCallKit kit(jvms, this);
428 Compile* C = kit.C;
429 int nodes = C->unique();
430 _last_predicate = predicate;
431 #ifndef PRODUCT
432 assert(is_predicated() && predicate < predicates_count(), "sanity");
433 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
434 char buf[1000];
435 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
436 tty->print_cr("Predicate for intrinsic %s", str);
437 }
438 #endif
439 ciMethod* callee = kit.callee();
440 const int bci = kit.bci();
441
442 Node* slow_ctl = kit.try_to_predicate(predicate);
443 if (!kit.failing()) {
444 if (C->print_intrinsics() || C->print_inlining()) {
445 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
446 }
447 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
448 if (C->log()) {
449 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
450 vmIntrinsics::name_at(intrinsic_id()),
451 (is_virtual() ? " virtual='1'" : ""),
452 C->unique() - nodes);
453 }
454 return slow_ctl; // Could be NULL if the check folds.
455 }
456
457 // The intrinsic bailed out
458 if (C->print_intrinsics() || C->print_inlining()) {
459 if (jvms->has_method()) {
460 // Not a root compile.
461 const char* msg = "failed to generate predicate for intrinsic";
462 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
463 } else {
464 // Root compile
465 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
466 vmIntrinsics::name_at(intrinsic_id()),
467 (is_virtual() ? " (virtual)" : ""), bci);
468 }
469 }
470 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
471 return NULL;
472 }
473
474 bool LibraryCallKit::try_to_inline(int predicate) {
475 // Handle symbolic names for otherwise undistinguished boolean switches:
476 const bool is_store = true;
477 const bool is_compress = true;
478 const bool is_static = true;
479 const bool is_volatile = true;
480
481 if (!jvms()->has_method()) {
482 // Root JVMState has a null method.
483 assert(map()->memory()->Opcode() == Op_Parm, "");
484 // Insert the memory aliasing node
485 set_all_memory(reset_memory());
486 }
487 assert(merged_memory(), "");
488
489
490 switch (intrinsic_id()) {
491 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
492 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
493 case vmIntrinsics::_getClass: return inline_native_getClass();
494
495 case vmIntrinsics::_dsin:
496 case vmIntrinsics::_dcos:
497 case vmIntrinsics::_dtan:
498 case vmIntrinsics::_dabs:
499 case vmIntrinsics::_datan2:
500 case vmIntrinsics::_dsqrt:
501 case vmIntrinsics::_dexp:
502 case vmIntrinsics::_dlog:
503 case vmIntrinsics::_dlog10:
504 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
505
506 case vmIntrinsics::_min:
507 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
508
509 case vmIntrinsics::_notify:
510 case vmIntrinsics::_notifyAll:
511 if (InlineNotify) {
512 return inline_notify(intrinsic_id());
513 }
514 return false;
515
516 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
517 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
518 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
519 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
520 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
521 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
522 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
523 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
524 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
525 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
526 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
527 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
528
529 case vmIntrinsics::_arraycopy: return inline_arraycopy();
530
531 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
532 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
533 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
534 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
535
536 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
537 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
538 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
539 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
540 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
541 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
542 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
543
544 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
545 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
546
547 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
548 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
549 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
550 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
551
552 case vmIntrinsics::_compressStringC:
553 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
554 case vmIntrinsics::_inflateStringC:
555 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
556
557 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
558 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
559 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
560 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
561 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
562 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
563 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
564 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
565 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
566
567 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
568 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
569 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
570 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
571 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
572 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
573 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
574 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
575 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
576
577 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
578 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
579 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
580 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
581 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
582 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
583 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
584 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
585 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
586
587 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
588 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
589 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
590 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
591 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
592 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
593 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
594 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
595 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
596
597 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
598 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
599 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
600 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
601
602 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
603 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
604 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
605 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
606
607 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
608 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
609 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
610 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
611 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
612 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
613 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
614 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
615 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
616
617 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
618 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
619 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
620 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
621 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
622 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
623 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
624 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
625 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
626
627 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
628 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
629 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
630 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
631 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
632 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
633 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
634 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
635 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
636
637 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
638 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
639 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
640 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
641 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
642 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
643 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
644 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
645 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
646
647 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
648 case vmIntrinsics::_compareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
649 case vmIntrinsics::_compareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
650 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
651 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
652
653 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
654 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
655 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
656 case vmIntrinsics::_weakCompareAndSwapObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
657 case vmIntrinsics::_weakCompareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
658 case vmIntrinsics::_weakCompareAndSwapByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
659 case vmIntrinsics::_weakCompareAndSwapByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
660 case vmIntrinsics::_weakCompareAndSwapByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
661 case vmIntrinsics::_weakCompareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
662 case vmIntrinsics::_weakCompareAndSwapShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
663 case vmIntrinsics::_weakCompareAndSwapShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
664 case vmIntrinsics::_weakCompareAndSwapShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
665 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
666 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
667 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
668 case vmIntrinsics::_weakCompareAndSwapIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
669 case vmIntrinsics::_weakCompareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
670 case vmIntrinsics::_weakCompareAndSwapLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
671 case vmIntrinsics::_weakCompareAndSwapLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
672 case vmIntrinsics::_weakCompareAndSwapLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
673
674 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
675 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
676 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
677 case vmIntrinsics::_compareAndExchangeByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
678 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
679 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
680 case vmIntrinsics::_compareAndExchangeShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
681 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
682 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
683 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
684 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
685 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
686 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
687 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
688 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
689
690 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
691 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
692 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
693 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
694
695 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
696 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
697 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
698 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
699 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
700
701 case vmIntrinsics::_loadFence:
702 case vmIntrinsics::_storeFence:
703 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
704
705 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
706
707 case vmIntrinsics::_currentThread: return inline_native_currentThread();
708 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
709
710 #ifdef TRACE_HAVE_INTRINSICS
711 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
712 #endif
713 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
714 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
715 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
716 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
717 case vmIntrinsics::_getLength: return inline_native_getLength();
718 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
719 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
720 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
721 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
722 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
723 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
724
725 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
726 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
727
728 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
729
730 case vmIntrinsics::_isInstance:
731 case vmIntrinsics::_getModifiers:
732 case vmIntrinsics::_isInterface:
733 case vmIntrinsics::_isArray:
734 case vmIntrinsics::_isPrimitive:
735 case vmIntrinsics::_getSuperclass:
736 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
737
738 case vmIntrinsics::_floatToRawIntBits:
739 case vmIntrinsics::_floatToIntBits:
740 case vmIntrinsics::_intBitsToFloat:
741 case vmIntrinsics::_doubleToRawLongBits:
742 case vmIntrinsics::_doubleToLongBits:
743 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
744
745 case vmIntrinsics::_numberOfLeadingZeros_i:
746 case vmIntrinsics::_numberOfLeadingZeros_l:
747 case vmIntrinsics::_numberOfTrailingZeros_i:
748 case vmIntrinsics::_numberOfTrailingZeros_l:
749 case vmIntrinsics::_bitCount_i:
750 case vmIntrinsics::_bitCount_l:
751 case vmIntrinsics::_reverseBytes_i:
752 case vmIntrinsics::_reverseBytes_l:
753 case vmIntrinsics::_reverseBytes_s:
754 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
755
756 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
757
758 case vmIntrinsics::_Reference_get: return inline_reference_get();
759
760 case vmIntrinsics::_Class_cast: return inline_Class_cast();
761
762 case vmIntrinsics::_aescrypt_encryptBlock:
763 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
764
765 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
766 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
767 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
768
769 case vmIntrinsics::_counterMode_AESCrypt:
770 return inline_counterMode_AESCrypt(intrinsic_id());
771
772 case vmIntrinsics::_sha_implCompress:
773 case vmIntrinsics::_sha2_implCompress:
774 case vmIntrinsics::_sha5_implCompress:
775 return inline_sha_implCompress(intrinsic_id());
776
777 case vmIntrinsics::_digestBase_implCompressMB:
778 return inline_digestBase_implCompressMB(predicate);
779
780 case vmIntrinsics::_multiplyToLen:
781 return inline_multiplyToLen();
782
783 case vmIntrinsics::_squareToLen:
784 return inline_squareToLen();
785
786 case vmIntrinsics::_mulAdd:
787 return inline_mulAdd();
788
789 case vmIntrinsics::_montgomeryMultiply:
790 return inline_montgomeryMultiply();
791 case vmIntrinsics::_montgomerySquare:
792 return inline_montgomerySquare();
793
794 case vmIntrinsics::_vectorizedMismatch:
795 return inline_vectorizedMismatch();
796
797 case vmIntrinsics::_ghash_processBlocks:
798 return inline_ghash_processBlocks();
799
800 case vmIntrinsics::_encodeISOArray:
801 case vmIntrinsics::_encodeByteISOArray:
802 return inline_encodeISOArray();
803
804 case vmIntrinsics::_updateCRC32:
805 return inline_updateCRC32();
806 case vmIntrinsics::_updateBytesCRC32:
807 return inline_updateBytesCRC32();
808 case vmIntrinsics::_updateByteBufferCRC32:
809 return inline_updateByteBufferCRC32();
810
811 case vmIntrinsics::_updateBytesCRC32C:
812 return inline_updateBytesCRC32C();
813 case vmIntrinsics::_updateDirectByteBufferCRC32C:
814 return inline_updateDirectByteBufferCRC32C();
815
816 case vmIntrinsics::_updateBytesAdler32:
817 return inline_updateBytesAdler32();
818 case vmIntrinsics::_updateByteBufferAdler32:
819 return inline_updateByteBufferAdler32();
820
821 case vmIntrinsics::_profileBoolean:
822 return inline_profileBoolean();
823 case vmIntrinsics::_isCompileConstant:
824 return inline_isCompileConstant();
825
826 case vmIntrinsics::_hasNegatives:
827 return inline_hasNegatives();
828
829 default:
830 // If you get here, it may be that someone has added a new intrinsic
831 // to the list in vmSymbols.hpp without implementing it here.
832 #ifndef PRODUCT
833 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
834 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
835 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
836 }
837 #endif
838 return false;
839 }
840 }
841
842 Node* LibraryCallKit::try_to_predicate(int predicate) {
843 if (!jvms()->has_method()) {
844 // Root JVMState has a null method.
845 assert(map()->memory()->Opcode() == Op_Parm, "");
846 // Insert the memory aliasing node
847 set_all_memory(reset_memory());
848 }
849 assert(merged_memory(), "");
850
851 switch (intrinsic_id()) {
852 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
853 return inline_cipherBlockChaining_AESCrypt_predicate(false);
854 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
855 return inline_cipherBlockChaining_AESCrypt_predicate(true);
856 case vmIntrinsics::_counterMode_AESCrypt:
857 return inline_counterMode_AESCrypt_predicate();
858 case vmIntrinsics::_digestBase_implCompressMB:
859 return inline_digestBase_implCompressMB_predicate(predicate);
860
861 default:
862 // If you get here, it may be that someone has added a new intrinsic
863 // to the list in vmSymbols.hpp without implementing it here.
864 #ifndef PRODUCT
865 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
866 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
867 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
868 }
869 #endif
870 Node* slow_ctl = control();
871 set_control(top()); // No fast path instrinsic
872 return slow_ctl;
873 }
874 }
875
876 //------------------------------set_result-------------------------------
877 // Helper function for finishing intrinsics.
878 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
879 record_for_igvn(region);
880 set_control(_gvn.transform(region));
881 set_result( _gvn.transform(value));
882 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
883 }
884
885 //------------------------------generate_guard---------------------------
886 // Helper function for generating guarded fast-slow graph structures.
887 // The given 'test', if true, guards a slow path. If the test fails
888 // then a fast path can be taken. (We generally hope it fails.)
889 // In all cases, GraphKit::control() is updated to the fast path.
890 // The returned value represents the control for the slow path.
891 // The return value is never 'top'; it is either a valid control
892 // or NULL if it is obvious that the slow path can never be taken.
893 // Also, if region and the slow control are not NULL, the slow edge
894 // is appended to the region.
895 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
896 if (stopped()) {
897 // Already short circuited.
898 return NULL;
899 }
900
901 // Build an if node and its projections.
902 // If test is true we take the slow path, which we assume is uncommon.
903 if (_gvn.type(test) == TypeInt::ZERO) {
904 // The slow branch is never taken. No need to build this guard.
905 return NULL;
906 }
907
908 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
909
910 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
911 if (if_slow == top()) {
912 // The slow branch is never taken. No need to build this guard.
913 return NULL;
914 }
915
916 if (region != NULL)
917 region->add_req(if_slow);
918
919 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
920 set_control(if_fast);
921
922 return if_slow;
923 }
924
925 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
926 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
927 }
928 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
929 return generate_guard(test, region, PROB_FAIR);
930 }
931
932 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
933 Node* *pos_index) {
934 if (stopped())
935 return NULL; // already stopped
936 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
937 return NULL; // index is already adequately typed
938 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
939 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
940 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
941 if (is_neg != NULL && pos_index != NULL) {
942 // Emulate effect of Parse::adjust_map_after_if.
943 Node* ccast = new CastIINode(index, TypeInt::POS);
944 ccast->set_req(0, control());
945 (*pos_index) = _gvn.transform(ccast);
946 }
947 return is_neg;
948 }
949
950 // Make sure that 'position' is a valid limit index, in [0..length].
951 // There are two equivalent plans for checking this:
952 // A. (offset + copyLength) unsigned<= arrayLength
953 // B. offset <= (arrayLength - copyLength)
954 // We require that all of the values above, except for the sum and
955 // difference, are already known to be non-negative.
956 // Plan A is robust in the face of overflow, if offset and copyLength
957 // are both hugely positive.
958 //
959 // Plan B is less direct and intuitive, but it does not overflow at
960 // all, since the difference of two non-negatives is always
961 // representable. Whenever Java methods must perform the equivalent
962 // check they generally use Plan B instead of Plan A.
963 // For the moment we use Plan A.
964 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
965 Node* subseq_length,
966 Node* array_length,
967 RegionNode* region) {
968 if (stopped())
969 return NULL; // already stopped
970 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
971 if (zero_offset && subseq_length->eqv_uncast(array_length))
972 return NULL; // common case of whole-array copy
973 Node* last = subseq_length;
974 if (!zero_offset) // last += offset
975 last = _gvn.transform(new AddINode(last, offset));
976 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
977 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
978 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
979 return is_over;
980 }
981
982 // Emit range checks for the given String.value byte array
983 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
984 if (stopped()) {
985 return; // already stopped
986 }
987 RegionNode* bailout = new RegionNode(1);
988 record_for_igvn(bailout);
989 if (char_count) {
990 // Convert char count to byte count
991 count = _gvn.transform(new LShiftINode(count, intcon(1)));
992 }
993
994 // Offset and count must not be negative
995 generate_negative_guard(offset, bailout);
996 generate_negative_guard(count, bailout);
997 // Offset + count must not exceed length of array
998 generate_limit_guard(offset, count, load_array_length(array), bailout);
999
1000 if (bailout->req() > 1) {
1001 PreserveJVMState pjvms(this);
1002 set_control(_gvn.transform(bailout));
1003 uncommon_trap(Deoptimization::Reason_intrinsic,
1004 Deoptimization::Action_maybe_recompile);
1005 }
1006 }
1007
1008 //--------------------------generate_current_thread--------------------
1009 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1010 ciKlass* thread_klass = env()->Thread_klass();
1011 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1012 Node* thread = _gvn.transform(new ThreadLocalNode());
1013 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1014 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1015 tls_output = thread;
1016 return threadObj;
1017 }
1018
1019
1020 //------------------------------make_string_method_node------------------------
1021 // Helper method for String intrinsic functions. This version is called with
1022 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1023 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1024 // containing the lengths of str1 and str2.
1025 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1026 Node* result = NULL;
1027 switch (opcode) {
1028 case Op_StrIndexOf:
1029 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1030 str1_start, cnt1, str2_start, cnt2, ae);
1031 break;
1032 case Op_StrComp:
1033 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1034 str1_start, cnt1, str2_start, cnt2, ae);
1035 break;
1036 case Op_StrEquals:
1037 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1038 // Use the constant length if there is one because optimized match rule may exist.
1039 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1040 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1041 break;
1042 default:
1043 ShouldNotReachHere();
1044 return NULL;
1045 }
1046
1047 // All these intrinsics have checks.
1048 C->set_has_split_ifs(true); // Has chance for split-if optimization
1049
1050 return _gvn.transform(result);
1051 }
1052
1053 //------------------------------inline_string_compareTo------------------------
1054 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1055 Node* arg1 = argument(0);
1056 Node* arg2 = argument(1);
1057
1058 // Get start addr and length of first argument
1059 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1060 Node* arg1_cnt = load_array_length(arg1);
1061
1062 // Get start addr and length of second argument
1063 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1064 Node* arg2_cnt = load_array_length(arg2);
1065
1066 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1067 set_result(result);
1068 return true;
1069 }
1070
1071 //------------------------------inline_string_equals------------------------
1072 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1073 Node* arg1 = argument(0);
1074 Node* arg2 = argument(1);
1075
1076 // paths (plus control) merge
1077 RegionNode* region = new RegionNode(3);
1078 Node* phi = new PhiNode(region, TypeInt::BOOL);
1079
1080 if (!stopped()) {
1081 // Get start addr and length of first argument
1082 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1083 Node* arg1_cnt = load_array_length(arg1);
1084
1085 // Get start addr and length of second argument
1086 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1087 Node* arg2_cnt = load_array_length(arg2);
1088
1089 // Check for arg1_cnt != arg2_cnt
1090 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1091 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1092 Node* if_ne = generate_slow_guard(bol, NULL);
1093 if (if_ne != NULL) {
1094 phi->init_req(2, intcon(0));
1095 region->init_req(2, if_ne);
1096 }
1097
1098 // Check for count == 0 is done by assembler code for StrEquals.
1099
1100 if (!stopped()) {
1101 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1102 phi->init_req(1, equals);
1103 region->init_req(1, control());
1104 }
1105 }
1106
1107 // post merge
1108 set_control(_gvn.transform(region));
1109 record_for_igvn(region);
1110
1111 set_result(_gvn.transform(phi));
1112 return true;
1113 }
1114
1115 //------------------------------inline_array_equals----------------------------
1116 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1117 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1118 Node* arg1 = argument(0);
1119 Node* arg2 = argument(1);
1120
1121 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1122 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1123 return true;
1124 }
1125
1126 //------------------------------inline_hasNegatives------------------------------
1127 bool LibraryCallKit::inline_hasNegatives() {
1128 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1129 return false;
1130 }
1131
1132 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1133 // no receiver since it is static method
1134 Node* ba = argument(0);
1135 Node* offset = argument(1);
1136 Node* len = argument(2);
1137
1138 // Range checks
1139 generate_string_range_check(ba, offset, len, false);
1140 if (stopped()) {
1141 return true;
1142 }
1143 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1144 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1145 set_result(_gvn.transform(result));
1146 return true;
1147 }
1148
1149 bool LibraryCallKit::inline_preconditions_checkIndex() {
1150 Node* index = argument(0);
1151 Node* length = argument(1);
1152 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1153 return false;
1154 }
1155
1156 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1157 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1158
1159 {
1160 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1161 uncommon_trap(Deoptimization::Reason_intrinsic,
1162 Deoptimization::Action_make_not_entrant);
1163 }
1164
1165 if (stopped()) {
1166 return false;
1167 }
1168
1169 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1170 BoolTest::mask btest = BoolTest::lt;
1171 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1172 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1173 _gvn.set_type(rc, rc->Value(&_gvn));
1174 if (!rc_bool->is_Con()) {
1175 record_for_igvn(rc);
1176 }
1177 set_control(_gvn.transform(new IfTrueNode(rc)));
1178 {
1179 PreserveJVMState pjvms(this);
1180 set_control(_gvn.transform(new IfFalseNode(rc)));
1181 uncommon_trap(Deoptimization::Reason_range_check,
1182 Deoptimization::Action_make_not_entrant);
1183 }
1184
1185 if (stopped()) {
1186 return false;
1187 }
1188
1189 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1190 result->set_req(0, control());
1191 result = _gvn.transform(result);
1192 set_result(result);
1193 replace_in_map(index, result);
1194 return true;
1195 }
1196
1197 //------------------------------inline_string_indexOf------------------------
1198 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1199 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1200 return false;
1201 }
1202 Node* src = argument(0);
1203 Node* tgt = argument(1);
1204
1205 // Make the merge point
1206 RegionNode* result_rgn = new RegionNode(4);
1207 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1208
1209 // Get start addr and length of source string
1210 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1211 Node* src_count = load_array_length(src);
1212
1213 // Get start addr and length of substring
1214 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1215 Node* tgt_count = load_array_length(tgt);
1216
1217 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1218 // Divide src size by 2 if String is UTF16 encoded
1219 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1220 }
1221 if (ae == StrIntrinsicNode::UU) {
1222 // Divide substring size by 2 if String is UTF16 encoded
1223 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1224 }
1225
1226 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1227 if (result != NULL) {
1228 result_phi->init_req(3, result);
1229 result_rgn->init_req(3, control());
1230 }
1231 set_control(_gvn.transform(result_rgn));
1232 record_for_igvn(result_rgn);
1233 set_result(_gvn.transform(result_phi));
1234
1235 return true;
1236 }
1237
1238 //-----------------------------inline_string_indexOf-----------------------
1239 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1240 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1241 return false;
1242 }
1243 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1244 return false;
1245 }
1246 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1247 Node* src = argument(0); // byte[]
1248 Node* src_count = argument(1); // char count
1249 Node* tgt = argument(2); // byte[]
1250 Node* tgt_count = argument(3); // char count
1251 Node* from_index = argument(4); // char index
1252
1253 // Multiply byte array index by 2 if String is UTF16 encoded
1254 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1255 src_count = _gvn.transform(new SubINode(src_count, from_index));
1256 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1257 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1258
1259 // Range checks
1260 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1261 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1262 if (stopped()) {
1263 return true;
1264 }
1265
1266 RegionNode* region = new RegionNode(5);
1267 Node* phi = new PhiNode(region, TypeInt::INT);
1268
1269 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1270 if (result != NULL) {
1271 // The result is index relative to from_index if substring was found, -1 otherwise.
1272 // Generate code which will fold into cmove.
1273 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1274 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1275
1276 Node* if_lt = generate_slow_guard(bol, NULL);
1277 if (if_lt != NULL) {
1278 // result == -1
1279 phi->init_req(3, result);
1280 region->init_req(3, if_lt);
1281 }
1282 if (!stopped()) {
1283 result = _gvn.transform(new AddINode(result, from_index));
1284 phi->init_req(4, result);
1285 region->init_req(4, control());
1286 }
1287 }
1288
1289 set_control(_gvn.transform(region));
1290 record_for_igvn(region);
1291 set_result(_gvn.transform(phi));
1292
1293 return true;
1294 }
1295
1296 // Create StrIndexOfNode with fast path checks
1297 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1298 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1299 // Check for substr count > string count
1300 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1301 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1302 Node* if_gt = generate_slow_guard(bol, NULL);
1303 if (if_gt != NULL) {
1304 phi->init_req(1, intcon(-1));
1305 region->init_req(1, if_gt);
1306 }
1307 if (!stopped()) {
1308 // Check for substr count == 0
1309 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1310 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1311 Node* if_zero = generate_slow_guard(bol, NULL);
1312 if (if_zero != NULL) {
1313 phi->init_req(2, intcon(0));
1314 region->init_req(2, if_zero);
1315 }
1316 }
1317 if (!stopped()) {
1318 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1319 }
1320 return NULL;
1321 }
1322
1323 //-----------------------------inline_string_indexOfChar-----------------------
1324 bool LibraryCallKit::inline_string_indexOfChar() {
1325 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1326 return false;
1327 }
1328 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1329 return false;
1330 }
1331 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1332 Node* src = argument(0); // byte[]
1333 Node* tgt = argument(1); // tgt is int ch
1334 Node* from_index = argument(2);
1335 Node* max = argument(3);
1336
1337 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1338 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1339 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1340
1341 // Range checks
1342 generate_string_range_check(src, src_offset, src_count, true);
1343 if (stopped()) {
1344 return true;
1345 }
1346
1347 RegionNode* region = new RegionNode(3);
1348 Node* phi = new PhiNode(region, TypeInt::INT);
1349
1350 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1351 C->set_has_split_ifs(true); // Has chance for split-if optimization
1352 _gvn.transform(result);
1353
1354 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1355 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1356
1357 Node* if_lt = generate_slow_guard(bol, NULL);
1358 if (if_lt != NULL) {
1359 // result == -1
1360 phi->init_req(2, result);
1361 region->init_req(2, if_lt);
1362 }
1363 if (!stopped()) {
1364 result = _gvn.transform(new AddINode(result, from_index));
1365 phi->init_req(1, result);
1366 region->init_req(1, control());
1367 }
1368 set_control(_gvn.transform(region));
1369 record_for_igvn(region);
1370 set_result(_gvn.transform(phi));
1371
1372 return true;
1373 }
1374 //---------------------------inline_string_copy---------------------
1375 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1376 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1377 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1378 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1379 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1380 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1381 bool LibraryCallKit::inline_string_copy(bool compress) {
1382 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1383 return false;
1384 }
1385 int nargs = 5; // 2 oops, 3 ints
1386 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1387
1388 Node* src = argument(0);
1389 Node* src_offset = argument(1);
1390 Node* dst = argument(2);
1391 Node* dst_offset = argument(3);
1392 Node* length = argument(4);
1393
1394 // Check for allocation before we add nodes that would confuse
1395 // tightly_coupled_allocation()
1396 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1397
1398 // Figure out the size and type of the elements we will be copying.
1399 const Type* src_type = src->Value(&_gvn);
1400 const Type* dst_type = dst->Value(&_gvn);
1401 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1402 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1403 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1404 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1405 "Unsupported array types for inline_string_copy");
1406
1407 // Range checks
1408 generate_string_range_check(src, src_offset, length, compress && src_elem == T_BYTE);
1409 generate_string_range_check(dst, dst_offset, length, !compress && dst_elem == T_BYTE);
1410 if (stopped()) {
1411 return true;
1412 }
1413
1414 // Convert char[] offsets to byte[] offsets
1415 if (compress && src_elem == T_BYTE) {
1416 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1417 } else if (!compress && dst_elem == T_BYTE) {
1418 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1419 }
1420
1421 Node* src_start = array_element_address(src, src_offset, src_elem);
1422 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1423 // 'src_start' points to src array + scaled offset
1424 // 'dst_start' points to dst array + scaled offset
1425 Node* count = NULL;
1426 if (compress) {
1427 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1428 } else {
1429 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1430 }
1431
1432 if (alloc != NULL) {
1433 if (alloc->maybe_set_complete(&_gvn)) {
1434 // "You break it, you buy it."
1435 InitializeNode* init = alloc->initialization();
1436 assert(init->is_complete(), "we just did this");
1437 init->set_complete_with_arraycopy();
1438 assert(dst->is_CheckCastPP(), "sanity");
1439 assert(dst->in(0)->in(0) == init, "dest pinned");
1440 }
1441 // Do not let stores that initialize this object be reordered with
1442 // a subsequent store that would make this object accessible by
1443 // other threads.
1444 // Record what AllocateNode this StoreStore protects so that
1445 // escape analysis can go from the MemBarStoreStoreNode to the
1446 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1447 // based on the escape status of the AllocateNode.
1448 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1449 }
1450 if (compress) {
1451 set_result(_gvn.transform(count));
1452 }
1453 return true;
1454 }
1455
1456 #ifdef _LP64
1457 #define XTOP ,top() /*additional argument*/
1458 #else //_LP64
1459 #define XTOP /*no additional argument*/
1460 #endif //_LP64
1461
1462 //------------------------inline_string_toBytesU--------------------------
1463 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1464 bool LibraryCallKit::inline_string_toBytesU() {
1465 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1466 return false;
1467 }
1468 // Get the arguments.
1469 Node* value = argument(0);
1470 Node* offset = argument(1);
1471 Node* length = argument(2);
1472
1473 Node* newcopy = NULL;
1474
1475 // Set the original stack and the reexecute bit for the interpreter to reexecute
1476 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1477 { PreserveReexecuteState preexecs(this);
1478 jvms()->set_should_reexecute(true);
1479
1480 // Check if a null path was taken unconditionally.
1481 value = null_check(value);
1482
1483 RegionNode* bailout = new RegionNode(1);
1484 record_for_igvn(bailout);
1485
1486 // Range checks
1487 generate_negative_guard(offset, bailout);
1488 generate_negative_guard(length, bailout);
1489 generate_limit_guard(offset, length, load_array_length(value), bailout);
1490 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1491 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1492
1493 if (bailout->req() > 1) {
1494 PreserveJVMState pjvms(this);
1495 set_control(_gvn.transform(bailout));
1496 uncommon_trap(Deoptimization::Reason_intrinsic,
1497 Deoptimization::Action_maybe_recompile);
1498 }
1499 if (stopped()) {
1500 return true;
1501 }
1502
1503 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1504 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1505 newcopy = new_array(klass_node, size, 0); // no arguments to push
1506 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1507
1508 // Calculate starting addresses.
1509 Node* src_start = array_element_address(value, offset, T_CHAR);
1510 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1511
1512 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1513 const TypeInt* toffset = gvn().type(offset)->is_int();
1514 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1515
1516 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1517 const char* copyfunc_name = "arraycopy";
1518 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1519 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1520 OptoRuntime::fast_arraycopy_Type(),
1521 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1522 src_start, dst_start, ConvI2X(length) XTOP);
1523 // Do not let reads from the cloned object float above the arraycopy.
1524 if (alloc != NULL) {
1525 if (alloc->maybe_set_complete(&_gvn)) {
1526 // "You break it, you buy it."
1527 InitializeNode* init = alloc->initialization();
1528 assert(init->is_complete(), "we just did this");
1529 init->set_complete_with_arraycopy();
1530 assert(newcopy->is_CheckCastPP(), "sanity");
1531 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1532 }
1533 // Do not let stores that initialize this object be reordered with
1534 // a subsequent store that would make this object accessible by
1535 // other threads.
1536 // Record what AllocateNode this StoreStore protects so that
1537 // escape analysis can go from the MemBarStoreStoreNode to the
1538 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1539 // based on the escape status of the AllocateNode.
1540 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1541 } else {
1542 insert_mem_bar(Op_MemBarCPUOrder);
1543 }
1544 } // original reexecute is set back here
1545
1546 C->set_has_split_ifs(true); // Has chance for split-if optimization
1547 if (!stopped()) {
1548 set_result(newcopy);
1549 }
1550 return true;
1551 }
1552
1553 //------------------------inline_string_getCharsU--------------------------
1554 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1555 bool LibraryCallKit::inline_string_getCharsU() {
1556 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1557 return false;
1558 }
1559
1560 // Get the arguments.
1561 Node* src = argument(0);
1562 Node* src_begin = argument(1);
1563 Node* src_end = argument(2); // exclusive offset (i < src_end)
1564 Node* dst = argument(3);
1565 Node* dst_begin = argument(4);
1566
1567 // Check for allocation before we add nodes that would confuse
1568 // tightly_coupled_allocation()
1569 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1570
1571 // Check if a null path was taken unconditionally.
1572 src = null_check(src);
1573 dst = null_check(dst);
1574 if (stopped()) {
1575 return true;
1576 }
1577
1578 // Get length and convert char[] offset to byte[] offset
1579 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1580 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1581
1582 // Range checks
1583 generate_string_range_check(src, src_begin, length, true);
1584 generate_string_range_check(dst, dst_begin, length, false);
1585 if (stopped()) {
1586 return true;
1587 }
1588
1589 if (!stopped()) {
1590 // Calculate starting addresses.
1591 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1592 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1593
1594 // Check if array addresses are aligned to HeapWordSize
1595 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1596 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1597 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1598 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1599
1600 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1601 const char* copyfunc_name = "arraycopy";
1602 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1603 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1604 OptoRuntime::fast_arraycopy_Type(),
1605 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1606 src_start, dst_start, ConvI2X(length) XTOP);
1607 // Do not let reads from the cloned object float above the arraycopy.
1608 if (alloc != NULL) {
1609 if (alloc->maybe_set_complete(&_gvn)) {
1610 // "You break it, you buy it."
1611 InitializeNode* init = alloc->initialization();
1612 assert(init->is_complete(), "we just did this");
1613 init->set_complete_with_arraycopy();
1614 assert(dst->is_CheckCastPP(), "sanity");
1615 assert(dst->in(0)->in(0) == init, "dest pinned");
1616 }
1617 // Do not let stores that initialize this object be reordered with
1618 // a subsequent store that would make this object accessible by
1619 // other threads.
1620 // Record what AllocateNode this StoreStore protects so that
1621 // escape analysis can go from the MemBarStoreStoreNode to the
1622 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1623 // based on the escape status of the AllocateNode.
1624 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1625 } else {
1626 insert_mem_bar(Op_MemBarCPUOrder);
1627 }
1628 }
1629
1630 C->set_has_split_ifs(true); // Has chance for split-if optimization
1631 return true;
1632 }
1633
1634 //----------------------inline_string_char_access----------------------------
1635 // Store/Load char to/from byte[] array.
1636 // static void StringUTF16.putChar(byte[] val, int index, int c)
1637 // static char StringUTF16.getChar(byte[] val, int index)
1638 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1639 Node* value = argument(0);
1640 Node* index = argument(1);
1641 Node* ch = is_store ? argument(2) : NULL;
1642
1643 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1644 // correctly requires matched array shapes.
1645 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1646 "sanity: byte[] and char[] bases agree");
1647 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1648 "sanity: byte[] and char[] scales agree");
1649
1650 // Bail when getChar over constants is requested: constant folding would
1651 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1652 // Java method would constant fold nicely instead.
1653 if (!is_store && value->is_Con() && index->is_Con()) {
1654 return false;
1655 }
1656
1657 Node* adr = array_element_address(value, index, T_CHAR);
1658 if (is_store) {
1659 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1660 false, false, true /* mismatched */);
1661 } else {
1662 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1663 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1664 set_result(ch);
1665 }
1666 return true;
1667 }
1668
1669 //--------------------------round_double_node--------------------------------
1670 // Round a double node if necessary.
1671 Node* LibraryCallKit::round_double_node(Node* n) {
1672 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1673 n = _gvn.transform(new RoundDoubleNode(0, n));
1674 return n;
1675 }
1676
1677 //------------------------------inline_math-----------------------------------
1678 // public static double Math.abs(double)
1679 // public static double Math.sqrt(double)
1680 // public static double Math.log(double)
1681 // public static double Math.log10(double)
1682 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1683 Node* arg = round_double_node(argument(0));
1684 Node* n = NULL;
1685 switch (id) {
1686 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1687 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1688 default: fatal_unexpected_iid(id); break;
1689 }
1690 set_result(_gvn.transform(n));
1691 return true;
1692 }
1693
1694 //------------------------------inline_trig----------------------------------
1695 // Inline sin/cos/tan instructions, if possible. If rounding is required, do
1696 // argument reduction which will turn into a fast/slow diamond.
1697 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1698 Node* arg = round_double_node(argument(0));
1699 Node* n = NULL;
1700
1701 n = _gvn.transform(n);
1702
1703 // Rounding required? Check for argument reduction!
1704 if (Matcher::strict_fp_requires_explicit_rounding) {
1705 static const double pi_4 = 0.7853981633974483;
1706 static const double neg_pi_4 = -0.7853981633974483;
1707 // pi/2 in 80-bit extended precision
1708 // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
1709 // -pi/2 in 80-bit extended precision
1710 // 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};
1711 // Cutoff value for using this argument reduction technique
1712 //static const double pi_2_minus_epsilon = 1.564660403643354;
1713 //static const double neg_pi_2_plus_epsilon = -1.564660403643354;
1714
1715 // Pseudocode for sin:
1716 // if (x <= Math.PI / 4.0) {
1717 // if (x >= -Math.PI / 4.0) return fsin(x);
1718 // if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
1719 // } else {
1720 // if (x <= Math.PI / 2.0) return fcos(x - Math.PI / 2.0);
1721 // }
1722 // return StrictMath.sin(x);
1723
1724 // Pseudocode for cos:
1725 // if (x <= Math.PI / 4.0) {
1726 // if (x >= -Math.PI / 4.0) return fcos(x);
1727 // if (x >= -Math.PI / 2.0) return fsin(x + Math.PI / 2.0);
1728 // } else {
1729 // if (x <= Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
1730 // }
1731 // return StrictMath.cos(x);
1732
1733 // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
1734 // requires a special machine instruction to load it. Instead we'll try
1735 // the 'easy' case. If we really need the extra range +/- PI/2 we'll
1736 // probably do the math inside the SIN encoding.
1737
1738 // Make the merge point
1739 RegionNode* r = new RegionNode(3);
1740 Node* phi = new PhiNode(r, Type::DOUBLE);
1741
1742 // Flatten arg so we need only 1 test
1743 Node *abs = _gvn.transform(new AbsDNode(arg));
1744 // Node for PI/4 constant
1745 Node *pi4 = makecon(TypeD::make(pi_4));
1746 // Check PI/4 : abs(arg)
1747 Node *cmp = _gvn.transform(new CmpDNode(pi4,abs));
1748 // Check: If PI/4 < abs(arg) then go slow
1749 Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt ));
1750 // Branch either way
1751 IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1752 set_control(opt_iff(r,iff));
1753
1754 // Set fast path result
1755 phi->init_req(2, n);
1756
1757 // Slow path - non-blocking leaf call
1758 Node* call = NULL;
1759 switch (id) {
1760 case vmIntrinsics::_dtan:
1761 call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1762 CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
1763 "Tan", NULL, arg, top());
1764 break;
1765 }
1766 assert(control()->in(0) == call, "");
1767 Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1768 r->init_req(1, control());
1769 phi->init_req(1, slow_result);
1770
1771 // Post-merge
1772 set_control(_gvn.transform(r));
1773 record_for_igvn(r);
1774 n = _gvn.transform(phi);
1775
1776 C->set_has_split_ifs(true); // Has chance for split-if optimization
1777 }
1778 set_result(n);
1779 return true;
1780 }
1781
1782 //------------------------------runtime_math-----------------------------
1783 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1784 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1785 "must be (DD)D or (D)D type");
1786
1787 // Inputs
1788 Node* a = round_double_node(argument(0));
1789 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1790
1791 const TypePtr* no_memory_effects = NULL;
1792 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1793 no_memory_effects,
1794 a, top(), b, b ? top() : NULL);
1795 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1796 #ifdef ASSERT
1797 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1798 assert(value_top == top(), "second value must be top");
1799 #endif
1800
1801 set_result(value);
1802 return true;
1803 }
1804
1805 //------------------------------inline_math_native-----------------------------
1806 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1807 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1808 switch (id) {
1809 // These intrinsics are not properly supported on all hardware
1810 case vmIntrinsics::_dsin:
1811 return StubRoutines::dsin() != NULL ?
1812 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1813 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1814 case vmIntrinsics::_dcos:
1815 return StubRoutines::dcos() != NULL ?
1816 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1817 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1818 case vmIntrinsics::_dtan:
1819 return StubRoutines::dtan() != NULL ?
1820 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1821 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1822 case vmIntrinsics::_dlog:
1823 return StubRoutines::dlog() != NULL ?
1824 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1825 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1826 case vmIntrinsics::_dlog10:
1827 return StubRoutines::dlog10() != NULL ?
1828 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1829 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1830
1831 // These intrinsics are supported on all hardware
1832 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1833 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1834
1835 case vmIntrinsics::_dexp:
1836 return StubRoutines::dexp() != NULL ?
1837 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1838 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1839 case vmIntrinsics::_dpow:
1840 return StubRoutines::dpow() != NULL ?
1841 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1842 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1843 #undef FN_PTR
1844
1845 // These intrinsics are not yet correctly implemented
1846 case vmIntrinsics::_datan2:
1847 return false;
1848
1849 default:
1850 fatal_unexpected_iid(id);
1851 return false;
1852 }
1853 }
1854
1855 static bool is_simple_name(Node* n) {
1856 return (n->req() == 1 // constant
1857 || (n->is_Type() && n->as_Type()->type()->singleton())
1858 || n->is_Proj() // parameter or return value
1859 || n->is_Phi() // local of some sort
1860 );
1861 }
1862
1863 //----------------------------inline_notify-----------------------------------*
1864 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1865 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1866 address func;
1867 if (id == vmIntrinsics::_notify) {
1868 func = OptoRuntime::monitor_notify_Java();
1869 } else {
1870 func = OptoRuntime::monitor_notifyAll_Java();
1871 }
1872 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1873 make_slow_call_ex(call, env()->Throwable_klass(), false);
1874 return true;
1875 }
1876
1877
1878 //----------------------------inline_min_max-----------------------------------
1879 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1880 set_result(generate_min_max(id, argument(0), argument(1)));
1881 return true;
1882 }
1883
1884 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1885 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1886 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1887 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1888 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1889
1890 {
1891 PreserveJVMState pjvms(this);
1892 PreserveReexecuteState preexecs(this);
1893 jvms()->set_should_reexecute(true);
1894
1895 set_control(slow_path);
1896 set_i_o(i_o());
1897
1898 uncommon_trap(Deoptimization::Reason_intrinsic,
1899 Deoptimization::Action_none);
1900 }
1901
1902 set_control(fast_path);
1903 set_result(math);
1904 }
1905
1906 template <typename OverflowOp>
1907 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1908 typedef typename OverflowOp::MathOp MathOp;
1909
1910 MathOp* mathOp = new MathOp(arg1, arg2);
1911 Node* operation = _gvn.transform( mathOp );
1912 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1913 inline_math_mathExact(operation, ofcheck);
1914 return true;
1915 }
1916
1917 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1918 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1919 }
1920
1921 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1922 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1923 }
1924
1925 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1926 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1927 }
1928
1929 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1930 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1931 }
1932
1933 bool LibraryCallKit::inline_math_negateExactI() {
1934 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1935 }
1936
1937 bool LibraryCallKit::inline_math_negateExactL() {
1938 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1939 }
1940
1941 bool LibraryCallKit::inline_math_multiplyExactI() {
1942 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1943 }
1944
1945 bool LibraryCallKit::inline_math_multiplyExactL() {
1946 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1947 }
1948
1949 Node*
1950 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1951 // These are the candidate return value:
1952 Node* xvalue = x0;
1953 Node* yvalue = y0;
1954
1955 if (xvalue == yvalue) {
1956 return xvalue;
1957 }
1958
1959 bool want_max = (id == vmIntrinsics::_max);
1960
1961 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1962 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1963 if (txvalue == NULL || tyvalue == NULL) return top();
1964 // This is not really necessary, but it is consistent with a
1965 // hypothetical MaxINode::Value method:
1966 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1967
1968 // %%% This folding logic should (ideally) be in a different place.
1969 // Some should be inside IfNode, and there to be a more reliable
1970 // transformation of ?: style patterns into cmoves. We also want
1971 // more powerful optimizations around cmove and min/max.
1972
1973 // Try to find a dominating comparison of these guys.
1974 // It can simplify the index computation for Arrays.copyOf
1975 // and similar uses of System.arraycopy.
1976 // First, compute the normalized version of CmpI(x, y).
1977 int cmp_op = Op_CmpI;
1978 Node* xkey = xvalue;
1979 Node* ykey = yvalue;
1980 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1981 if (ideal_cmpxy->is_Cmp()) {
1982 // E.g., if we have CmpI(length - offset, count),
1983 // it might idealize to CmpI(length, count + offset)
1984 cmp_op = ideal_cmpxy->Opcode();
1985 xkey = ideal_cmpxy->in(1);
1986 ykey = ideal_cmpxy->in(2);
1987 }
1988
1989 // Start by locating any relevant comparisons.
1990 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1991 Node* cmpxy = NULL;
1992 Node* cmpyx = NULL;
1993 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1994 Node* cmp = start_from->fast_out(k);
1995 if (cmp->outcnt() > 0 && // must have prior uses
1996 cmp->in(0) == NULL && // must be context-independent
1997 cmp->Opcode() == cmp_op) { // right kind of compare
1998 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1999 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2000 }
2001 }
2002
2003 const int NCMPS = 2;
2004 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2005 int cmpn;
2006 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2007 if (cmps[cmpn] != NULL) break; // find a result
2008 }
2009 if (cmpn < NCMPS) {
2010 // Look for a dominating test that tells us the min and max.
2011 int depth = 0; // Limit search depth for speed
2012 Node* dom = control();
2013 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2014 if (++depth >= 100) break;
2015 Node* ifproj = dom;
2016 if (!ifproj->is_Proj()) continue;
2017 Node* iff = ifproj->in(0);
2018 if (!iff->is_If()) continue;
2019 Node* bol = iff->in(1);
2020 if (!bol->is_Bool()) continue;
2021 Node* cmp = bol->in(1);
2022 if (cmp == NULL) continue;
2023 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2024 if (cmps[cmpn] == cmp) break;
2025 if (cmpn == NCMPS) continue;
2026 BoolTest::mask btest = bol->as_Bool()->_test._test;
2027 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2028 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2029 // At this point, we know that 'x btest y' is true.
2030 switch (btest) {
2031 case BoolTest::eq:
2032 // They are proven equal, so we can collapse the min/max.
2033 // Either value is the answer. Choose the simpler.
2034 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2035 return yvalue;
2036 return xvalue;
2037 case BoolTest::lt: // x < y
2038 case BoolTest::le: // x <= y
2039 return (want_max ? yvalue : xvalue);
2040 case BoolTest::gt: // x > y
2041 case BoolTest::ge: // x >= y
2042 return (want_max ? xvalue : yvalue);
2043 }
2044 }
2045 }
2046
2047 // We failed to find a dominating test.
2048 // Let's pick a test that might GVN with prior tests.
2049 Node* best_bol = NULL;
2050 BoolTest::mask best_btest = BoolTest::illegal;
2051 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2052 Node* cmp = cmps[cmpn];
2053 if (cmp == NULL) continue;
2054 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2055 Node* bol = cmp->fast_out(j);
2056 if (!bol->is_Bool()) continue;
2057 BoolTest::mask btest = bol->as_Bool()->_test._test;
2058 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2059 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2060 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2061 best_bol = bol->as_Bool();
2062 best_btest = btest;
2063 }
2064 }
2065 }
2066
2067 Node* answer_if_true = NULL;
2068 Node* answer_if_false = NULL;
2069 switch (best_btest) {
2070 default:
2071 if (cmpxy == NULL)
2072 cmpxy = ideal_cmpxy;
2073 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2074 // and fall through:
2075 case BoolTest::lt: // x < y
2076 case BoolTest::le: // x <= y
2077 answer_if_true = (want_max ? yvalue : xvalue);
2078 answer_if_false = (want_max ? xvalue : yvalue);
2079 break;
2080 case BoolTest::gt: // x > y
2081 case BoolTest::ge: // x >= y
2082 answer_if_true = (want_max ? xvalue : yvalue);
2083 answer_if_false = (want_max ? yvalue : xvalue);
2084 break;
2085 }
2086
2087 jint hi, lo;
2088 if (want_max) {
2089 // We can sharpen the minimum.
2090 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2091 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2092 } else {
2093 // We can sharpen the maximum.
2094 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2095 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2096 }
2097
2098 // Use a flow-free graph structure, to avoid creating excess control edges
2099 // which could hinder other optimizations.
2100 // Since Math.min/max is often used with arraycopy, we want
2101 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2102 Node* cmov = CMoveNode::make(NULL, best_bol,
2103 answer_if_false, answer_if_true,
2104 TypeInt::make(lo, hi, widen));
2105
2106 return _gvn.transform(cmov);
2107
2108 /*
2109 // This is not as desirable as it may seem, since Min and Max
2110 // nodes do not have a full set of optimizations.
2111 // And they would interfere, anyway, with 'if' optimizations
2112 // and with CMoveI canonical forms.
2113 switch (id) {
2114 case vmIntrinsics::_min:
2115 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2116 case vmIntrinsics::_max:
2117 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2118 default:
2119 ShouldNotReachHere();
2120 }
2121 */
2122 }
2123
2124 inline int
2125 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2126 const TypePtr* base_type = TypePtr::NULL_PTR;
2127 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2128 if (base_type == NULL) {
2129 // Unknown type.
2130 return Type::AnyPtr;
2131 } else if (base_type == TypePtr::NULL_PTR) {
2132 // Since this is a NULL+long form, we have to switch to a rawptr.
2133 base = _gvn.transform(new CastX2PNode(offset));
2134 offset = MakeConX(0);
2135 return Type::RawPtr;
2136 } else if (base_type->base() == Type::RawPtr) {
2137 return Type::RawPtr;
2138 } else if (base_type->isa_oopptr()) {
2139 // Base is never null => always a heap address.
2140 if (base_type->ptr() == TypePtr::NotNull) {
2141 return Type::OopPtr;
2142 }
2143 // Offset is small => always a heap address.
2144 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2145 if (offset_type != NULL &&
2146 base_type->offset() == 0 && // (should always be?)
2147 offset_type->_lo >= 0 &&
2148 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2149 return Type::OopPtr;
2150 }
2151 // Otherwise, it might either be oop+off or NULL+addr.
2152 return Type::AnyPtr;
2153 } else {
2154 // No information:
2155 return Type::AnyPtr;
2156 }
2157 }
2158
2159 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2160 int kind = classify_unsafe_addr(base, offset);
2161 if (kind == Type::RawPtr) {
2162 return basic_plus_adr(top(), base, offset);
2163 } else {
2164 return basic_plus_adr(base, offset);
2165 }
2166 }
2167
2168 //--------------------------inline_number_methods-----------------------------
2169 // inline int Integer.numberOfLeadingZeros(int)
2170 // inline int Long.numberOfLeadingZeros(long)
2171 //
2172 // inline int Integer.numberOfTrailingZeros(int)
2173 // inline int Long.numberOfTrailingZeros(long)
2174 //
2175 // inline int Integer.bitCount(int)
2176 // inline int Long.bitCount(long)
2177 //
2178 // inline char Character.reverseBytes(char)
2179 // inline short Short.reverseBytes(short)
2180 // inline int Integer.reverseBytes(int)
2181 // inline long Long.reverseBytes(long)
2182 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2183 Node* arg = argument(0);
2184 Node* n = NULL;
2185 switch (id) {
2186 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2187 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2188 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2189 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2190 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2191 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2192 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2193 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2194 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2195 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2196 default: fatal_unexpected_iid(id); break;
2197 }
2198 set_result(_gvn.transform(n));
2199 return true;
2200 }
2201
2202 //----------------------------inline_unsafe_access----------------------------
2203
2204 // Helper that guards and inserts a pre-barrier.
2205 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2206 Node* pre_val, bool need_mem_bar) {
2207 // We could be accessing the referent field of a reference object. If so, when G1
2208 // is enabled, we need to log the value in the referent field in an SATB buffer.
2209 // This routine performs some compile time filters and generates suitable
2210 // runtime filters that guard the pre-barrier code.
2211 // Also add memory barrier for non volatile load from the referent field
2212 // to prevent commoning of loads across safepoint.
2213 if (!UseG1GC && !need_mem_bar)
2214 return;
2215
2216 // Some compile time checks.
2217
2218 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2219 const TypeX* otype = offset->find_intptr_t_type();
2220 if (otype != NULL && otype->is_con() &&
2221 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2222 // Constant offset but not the reference_offset so just return
2223 return;
2224 }
2225
2226 // We only need to generate the runtime guards for instances.
2227 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2228 if (btype != NULL) {
2229 if (btype->isa_aryptr()) {
2230 // Array type so nothing to do
2231 return;
2232 }
2233
2234 const TypeInstPtr* itype = btype->isa_instptr();
2235 if (itype != NULL) {
2236 // Can the klass of base_oop be statically determined to be
2237 // _not_ a sub-class of Reference and _not_ Object?
2238 ciKlass* klass = itype->klass();
2239 if ( klass->is_loaded() &&
2240 !klass->is_subtype_of(env()->Reference_klass()) &&
2241 !env()->Object_klass()->is_subtype_of(klass)) {
2242 return;
2243 }
2244 }
2245 }
2246
2247 // The compile time filters did not reject base_oop/offset so
2248 // we need to generate the following runtime filters
2249 //
2250 // if (offset == java_lang_ref_Reference::_reference_offset) {
2251 // if (instance_of(base, java.lang.ref.Reference)) {
2252 // pre_barrier(_, pre_val, ...);
2253 // }
2254 // }
2255
2256 float likely = PROB_LIKELY( 0.999);
2257 float unlikely = PROB_UNLIKELY(0.999);
2258
2259 IdealKit ideal(this);
2260 #define __ ideal.
2261
2262 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2263
2264 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2265 // Update graphKit memory and control from IdealKit.
2266 sync_kit(ideal);
2267
2268 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2269 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2270
2271 // Update IdealKit memory and control from graphKit.
2272 __ sync_kit(this);
2273
2274 Node* one = __ ConI(1);
2275 // is_instof == 0 if base_oop == NULL
2276 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2277
2278 // Update graphKit from IdeakKit.
2279 sync_kit(ideal);
2280
2281 // Use the pre-barrier to record the value in the referent field
2282 pre_barrier(false /* do_load */,
2283 __ ctrl(),
2284 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2285 pre_val /* pre_val */,
2286 T_OBJECT);
2287 if (need_mem_bar) {
2288 // Add memory barrier to prevent commoning reads from this field
2289 // across safepoint since GC can change its value.
2290 insert_mem_bar(Op_MemBarCPUOrder);
2291 }
2292 // Update IdealKit from graphKit.
2293 __ sync_kit(this);
2294
2295 } __ end_if(); // _ref_type != ref_none
2296 } __ end_if(); // offset == referent_offset
2297
2298 // Final sync IdealKit and GraphKit.
2299 final_sync(ideal);
2300 #undef __
2301 }
2302
2303
2304 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2305 // Attempt to infer a sharper value type from the offset and base type.
2306 ciKlass* sharpened_klass = NULL;
2307
2308 // See if it is an instance field, with an object type.
2309 if (alias_type->field() != NULL) {
2310 if (alias_type->field()->type()->is_klass()) {
2311 sharpened_klass = alias_type->field()->type()->as_klass();
2312 }
2313 }
2314
2315 // See if it is a narrow oop array.
2316 if (adr_type->isa_aryptr()) {
2317 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2318 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2319 if (elem_type != NULL) {
2320 sharpened_klass = elem_type->klass();
2321 }
2322 }
2323 }
2324
2325 // The sharpened class might be unloaded if there is no class loader
2326 // contraint in place.
2327 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2328 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2329
2330 #ifndef PRODUCT
2331 if (C->print_intrinsics() || C->print_inlining()) {
2332 tty->print(" from base type: "); adr_type->dump();
2333 tty->print(" sharpened value: "); tjp->dump();
2334 }
2335 #endif
2336 // Sharpen the value type.
2337 return tjp;
2338 }
2339 return NULL;
2340 }
2341
2342 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2343 if (callee()->is_static()) return false; // caller must have the capability!
2344 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2345 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2346 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2347
2348 #ifndef PRODUCT
2349 {
2350 ResourceMark rm;
2351 // Check the signatures.
2352 ciSignature* sig = callee()->signature();
2353 #ifdef ASSERT
2354 if (!is_store) {
2355 // Object getObject(Object base, int/long offset), etc.
2356 BasicType rtype = sig->return_type()->basic_type();
2357 assert(rtype == type, "getter must return the expected value");
2358 assert(sig->count() == 2, "oop getter has 2 arguments");
2359 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2360 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2361 } else {
2362 // void putObject(Object base, int/long offset, Object x), etc.
2363 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2364 assert(sig->count() == 3, "oop putter has 3 arguments");
2365 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2366 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2367 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2368 assert(vtype == type, "putter must accept the expected value");
2369 }
2370 #endif // ASSERT
2371 }
2372 #endif //PRODUCT
2373
2374 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2375
2376 Node* receiver = argument(0); // type: oop
2377
2378 // Build address expression.
2379 Node* adr;
2380 Node* heap_base_oop = top();
2381 Node* offset = top();
2382 Node* val;
2383
2384 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2385 Node* base = argument(1); // type: oop
2386 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2387 offset = argument(2); // type: long
2388 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2389 // to be plain byte offsets, which are also the same as those accepted
2390 // by oopDesc::field_base.
2391 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2392 "fieldOffset must be byte-scaled");
2393 // 32-bit machines ignore the high half!
2394 offset = ConvL2X(offset);
2395 adr = make_unsafe_address(base, offset);
2396 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2397 heap_base_oop = base;
2398 }
2399 val = is_store ? argument(4) : NULL;
2400
2401 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2402
2403 // Try to categorize the address. If it comes up as TypeJavaPtr::BOTTOM,
2404 // there was not enough information to nail it down.
2405 Compile::AliasType* alias_type = C->alias_type(adr_type);
2406 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2407
2408 assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2409 alias_type->basic_type() != T_ILLEGAL, "field, array element or unknown");
2410 bool mismatched = false;
2411 BasicType bt = alias_type->basic_type();
2412 if (bt != T_ILLEGAL) {
2413 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2414 // Alias type doesn't differentiate between byte[] and boolean[]).
2415 // Use address type to get the element type.
2416 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2417 }
2418 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2419 // accessing an array field with getObject is not a mismatch
2420 bt = T_OBJECT;
2421 }
2422 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2423 // Don't intrinsify mismatched object accesses
2424 return false;
2425 }
2426 mismatched = (bt != type);
2427 }
2428
2429 // First guess at the value type.
2430 const Type *value_type = Type::get_const_basic_type(type);
2431
2432 // We will need memory barriers unless we can determine a unique
2433 // alias category for this reference. (Note: If for some reason
2434 // the barriers get omitted and the unsafe reference begins to "pollute"
2435 // the alias analysis of the rest of the graph, either Compile::can_alias
2436 // or Compile::must_alias will throw a diagnostic assert.)
2437 bool need_mem_bar;
2438 switch (kind) {
2439 case Relaxed:
2440 need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2441 break;
2442 case Opaque:
2443 // Opaque uses CPUOrder membars for protection against code movement.
2444 case Acquire:
2445 case Release:
2446 case Volatile:
2447 need_mem_bar = true;
2448 break;
2449 default:
2450 ShouldNotReachHere();
2451 }
2452
2453 // Some accesses require access atomicity for all types, notably longs and doubles.
2454 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2455 bool requires_atomic_access = false;
2456 switch (kind) {
2457 case Relaxed:
2458 requires_atomic_access = AlwaysAtomicAccesses;
2459 break;
2460 case Opaque:
2461 // Opaque accesses are atomic.
2462 case Acquire:
2463 case Release:
2464 case Volatile:
2465 requires_atomic_access = true;
2466 break;
2467 default:
2468 ShouldNotReachHere();
2469 }
2470
2471 // Figure out the memory ordering.
2472 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2473 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2474
2475 // If we are reading the value of the referent field of a Reference
2476 // object (either by using Unsafe directly or through reflection)
2477 // then, if G1 is enabled, we need to record the referent in an
2478 // SATB log buffer using the pre-barrier mechanism.
2479 // Also we need to add memory barrier to prevent commoning reads
2480 // from this field across safepoint since GC can change its value.
2481 bool need_read_barrier = !is_store &&
2482 offset != top() && heap_base_oop != top();
2483
2484 if (!is_store && type == T_OBJECT) {
2485 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2486 if (tjp != NULL) {
2487 value_type = tjp;
2488 }
2489 }
2490
2491 receiver = null_check(receiver);
2492 if (stopped()) {
2493 return true;
2494 }
2495 // Heap pointers get a null-check from the interpreter,
2496 // as a courtesy. However, this is not guaranteed by Unsafe,
2497 // and it is not possible to fully distinguish unintended nulls
2498 // from intended ones in this API.
2499
2500 // We need to emit leading and trailing CPU membars (see below) in
2501 // addition to memory membars for special access modes. This is a little
2502 // too strong, but avoids the need to insert per-alias-type
2503 // volatile membars (for stores; compare Parse::do_put_xxx), which
2504 // we cannot do effectively here because we probably only have a
2505 // rough approximation of type.
2506
2507 switch(kind) {
2508 case Relaxed:
2509 case Opaque:
2510 case Acquire:
2511 break;
2512 case Release:
2513 case Volatile:
2514 if (is_store) {
2515 insert_mem_bar(Op_MemBarRelease);
2516 } else {
2517 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2518 insert_mem_bar(Op_MemBarVolatile);
2519 }
2520 }
2521 break;
2522 default:
2523 ShouldNotReachHere();
2524 }
2525
2526 // Memory barrier to prevent normal and 'unsafe' accesses from
2527 // bypassing each other. Happens after null checks, so the
2528 // exception paths do not take memory state from the memory barrier,
2529 // so there's no problems making a strong assert about mixing users
2530 // of safe & unsafe memory.
2531 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2532
2533 if (!is_store) {
2534 Node* p = NULL;
2535 // Try to constant fold a load from a constant field
2536 ciField* field = alias_type->field();
2537 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2538 // final or stable field
2539 p = make_constant_from_field(field, heap_base_oop);
2540 }
2541 if (p == NULL) {
2542 // To be valid, unsafe loads may depend on other conditions than
2543 // the one that guards them: pin the Load node
2544 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2545 // load value
2546 switch (type) {
2547 case T_BOOLEAN:
2548 case T_CHAR:
2549 case T_BYTE:
2550 case T_SHORT:
2551 case T_INT:
2552 case T_LONG:
2553 case T_FLOAT:
2554 case T_DOUBLE:
2555 break;
2556 case T_OBJECT:
2557 if (need_read_barrier) {
2558 // We do not require a mem bar inside pre_barrier if need_mem_bar
2559 // is set: the barriers would be emitted by us.
2560 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2561 }
2562 break;
2563 case T_ADDRESS:
2564 // Cast to an int type.
2565 p = _gvn.transform(new CastP2XNode(NULL, p));
2566 p = ConvX2UL(p);
2567 break;
2568 default:
2569 fatal("unexpected type %d: %s", type, type2name(type));
2570 break;
2571 }
2572 }
2573 // The load node has the control of the preceding MemBarCPUOrder. All
2574 // following nodes will have the control of the MemBarCPUOrder inserted at
2575 // the end of this method. So, pushing the load onto the stack at a later
2576 // point is fine.
2577 set_result(p);
2578 } else {
2579 // place effect of store into memory
2580 switch (type) {
2581 case T_DOUBLE:
2582 val = dstore_rounding(val);
2583 break;
2584 case T_ADDRESS:
2585 // Repackage the long as a pointer.
2586 val = ConvL2X(val);
2587 val = _gvn.transform(new CastX2PNode(val));
2588 break;
2589 }
2590
2591 if (type != T_OBJECT) {
2592 (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2593 } else {
2594 // Possibly an oop being stored to Java heap or native memory
2595 if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2596 // oop to Java heap.
2597 (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2598 } else {
2599 // We can't tell at compile time if we are storing in the Java heap or outside
2600 // of it. So we need to emit code to conditionally do the proper type of
2601 // store.
2602
2603 IdealKit ideal(this);
2604 #define __ ideal.
2605 // QQQ who knows what probability is here??
2606 __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2607 // Sync IdealKit and graphKit.
2608 sync_kit(ideal);
2609 Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2610 // Update IdealKit memory.
2611 __ sync_kit(this);
2612 } __ else_(); {
2613 __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched);
2614 } __ end_if();
2615 // Final sync IdealKit and GraphKit.
2616 final_sync(ideal);
2617 #undef __
2618 }
2619 }
2620 }
2621
2622 switch(kind) {
2623 case Relaxed:
2624 case Opaque:
2625 case Release:
2626 break;
2627 case Acquire:
2628 case Volatile:
2629 if (!is_store) {
2630 insert_mem_bar(Op_MemBarAcquire);
2631 } else {
2632 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2633 insert_mem_bar(Op_MemBarVolatile);
2634 }
2635 }
2636 break;
2637 default:
2638 ShouldNotReachHere();
2639 }
2640
2641 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2642
2643 return true;
2644 }
2645
2646 //----------------------------inline_unsafe_load_store----------------------------
2647 // This method serves a couple of different customers (depending on LoadStoreKind):
2648 //
2649 // LS_cmp_swap:
2650 //
2651 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2652 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2653 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2654 //
2655 // LS_cmp_swap_weak:
2656 //
2657 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2658 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2659 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2660 //
2661 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2662 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2663 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2664 //
2665 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2666 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2667 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2668 //
2669 // LS_cmp_exchange:
2670 //
2671 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2672 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2673 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2674 //
2675 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2676 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2677 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2678 //
2679 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2680 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2681 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2682 //
2683 // LS_get_add:
2684 //
2685 // int getAndAddInt( Object o, long offset, int delta)
2686 // long getAndAddLong(Object o, long offset, long delta)
2687 //
2688 // LS_get_set:
2689 //
2690 // int getAndSet(Object o, long offset, int newValue)
2691 // long getAndSet(Object o, long offset, long newValue)
2692 // Object getAndSet(Object o, long offset, Object newValue)
2693 //
2694 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2695 // This basic scheme here is the same as inline_unsafe_access, but
2696 // differs in enough details that combining them would make the code
2697 // overly confusing. (This is a true fact! I originally combined
2698 // them, but even I was confused by it!) As much code/comments as
2699 // possible are retained from inline_unsafe_access though to make
2700 // the correspondences clearer. - dl
2701
2702 if (callee()->is_static()) return false; // caller must have the capability!
2703
2704 #ifndef PRODUCT
2705 BasicType rtype;
2706 {
2707 ResourceMark rm;
2708 // Check the signatures.
2709 ciSignature* sig = callee()->signature();
2710 rtype = sig->return_type()->basic_type();
2711 switch(kind) {
2712 case LS_get_add:
2713 case LS_get_set: {
2714 // Check the signatures.
2715 #ifdef ASSERT
2716 assert(rtype == type, "get and set must return the expected type");
2717 assert(sig->count() == 3, "get and set has 3 arguments");
2718 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2719 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2720 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2721 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2722 #endif // ASSERT
2723 break;
2724 }
2725 case LS_cmp_swap:
2726 case LS_cmp_swap_weak: {
2727 // Check the signatures.
2728 #ifdef ASSERT
2729 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2730 assert(sig->count() == 4, "CAS has 4 arguments");
2731 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2732 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2733 #endif // ASSERT
2734 break;
2735 }
2736 case LS_cmp_exchange: {
2737 // Check the signatures.
2738 #ifdef ASSERT
2739 assert(rtype == type, "CAS must return the expected type");
2740 assert(sig->count() == 4, "CAS has 4 arguments");
2741 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2742 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2743 #endif // ASSERT
2744 break;
2745 }
2746 default:
2747 ShouldNotReachHere();
2748 }
2749 }
2750 #endif //PRODUCT
2751
2752 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2753
2754 // Get arguments:
2755 Node* receiver = NULL;
2756 Node* base = NULL;
2757 Node* offset = NULL;
2758 Node* oldval = NULL;
2759 Node* newval = NULL;
2760 switch(kind) {
2761 case LS_cmp_swap:
2762 case LS_cmp_swap_weak:
2763 case LS_cmp_exchange: {
2764 const bool two_slot_type = type2size[type] == 2;
2765 receiver = argument(0); // type: oop
2766 base = argument(1); // type: oop
2767 offset = argument(2); // type: long
2768 oldval = argument(4); // type: oop, int, or long
2769 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2770 break;
2771 }
2772 case LS_get_add:
2773 case LS_get_set: {
2774 receiver = argument(0); // type: oop
2775 base = argument(1); // type: oop
2776 offset = argument(2); // type: long
2777 oldval = NULL;
2778 newval = argument(4); // type: oop, int, or long
2779 break;
2780 }
2781 default:
2782 ShouldNotReachHere();
2783 }
2784
2785 // Null check receiver.
2786 receiver = null_check(receiver);
2787 if (stopped()) {
2788 return true;
2789 }
2790
2791 // Build field offset expression.
2792 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2793 // to be plain byte offsets, which are also the same as those accepted
2794 // by oopDesc::field_base.
2795 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2796 // 32-bit machines ignore the high half of long offsets
2797 offset = ConvL2X(offset);
2798 Node* adr = make_unsafe_address(base, offset);
2799 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2800
2801 Compile::AliasType* alias_type = C->alias_type(adr_type);
2802 assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2803 alias_type->basic_type() != T_ILLEGAL, "field, array element or unknown");
2804 BasicType bt = alias_type->basic_type();
2805 if (bt != T_ILLEGAL &&
2806 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2807 // Don't intrinsify mismatched object accesses.
2808 return false;
2809 }
2810
2811 // For CAS, unlike inline_unsafe_access, there seems no point in
2812 // trying to refine types. Just use the coarse types here.
2813 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2814 const Type *value_type = Type::get_const_basic_type(type);
2815
2816 switch (kind) {
2817 case LS_get_set:
2818 case LS_cmp_exchange: {
2819 if (type == T_OBJECT) {
2820 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2821 if (tjp != NULL) {
2822 value_type = tjp;
2823 }
2824 }
2825 break;
2826 }
2827 case LS_cmp_swap:
2828 case LS_cmp_swap_weak:
2829 case LS_get_add:
2830 break;
2831 default:
2832 ShouldNotReachHere();
2833 }
2834
2835 int alias_idx = C->get_alias_index(adr_type);
2836
2837 // Memory-model-wise, a LoadStore acts like a little synchronized
2838 // block, so needs barriers on each side. These don't translate
2839 // into actual barriers on most machines, but we still need rest of
2840 // compiler to respect ordering.
2841
2842 switch (access_kind) {
2843 case Relaxed:
2844 case Acquire:
2845 break;
2846 case Release:
2847 insert_mem_bar(Op_MemBarRelease);
2848 break;
2849 case Volatile:
2850 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2851 insert_mem_bar(Op_MemBarVolatile);
2852 } else {
2853 insert_mem_bar(Op_MemBarRelease);
2854 }
2855 break;
2856 default:
2857 ShouldNotReachHere();
2858 }
2859 insert_mem_bar(Op_MemBarCPUOrder);
2860
2861 // Figure out the memory ordering.
2862 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2863
2864 // 4984716: MemBars must be inserted before this
2865 // memory node in order to avoid a false
2866 // dependency which will confuse the scheduler.
2867 Node *mem = memory(alias_idx);
2868
2869 // For now, we handle only those cases that actually exist: ints,
2870 // longs, and Object. Adding others should be straightforward.
2871 Node* load_store = NULL;
2872 switch(type) {
2873 case T_BYTE:
2874 switch(kind) {
2875 case LS_get_add:
2876 load_store = _gvn.transform(new GetAndAddBNode(control(), mem, adr, newval, adr_type));
2877 break;
2878 case LS_get_set:
2879 load_store = _gvn.transform(new GetAndSetBNode(control(), mem, adr, newval, adr_type));
2880 break;
2881 case LS_cmp_swap_weak:
2882 load_store = _gvn.transform(new WeakCompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2883 break;
2884 case LS_cmp_swap:
2885 load_store = _gvn.transform(new CompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2886 break;
2887 case LS_cmp_exchange:
2888 load_store = _gvn.transform(new CompareAndExchangeBNode(control(), mem, adr, newval, oldval, adr_type, mo));
2889 break;
2890 default:
2891 ShouldNotReachHere();
2892 }
2893 break;
2894 case T_SHORT:
2895 switch(kind) {
2896 case LS_get_add:
2897 load_store = _gvn.transform(new GetAndAddSNode(control(), mem, adr, newval, adr_type));
2898 break;
2899 case LS_get_set:
2900 load_store = _gvn.transform(new GetAndSetSNode(control(), mem, adr, newval, adr_type));
2901 break;
2902 case LS_cmp_swap_weak:
2903 load_store = _gvn.transform(new WeakCompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2904 break;
2905 case LS_cmp_swap:
2906 load_store = _gvn.transform(new CompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2907 break;
2908 case LS_cmp_exchange:
2909 load_store = _gvn.transform(new CompareAndExchangeSNode(control(), mem, adr, newval, oldval, adr_type, mo));
2910 break;
2911 default:
2912 ShouldNotReachHere();
2913 }
2914 break;
2915 case T_INT:
2916 switch(kind) {
2917 case LS_get_add:
2918 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2919 break;
2920 case LS_get_set:
2921 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2922 break;
2923 case LS_cmp_swap_weak:
2924 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2925 break;
2926 case LS_cmp_swap:
2927 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2928 break;
2929 case LS_cmp_exchange:
2930 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2931 break;
2932 default:
2933 ShouldNotReachHere();
2934 }
2935 break;
2936 case T_LONG:
2937 switch(kind) {
2938 case LS_get_add:
2939 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2940 break;
2941 case LS_get_set:
2942 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2943 break;
2944 case LS_cmp_swap_weak:
2945 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2946 break;
2947 case LS_cmp_swap:
2948 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2949 break;
2950 case LS_cmp_exchange:
2951 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2952 break;
2953 default:
2954 ShouldNotReachHere();
2955 }
2956 break;
2957 case T_OBJECT:
2958 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2959 // could be delayed during Parse (for example, in adjust_map_after_if()).
2960 // Execute transformation here to avoid barrier generation in such case.
2961 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2962 newval = _gvn.makecon(TypePtr::NULL_PTR);
2963
2964 // Reference stores need a store barrier.
2965 switch(kind) {
2966 case LS_get_set: {
2967 // If pre-barrier must execute before the oop store, old value will require do_load here.
2968 if (!can_move_pre_barrier()) {
2969 pre_barrier(true /* do_load*/,
2970 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2971 NULL /* pre_val*/,
2972 T_OBJECT);
2973 } // Else move pre_barrier to use load_store value, see below.
2974 break;
2975 }
2976 case LS_cmp_swap_weak:
2977 case LS_cmp_swap:
2978 case LS_cmp_exchange: {
2979 // Same as for newval above:
2980 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2981 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2982 }
2983 // The only known value which might get overwritten is oldval.
2984 pre_barrier(false /* do_load */,
2985 control(), NULL, NULL, max_juint, NULL, NULL,
2986 oldval /* pre_val */,
2987 T_OBJECT);
2988 break;
2989 }
2990 default:
2991 ShouldNotReachHere();
2992 }
2993
2994 #ifdef _LP64
2995 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2996 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2997
2998 switch(kind) {
2999 case LS_get_set:
3000 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
3001 break;
3002 case LS_cmp_swap_weak: {
3003 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
3004 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
3005 break;
3006 }
3007 case LS_cmp_swap: {
3008 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
3009 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
3010 break;
3011 }
3012 case LS_cmp_exchange: {
3013 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
3014 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
3015 break;
3016 }
3017 default:
3018 ShouldNotReachHere();
3019 }
3020 } else
3021 #endif
3022 switch (kind) {
3023 case LS_get_set:
3024 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
3025 break;
3026 case LS_cmp_swap_weak:
3027 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
3028 break;
3029 case LS_cmp_swap:
3030 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
3031 break;
3032 case LS_cmp_exchange:
3033 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
3034 break;
3035 default:
3036 ShouldNotReachHere();
3037 }
3038
3039 // Emit the post barrier only when the actual store happened. This makes sense
3040 // to check only for LS_cmp_* that can fail to set the value.
3041 // LS_cmp_exchange does not produce any branches by default, so there is no
3042 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
3043 // CompareAndExchange and move branches here, it would make sense to conditionalize
3044 // post_barriers for LS_cmp_exchange as well.
3045 //
3046 // CAS success path is marked more likely since we anticipate this is a performance
3047 // critical path, while CAS failure path can use the penalty for going through unlikely
3048 // path as backoff. Which is still better than doing a store barrier there.
3049 switch (kind) {
3050 case LS_get_set:
3051 case LS_cmp_exchange: {
3052 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3053 break;
3054 }
3055 case LS_cmp_swap_weak:
3056 case LS_cmp_swap: {
3057 IdealKit ideal(this);
3058 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3059 sync_kit(ideal);
3060 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3061 ideal.sync_kit(this);
3062 } ideal.end_if();
3063 final_sync(ideal);
3064 break;
3065 }
3066 default:
3067 ShouldNotReachHere();
3068 }
3069 break;
3070 default:
3071 fatal("unexpected type %d: %s", type, type2name(type));
3072 break;
3073 }
3074
3075 // SCMemProjNodes represent the memory state of a LoadStore. Their
3076 // main role is to prevent LoadStore nodes from being optimized away
3077 // when their results aren't used.
3078 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3079 set_memory(proj, alias_idx);
3080
3081 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3082 #ifdef _LP64
3083 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3084 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3085 }
3086 #endif
3087 if (can_move_pre_barrier()) {
3088 // Don't need to load pre_val. The old value is returned by load_store.
3089 // The pre_barrier can execute after the xchg as long as no safepoint
3090 // gets inserted between them.
3091 pre_barrier(false /* do_load */,
3092 control(), NULL, NULL, max_juint, NULL, NULL,
3093 load_store /* pre_val */,
3094 T_OBJECT);
3095 }
3096 }
3097
3098 // Add the trailing membar surrounding the access
3099 insert_mem_bar(Op_MemBarCPUOrder);
3100
3101 switch (access_kind) {
3102 case Relaxed:
3103 case Release:
3104 break; // do nothing
3105 case Acquire:
3106 case Volatile:
3107 insert_mem_bar(Op_MemBarAcquire);
3108 // !support_IRIW_for_not_multiple_copy_atomic_cpu handled in platform code
3109 break;
3110 default:
3111 ShouldNotReachHere();
3112 }
3113
3114 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3115 set_result(load_store);
3116 return true;
3117 }
3118
3119 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3120 MemNode::MemOrd mo = MemNode::unset;
3121 switch(kind) {
3122 case Opaque:
3123 case Relaxed: mo = MemNode::unordered; break;
3124 case Acquire: mo = MemNode::acquire; break;
3125 case Release: mo = MemNode::release; break;
3126 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3127 default:
3128 ShouldNotReachHere();
3129 }
3130 guarantee(mo != MemNode::unset, "Should select memory ordering");
3131 return mo;
3132 }
3133
3134 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3135 MemNode::MemOrd mo = MemNode::unset;
3136 switch(kind) {
3137 case Opaque:
3138 case Relaxed: mo = MemNode::unordered; break;
3139 case Acquire: mo = MemNode::acquire; break;
3140 case Release: mo = MemNode::release; break;
3141 case Volatile: mo = MemNode::seqcst; break;
3142 default:
3143 ShouldNotReachHere();
3144 }
3145 guarantee(mo != MemNode::unset, "Should select memory ordering");
3146 return mo;
3147 }
3148
3149 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3150 // Regardless of form, don't allow previous ld/st to move down,
3151 // then issue acquire, release, or volatile mem_bar.
3152 insert_mem_bar(Op_MemBarCPUOrder);
3153 switch(id) {
3154 case vmIntrinsics::_loadFence:
3155 insert_mem_bar(Op_LoadFence);
3156 return true;
3157 case vmIntrinsics::_storeFence:
3158 insert_mem_bar(Op_StoreFence);
3159 return true;
3160 case vmIntrinsics::_fullFence:
3161 insert_mem_bar(Op_MemBarVolatile);
3162 return true;
3163 default:
3164 fatal_unexpected_iid(id);
3165 return false;
3166 }
3167 }
3168
3169 bool LibraryCallKit::inline_onspinwait() {
3170 insert_mem_bar(Op_OnSpinWait);
3171 return true;
3172 }
3173
3174 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3175 if (!kls->is_Con()) {
3176 return true;
3177 }
3178 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3179 if (klsptr == NULL) {
3180 return true;
3181 }
3182 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3183 // don't need a guard for a klass that is already initialized
3184 return !ik->is_initialized();
3185 }
3186
3187 //----------------------------inline_unsafe_allocate---------------------------
3188 // public native Object Unsafe.allocateInstance(Class<?> cls);
3189 bool LibraryCallKit::inline_unsafe_allocate() {
3190 if (callee()->is_static()) return false; // caller must have the capability!
3191
3192 null_check_receiver(); // null-check, then ignore
3193 Node* cls = null_check(argument(1));
3194 if (stopped()) return true;
3195
3196 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3197 kls = null_check(kls);
3198 if (stopped()) return true; // argument was like int.class
3199
3200 Node* test = NULL;
3201 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3202 // Note: The argument might still be an illegal value like
3203 // Serializable.class or Object[].class. The runtime will handle it.
3204 // But we must make an explicit check for initialization.
3205 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3206 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3207 // can generate code to load it as unsigned byte.
3208 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3209 Node* bits = intcon(InstanceKlass::fully_initialized);
3210 test = _gvn.transform(new SubINode(inst, bits));
3211 // The 'test' is non-zero if we need to take a slow path.
3212 }
3213
3214 Node* obj = new_instance(kls, test);
3215 set_result(obj);
3216 return true;
3217 }
3218
3219 //------------------------inline_native_time_funcs--------------
3220 // inline code for System.currentTimeMillis() and System.nanoTime()
3221 // these have the same type and signature
3222 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3223 const TypeFunc* tf = OptoRuntime::void_long_Type();
3224 const TypePtr* no_memory_effects = NULL;
3225 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3226 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3227 #ifdef ASSERT
3228 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3229 assert(value_top == top(), "second value must be top");
3230 #endif
3231 set_result(value);
3232 return true;
3233 }
3234
3235 //------------------------inline_native_currentThread------------------
3236 bool LibraryCallKit::inline_native_currentThread() {
3237 Node* junk = NULL;
3238 set_result(generate_current_thread(junk));
3239 return true;
3240 }
3241
3242 //------------------------inline_native_isInterrupted------------------
3243 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3244 bool LibraryCallKit::inline_native_isInterrupted() {
3245 // Add a fast path to t.isInterrupted(clear_int):
3246 // (t == Thread.current() &&
3247 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3248 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3249 // So, in the common case that the interrupt bit is false,
3250 // we avoid making a call into the VM. Even if the interrupt bit
3251 // is true, if the clear_int argument is false, we avoid the VM call.
3252 // However, if the receiver is not currentThread, we must call the VM,
3253 // because there must be some locking done around the operation.
3254
3255 // We only go to the fast case code if we pass two guards.
3256 // Paths which do not pass are accumulated in the slow_region.
3257
3258 enum {
3259 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3260 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3261 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3262 PATH_LIMIT
3263 };
3264
3265 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3266 // out of the function.
3267 insert_mem_bar(Op_MemBarCPUOrder);
3268
3269 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3270 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3271
3272 RegionNode* slow_region = new RegionNode(1);
3273 record_for_igvn(slow_region);
3274
3275 // (a) Receiving thread must be the current thread.
3276 Node* rec_thr = argument(0);
3277 Node* tls_ptr = NULL;
3278 Node* cur_thr = generate_current_thread(tls_ptr);
3279 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3280 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3281
3282 generate_slow_guard(bol_thr, slow_region);
3283
3284 // (b) Interrupt bit on TLS must be false.
3285 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3286 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3287 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3288
3289 // Set the control input on the field _interrupted read to prevent it floating up.
3290 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3291 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3292 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3293
3294 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3295
3296 // First fast path: if (!TLS._interrupted) return false;
3297 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3298 result_rgn->init_req(no_int_result_path, false_bit);
3299 result_val->init_req(no_int_result_path, intcon(0));
3300
3301 // drop through to next case
3302 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3303
3304 #ifndef TARGET_OS_FAMILY_windows
3305 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3306 Node* clr_arg = argument(1);
3307 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3308 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3309 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3310
3311 // Second fast path: ... else if (!clear_int) return true;
3312 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3313 result_rgn->init_req(no_clear_result_path, false_arg);
3314 result_val->init_req(no_clear_result_path, intcon(1));
3315
3316 // drop through to next case
3317 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3318 #else
3319 // To return true on Windows you must read the _interrupted field
3320 // and check the event state i.e. take the slow path.
3321 #endif // TARGET_OS_FAMILY_windows
3322
3323 // (d) Otherwise, go to the slow path.
3324 slow_region->add_req(control());
3325 set_control( _gvn.transform(slow_region));
3326
3327 if (stopped()) {
3328 // There is no slow path.
3329 result_rgn->init_req(slow_result_path, top());
3330 result_val->init_req(slow_result_path, top());
3331 } else {
3332 // non-virtual because it is a private non-static
3333 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3334
3335 Node* slow_val = set_results_for_java_call(slow_call);
3336 // this->control() comes from set_results_for_java_call
3337
3338 Node* fast_io = slow_call->in(TypeFunc::I_O);
3339 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3340
3341 // These two phis are pre-filled with copies of of the fast IO and Memory
3342 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3343 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3344
3345 result_rgn->init_req(slow_result_path, control());
3346 result_io ->init_req(slow_result_path, i_o());
3347 result_mem->init_req(slow_result_path, reset_memory());
3348 result_val->init_req(slow_result_path, slow_val);
3349
3350 set_all_memory(_gvn.transform(result_mem));
3351 set_i_o( _gvn.transform(result_io));
3352 }
3353
3354 C->set_has_split_ifs(true); // Has chance for split-if optimization
3355 set_result(result_rgn, result_val);
3356 return true;
3357 }
3358
3359 //---------------------------load_mirror_from_klass----------------------------
3360 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3361 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3362 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3363 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3364 }
3365
3366 //-----------------------load_klass_from_mirror_common-------------------------
3367 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3368 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3369 // and branch to the given path on the region.
3370 // If never_see_null, take an uncommon trap on null, so we can optimistically
3371 // compile for the non-null case.
3372 // If the region is NULL, force never_see_null = true.
3373 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3374 bool never_see_null,
3375 RegionNode* region,
3376 int null_path,
3377 int offset) {
3378 if (region == NULL) never_see_null = true;
3379 Node* p = basic_plus_adr(mirror, offset);
3380 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3381 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3382 Node* null_ctl = top();
3383 kls = null_check_oop(kls, &null_ctl, never_see_null);
3384 if (region != NULL) {
3385 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3386 region->init_req(null_path, null_ctl);
3387 } else {
3388 assert(null_ctl == top(), "no loose ends");
3389 }
3390 return kls;
3391 }
3392
3393 //--------------------(inline_native_Class_query helpers)---------------------
3394 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3395 // Fall through if (mods & mask) == bits, take the guard otherwise.
3396 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3397 // Branch around if the given klass has the given modifier bit set.
3398 // Like generate_guard, adds a new path onto the region.
3399 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3400 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3401 Node* mask = intcon(modifier_mask);
3402 Node* bits = intcon(modifier_bits);
3403 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3404 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3405 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3406 return generate_fair_guard(bol, region);
3407 }
3408 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3409 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3410 }
3411
3412 //-------------------------inline_native_Class_query-------------------
3413 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3414 const Type* return_type = TypeInt::BOOL;
3415 Node* prim_return_value = top(); // what happens if it's a primitive class?
3416 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3417 bool expect_prim = false; // most of these guys expect to work on refs
3418
3419 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3420
3421 Node* mirror = argument(0);
3422 Node* obj = top();
3423
3424 switch (id) {
3425 case vmIntrinsics::_isInstance:
3426 // nothing is an instance of a primitive type
3427 prim_return_value = intcon(0);
3428 obj = argument(1);
3429 break;
3430 case vmIntrinsics::_getModifiers:
3431 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3432 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3433 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3434 break;
3435 case vmIntrinsics::_isInterface:
3436 prim_return_value = intcon(0);
3437 break;
3438 case vmIntrinsics::_isArray:
3439 prim_return_value = intcon(0);
3440 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3441 break;
3442 case vmIntrinsics::_isPrimitive:
3443 prim_return_value = intcon(1);
3444 expect_prim = true; // obviously
3445 break;
3446 case vmIntrinsics::_getSuperclass:
3447 prim_return_value = null();
3448 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3449 break;
3450 case vmIntrinsics::_getClassAccessFlags:
3451 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3452 return_type = TypeInt::INT; // not bool! 6297094
3453 break;
3454 default:
3455 fatal_unexpected_iid(id);
3456 break;
3457 }
3458
3459 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3460 if (mirror_con == NULL) return false; // cannot happen?
3461
3462 #ifndef PRODUCT
3463 if (C->print_intrinsics() || C->print_inlining()) {
3464 ciType* k = mirror_con->java_mirror_type();
3465 if (k) {
3466 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3467 k->print_name();
3468 tty->cr();
3469 }
3470 }
3471 #endif
3472
3473 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3474 RegionNode* region = new RegionNode(PATH_LIMIT);
3475 record_for_igvn(region);
3476 PhiNode* phi = new PhiNode(region, return_type);
3477
3478 // The mirror will never be null of Reflection.getClassAccessFlags, however
3479 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3480 // if it is. See bug 4774291.
3481
3482 // For Reflection.getClassAccessFlags(), the null check occurs in
3483 // the wrong place; see inline_unsafe_access(), above, for a similar
3484 // situation.
3485 mirror = null_check(mirror);
3486 // If mirror or obj is dead, only null-path is taken.
3487 if (stopped()) return true;
3488
3489 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3490
3491 // Now load the mirror's klass metaobject, and null-check it.
3492 // Side-effects region with the control path if the klass is null.
3493 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3494 // If kls is null, we have a primitive mirror.
3495 phi->init_req(_prim_path, prim_return_value);
3496 if (stopped()) { set_result(region, phi); return true; }
3497 bool safe_for_replace = (region->in(_prim_path) == top());
3498
3499 Node* p; // handy temp
3500 Node* null_ctl;
3501
3502 // Now that we have the non-null klass, we can perform the real query.
3503 // For constant classes, the query will constant-fold in LoadNode::Value.
3504 Node* query_value = top();
3505 switch (id) {
3506 case vmIntrinsics::_isInstance:
3507 // nothing is an instance of a primitive type
3508 query_value = gen_instanceof(obj, kls, safe_for_replace);
3509 break;
3510
3511 case vmIntrinsics::_getModifiers:
3512 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3513 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3514 break;
3515
3516 case vmIntrinsics::_isInterface:
3517 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3518 if (generate_interface_guard(kls, region) != NULL)
3519 // A guard was added. If the guard is taken, it was an interface.
3520 phi->add_req(intcon(1));
3521 // If we fall through, it's a plain class.
3522 query_value = intcon(0);
3523 break;
3524
3525 case vmIntrinsics::_isArray:
3526 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3527 if (generate_array_guard(kls, region) != NULL)
3528 // A guard was added. If the guard is taken, it was an array.
3529 phi->add_req(intcon(1));
3530 // If we fall through, it's a plain class.
3531 query_value = intcon(0);
3532 break;
3533
3534 case vmIntrinsics::_isPrimitive:
3535 query_value = intcon(0); // "normal" path produces false
3536 break;
3537
3538 case vmIntrinsics::_getSuperclass:
3539 // The rules here are somewhat unfortunate, but we can still do better
3540 // with random logic than with a JNI call.
3541 // Interfaces store null or Object as _super, but must report null.
3542 // Arrays store an intermediate super as _super, but must report Object.
3543 // Other types can report the actual _super.
3544 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3545 if (generate_interface_guard(kls, region) != NULL)
3546 // A guard was added. If the guard is taken, it was an interface.
3547 phi->add_req(null());
3548 if (generate_array_guard(kls, region) != NULL)
3549 // A guard was added. If the guard is taken, it was an array.
3550 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3551 // If we fall through, it's a plain class. Get its _super.
3552 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3553 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3554 null_ctl = top();
3555 kls = null_check_oop(kls, &null_ctl);
3556 if (null_ctl != top()) {
3557 // If the guard is taken, Object.superClass is null (both klass and mirror).
3558 region->add_req(null_ctl);
3559 phi ->add_req(null());
3560 }
3561 if (!stopped()) {
3562 query_value = load_mirror_from_klass(kls);
3563 }
3564 break;
3565
3566 case vmIntrinsics::_getClassAccessFlags:
3567 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3568 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3569 break;
3570
3571 default:
3572 fatal_unexpected_iid(id);
3573 break;
3574 }
3575
3576 // Fall-through is the normal case of a query to a real class.
3577 phi->init_req(1, query_value);
3578 region->init_req(1, control());
3579
3580 C->set_has_split_ifs(true); // Has chance for split-if optimization
3581 set_result(region, phi);
3582 return true;
3583 }
3584
3585 //-------------------------inline_Class_cast-------------------
3586 bool LibraryCallKit::inline_Class_cast() {
3587 Node* mirror = argument(0); // Class
3588 Node* obj = argument(1);
3589 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3590 if (mirror_con == NULL) {
3591 return false; // dead path (mirror->is_top()).
3592 }
3593 if (obj == NULL || obj->is_top()) {
3594 return false; // dead path
3595 }
3596 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3597
3598 // First, see if Class.cast() can be folded statically.
3599 // java_mirror_type() returns non-null for compile-time Class constants.
3600 ciType* tm = mirror_con->java_mirror_type();
3601 if (tm != NULL && tm->is_klass() &&
3602 tp != NULL && tp->klass() != NULL) {
3603 if (!tp->klass()->is_loaded()) {
3604 // Don't use intrinsic when class is not loaded.
3605 return false;
3606 } else {
3607 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3608 if (static_res == Compile::SSC_always_true) {
3609 // isInstance() is true - fold the code.
3610 set_result(obj);
3611 return true;
3612 } else if (static_res == Compile::SSC_always_false) {
3613 // Don't use intrinsic, have to throw ClassCastException.
3614 // If the reference is null, the non-intrinsic bytecode will
3615 // be optimized appropriately.
3616 return false;
3617 }
3618 }
3619 }
3620
3621 // Bailout intrinsic and do normal inlining if exception path is frequent.
3622 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3623 return false;
3624 }
3625
3626 // Generate dynamic checks.
3627 // Class.cast() is java implementation of _checkcast bytecode.
3628 // Do checkcast (Parse::do_checkcast()) optimizations here.
3629
3630 mirror = null_check(mirror);
3631 // If mirror is dead, only null-path is taken.
3632 if (stopped()) {
3633 return true;
3634 }
3635
3636 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3637 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3638 RegionNode* region = new RegionNode(PATH_LIMIT);
3639 record_for_igvn(region);
3640
3641 // Now load the mirror's klass metaobject, and null-check it.
3642 // If kls is null, we have a primitive mirror and
3643 // nothing is an instance of a primitive type.
3644 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3645
3646 Node* res = top();
3647 if (!stopped()) {
3648 Node* bad_type_ctrl = top();
3649 // Do checkcast optimizations.
3650 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3651 region->init_req(_bad_type_path, bad_type_ctrl);
3652 }
3653 if (region->in(_prim_path) != top() ||
3654 region->in(_bad_type_path) != top()) {
3655 // Let Interpreter throw ClassCastException.
3656 PreserveJVMState pjvms(this);
3657 set_control(_gvn.transform(region));
3658 uncommon_trap(Deoptimization::Reason_intrinsic,
3659 Deoptimization::Action_maybe_recompile);
3660 }
3661 if (!stopped()) {
3662 set_result(res);
3663 }
3664 return true;
3665 }
3666
3667
3668 //--------------------------inline_native_subtype_check------------------------
3669 // This intrinsic takes the JNI calls out of the heart of
3670 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3671 bool LibraryCallKit::inline_native_subtype_check() {
3672 // Pull both arguments off the stack.
3673 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3674 args[0] = argument(0);
3675 args[1] = argument(1);
3676 Node* klasses[2]; // corresponding Klasses: superk, subk
3677 klasses[0] = klasses[1] = top();
3678
3679 enum {
3680 // A full decision tree on {superc is prim, subc is prim}:
3681 _prim_0_path = 1, // {P,N} => false
3682 // {P,P} & superc!=subc => false
3683 _prim_same_path, // {P,P} & superc==subc => true
3684 _prim_1_path, // {N,P} => false
3685 _ref_subtype_path, // {N,N} & subtype check wins => true
3686 _both_ref_path, // {N,N} & subtype check loses => false
3687 PATH_LIMIT
3688 };
3689
3690 RegionNode* region = new RegionNode(PATH_LIMIT);
3691 Node* phi = new PhiNode(region, TypeInt::BOOL);
3692 record_for_igvn(region);
3693
3694 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3695 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3696 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3697
3698 // First null-check both mirrors and load each mirror's klass metaobject.
3699 int which_arg;
3700 for (which_arg = 0; which_arg <= 1; which_arg++) {
3701 Node* arg = args[which_arg];
3702 arg = null_check(arg);
3703 if (stopped()) break;
3704 args[which_arg] = arg;
3705
3706 Node* p = basic_plus_adr(arg, class_klass_offset);
3707 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3708 klasses[which_arg] = _gvn.transform(kls);
3709 }
3710
3711 // Having loaded both klasses, test each for null.
3712 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3713 for (which_arg = 0; which_arg <= 1; which_arg++) {
3714 Node* kls = klasses[which_arg];
3715 Node* null_ctl = top();
3716 kls = null_check_oop(kls, &null_ctl, never_see_null);
3717 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3718 region->init_req(prim_path, null_ctl);
3719 if (stopped()) break;
3720 klasses[which_arg] = kls;
3721 }
3722
3723 if (!stopped()) {
3724 // now we have two reference types, in klasses[0..1]
3725 Node* subk = klasses[1]; // the argument to isAssignableFrom
3726 Node* superk = klasses[0]; // the receiver
3727 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3728 // now we have a successful reference subtype check
3729 region->set_req(_ref_subtype_path, control());
3730 }
3731
3732 // If both operands are primitive (both klasses null), then
3733 // we must return true when they are identical primitives.
3734 // It is convenient to test this after the first null klass check.
3735 set_control(region->in(_prim_0_path)); // go back to first null check
3736 if (!stopped()) {
3737 // Since superc is primitive, make a guard for the superc==subc case.
3738 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3739 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3740 generate_guard(bol_eq, region, PROB_FAIR);
3741 if (region->req() == PATH_LIMIT+1) {
3742 // A guard was added. If the added guard is taken, superc==subc.
3743 region->swap_edges(PATH_LIMIT, _prim_same_path);
3744 region->del_req(PATH_LIMIT);
3745 }
3746 region->set_req(_prim_0_path, control()); // Not equal after all.
3747 }
3748
3749 // these are the only paths that produce 'true':
3750 phi->set_req(_prim_same_path, intcon(1));
3751 phi->set_req(_ref_subtype_path, intcon(1));
3752
3753 // pull together the cases:
3754 assert(region->req() == PATH_LIMIT, "sane region");
3755 for (uint i = 1; i < region->req(); i++) {
3756 Node* ctl = region->in(i);
3757 if (ctl == NULL || ctl == top()) {
3758 region->set_req(i, top());
3759 phi ->set_req(i, top());
3760 } else if (phi->in(i) == NULL) {
3761 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3762 }
3763 }
3764
3765 set_control(_gvn.transform(region));
3766 set_result(_gvn.transform(phi));
3767 return true;
3768 }
3769
3770 //---------------------generate_array_guard_common------------------------
3771 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3772 bool obj_array, bool not_array) {
3773
3774 if (stopped()) {
3775 return NULL;
3776 }
3777
3778 // If obj_array/non_array==false/false:
3779 // Branch around if the given klass is in fact an array (either obj or prim).
3780 // If obj_array/non_array==false/true:
3781 // Branch around if the given klass is not an array klass of any kind.
3782 // If obj_array/non_array==true/true:
3783 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3784 // If obj_array/non_array==true/false:
3785 // Branch around if the kls is an oop array (Object[] or subtype)
3786 //
3787 // Like generate_guard, adds a new path onto the region.
3788 jint layout_con = 0;
3789 Node* layout_val = get_layout_helper(kls, layout_con);
3790 if (layout_val == NULL) {
3791 bool query = (obj_array
3792 ? Klass::layout_helper_is_objArray(layout_con)
3793 : Klass::layout_helper_is_array(layout_con));
3794 if (query == not_array) {
3795 return NULL; // never a branch
3796 } else { // always a branch
3797 Node* always_branch = control();
3798 if (region != NULL)
3799 region->add_req(always_branch);
3800 set_control(top());
3801 return always_branch;
3802 }
3803 }
3804 // Now test the correct condition.
3805 jint nval = (obj_array
3806 ? (jint)(Klass::_lh_array_tag_type_value
3807 << Klass::_lh_array_tag_shift)
3808 : Klass::_lh_neutral_value);
3809 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3810 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3811 // invert the test if we are looking for a non-array
3812 if (not_array) btest = BoolTest(btest).negate();
3813 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3814 return generate_fair_guard(bol, region);
3815 }
3816
3817
3818 //-----------------------inline_native_newArray--------------------------
3819 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3820 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3821 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3822 Node* mirror;
3823 Node* count_val;
3824 if (uninitialized) {
3825 mirror = argument(1);
3826 count_val = argument(2);
3827 } else {
3828 mirror = argument(0);
3829 count_val = argument(1);
3830 }
3831
3832 mirror = null_check(mirror);
3833 // If mirror or obj is dead, only null-path is taken.
3834 if (stopped()) return true;
3835
3836 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3837 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3838 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3839 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3840 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3841
3842 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3843 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3844 result_reg, _slow_path);
3845 Node* normal_ctl = control();
3846 Node* no_array_ctl = result_reg->in(_slow_path);
3847
3848 // Generate code for the slow case. We make a call to newArray().
3849 set_control(no_array_ctl);
3850 if (!stopped()) {
3851 // Either the input type is void.class, or else the
3852 // array klass has not yet been cached. Either the
3853 // ensuing call will throw an exception, or else it
3854 // will cache the array klass for next time.
3855 PreserveJVMState pjvms(this);
3856 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3857 Node* slow_result = set_results_for_java_call(slow_call);
3858 // this->control() comes from set_results_for_java_call
3859 result_reg->set_req(_slow_path, control());
3860 result_val->set_req(_slow_path, slow_result);
3861 result_io ->set_req(_slow_path, i_o());
3862 result_mem->set_req(_slow_path, reset_memory());
3863 }
3864
3865 set_control(normal_ctl);
3866 if (!stopped()) {
3867 // Normal case: The array type has been cached in the java.lang.Class.
3868 // The following call works fine even if the array type is polymorphic.
3869 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3870 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3871 result_reg->init_req(_normal_path, control());
3872 result_val->init_req(_normal_path, obj);
3873 result_io ->init_req(_normal_path, i_o());
3874 result_mem->init_req(_normal_path, reset_memory());
3875
3876 if (uninitialized) {
3877 // Mark the allocation so that zeroing is skipped
3878 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3879 alloc->maybe_set_complete(&_gvn);
3880 }
3881 }
3882
3883 // Return the combined state.
3884 set_i_o( _gvn.transform(result_io) );
3885 set_all_memory( _gvn.transform(result_mem));
3886
3887 C->set_has_split_ifs(true); // Has chance for split-if optimization
3888 set_result(result_reg, result_val);
3889 return true;
3890 }
3891
3892 //----------------------inline_native_getLength--------------------------
3893 // public static native int java.lang.reflect.Array.getLength(Object array);
3894 bool LibraryCallKit::inline_native_getLength() {
3895 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3896
3897 Node* array = null_check(argument(0));
3898 // If array is dead, only null-path is taken.
3899 if (stopped()) return true;
3900
3901 // Deoptimize if it is a non-array.
3902 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3903
3904 if (non_array != NULL) {
3905 PreserveJVMState pjvms(this);
3906 set_control(non_array);
3907 uncommon_trap(Deoptimization::Reason_intrinsic,
3908 Deoptimization::Action_maybe_recompile);
3909 }
3910
3911 // If control is dead, only non-array-path is taken.
3912 if (stopped()) return true;
3913
3914 // The works fine even if the array type is polymorphic.
3915 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3916 Node* result = load_array_length(array);
3917
3918 C->set_has_split_ifs(true); // Has chance for split-if optimization
3919 set_result(result);
3920 return true;
3921 }
3922
3923 //------------------------inline_array_copyOf----------------------------
3924 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3925 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3926 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3927 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3928
3929 // Get the arguments.
3930 Node* original = argument(0);
3931 Node* start = is_copyOfRange? argument(1): intcon(0);
3932 Node* end = is_copyOfRange? argument(2): argument(1);
3933 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3934
3935 Node* newcopy = NULL;
3936
3937 // Set the original stack and the reexecute bit for the interpreter to reexecute
3938 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3939 { PreserveReexecuteState preexecs(this);
3940 jvms()->set_should_reexecute(true);
3941
3942 array_type_mirror = null_check(array_type_mirror);
3943 original = null_check(original);
3944
3945 // Check if a null path was taken unconditionally.
3946 if (stopped()) return true;
3947
3948 Node* orig_length = load_array_length(original);
3949
3950 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3951 klass_node = null_check(klass_node);
3952
3953 RegionNode* bailout = new RegionNode(1);
3954 record_for_igvn(bailout);
3955
3956 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3957 // Bail out if that is so.
3958 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3959 if (not_objArray != NULL) {
3960 // Improve the klass node's type from the new optimistic assumption:
3961 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3962 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3963 Node* cast = new CastPPNode(klass_node, akls);
3964 cast->init_req(0, control());
3965 klass_node = _gvn.transform(cast);
3966 }
3967
3968 // Bail out if either start or end is negative.
3969 generate_negative_guard(start, bailout, &start);
3970 generate_negative_guard(end, bailout, &end);
3971
3972 Node* length = end;
3973 if (_gvn.type(start) != TypeInt::ZERO) {
3974 length = _gvn.transform(new SubINode(end, start));
3975 }
3976
3977 // Bail out if length is negative.
3978 // Without this the new_array would throw
3979 // NegativeArraySizeException but IllegalArgumentException is what
3980 // should be thrown
3981 generate_negative_guard(length, bailout, &length);
3982
3983 if (bailout->req() > 1) {
3984 PreserveJVMState pjvms(this);
3985 set_control(_gvn.transform(bailout));
3986 uncommon_trap(Deoptimization::Reason_intrinsic,
3987 Deoptimization::Action_maybe_recompile);
3988 }
3989
3990 if (!stopped()) {
3991 // How many elements will we copy from the original?
3992 // The answer is MinI(orig_length - start, length).
3993 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3994 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3995
3996 // Generate a direct call to the right arraycopy function(s).
3997 // We know the copy is disjoint but we might not know if the
3998 // oop stores need checking.
3999 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4000 // This will fail a store-check if x contains any non-nulls.
4001
4002 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4003 // loads/stores but it is legal only if we're sure the
4004 // Arrays.copyOf would succeed. So we need all input arguments
4005 // to the copyOf to be validated, including that the copy to the
4006 // new array won't trigger an ArrayStoreException. That subtype
4007 // check can be optimized if we know something on the type of
4008 // the input array from type speculation.
4009 if (_gvn.type(klass_node)->singleton()) {
4010 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
4011 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4012
4013 int test = C->static_subtype_check(superk, subk);
4014 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4015 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4016 if (t_original->speculative_type() != NULL) {
4017 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4018 }
4019 }
4020 }
4021
4022 bool validated = false;
4023 // Reason_class_check rather than Reason_intrinsic because we
4024 // want to intrinsify even if this traps.
4025 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4026 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
4027 klass_node);
4028
4029 if (not_subtype_ctrl != top()) {
4030 PreserveJVMState pjvms(this);
4031 set_control(not_subtype_ctrl);
4032 uncommon_trap(Deoptimization::Reason_class_check,
4033 Deoptimization::Action_make_not_entrant);
4034 assert(stopped(), "Should be stopped");
4035 }
4036 validated = true;
4037 }
4038
4039 if (!stopped()) {
4040 newcopy = new_array(klass_node, length, 0); // no arguments to push
4041
4042 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
4043 load_object_klass(original), klass_node);
4044 if (!is_copyOfRange) {
4045 ac->set_copyof(validated);
4046 } else {
4047 ac->set_copyofrange(validated);
4048 }
4049 Node* n = _gvn.transform(ac);
4050 if (n == ac) {
4051 ac->connect_outputs(this);
4052 } else {
4053 assert(validated, "shouldn't transform if all arguments not validated");
4054 set_all_memory(n);
4055 }
4056 }
4057 }
4058 } // original reexecute is set back here
4059
4060 C->set_has_split_ifs(true); // Has chance for split-if optimization
4061 if (!stopped()) {
4062 set_result(newcopy);
4063 }
4064 return true;
4065 }
4066
4067
4068 //----------------------generate_virtual_guard---------------------------
4069 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4070 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4071 RegionNode* slow_region) {
4072 ciMethod* method = callee();
4073 int vtable_index = method->vtable_index();
4074 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4075 "bad index %d", vtable_index);
4076 // Get the Method* out of the appropriate vtable entry.
4077 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4078 vtable_index*vtableEntry::size_in_bytes() +
4079 vtableEntry::method_offset_in_bytes();
4080 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4081 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4082
4083 // Compare the target method with the expected method (e.g., Object.hashCode).
4084 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4085
4086 Node* native_call = makecon(native_call_addr);
4087 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4088 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4089
4090 return generate_slow_guard(test_native, slow_region);
4091 }
4092
4093 //-----------------------generate_method_call----------------------------
4094 // Use generate_method_call to make a slow-call to the real
4095 // method if the fast path fails. An alternative would be to
4096 // use a stub like OptoRuntime::slow_arraycopy_Java.
4097 // This only works for expanding the current library call,
4098 // not another intrinsic. (E.g., don't use this for making an
4099 // arraycopy call inside of the copyOf intrinsic.)
4100 CallJavaNode*
4101 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4102 // When compiling the intrinsic method itself, do not use this technique.
4103 guarantee(callee() != C->method(), "cannot make slow-call to self");
4104
4105 ciMethod* method = callee();
4106 // ensure the JVMS we have will be correct for this call
4107 guarantee(method_id == method->intrinsic_id(), "must match");
4108
4109 const TypeFunc* tf = TypeFunc::make(method);
4110 CallJavaNode* slow_call;
4111 if (is_static) {
4112 assert(!is_virtual, "");
4113 slow_call = new CallStaticJavaNode(C, tf,
4114 SharedRuntime::get_resolve_static_call_stub(),
4115 method, bci());
4116 } else if (is_virtual) {
4117 null_check_receiver();
4118 int vtable_index = Method::invalid_vtable_index;
4119 if (UseInlineCaches) {
4120 // Suppress the vtable call
4121 } else {
4122 // hashCode and clone are not a miranda methods,
4123 // so the vtable index is fixed.
4124 // No need to use the linkResolver to get it.
4125 vtable_index = method->vtable_index();
4126 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4127 "bad index %d", vtable_index);
4128 }
4129 slow_call = new CallDynamicJavaNode(tf,
4130 SharedRuntime::get_resolve_virtual_call_stub(),
4131 method, vtable_index, bci());
4132 } else { // neither virtual nor static: opt_virtual
4133 null_check_receiver();
4134 slow_call = new CallStaticJavaNode(C, tf,
4135 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4136 method, bci());
4137 slow_call->set_optimized_virtual(true);
4138 }
4139 set_arguments_for_java_call(slow_call);
4140 set_edges_for_java_call(slow_call);
4141 return slow_call;
4142 }
4143
4144
4145 /**
4146 * Build special case code for calls to hashCode on an object. This call may
4147 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4148 * slightly different code.
4149 */
4150 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4151 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4152 assert(!(is_virtual && is_static), "either virtual, special, or static");
4153
4154 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4155
4156 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4157 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4158 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4159 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4160 Node* obj = NULL;
4161 if (!is_static) {
4162 // Check for hashing null object
4163 obj = null_check_receiver();
4164 if (stopped()) return true; // unconditionally null
4165 result_reg->init_req(_null_path, top());
4166 result_val->init_req(_null_path, top());
4167 } else {
4168 // Do a null check, and return zero if null.
4169 // System.identityHashCode(null) == 0
4170 obj = argument(0);
4171 Node* null_ctl = top();
4172 obj = null_check_oop(obj, &null_ctl);
4173 result_reg->init_req(_null_path, null_ctl);
4174 result_val->init_req(_null_path, _gvn.intcon(0));
4175 }
4176
4177 // Unconditionally null? Then return right away.
4178 if (stopped()) {
4179 set_control( result_reg->in(_null_path));
4180 if (!stopped())
4181 set_result(result_val->in(_null_path));
4182 return true;
4183 }
4184
4185 // We only go to the fast case code if we pass a number of guards. The
4186 // paths which do not pass are accumulated in the slow_region.
4187 RegionNode* slow_region = new RegionNode(1);
4188 record_for_igvn(slow_region);
4189
4190 // If this is a virtual call, we generate a funny guard. We pull out
4191 // the vtable entry corresponding to hashCode() from the target object.
4192 // If the target method which we are calling happens to be the native
4193 // Object hashCode() method, we pass the guard. We do not need this
4194 // guard for non-virtual calls -- the caller is known to be the native
4195 // Object hashCode().
4196 if (is_virtual) {
4197 // After null check, get the object's klass.
4198 Node* obj_klass = load_object_klass(obj);
4199 generate_virtual_guard(obj_klass, slow_region);
4200 }
4201
4202 // Get the header out of the object, use LoadMarkNode when available
4203 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4204 // The control of the load must be NULL. Otherwise, the load can move before
4205 // the null check after castPP removal.
4206 Node* no_ctrl = NULL;
4207 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4208
4209 // Test the header to see if it is unlocked.
4210 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4211 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4212 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4213 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4214 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4215
4216 generate_slow_guard(test_unlocked, slow_region);
4217
4218 // Get the hash value and check to see that it has been properly assigned.
4219 // We depend on hash_mask being at most 32 bits and avoid the use of
4220 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4221 // vm: see markOop.hpp.
4222 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4223 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4224 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4225 // This hack lets the hash bits live anywhere in the mark object now, as long
4226 // as the shift drops the relevant bits into the low 32 bits. Note that
4227 // Java spec says that HashCode is an int so there's no point in capturing
4228 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4229 hshifted_header = ConvX2I(hshifted_header);
4230 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4231
4232 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4233 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4234 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4235
4236 generate_slow_guard(test_assigned, slow_region);
4237
4238 Node* init_mem = reset_memory();
4239 // fill in the rest of the null path:
4240 result_io ->init_req(_null_path, i_o());
4241 result_mem->init_req(_null_path, init_mem);
4242
4243 result_val->init_req(_fast_path, hash_val);
4244 result_reg->init_req(_fast_path, control());
4245 result_io ->init_req(_fast_path, i_o());
4246 result_mem->init_req(_fast_path, init_mem);
4247
4248 // Generate code for the slow case. We make a call to hashCode().
4249 set_control(_gvn.transform(slow_region));
4250 if (!stopped()) {
4251 // No need for PreserveJVMState, because we're using up the present state.
4252 set_all_memory(init_mem);
4253 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4254 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4255 Node* slow_result = set_results_for_java_call(slow_call);
4256 // this->control() comes from set_results_for_java_call
4257 result_reg->init_req(_slow_path, control());
4258 result_val->init_req(_slow_path, slow_result);
4259 result_io ->set_req(_slow_path, i_o());
4260 result_mem ->set_req(_slow_path, reset_memory());
4261 }
4262
4263 // Return the combined state.
4264 set_i_o( _gvn.transform(result_io) );
4265 set_all_memory( _gvn.transform(result_mem));
4266
4267 set_result(result_reg, result_val);
4268 return true;
4269 }
4270
4271 //---------------------------inline_native_getClass----------------------------
4272 // public final native Class<?> java.lang.Object.getClass();
4273 //
4274 // Build special case code for calls to getClass on an object.
4275 bool LibraryCallKit::inline_native_getClass() {
4276 Node* obj = null_check_receiver();
4277 if (stopped()) return true;
4278 set_result(load_mirror_from_klass(load_object_klass(obj)));
4279 return true;
4280 }
4281
4282 //-----------------inline_native_Reflection_getCallerClass---------------------
4283 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4284 //
4285 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4286 //
4287 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4288 // in that it must skip particular security frames and checks for
4289 // caller sensitive methods.
4290 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4291 #ifndef PRODUCT
4292 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4293 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4294 }
4295 #endif
4296
4297 if (!jvms()->has_method()) {
4298 #ifndef PRODUCT
4299 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4300 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4301 }
4302 #endif
4303 return false;
4304 }
4305
4306 // Walk back up the JVM state to find the caller at the required
4307 // depth.
4308 JVMState* caller_jvms = jvms();
4309
4310 // Cf. JVM_GetCallerClass
4311 // NOTE: Start the loop at depth 1 because the current JVM state does
4312 // not include the Reflection.getCallerClass() frame.
4313 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4314 ciMethod* m = caller_jvms->method();
4315 switch (n) {
4316 case 0:
4317 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4318 break;
4319 case 1:
4320 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4321 if (!m->caller_sensitive()) {
4322 #ifndef PRODUCT
4323 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4324 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4325 }
4326 #endif
4327 return false; // bail-out; let JVM_GetCallerClass do the work
4328 }
4329 break;
4330 default:
4331 if (!m->is_ignored_by_security_stack_walk()) {
4332 // We have reached the desired frame; return the holder class.
4333 // Acquire method holder as java.lang.Class and push as constant.
4334 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4335 ciInstance* caller_mirror = caller_klass->java_mirror();
4336 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4337
4338 #ifndef PRODUCT
4339 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4340 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());
4341 tty->print_cr(" JVM state at this point:");
4342 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4343 ciMethod* m = jvms()->of_depth(i)->method();
4344 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4345 }
4346 }
4347 #endif
4348 return true;
4349 }
4350 break;
4351 }
4352 }
4353
4354 #ifndef PRODUCT
4355 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4356 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4357 tty->print_cr(" JVM state at this point:");
4358 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4359 ciMethod* m = jvms()->of_depth(i)->method();
4360 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4361 }
4362 }
4363 #endif
4364
4365 return false; // bail-out; let JVM_GetCallerClass do the work
4366 }
4367
4368 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4369 Node* arg = argument(0);
4370 Node* result = NULL;
4371
4372 switch (id) {
4373 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4374 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4375 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4376 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4377
4378 case vmIntrinsics::_doubleToLongBits: {
4379 // two paths (plus control) merge in a wood
4380 RegionNode *r = new RegionNode(3);
4381 Node *phi = new PhiNode(r, TypeLong::LONG);
4382
4383 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4384 // Build the boolean node
4385 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4386
4387 // Branch either way.
4388 // NaN case is less traveled, which makes all the difference.
4389 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4390 Node *opt_isnan = _gvn.transform(ifisnan);
4391 assert( opt_isnan->is_If(), "Expect an IfNode");
4392 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4393 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4394
4395 set_control(iftrue);
4396
4397 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4398 Node *slow_result = longcon(nan_bits); // return NaN
4399 phi->init_req(1, _gvn.transform( slow_result ));
4400 r->init_req(1, iftrue);
4401
4402 // Else fall through
4403 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4404 set_control(iffalse);
4405
4406 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4407 r->init_req(2, iffalse);
4408
4409 // Post merge
4410 set_control(_gvn.transform(r));
4411 record_for_igvn(r);
4412
4413 C->set_has_split_ifs(true); // Has chance for split-if optimization
4414 result = phi;
4415 assert(result->bottom_type()->isa_long(), "must be");
4416 break;
4417 }
4418
4419 case vmIntrinsics::_floatToIntBits: {
4420 // two paths (plus control) merge in a wood
4421 RegionNode *r = new RegionNode(3);
4422 Node *phi = new PhiNode(r, TypeInt::INT);
4423
4424 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4425 // Build the boolean node
4426 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4427
4428 // Branch either way.
4429 // NaN case is less traveled, which makes all the difference.
4430 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4431 Node *opt_isnan = _gvn.transform(ifisnan);
4432 assert( opt_isnan->is_If(), "Expect an IfNode");
4433 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4434 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4435
4436 set_control(iftrue);
4437
4438 static const jint nan_bits = 0x7fc00000;
4439 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4440 phi->init_req(1, _gvn.transform( slow_result ));
4441 r->init_req(1, iftrue);
4442
4443 // Else fall through
4444 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4445 set_control(iffalse);
4446
4447 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4448 r->init_req(2, iffalse);
4449
4450 // Post merge
4451 set_control(_gvn.transform(r));
4452 record_for_igvn(r);
4453
4454 C->set_has_split_ifs(true); // Has chance for split-if optimization
4455 result = phi;
4456 assert(result->bottom_type()->isa_int(), "must be");
4457 break;
4458 }
4459
4460 default:
4461 fatal_unexpected_iid(id);
4462 break;
4463 }
4464 set_result(_gvn.transform(result));
4465 return true;
4466 }
4467
4468 //----------------------inline_unsafe_copyMemory-------------------------
4469 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4470 bool LibraryCallKit::inline_unsafe_copyMemory() {
4471 if (callee()->is_static()) return false; // caller must have the capability!
4472 null_check_receiver(); // null-check receiver
4473 if (stopped()) return true;
4474
4475 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4476
4477 Node* src_ptr = argument(1); // type: oop
4478 Node* src_off = ConvL2X(argument(2)); // type: long
4479 Node* dst_ptr = argument(4); // type: oop
4480 Node* dst_off = ConvL2X(argument(5)); // type: long
4481 Node* size = ConvL2X(argument(7)); // type: long
4482
4483 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4484 "fieldOffset must be byte-scaled");
4485
4486 Node* src = make_unsafe_address(src_ptr, src_off);
4487 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4488
4489 // Conservatively insert a memory barrier on all memory slices.
4490 // Do not let writes of the copy source or destination float below the copy.
4491 insert_mem_bar(Op_MemBarCPUOrder);
4492
4493 // Call it. Note that the length argument is not scaled.
4494 make_runtime_call(RC_LEAF|RC_NO_FP,
4495 OptoRuntime::fast_arraycopy_Type(),
4496 StubRoutines::unsafe_arraycopy(),
4497 "unsafe_arraycopy",
4498 TypeRawPtr::BOTTOM,
4499 src, dst, size XTOP);
4500
4501 // Do not let reads of the copy destination float above the copy.
4502 insert_mem_bar(Op_MemBarCPUOrder);
4503
4504 return true;
4505 }
4506
4507 //------------------------clone_coping-----------------------------------
4508 // Helper function for inline_native_clone.
4509 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4510 assert(obj_size != NULL, "");
4511 Node* raw_obj = alloc_obj->in(1);
4512 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4513
4514 AllocateNode* alloc = NULL;
4515 if (ReduceBulkZeroing) {
4516 // We will be completely responsible for initializing this object -
4517 // mark Initialize node as complete.
4518 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4519 // The object was just allocated - there should be no any stores!
4520 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4521 // Mark as complete_with_arraycopy so that on AllocateNode
4522 // expansion, we know this AllocateNode is initialized by an array
4523 // copy and a StoreStore barrier exists after the array copy.
4524 alloc->initialization()->set_complete_with_arraycopy();
4525 }
4526
4527 // Copy the fastest available way.
4528 // TODO: generate fields copies for small objects instead.
4529 Node* src = obj;
4530 Node* dest = alloc_obj;
4531 Node* size = _gvn.transform(obj_size);
4532
4533 // Exclude the header but include array length to copy by 8 bytes words.
4534 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4535 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4536 instanceOopDesc::base_offset_in_bytes();
4537 // base_off:
4538 // 8 - 32-bit VM
4539 // 12 - 64-bit VM, compressed klass
4540 // 16 - 64-bit VM, normal klass
4541 if (base_off % BytesPerLong != 0) {
4542 assert(UseCompressedClassPointers, "");
4543 if (is_array) {
4544 // Exclude length to copy by 8 bytes words.
4545 base_off += sizeof(int);
4546 } else {
4547 // Include klass to copy by 8 bytes words.
4548 base_off = instanceOopDesc::klass_offset_in_bytes();
4549 }
4550 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4551 }
4552 src = basic_plus_adr(src, base_off);
4553 dest = basic_plus_adr(dest, base_off);
4554
4555 // Compute the length also, if needed:
4556 Node* countx = size;
4557 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4558 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4559
4560 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4561
4562 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4563 ac->set_clonebasic();
4564 Node* n = _gvn.transform(ac);
4565 if (n == ac) {
4566 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4567 } else {
4568 set_all_memory(n);
4569 }
4570
4571 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4572 if (card_mark) {
4573 assert(!is_array, "");
4574 // Put in store barrier for any and all oops we are sticking
4575 // into this object. (We could avoid this if we could prove
4576 // that the object type contains no oop fields at all.)
4577 Node* no_particular_value = NULL;
4578 Node* no_particular_field = NULL;
4579 int raw_adr_idx = Compile::AliasIdxRaw;
4580 post_barrier(control(),
4581 memory(raw_adr_type),
4582 alloc_obj,
4583 no_particular_field,
4584 raw_adr_idx,
4585 no_particular_value,
4586 T_OBJECT,
4587 false);
4588 }
4589
4590 // Do not let reads from the cloned object float above the arraycopy.
4591 if (alloc != NULL) {
4592 // Do not let stores that initialize this object be reordered with
4593 // a subsequent store that would make this object accessible by
4594 // other threads.
4595 // Record what AllocateNode this StoreStore protects so that
4596 // escape analysis can go from the MemBarStoreStoreNode to the
4597 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4598 // based on the escape status of the AllocateNode.
4599 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4600 } else {
4601 insert_mem_bar(Op_MemBarCPUOrder);
4602 }
4603 }
4604
4605 //------------------------inline_native_clone----------------------------
4606 // protected native Object java.lang.Object.clone();
4607 //
4608 // Here are the simple edge cases:
4609 // null receiver => normal trap
4610 // virtual and clone was overridden => slow path to out-of-line clone
4611 // not cloneable or finalizer => slow path to out-of-line Object.clone
4612 //
4613 // The general case has two steps, allocation and copying.
4614 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4615 //
4616 // Copying also has two cases, oop arrays and everything else.
4617 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4618 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4619 //
4620 // These steps fold up nicely if and when the cloned object's klass
4621 // can be sharply typed as an object array, a type array, or an instance.
4622 //
4623 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4624 PhiNode* result_val;
4625
4626 // Set the reexecute bit for the interpreter to reexecute
4627 // the bytecode that invokes Object.clone if deoptimization happens.
4628 { PreserveReexecuteState preexecs(this);
4629 jvms()->set_should_reexecute(true);
4630
4631 Node* obj = null_check_receiver();
4632 if (stopped()) return true;
4633
4634 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4635
4636 // If we are going to clone an instance, we need its exact type to
4637 // know the number and types of fields to convert the clone to
4638 // loads/stores. Maybe a speculative type can help us.
4639 if (!obj_type->klass_is_exact() &&
4640 obj_type->speculative_type() != NULL &&
4641 obj_type->speculative_type()->is_instance_klass()) {
4642 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4643 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4644 !spec_ik->has_injected_fields()) {
4645 ciKlass* k = obj_type->klass();
4646 if (!k->is_instance_klass() ||
4647 k->as_instance_klass()->is_interface() ||
4648 k->as_instance_klass()->has_subklass()) {
4649 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4650 }
4651 }
4652 }
4653
4654 Node* obj_klass = load_object_klass(obj);
4655 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4656 const TypeOopPtr* toop = ((tklass != NULL)
4657 ? tklass->as_instance_type()
4658 : TypeInstPtr::NOTNULL);
4659
4660 // Conservatively insert a memory barrier on all memory slices.
4661 // Do not let writes into the original float below the clone.
4662 insert_mem_bar(Op_MemBarCPUOrder);
4663
4664 // paths into result_reg:
4665 enum {
4666 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4667 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4668 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4669 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4670 PATH_LIMIT
4671 };
4672 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4673 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4674 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4675 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4676 record_for_igvn(result_reg);
4677
4678 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4679 int raw_adr_idx = Compile::AliasIdxRaw;
4680
4681 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4682 if (array_ctl != NULL) {
4683 // It's an array.
4684 PreserveJVMState pjvms(this);
4685 set_control(array_ctl);
4686 Node* obj_length = load_array_length(obj);
4687 Node* obj_size = NULL;
4688 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4689
4690 if (!use_ReduceInitialCardMarks()) {
4691 // If it is an oop array, it requires very special treatment,
4692 // because card marking is required on each card of the array.
4693 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4694 if (is_obja != NULL) {
4695 PreserveJVMState pjvms2(this);
4696 set_control(is_obja);
4697 // Generate a direct call to the right arraycopy function(s).
4698 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4699 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4700 ac->set_cloneoop();
4701 Node* n = _gvn.transform(ac);
4702 assert(n == ac, "cannot disappear");
4703 ac->connect_outputs(this);
4704
4705 result_reg->init_req(_objArray_path, control());
4706 result_val->init_req(_objArray_path, alloc_obj);
4707 result_i_o ->set_req(_objArray_path, i_o());
4708 result_mem ->set_req(_objArray_path, reset_memory());
4709 }
4710 }
4711 // Otherwise, there are no card marks to worry about.
4712 // (We can dispense with card marks if we know the allocation
4713 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4714 // causes the non-eden paths to take compensating steps to
4715 // simulate a fresh allocation, so that no further
4716 // card marks are required in compiled code to initialize
4717 // the object.)
4718
4719 if (!stopped()) {
4720 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4721
4722 // Present the results of the copy.
4723 result_reg->init_req(_array_path, control());
4724 result_val->init_req(_array_path, alloc_obj);
4725 result_i_o ->set_req(_array_path, i_o());
4726 result_mem ->set_req(_array_path, reset_memory());
4727 }
4728 }
4729
4730 // We only go to the instance fast case code if we pass a number of guards.
4731 // The paths which do not pass are accumulated in the slow_region.
4732 RegionNode* slow_region = new RegionNode(1);
4733 record_for_igvn(slow_region);
4734 if (!stopped()) {
4735 // It's an instance (we did array above). Make the slow-path tests.
4736 // If this is a virtual call, we generate a funny guard. We grab
4737 // the vtable entry corresponding to clone() from the target object.
4738 // If the target method which we are calling happens to be the
4739 // Object clone() method, we pass the guard. We do not need this
4740 // guard for non-virtual calls; the caller is known to be the native
4741 // Object clone().
4742 if (is_virtual) {
4743 generate_virtual_guard(obj_klass, slow_region);
4744 }
4745
4746 // The object must be easily cloneable and must not have a finalizer.
4747 // Both of these conditions may be checked in a single test.
4748 // We could optimize the test further, but we don't care.
4749 generate_access_flags_guard(obj_klass,
4750 // Test both conditions:
4751 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4752 // Must be cloneable but not finalizer:
4753 JVM_ACC_IS_CLONEABLE_FAST,
4754 slow_region);
4755 }
4756
4757 if (!stopped()) {
4758 // It's an instance, and it passed the slow-path tests.
4759 PreserveJVMState pjvms(this);
4760 Node* obj_size = NULL;
4761 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4762 // is reexecuted if deoptimization occurs and there could be problems when merging
4763 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4764 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4765
4766 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4767
4768 // Present the results of the slow call.
4769 result_reg->init_req(_instance_path, control());
4770 result_val->init_req(_instance_path, alloc_obj);
4771 result_i_o ->set_req(_instance_path, i_o());
4772 result_mem ->set_req(_instance_path, reset_memory());
4773 }
4774
4775 // Generate code for the slow case. We make a call to clone().
4776 set_control(_gvn.transform(slow_region));
4777 if (!stopped()) {
4778 PreserveJVMState pjvms(this);
4779 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4780 Node* slow_result = set_results_for_java_call(slow_call);
4781 // this->control() comes from set_results_for_java_call
4782 result_reg->init_req(_slow_path, control());
4783 result_val->init_req(_slow_path, slow_result);
4784 result_i_o ->set_req(_slow_path, i_o());
4785 result_mem ->set_req(_slow_path, reset_memory());
4786 }
4787
4788 // Return the combined state.
4789 set_control( _gvn.transform(result_reg));
4790 set_i_o( _gvn.transform(result_i_o));
4791 set_all_memory( _gvn.transform(result_mem));
4792 } // original reexecute is set back here
4793
4794 set_result(_gvn.transform(result_val));
4795 return true;
4796 }
4797
4798 // If we have a tighly coupled allocation, the arraycopy may take care
4799 // of the array initialization. If one of the guards we insert between
4800 // the allocation and the arraycopy causes a deoptimization, an
4801 // unitialized array will escape the compiled method. To prevent that
4802 // we set the JVM state for uncommon traps between the allocation and
4803 // the arraycopy to the state before the allocation so, in case of
4804 // deoptimization, we'll reexecute the allocation and the
4805 // initialization.
4806 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4807 if (alloc != NULL) {
4808 ciMethod* trap_method = alloc->jvms()->method();
4809 int trap_bci = alloc->jvms()->bci();
4810
4811 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4812 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4813 // Make sure there's no store between the allocation and the
4814 // arraycopy otherwise visible side effects could be rexecuted
4815 // in case of deoptimization and cause incorrect execution.
4816 bool no_interfering_store = true;
4817 Node* mem = alloc->in(TypeFunc::Memory);
4818 if (mem->is_MergeMem()) {
4819 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4820 Node* n = mms.memory();
4821 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4822 assert(n->is_Store(), "what else?");
4823 no_interfering_store = false;
4824 break;
4825 }
4826 }
4827 } else {
4828 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4829 Node* n = mms.memory();
4830 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4831 assert(n->is_Store(), "what else?");
4832 no_interfering_store = false;
4833 break;
4834 }
4835 }
4836 }
4837
4838 if (no_interfering_store) {
4839 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4840 uint size = alloc->req();
4841 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4842 old_jvms->set_map(sfpt);
4843 for (uint i = 0; i < size; i++) {
4844 sfpt->init_req(i, alloc->in(i));
4845 }
4846 // re-push array length for deoptimization
4847 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4848 old_jvms->set_sp(old_jvms->sp()+1);
4849 old_jvms->set_monoff(old_jvms->monoff()+1);
4850 old_jvms->set_scloff(old_jvms->scloff()+1);
4851 old_jvms->set_endoff(old_jvms->endoff()+1);
4852 old_jvms->set_should_reexecute(true);
4853
4854 sfpt->set_i_o(map()->i_o());
4855 sfpt->set_memory(map()->memory());
4856 sfpt->set_control(map()->control());
4857
4858 JVMState* saved_jvms = jvms();
4859 saved_reexecute_sp = _reexecute_sp;
4860
4861 set_jvms(sfpt->jvms());
4862 _reexecute_sp = jvms()->sp();
4863
4864 return saved_jvms;
4865 }
4866 }
4867 }
4868 return NULL;
4869 }
4870
4871 // In case of a deoptimization, we restart execution at the
4872 // allocation, allocating a new array. We would leave an uninitialized
4873 // array in the heap that GCs wouldn't expect. Move the allocation
4874 // after the traps so we don't allocate the array if we
4875 // deoptimize. This is possible because tightly_coupled_allocation()
4876 // guarantees there's no observer of the allocated array at this point
4877 // and the control flow is simple enough.
4878 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4879 if (saved_jvms != NULL && !stopped()) {
4880 assert(alloc != NULL, "only with a tightly coupled allocation");
4881 // restore JVM state to the state at the arraycopy
4882 saved_jvms->map()->set_control(map()->control());
4883 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4884 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4885 // If we've improved the types of some nodes (null check) while
4886 // emitting the guards, propagate them to the current state
4887 map()->replaced_nodes().apply(saved_jvms->map());
4888 set_jvms(saved_jvms);
4889 _reexecute_sp = saved_reexecute_sp;
4890
4891 // Remove the allocation from above the guards
4892 CallProjections callprojs;
4893 alloc->extract_projections(&callprojs, true);
4894 InitializeNode* init = alloc->initialization();
4895 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4896 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4897 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4898 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4899
4900 // move the allocation here (after the guards)
4901 _gvn.hash_delete(alloc);
4902 alloc->set_req(TypeFunc::Control, control());
4903 alloc->set_req(TypeFunc::I_O, i_o());
4904 Node *mem = reset_memory();
4905 set_all_memory(mem);
4906 alloc->set_req(TypeFunc::Memory, mem);
4907 set_control(init->proj_out(TypeFunc::Control));
4908 set_i_o(callprojs.fallthrough_ioproj);
4909
4910 // Update memory as done in GraphKit::set_output_for_allocation()
4911 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4912 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4913 if (ary_type->isa_aryptr() && length_type != NULL) {
4914 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4915 }
4916 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4917 int elemidx = C->get_alias_index(telemref);
4918 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4919 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4920
4921 Node* allocx = _gvn.transform(alloc);
4922 assert(allocx == alloc, "where has the allocation gone?");
4923 assert(dest->is_CheckCastPP(), "not an allocation result?");
4924
4925 _gvn.hash_delete(dest);
4926 dest->set_req(0, control());
4927 Node* destx = _gvn.transform(dest);
4928 assert(destx == dest, "where has the allocation result gone?");
4929 }
4930 }
4931
4932
4933 //------------------------------inline_arraycopy-----------------------
4934 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4935 // Object dest, int destPos,
4936 // int length);
4937 bool LibraryCallKit::inline_arraycopy() {
4938 // Get the arguments.
4939 Node* src = argument(0); // type: oop
4940 Node* src_offset = argument(1); // type: int
4941 Node* dest = argument(2); // type: oop
4942 Node* dest_offset = argument(3); // type: int
4943 Node* length = argument(4); // type: int
4944
4945
4946 // Check for allocation before we add nodes that would confuse
4947 // tightly_coupled_allocation()
4948 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4949
4950 int saved_reexecute_sp = -1;
4951 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4952 // See arraycopy_restore_alloc_state() comment
4953 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4954 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4955 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4956 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4957
4958 // The following tests must be performed
4959 // (1) src and dest are arrays.
4960 // (2) src and dest arrays must have elements of the same BasicType
4961 // (3) src and dest must not be null.
4962 // (4) src_offset must not be negative.
4963 // (5) dest_offset must not be negative.
4964 // (6) length must not be negative.
4965 // (7) src_offset + length must not exceed length of src.
4966 // (8) dest_offset + length must not exceed length of dest.
4967 // (9) each element of an oop array must be assignable
4968
4969 // (3) src and dest must not be null.
4970 // always do this here because we need the JVM state for uncommon traps
4971 Node* null_ctl = top();
4972 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4973 assert(null_ctl->is_top(), "no null control here");
4974 dest = null_check(dest, T_ARRAY);
4975
4976 if (!can_emit_guards) {
4977 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4978 // guards but the arraycopy node could still take advantage of a
4979 // tightly allocated allocation. tightly_coupled_allocation() is
4980 // called again to make sure it takes the null check above into
4981 // account: the null check is mandatory and if it caused an
4982 // uncommon trap to be emitted then the allocation can't be
4983 // considered tightly coupled in this context.
4984 alloc = tightly_coupled_allocation(dest, NULL);
4985 }
4986
4987 bool validated = false;
4988
4989 const Type* src_type = _gvn.type(src);
4990 const Type* dest_type = _gvn.type(dest);
4991 const TypeAryPtr* top_src = src_type->isa_aryptr();
4992 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4993
4994 // Do we have the type of src?
4995 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4996 // Do we have the type of dest?
4997 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4998 // Is the type for src from speculation?
4999 bool src_spec = false;
5000 // Is the type for dest from speculation?
5001 bool dest_spec = false;
5002
5003 if ((!has_src || !has_dest) && can_emit_guards) {
5004 // We don't have sufficient type information, let's see if
5005 // speculative types can help. We need to have types for both src
5006 // and dest so that it pays off.
5007
5008 // Do we already have or could we have type information for src
5009 bool could_have_src = has_src;
5010 // Do we already have or could we have type information for dest
5011 bool could_have_dest = has_dest;
5012
5013 ciKlass* src_k = NULL;
5014 if (!has_src) {
5015 src_k = src_type->speculative_type_not_null();
5016 if (src_k != NULL && src_k->is_array_klass()) {
5017 could_have_src = true;
5018 }
5019 }
5020
5021 ciKlass* dest_k = NULL;
5022 if (!has_dest) {
5023 dest_k = dest_type->speculative_type_not_null();
5024 if (dest_k != NULL && dest_k->is_array_klass()) {
5025 could_have_dest = true;
5026 }
5027 }
5028
5029 if (could_have_src && could_have_dest) {
5030 // This is going to pay off so emit the required guards
5031 if (!has_src) {
5032 src = maybe_cast_profiled_obj(src, src_k, true);
5033 src_type = _gvn.type(src);
5034 top_src = src_type->isa_aryptr();
5035 has_src = (top_src != NULL && top_src->klass() != NULL);
5036 src_spec = true;
5037 }
5038 if (!has_dest) {
5039 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5040 dest_type = _gvn.type(dest);
5041 top_dest = dest_type->isa_aryptr();
5042 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5043 dest_spec = true;
5044 }
5045 }
5046 }
5047
5048 if (has_src && has_dest && can_emit_guards) {
5049 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5050 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5051 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5052 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5053
5054 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5055 // If both arrays are object arrays then having the exact types
5056 // for both will remove the need for a subtype check at runtime
5057 // before the call and may make it possible to pick a faster copy
5058 // routine (without a subtype check on every element)
5059 // Do we have the exact type of src?
5060 bool could_have_src = src_spec;
5061 // Do we have the exact type of dest?
5062 bool could_have_dest = dest_spec;
5063 ciKlass* src_k = top_src->klass();
5064 ciKlass* dest_k = top_dest->klass();
5065 if (!src_spec) {
5066 src_k = src_type->speculative_type_not_null();
5067 if (src_k != NULL && src_k->is_array_klass()) {
5068 could_have_src = true;
5069 }
5070 }
5071 if (!dest_spec) {
5072 dest_k = dest_type->speculative_type_not_null();
5073 if (dest_k != NULL && dest_k->is_array_klass()) {
5074 could_have_dest = true;
5075 }
5076 }
5077 if (could_have_src && could_have_dest) {
5078 // If we can have both exact types, emit the missing guards
5079 if (could_have_src && !src_spec) {
5080 src = maybe_cast_profiled_obj(src, src_k, true);
5081 }
5082 if (could_have_dest && !dest_spec) {
5083 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5084 }
5085 }
5086 }
5087 }
5088
5089 ciMethod* trap_method = method();
5090 int trap_bci = bci();
5091 if (saved_jvms != NULL) {
5092 trap_method = alloc->jvms()->method();
5093 trap_bci = alloc->jvms()->bci();
5094 }
5095
5096 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5097 can_emit_guards &&
5098 !src->is_top() && !dest->is_top()) {
5099 // validate arguments: enables transformation the ArrayCopyNode
5100 validated = true;
5101
5102 RegionNode* slow_region = new RegionNode(1);
5103 record_for_igvn(slow_region);
5104
5105 // (1) src and dest are arrays.
5106 generate_non_array_guard(load_object_klass(src), slow_region);
5107 generate_non_array_guard(load_object_klass(dest), slow_region);
5108
5109 // (2) src and dest arrays must have elements of the same BasicType
5110 // done at macro expansion or at Ideal transformation time
5111
5112 // (4) src_offset must not be negative.
5113 generate_negative_guard(src_offset, slow_region);
5114
5115 // (5) dest_offset must not be negative.
5116 generate_negative_guard(dest_offset, slow_region);
5117
5118 // (7) src_offset + length must not exceed length of src.
5119 generate_limit_guard(src_offset, length,
5120 load_array_length(src),
5121 slow_region);
5122
5123 // (8) dest_offset + length must not exceed length of dest.
5124 generate_limit_guard(dest_offset, length,
5125 load_array_length(dest),
5126 slow_region);
5127
5128 // (9) each element of an oop array must be assignable
5129 Node* src_klass = load_object_klass(src);
5130 Node* dest_klass = load_object_klass(dest);
5131 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5132
5133 if (not_subtype_ctrl != top()) {
5134 PreserveJVMState pjvms(this);
5135 set_control(not_subtype_ctrl);
5136 uncommon_trap(Deoptimization::Reason_intrinsic,
5137 Deoptimization::Action_make_not_entrant);
5138 assert(stopped(), "Should be stopped");
5139 }
5140 {
5141 PreserveJVMState pjvms(this);
5142 set_control(_gvn.transform(slow_region));
5143 uncommon_trap(Deoptimization::Reason_intrinsic,
5144 Deoptimization::Action_make_not_entrant);
5145 assert(stopped(), "Should be stopped");
5146 }
5147 }
5148
5149 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5150
5151 if (stopped()) {
5152 return true;
5153 }
5154
5155 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5156 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5157 // so the compiler has a chance to eliminate them: during macro expansion,
5158 // we have to set their control (CastPP nodes are eliminated).
5159 load_object_klass(src), load_object_klass(dest),
5160 load_array_length(src), load_array_length(dest));
5161
5162 ac->set_arraycopy(validated);
5163
5164 Node* n = _gvn.transform(ac);
5165 if (n == ac) {
5166 ac->connect_outputs(this);
5167 } else {
5168 assert(validated, "shouldn't transform if all arguments not validated");
5169 set_all_memory(n);
5170 }
5171
5172 return true;
5173 }
5174
5175
5176 // Helper function which determines if an arraycopy immediately follows
5177 // an allocation, with no intervening tests or other escapes for the object.
5178 AllocateArrayNode*
5179 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5180 RegionNode* slow_region) {
5181 if (stopped()) return NULL; // no fast path
5182 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5183
5184 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5185 if (alloc == NULL) return NULL;
5186
5187 Node* rawmem = memory(Compile::AliasIdxRaw);
5188 // Is the allocation's memory state untouched?
5189 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5190 // Bail out if there have been raw-memory effects since the allocation.
5191 // (Example: There might have been a call or safepoint.)
5192 return NULL;
5193 }
5194 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5195 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5196 return NULL;
5197 }
5198
5199 // There must be no unexpected observers of this allocation.
5200 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5201 Node* obs = ptr->fast_out(i);
5202 if (obs != this->map()) {
5203 return NULL;
5204 }
5205 }
5206
5207 // This arraycopy must unconditionally follow the allocation of the ptr.
5208 Node* alloc_ctl = ptr->in(0);
5209 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5210
5211 Node* ctl = control();
5212 while (ctl != alloc_ctl) {
5213 // There may be guards which feed into the slow_region.
5214 // Any other control flow means that we might not get a chance
5215 // to finish initializing the allocated object.
5216 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5217 IfNode* iff = ctl->in(0)->as_If();
5218 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5219 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5220 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5221 ctl = iff->in(0); // This test feeds the known slow_region.
5222 continue;
5223 }
5224 // One more try: Various low-level checks bottom out in
5225 // uncommon traps. If the debug-info of the trap omits
5226 // any reference to the allocation, as we've already
5227 // observed, then there can be no objection to the trap.
5228 bool found_trap = false;
5229 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5230 Node* obs = not_ctl->fast_out(j);
5231 if (obs->in(0) == not_ctl && obs->is_Call() &&
5232 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5233 found_trap = true; break;
5234 }
5235 }
5236 if (found_trap) {
5237 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5238 continue;
5239 }
5240 }
5241 return NULL;
5242 }
5243
5244 // If we get this far, we have an allocation which immediately
5245 // precedes the arraycopy, and we can take over zeroing the new object.
5246 // The arraycopy will finish the initialization, and provide
5247 // a new control state to which we will anchor the destination pointer.
5248
5249 return alloc;
5250 }
5251
5252 //-------------inline_encodeISOArray-----------------------------------
5253 // encode char[] to byte[] in ISO_8859_1
5254 bool LibraryCallKit::inline_encodeISOArray() {
5255 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5256 // no receiver since it is static method
5257 Node *src = argument(0);
5258 Node *src_offset = argument(1);
5259 Node *dst = argument(2);
5260 Node *dst_offset = argument(3);
5261 Node *length = argument(4);
5262
5263 const Type* src_type = src->Value(&_gvn);
5264 const Type* dst_type = dst->Value(&_gvn);
5265 const TypeAryPtr* top_src = src_type->isa_aryptr();
5266 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5267 if (top_src == NULL || top_src->klass() == NULL ||
5268 top_dest == NULL || top_dest->klass() == NULL) {
5269 // failed array check
5270 return false;
5271 }
5272
5273 // Figure out the size and type of the elements we will be copying.
5274 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5275 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5276 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5277 return false;
5278 }
5279
5280 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5281 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5282 // 'src_start' points to src array + scaled offset
5283 // 'dst_start' points to dst array + scaled offset
5284
5285 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5286 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5287 enc = _gvn.transform(enc);
5288 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5289 set_memory(res_mem, mtype);
5290 set_result(enc);
5291 return true;
5292 }
5293
5294 //-------------inline_multiplyToLen-----------------------------------
5295 bool LibraryCallKit::inline_multiplyToLen() {
5296 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5297
5298 address stubAddr = StubRoutines::multiplyToLen();
5299 if (stubAddr == NULL) {
5300 return false; // Intrinsic's stub is not implemented on this platform
5301 }
5302 const char* stubName = "multiplyToLen";
5303
5304 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5305
5306 // no receiver because it is a static method
5307 Node* x = argument(0);
5308 Node* xlen = argument(1);
5309 Node* y = argument(2);
5310 Node* ylen = argument(3);
5311 Node* z = argument(4);
5312
5313 const Type* x_type = x->Value(&_gvn);
5314 const Type* y_type = y->Value(&_gvn);
5315 const TypeAryPtr* top_x = x_type->isa_aryptr();
5316 const TypeAryPtr* top_y = y_type->isa_aryptr();
5317 if (top_x == NULL || top_x->klass() == NULL ||
5318 top_y == NULL || top_y->klass() == NULL) {
5319 // failed array check
5320 return false;
5321 }
5322
5323 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5324 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5325 if (x_elem != T_INT || y_elem != T_INT) {
5326 return false;
5327 }
5328
5329 // Set the original stack and the reexecute bit for the interpreter to reexecute
5330 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5331 // on the return from z array allocation in runtime.
5332 { PreserveReexecuteState preexecs(this);
5333 jvms()->set_should_reexecute(true);
5334
5335 Node* x_start = array_element_address(x, intcon(0), x_elem);
5336 Node* y_start = array_element_address(y, intcon(0), y_elem);
5337 // 'x_start' points to x array + scaled xlen
5338 // 'y_start' points to y array + scaled ylen
5339
5340 // Allocate the result array
5341 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5342 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5343 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5344
5345 IdealKit ideal(this);
5346
5347 #define __ ideal.
5348 Node* one = __ ConI(1);
5349 Node* zero = __ ConI(0);
5350 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5351 __ set(need_alloc, zero);
5352 __ set(z_alloc, z);
5353 __ if_then(z, BoolTest::eq, null()); {
5354 __ increment (need_alloc, one);
5355 } __ else_(); {
5356 // Update graphKit memory and control from IdealKit.
5357 sync_kit(ideal);
5358 Node* zlen_arg = load_array_length(z);
5359 // Update IdealKit memory and control from graphKit.
5360 __ sync_kit(this);
5361 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5362 __ increment (need_alloc, one);
5363 } __ end_if();
5364 } __ end_if();
5365
5366 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5367 // Update graphKit memory and control from IdealKit.
5368 sync_kit(ideal);
5369 Node * narr = new_array(klass_node, zlen, 1);
5370 // Update IdealKit memory and control from graphKit.
5371 __ sync_kit(this);
5372 __ set(z_alloc, narr);
5373 } __ end_if();
5374
5375 sync_kit(ideal);
5376 z = __ value(z_alloc);
5377 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5378 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5379 // Final sync IdealKit and GraphKit.
5380 final_sync(ideal);
5381 #undef __
5382
5383 Node* z_start = array_element_address(z, intcon(0), T_INT);
5384
5385 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5386 OptoRuntime::multiplyToLen_Type(),
5387 stubAddr, stubName, TypePtr::BOTTOM,
5388 x_start, xlen, y_start, ylen, z_start, zlen);
5389 } // original reexecute is set back here
5390
5391 C->set_has_split_ifs(true); // Has chance for split-if optimization
5392 set_result(z);
5393 return true;
5394 }
5395
5396 //-------------inline_squareToLen------------------------------------
5397 bool LibraryCallKit::inline_squareToLen() {
5398 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5399
5400 address stubAddr = StubRoutines::squareToLen();
5401 if (stubAddr == NULL) {
5402 return false; // Intrinsic's stub is not implemented on this platform
5403 }
5404 const char* stubName = "squareToLen";
5405
5406 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5407
5408 Node* x = argument(0);
5409 Node* len = argument(1);
5410 Node* z = argument(2);
5411 Node* zlen = argument(3);
5412
5413 const Type* x_type = x->Value(&_gvn);
5414 const Type* z_type = z->Value(&_gvn);
5415 const TypeAryPtr* top_x = x_type->isa_aryptr();
5416 const TypeAryPtr* top_z = z_type->isa_aryptr();
5417 if (top_x == NULL || top_x->klass() == NULL ||
5418 top_z == NULL || top_z->klass() == NULL) {
5419 // failed array check
5420 return false;
5421 }
5422
5423 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5424 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5425 if (x_elem != T_INT || z_elem != T_INT) {
5426 return false;
5427 }
5428
5429
5430 Node* x_start = array_element_address(x, intcon(0), x_elem);
5431 Node* z_start = array_element_address(z, intcon(0), z_elem);
5432
5433 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5434 OptoRuntime::squareToLen_Type(),
5435 stubAddr, stubName, TypePtr::BOTTOM,
5436 x_start, len, z_start, zlen);
5437
5438 set_result(z);
5439 return true;
5440 }
5441
5442 //-------------inline_mulAdd------------------------------------------
5443 bool LibraryCallKit::inline_mulAdd() {
5444 assert(UseMulAddIntrinsic, "not implemented on this platform");
5445
5446 address stubAddr = StubRoutines::mulAdd();
5447 if (stubAddr == NULL) {
5448 return false; // Intrinsic's stub is not implemented on this platform
5449 }
5450 const char* stubName = "mulAdd";
5451
5452 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5453
5454 Node* out = argument(0);
5455 Node* in = argument(1);
5456 Node* offset = argument(2);
5457 Node* len = argument(3);
5458 Node* k = argument(4);
5459
5460 const Type* out_type = out->Value(&_gvn);
5461 const Type* in_type = in->Value(&_gvn);
5462 const TypeAryPtr* top_out = out_type->isa_aryptr();
5463 const TypeAryPtr* top_in = in_type->isa_aryptr();
5464 if (top_out == NULL || top_out->klass() == NULL ||
5465 top_in == NULL || top_in->klass() == NULL) {
5466 // failed array check
5467 return false;
5468 }
5469
5470 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5471 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5472 if (out_elem != T_INT || in_elem != T_INT) {
5473 return false;
5474 }
5475
5476 Node* outlen = load_array_length(out);
5477 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5478 Node* out_start = array_element_address(out, intcon(0), out_elem);
5479 Node* in_start = array_element_address(in, intcon(0), in_elem);
5480
5481 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5482 OptoRuntime::mulAdd_Type(),
5483 stubAddr, stubName, TypePtr::BOTTOM,
5484 out_start,in_start, new_offset, len, k);
5485 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5486 set_result(result);
5487 return true;
5488 }
5489
5490 //-------------inline_montgomeryMultiply-----------------------------------
5491 bool LibraryCallKit::inline_montgomeryMultiply() {
5492 address stubAddr = StubRoutines::montgomeryMultiply();
5493 if (stubAddr == NULL) {
5494 return false; // Intrinsic's stub is not implemented on this platform
5495 }
5496
5497 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5498 const char* stubName = "montgomery_square";
5499
5500 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5501
5502 Node* a = argument(0);
5503 Node* b = argument(1);
5504 Node* n = argument(2);
5505 Node* len = argument(3);
5506 Node* inv = argument(4);
5507 Node* m = argument(6);
5508
5509 const Type* a_type = a->Value(&_gvn);
5510 const TypeAryPtr* top_a = a_type->isa_aryptr();
5511 const Type* b_type = b->Value(&_gvn);
5512 const TypeAryPtr* top_b = b_type->isa_aryptr();
5513 const Type* n_type = a->Value(&_gvn);
5514 const TypeAryPtr* top_n = n_type->isa_aryptr();
5515 const Type* m_type = a->Value(&_gvn);
5516 const TypeAryPtr* top_m = m_type->isa_aryptr();
5517 if (top_a == NULL || top_a->klass() == NULL ||
5518 top_b == NULL || top_b->klass() == NULL ||
5519 top_n == NULL || top_n->klass() == NULL ||
5520 top_m == NULL || top_m->klass() == NULL) {
5521 // failed array check
5522 return false;
5523 }
5524
5525 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5526 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5527 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5528 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5529 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5530 return false;
5531 }
5532
5533 // Make the call
5534 {
5535 Node* a_start = array_element_address(a, intcon(0), a_elem);
5536 Node* b_start = array_element_address(b, intcon(0), b_elem);
5537 Node* n_start = array_element_address(n, intcon(0), n_elem);
5538 Node* m_start = array_element_address(m, intcon(0), m_elem);
5539
5540 Node* call = make_runtime_call(RC_LEAF,
5541 OptoRuntime::montgomeryMultiply_Type(),
5542 stubAddr, stubName, TypePtr::BOTTOM,
5543 a_start, b_start, n_start, len, inv, top(),
5544 m_start);
5545 set_result(m);
5546 }
5547
5548 return true;
5549 }
5550
5551 bool LibraryCallKit::inline_montgomerySquare() {
5552 address stubAddr = StubRoutines::montgomerySquare();
5553 if (stubAddr == NULL) {
5554 return false; // Intrinsic's stub is not implemented on this platform
5555 }
5556
5557 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5558 const char* stubName = "montgomery_square";
5559
5560 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5561
5562 Node* a = argument(0);
5563 Node* n = argument(1);
5564 Node* len = argument(2);
5565 Node* inv = argument(3);
5566 Node* m = argument(5);
5567
5568 const Type* a_type = a->Value(&_gvn);
5569 const TypeAryPtr* top_a = a_type->isa_aryptr();
5570 const Type* n_type = a->Value(&_gvn);
5571 const TypeAryPtr* top_n = n_type->isa_aryptr();
5572 const Type* m_type = a->Value(&_gvn);
5573 const TypeAryPtr* top_m = m_type->isa_aryptr();
5574 if (top_a == NULL || top_a->klass() == NULL ||
5575 top_n == NULL || top_n->klass() == NULL ||
5576 top_m == NULL || top_m->klass() == NULL) {
5577 // failed array check
5578 return false;
5579 }
5580
5581 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5582 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5583 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5584 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5585 return false;
5586 }
5587
5588 // Make the call
5589 {
5590 Node* a_start = array_element_address(a, intcon(0), a_elem);
5591 Node* n_start = array_element_address(n, intcon(0), n_elem);
5592 Node* m_start = array_element_address(m, intcon(0), m_elem);
5593
5594 Node* call = make_runtime_call(RC_LEAF,
5595 OptoRuntime::montgomerySquare_Type(),
5596 stubAddr, stubName, TypePtr::BOTTOM,
5597 a_start, n_start, len, inv, top(),
5598 m_start);
5599 set_result(m);
5600 }
5601
5602 return true;
5603 }
5604
5605 //-------------inline_vectorizedMismatch------------------------------
5606 bool LibraryCallKit::inline_vectorizedMismatch() {
5607 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5608
5609 address stubAddr = StubRoutines::vectorizedMismatch();
5610 if (stubAddr == NULL) {
5611 return false; // Intrinsic's stub is not implemented on this platform
5612 }
5613 const char* stubName = "vectorizedMismatch";
5614 int size_l = callee()->signature()->size();
5615 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5616
5617 Node* obja = argument(0);
5618 Node* aoffset = argument(1);
5619 Node* objb = argument(3);
5620 Node* boffset = argument(4);
5621 Node* length = argument(6);
5622 Node* scale = argument(7);
5623
5624 const Type* a_type = obja->Value(&_gvn);
5625 const Type* b_type = objb->Value(&_gvn);
5626 const TypeAryPtr* top_a = a_type->isa_aryptr();
5627 const TypeAryPtr* top_b = b_type->isa_aryptr();
5628 if (top_a == NULL || top_a->klass() == NULL ||
5629 top_b == NULL || top_b->klass() == NULL) {
5630 // failed array check
5631 return false;
5632 }
5633
5634 Node* call;
5635 jvms()->set_should_reexecute(true);
5636
5637 Node* obja_adr = make_unsafe_address(obja, aoffset);
5638 Node* objb_adr = make_unsafe_address(objb, boffset);
5639
5640 call = make_runtime_call(RC_LEAF,
5641 OptoRuntime::vectorizedMismatch_Type(),
5642 stubAddr, stubName, TypePtr::BOTTOM,
5643 obja_adr, objb_adr, length, scale);
5644
5645 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5646 set_result(result);
5647 return true;
5648 }
5649
5650 /**
5651 * Calculate CRC32 for byte.
5652 * int java.util.zip.CRC32.update(int crc, int b)
5653 */
5654 bool LibraryCallKit::inline_updateCRC32() {
5655 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5656 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5657 // no receiver since it is static method
5658 Node* crc = argument(0); // type: int
5659 Node* b = argument(1); // type: int
5660
5661 /*
5662 * int c = ~ crc;
5663 * b = timesXtoThe32[(b ^ c) & 0xFF];
5664 * b = b ^ (c >>> 8);
5665 * crc = ~b;
5666 */
5667
5668 Node* M1 = intcon(-1);
5669 crc = _gvn.transform(new XorINode(crc, M1));
5670 Node* result = _gvn.transform(new XorINode(crc, b));
5671 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5672
5673 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5674 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5675 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5676 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5677
5678 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5679 result = _gvn.transform(new XorINode(crc, result));
5680 result = _gvn.transform(new XorINode(result, M1));
5681 set_result(result);
5682 return true;
5683 }
5684
5685 /**
5686 * Calculate CRC32 for byte[] array.
5687 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5688 */
5689 bool LibraryCallKit::inline_updateBytesCRC32() {
5690 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5691 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5692 // no receiver since it is static method
5693 Node* crc = argument(0); // type: int
5694 Node* src = argument(1); // type: oop
5695 Node* offset = argument(2); // type: int
5696 Node* length = argument(3); // type: int
5697
5698 const Type* src_type = src->Value(&_gvn);
5699 const TypeAryPtr* top_src = src_type->isa_aryptr();
5700 if (top_src == NULL || top_src->klass() == NULL) {
5701 // failed array check
5702 return false;
5703 }
5704
5705 // Figure out the size and type of the elements we will be copying.
5706 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5707 if (src_elem != T_BYTE) {
5708 return false;
5709 }
5710
5711 // 'src_start' points to src array + scaled offset
5712 Node* src_start = array_element_address(src, offset, src_elem);
5713
5714 // We assume that range check is done by caller.
5715 // TODO: generate range check (offset+length < src.length) in debug VM.
5716
5717 // Call the stub.
5718 address stubAddr = StubRoutines::updateBytesCRC32();
5719 const char *stubName = "updateBytesCRC32";
5720
5721 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5722 stubAddr, stubName, TypePtr::BOTTOM,
5723 crc, src_start, length);
5724 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5725 set_result(result);
5726 return true;
5727 }
5728
5729 /**
5730 * Calculate CRC32 for ByteBuffer.
5731 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5732 */
5733 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5734 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5735 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5736 // no receiver since it is static method
5737 Node* crc = argument(0); // type: int
5738 Node* src = argument(1); // type: long
5739 Node* offset = argument(3); // type: int
5740 Node* length = argument(4); // type: int
5741
5742 src = ConvL2X(src); // adjust Java long to machine word
5743 Node* base = _gvn.transform(new CastX2PNode(src));
5744 offset = ConvI2X(offset);
5745
5746 // 'src_start' points to src array + scaled offset
5747 Node* src_start = basic_plus_adr(top(), base, offset);
5748
5749 // Call the stub.
5750 address stubAddr = StubRoutines::updateBytesCRC32();
5751 const char *stubName = "updateBytesCRC32";
5752
5753 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5754 stubAddr, stubName, TypePtr::BOTTOM,
5755 crc, src_start, length);
5756 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5757 set_result(result);
5758 return true;
5759 }
5760
5761 //------------------------------get_table_from_crc32c_class-----------------------
5762 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5763 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5764 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5765
5766 return table;
5767 }
5768
5769 //------------------------------inline_updateBytesCRC32C-----------------------
5770 //
5771 // Calculate CRC32C for byte[] array.
5772 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5773 //
5774 bool LibraryCallKit::inline_updateBytesCRC32C() {
5775 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5776 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5777 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5778 // no receiver since it is a static method
5779 Node* crc = argument(0); // type: int
5780 Node* src = argument(1); // type: oop
5781 Node* offset = argument(2); // type: int
5782 Node* end = argument(3); // type: int
5783
5784 Node* length = _gvn.transform(new SubINode(end, offset));
5785
5786 const Type* src_type = src->Value(&_gvn);
5787 const TypeAryPtr* top_src = src_type->isa_aryptr();
5788 if (top_src == NULL || top_src->klass() == NULL) {
5789 // failed array check
5790 return false;
5791 }
5792
5793 // Figure out the size and type of the elements we will be copying.
5794 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5795 if (src_elem != T_BYTE) {
5796 return false;
5797 }
5798
5799 // 'src_start' points to src array + scaled offset
5800 Node* src_start = array_element_address(src, offset, src_elem);
5801
5802 // static final int[] byteTable in class CRC32C
5803 Node* table = get_table_from_crc32c_class(callee()->holder());
5804 Node* table_start = array_element_address(table, intcon(0), T_INT);
5805
5806 // We assume that range check is done by caller.
5807 // TODO: generate range check (offset+length < src.length) in debug VM.
5808
5809 // Call the stub.
5810 address stubAddr = StubRoutines::updateBytesCRC32C();
5811 const char *stubName = "updateBytesCRC32C";
5812
5813 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5814 stubAddr, stubName, TypePtr::BOTTOM,
5815 crc, src_start, length, table_start);
5816 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5817 set_result(result);
5818 return true;
5819 }
5820
5821 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5822 //
5823 // Calculate CRC32C for DirectByteBuffer.
5824 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5825 //
5826 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5827 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5828 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5829 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5830 // no receiver since it is a static method
5831 Node* crc = argument(0); // type: int
5832 Node* src = argument(1); // type: long
5833 Node* offset = argument(3); // type: int
5834 Node* end = argument(4); // type: int
5835
5836 Node* length = _gvn.transform(new SubINode(end, offset));
5837
5838 src = ConvL2X(src); // adjust Java long to machine word
5839 Node* base = _gvn.transform(new CastX2PNode(src));
5840 offset = ConvI2X(offset);
5841
5842 // 'src_start' points to src array + scaled offset
5843 Node* src_start = basic_plus_adr(top(), base, offset);
5844
5845 // static final int[] byteTable in class CRC32C
5846 Node* table = get_table_from_crc32c_class(callee()->holder());
5847 Node* table_start = array_element_address(table, intcon(0), T_INT);
5848
5849 // Call the stub.
5850 address stubAddr = StubRoutines::updateBytesCRC32C();
5851 const char *stubName = "updateBytesCRC32C";
5852
5853 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5854 stubAddr, stubName, TypePtr::BOTTOM,
5855 crc, src_start, length, table_start);
5856 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5857 set_result(result);
5858 return true;
5859 }
5860
5861 //------------------------------inline_updateBytesAdler32----------------------
5862 //
5863 // Calculate Adler32 checksum for byte[] array.
5864 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5865 //
5866 bool LibraryCallKit::inline_updateBytesAdler32() {
5867 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5868 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5869 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5870 // no receiver since it is static method
5871 Node* crc = argument(0); // type: int
5872 Node* src = argument(1); // type: oop
5873 Node* offset = argument(2); // type: int
5874 Node* length = argument(3); // type: int
5875
5876 const Type* src_type = src->Value(&_gvn);
5877 const TypeAryPtr* top_src = src_type->isa_aryptr();
5878 if (top_src == NULL || top_src->klass() == NULL) {
5879 // failed array check
5880 return false;
5881 }
5882
5883 // Figure out the size and type of the elements we will be copying.
5884 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5885 if (src_elem != T_BYTE) {
5886 return false;
5887 }
5888
5889 // 'src_start' points to src array + scaled offset
5890 Node* src_start = array_element_address(src, offset, src_elem);
5891
5892 // We assume that range check is done by caller.
5893 // TODO: generate range check (offset+length < src.length) in debug VM.
5894
5895 // Call the stub.
5896 address stubAddr = StubRoutines::updateBytesAdler32();
5897 const char *stubName = "updateBytesAdler32";
5898
5899 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5900 stubAddr, stubName, TypePtr::BOTTOM,
5901 crc, src_start, length);
5902 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5903 set_result(result);
5904 return true;
5905 }
5906
5907 //------------------------------inline_updateByteBufferAdler32---------------
5908 //
5909 // Calculate Adler32 checksum for DirectByteBuffer.
5910 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5911 //
5912 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5913 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5914 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5915 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5916 // no receiver since it is static method
5917 Node* crc = argument(0); // type: int
5918 Node* src = argument(1); // type: long
5919 Node* offset = argument(3); // type: int
5920 Node* length = argument(4); // type: int
5921
5922 src = ConvL2X(src); // adjust Java long to machine word
5923 Node* base = _gvn.transform(new CastX2PNode(src));
5924 offset = ConvI2X(offset);
5925
5926 // 'src_start' points to src array + scaled offset
5927 Node* src_start = basic_plus_adr(top(), base, offset);
5928
5929 // Call the stub.
5930 address stubAddr = StubRoutines::updateBytesAdler32();
5931 const char *stubName = "updateBytesAdler32";
5932
5933 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5934 stubAddr, stubName, TypePtr::BOTTOM,
5935 crc, src_start, length);
5936
5937 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5938 set_result(result);
5939 return true;
5940 }
5941
5942 //----------------------------inline_reference_get----------------------------
5943 // public T java.lang.ref.Reference.get();
5944 bool LibraryCallKit::inline_reference_get() {
5945 const int referent_offset = java_lang_ref_Reference::referent_offset;
5946 guarantee(referent_offset > 0, "should have already been set");
5947
5948 // Get the argument:
5949 Node* reference_obj = null_check_receiver();
5950 if (stopped()) return true;
5951
5952 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5953
5954 ciInstanceKlass* klass = env()->Object_klass();
5955 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5956
5957 Node* no_ctrl = NULL;
5958 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5959
5960 // Use the pre-barrier to record the value in the referent field
5961 pre_barrier(false /* do_load */,
5962 control(),
5963 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5964 result /* pre_val */,
5965 T_OBJECT);
5966
5967 // Add memory barrier to prevent commoning reads from this field
5968 // across safepoint since GC can change its value.
5969 insert_mem_bar(Op_MemBarCPUOrder);
5970
5971 set_result(result);
5972 return true;
5973 }
5974
5975
5976 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5977 bool is_exact=true, bool is_static=false,
5978 ciInstanceKlass * fromKls=NULL) {
5979 if (fromKls == NULL) {
5980 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5981 assert(tinst != NULL, "obj is null");
5982 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5983 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5984 fromKls = tinst->klass()->as_instance_klass();
5985 } else {
5986 assert(is_static, "only for static field access");
5987 }
5988 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5989 ciSymbol::make(fieldTypeString),
5990 is_static);
5991
5992 assert (field != NULL, "undefined field");
5993 if (field == NULL) return (Node *) NULL;
5994
5995 if (is_static) {
5996 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5997 fromObj = makecon(tip);
5998 }
5999
6000 // Next code copied from Parse::do_get_xxx():
6001
6002 // Compute address and memory type.
6003 int offset = field->offset_in_bytes();
6004 bool is_vol = field->is_volatile();
6005 ciType* field_klass = field->type();
6006 assert(field_klass->is_loaded(), "should be loaded");
6007 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6008 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6009 BasicType bt = field->layout_type();
6010
6011 // Build the resultant type of the load
6012 const Type *type;
6013 if (bt == T_OBJECT) {
6014 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6015 } else {
6016 type = Type::get_const_basic_type(bt);
6017 }
6018
6019 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
6020 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
6021 }
6022 // Build the load.
6023 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
6024 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
6025 // If reference is volatile, prevent following memory ops from
6026 // floating up past the volatile read. Also prevents commoning
6027 // another volatile read.
6028 if (is_vol) {
6029 // Memory barrier includes bogus read of value to force load BEFORE membar
6030 insert_mem_bar(Op_MemBarAcquire, loadedField);
6031 }
6032 return loadedField;
6033 }
6034
6035 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6036 bool is_exact = true, bool is_static = false,
6037 ciInstanceKlass * fromKls = NULL) {
6038 if (fromKls == NULL) {
6039 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6040 assert(tinst != NULL, "obj is null");
6041 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6042 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6043 fromKls = tinst->klass()->as_instance_klass();
6044 }
6045 else {
6046 assert(is_static, "only for static field access");
6047 }
6048 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6049 ciSymbol::make(fieldTypeString),
6050 is_static);
6051
6052 assert(field != NULL, "undefined field");
6053 assert(!field->is_volatile(), "not defined for volatile fields");
6054
6055 if (is_static) {
6056 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6057 fromObj = makecon(tip);
6058 }
6059
6060 // Next code copied from Parse::do_get_xxx():
6061
6062 // Compute address and memory type.
6063 int offset = field->offset_in_bytes();
6064 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6065
6066 return adr;
6067 }
6068
6069 //------------------------------inline_aescrypt_Block-----------------------
6070 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6071 address stubAddr = NULL;
6072 const char *stubName;
6073 assert(UseAES, "need AES instruction support");
6074
6075 switch(id) {
6076 case vmIntrinsics::_aescrypt_encryptBlock:
6077 stubAddr = StubRoutines::aescrypt_encryptBlock();
6078 stubName = "aescrypt_encryptBlock";
6079 break;
6080 case vmIntrinsics::_aescrypt_decryptBlock:
6081 stubAddr = StubRoutines::aescrypt_decryptBlock();
6082 stubName = "aescrypt_decryptBlock";
6083 break;
6084 }
6085 if (stubAddr == NULL) return false;
6086
6087 Node* aescrypt_object = argument(0);
6088 Node* src = argument(1);
6089 Node* src_offset = argument(2);
6090 Node* dest = argument(3);
6091 Node* dest_offset = argument(4);
6092
6093 // (1) src and dest are arrays.
6094 const Type* src_type = src->Value(&_gvn);
6095 const Type* dest_type = dest->Value(&_gvn);
6096 const TypeAryPtr* top_src = src_type->isa_aryptr();
6097 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6098 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6099
6100 // for the quick and dirty code we will skip all the checks.
6101 // we are just trying to get the call to be generated.
6102 Node* src_start = src;
6103 Node* dest_start = dest;
6104 if (src_offset != NULL || dest_offset != NULL) {
6105 assert(src_offset != NULL && dest_offset != NULL, "");
6106 src_start = array_element_address(src, src_offset, T_BYTE);
6107 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6108 }
6109
6110 // now need to get the start of its expanded key array
6111 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6112 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6113 if (k_start == NULL) return false;
6114
6115 if (Matcher::pass_original_key_for_aes()) {
6116 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6117 // compatibility issues between Java key expansion and SPARC crypto instructions
6118 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6119 if (original_k_start == NULL) return false;
6120
6121 // Call the stub.
6122 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6123 stubAddr, stubName, TypePtr::BOTTOM,
6124 src_start, dest_start, k_start, original_k_start);
6125 } else {
6126 // Call the stub.
6127 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6128 stubAddr, stubName, TypePtr::BOTTOM,
6129 src_start, dest_start, k_start);
6130 }
6131
6132 return true;
6133 }
6134
6135 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6136 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6137 address stubAddr = NULL;
6138 const char *stubName = NULL;
6139
6140 assert(UseAES, "need AES instruction support");
6141
6142 switch(id) {
6143 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6144 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6145 stubName = "cipherBlockChaining_encryptAESCrypt";
6146 break;
6147 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6148 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6149 stubName = "cipherBlockChaining_decryptAESCrypt";
6150 break;
6151 }
6152 if (stubAddr == NULL) return false;
6153
6154 Node* cipherBlockChaining_object = argument(0);
6155 Node* src = argument(1);
6156 Node* src_offset = argument(2);
6157 Node* len = argument(3);
6158 Node* dest = argument(4);
6159 Node* dest_offset = argument(5);
6160
6161 // (1) src and dest are arrays.
6162 const Type* src_type = src->Value(&_gvn);
6163 const Type* dest_type = dest->Value(&_gvn);
6164 const TypeAryPtr* top_src = src_type->isa_aryptr();
6165 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6166 assert (top_src != NULL && top_src->klass() != NULL
6167 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6168
6169 // checks are the responsibility of the caller
6170 Node* src_start = src;
6171 Node* dest_start = dest;
6172 if (src_offset != NULL || dest_offset != NULL) {
6173 assert(src_offset != NULL && dest_offset != NULL, "");
6174 src_start = array_element_address(src, src_offset, T_BYTE);
6175 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6176 }
6177
6178 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6179 // (because of the predicated logic executed earlier).
6180 // so we cast it here safely.
6181 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6182
6183 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6184 if (embeddedCipherObj == NULL) return false;
6185
6186 // cast it to what we know it will be at runtime
6187 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6188 assert(tinst != NULL, "CBC obj is null");
6189 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6190 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6191 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6192
6193 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6194 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6195 const TypeOopPtr* xtype = aklass->as_instance_type();
6196 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6197 aescrypt_object = _gvn.transform(aescrypt_object);
6198
6199 // we need to get the start of the aescrypt_object's expanded key array
6200 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6201 if (k_start == NULL) return false;
6202
6203 // similarly, get the start address of the r vector
6204 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6205 if (objRvec == NULL) return false;
6206 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6207
6208 Node* cbcCrypt;
6209 if (Matcher::pass_original_key_for_aes()) {
6210 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6211 // compatibility issues between Java key expansion and SPARC crypto instructions
6212 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6213 if (original_k_start == NULL) return false;
6214
6215 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6216 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6217 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6218 stubAddr, stubName, TypePtr::BOTTOM,
6219 src_start, dest_start, k_start, r_start, len, original_k_start);
6220 } else {
6221 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6222 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6223 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6224 stubAddr, stubName, TypePtr::BOTTOM,
6225 src_start, dest_start, k_start, r_start, len);
6226 }
6227
6228 // return cipher length (int)
6229 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6230 set_result(retvalue);
6231 return true;
6232 }
6233
6234 //------------------------------inline_counterMode_AESCrypt-----------------------
6235 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6236 assert(UseAES, "need AES instruction support");
6237 if (!UseAESCTRIntrinsics) return false;
6238
6239 address stubAddr = NULL;
6240 const char *stubName = NULL;
6241 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6242 stubAddr = StubRoutines::counterMode_AESCrypt();
6243 stubName = "counterMode_AESCrypt";
6244 }
6245 if (stubAddr == NULL) return false;
6246
6247 Node* counterMode_object = argument(0);
6248 Node* src = argument(1);
6249 Node* src_offset = argument(2);
6250 Node* len = argument(3);
6251 Node* dest = argument(4);
6252 Node* dest_offset = argument(5);
6253
6254 // (1) src and dest are arrays.
6255 const Type* src_type = src->Value(&_gvn);
6256 const Type* dest_type = dest->Value(&_gvn);
6257 const TypeAryPtr* top_src = src_type->isa_aryptr();
6258 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6259 assert(top_src != NULL && top_src->klass() != NULL &&
6260 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6261
6262 // checks are the responsibility of the caller
6263 Node* src_start = src;
6264 Node* dest_start = dest;
6265 if (src_offset != NULL || dest_offset != NULL) {
6266 assert(src_offset != NULL && dest_offset != NULL, "");
6267 src_start = array_element_address(src, src_offset, T_BYTE);
6268 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6269 }
6270
6271 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6272 // (because of the predicated logic executed earlier).
6273 // so we cast it here safely.
6274 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6275 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6276 if (embeddedCipherObj == NULL) return false;
6277 // cast it to what we know it will be at runtime
6278 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6279 assert(tinst != NULL, "CTR obj is null");
6280 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6281 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6282 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6283 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6284 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6285 const TypeOopPtr* xtype = aklass->as_instance_type();
6286 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6287 aescrypt_object = _gvn.transform(aescrypt_object);
6288 // we need to get the start of the aescrypt_object's expanded key array
6289 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6290 if (k_start == NULL) return false;
6291 // similarly, get the start address of the r vector
6292 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6293 if (obj_counter == NULL) return false;
6294 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6295
6296 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6297 if (saved_encCounter == NULL) return false;
6298 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6299 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6300
6301 Node* ctrCrypt;
6302 if (Matcher::pass_original_key_for_aes()) {
6303 // no SPARC version for AES/CTR intrinsics now.
6304 return false;
6305 }
6306 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6307 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6308 OptoRuntime::counterMode_aescrypt_Type(),
6309 stubAddr, stubName, TypePtr::BOTTOM,
6310 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6311
6312 // return cipher length (int)
6313 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6314 set_result(retvalue);
6315 return true;
6316 }
6317
6318 //------------------------------get_key_start_from_aescrypt_object-----------------------
6319 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6320 #ifdef PPC64
6321 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6322 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6323 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6324 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6325 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6326 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6327 if (objSessionK == NULL) {
6328 return (Node *) NULL;
6329 }
6330 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6331 #else
6332 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6333 #endif // PPC64
6334 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6335 if (objAESCryptKey == NULL) return (Node *) NULL;
6336
6337 // now have the array, need to get the start address of the K array
6338 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6339 return k_start;
6340 }
6341
6342 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6343 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6344 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6345 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6346 if (objAESCryptKey == NULL) return (Node *) NULL;
6347
6348 // now have the array, need to get the start address of the lastKey array
6349 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6350 return original_k_start;
6351 }
6352
6353 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6354 // Return node representing slow path of predicate check.
6355 // the pseudo code we want to emulate with this predicate is:
6356 // for encryption:
6357 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6358 // for decryption:
6359 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6360 // note cipher==plain is more conservative than the original java code but that's OK
6361 //
6362 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6363 // The receiver was checked for NULL already.
6364 Node* objCBC = argument(0);
6365
6366 // Load embeddedCipher field of CipherBlockChaining object.
6367 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6368
6369 // get AESCrypt klass for instanceOf check
6370 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6371 // will have same classloader as CipherBlockChaining object
6372 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6373 assert(tinst != NULL, "CBCobj is null");
6374 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6375
6376 // we want to do an instanceof comparison against the AESCrypt class
6377 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6378 if (!klass_AESCrypt->is_loaded()) {
6379 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6380 Node* ctrl = control();
6381 set_control(top()); // no regular fast path
6382 return ctrl;
6383 }
6384 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6385
6386 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6387 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6388 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6389
6390 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6391
6392 // for encryption, we are done
6393 if (!decrypting)
6394 return instof_false; // even if it is NULL
6395
6396 // for decryption, we need to add a further check to avoid
6397 // taking the intrinsic path when cipher and plain are the same
6398 // see the original java code for why.
6399 RegionNode* region = new RegionNode(3);
6400 region->init_req(1, instof_false);
6401 Node* src = argument(1);
6402 Node* dest = argument(4);
6403 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6404 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6405 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6406 region->init_req(2, src_dest_conjoint);
6407
6408 record_for_igvn(region);
6409 return _gvn.transform(region);
6410 }
6411
6412 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6413 // Return node representing slow path of predicate check.
6414 // the pseudo code we want to emulate with this predicate is:
6415 // for encryption:
6416 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6417 // for decryption:
6418 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6419 // note cipher==plain is more conservative than the original java code but that's OK
6420 //
6421
6422 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6423 // The receiver was checked for NULL already.
6424 Node* objCTR = argument(0);
6425
6426 // Load embeddedCipher field of CipherBlockChaining object.
6427 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6428
6429 // get AESCrypt klass for instanceOf check
6430 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6431 // will have same classloader as CipherBlockChaining object
6432 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6433 assert(tinst != NULL, "CTRobj is null");
6434 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6435
6436 // we want to do an instanceof comparison against the AESCrypt class
6437 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6438 if (!klass_AESCrypt->is_loaded()) {
6439 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6440 Node* ctrl = control();
6441 set_control(top()); // no regular fast path
6442 return ctrl;
6443 }
6444
6445 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6446 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6447 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6448 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6449 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6450
6451 return instof_false; // even if it is NULL
6452 }
6453
6454 //------------------------------inline_ghash_processBlocks
6455 bool LibraryCallKit::inline_ghash_processBlocks() {
6456 address stubAddr;
6457 const char *stubName;
6458 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6459
6460 stubAddr = StubRoutines::ghash_processBlocks();
6461 stubName = "ghash_processBlocks";
6462
6463 Node* data = argument(0);
6464 Node* offset = argument(1);
6465 Node* len = argument(2);
6466 Node* state = argument(3);
6467 Node* subkeyH = argument(4);
6468
6469 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6470 assert(state_start, "state is NULL");
6471 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6472 assert(subkeyH_start, "subkeyH is NULL");
6473 Node* data_start = array_element_address(data, offset, T_BYTE);
6474 assert(data_start, "data is NULL");
6475
6476 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6477 OptoRuntime::ghash_processBlocks_Type(),
6478 stubAddr, stubName, TypePtr::BOTTOM,
6479 state_start, subkeyH_start, data_start, len);
6480 return true;
6481 }
6482
6483 //------------------------------inline_sha_implCompress-----------------------
6484 //
6485 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6486 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6487 //
6488 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6489 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6490 //
6491 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6492 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6493 //
6494 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6495 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6496
6497 Node* sha_obj = argument(0);
6498 Node* src = argument(1); // type oop
6499 Node* ofs = argument(2); // type int
6500
6501 const Type* src_type = src->Value(&_gvn);
6502 const TypeAryPtr* top_src = src_type->isa_aryptr();
6503 if (top_src == NULL || top_src->klass() == NULL) {
6504 // failed array check
6505 return false;
6506 }
6507 // Figure out the size and type of the elements we will be copying.
6508 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6509 if (src_elem != T_BYTE) {
6510 return false;
6511 }
6512 // 'src_start' points to src array + offset
6513 Node* src_start = array_element_address(src, ofs, src_elem);
6514 Node* state = NULL;
6515 address stubAddr;
6516 const char *stubName;
6517
6518 switch(id) {
6519 case vmIntrinsics::_sha_implCompress:
6520 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6521 state = get_state_from_sha_object(sha_obj);
6522 stubAddr = StubRoutines::sha1_implCompress();
6523 stubName = "sha1_implCompress";
6524 break;
6525 case vmIntrinsics::_sha2_implCompress:
6526 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6527 state = get_state_from_sha_object(sha_obj);
6528 stubAddr = StubRoutines::sha256_implCompress();
6529 stubName = "sha256_implCompress";
6530 break;
6531 case vmIntrinsics::_sha5_implCompress:
6532 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6533 state = get_state_from_sha5_object(sha_obj);
6534 stubAddr = StubRoutines::sha512_implCompress();
6535 stubName = "sha512_implCompress";
6536 break;
6537 default:
6538 fatal_unexpected_iid(id);
6539 return false;
6540 }
6541 if (state == NULL) return false;
6542
6543 // Call the stub.
6544 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6545 stubAddr, stubName, TypePtr::BOTTOM,
6546 src_start, state);
6547
6548 return true;
6549 }
6550
6551 //------------------------------inline_digestBase_implCompressMB-----------------------
6552 //
6553 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6554 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6555 //
6556 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6557 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6558 "need SHA1/SHA256/SHA512 instruction support");
6559 assert((uint)predicate < 3, "sanity");
6560 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6561
6562 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6563 Node* src = argument(1); // byte[] array
6564 Node* ofs = argument(2); // type int
6565 Node* limit = argument(3); // type int
6566
6567 const Type* src_type = src->Value(&_gvn);
6568 const TypeAryPtr* top_src = src_type->isa_aryptr();
6569 if (top_src == NULL || top_src->klass() == NULL) {
6570 // failed array check
6571 return false;
6572 }
6573 // Figure out the size and type of the elements we will be copying.
6574 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6575 if (src_elem != T_BYTE) {
6576 return false;
6577 }
6578 // 'src_start' points to src array + offset
6579 Node* src_start = array_element_address(src, ofs, src_elem);
6580
6581 const char* klass_SHA_name = NULL;
6582 const char* stub_name = NULL;
6583 address stub_addr = NULL;
6584 bool long_state = false;
6585
6586 switch (predicate) {
6587 case 0:
6588 if (UseSHA1Intrinsics) {
6589 klass_SHA_name = "sun/security/provider/SHA";
6590 stub_name = "sha1_implCompressMB";
6591 stub_addr = StubRoutines::sha1_implCompressMB();
6592 }
6593 break;
6594 case 1:
6595 if (UseSHA256Intrinsics) {
6596 klass_SHA_name = "sun/security/provider/SHA2";
6597 stub_name = "sha256_implCompressMB";
6598 stub_addr = StubRoutines::sha256_implCompressMB();
6599 }
6600 break;
6601 case 2:
6602 if (UseSHA512Intrinsics) {
6603 klass_SHA_name = "sun/security/provider/SHA5";
6604 stub_name = "sha512_implCompressMB";
6605 stub_addr = StubRoutines::sha512_implCompressMB();
6606 long_state = true;
6607 }
6608 break;
6609 default:
6610 fatal("unknown SHA intrinsic predicate: %d", predicate);
6611 }
6612 if (klass_SHA_name != NULL) {
6613 // get DigestBase klass to lookup for SHA klass
6614 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6615 assert(tinst != NULL, "digestBase_obj is not instance???");
6616 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6617
6618 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6619 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6620 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6621 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6622 }
6623 return false;
6624 }
6625 //------------------------------inline_sha_implCompressMB-----------------------
6626 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6627 bool long_state, address stubAddr, const char *stubName,
6628 Node* src_start, Node* ofs, Node* limit) {
6629 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6630 const TypeOopPtr* xtype = aklass->as_instance_type();
6631 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6632 sha_obj = _gvn.transform(sha_obj);
6633
6634 Node* state;
6635 if (long_state) {
6636 state = get_state_from_sha5_object(sha_obj);
6637 } else {
6638 state = get_state_from_sha_object(sha_obj);
6639 }
6640 if (state == NULL) return false;
6641
6642 // Call the stub.
6643 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6644 OptoRuntime::digestBase_implCompressMB_Type(),
6645 stubAddr, stubName, TypePtr::BOTTOM,
6646 src_start, state, ofs, limit);
6647 // return ofs (int)
6648 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6649 set_result(result);
6650
6651 return true;
6652 }
6653
6654 //------------------------------get_state_from_sha_object-----------------------
6655 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6656 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6657 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6658 if (sha_state == NULL) return (Node *) NULL;
6659
6660 // now have the array, need to get the start address of the state array
6661 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6662 return state;
6663 }
6664
6665 //------------------------------get_state_from_sha5_object-----------------------
6666 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6667 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6668 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6669 if (sha_state == NULL) return (Node *) NULL;
6670
6671 // now have the array, need to get the start address of the state array
6672 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6673 return state;
6674 }
6675
6676 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6677 // Return node representing slow path of predicate check.
6678 // the pseudo code we want to emulate with this predicate is:
6679 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6680 //
6681 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6682 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6683 "need SHA1/SHA256/SHA512 instruction support");
6684 assert((uint)predicate < 3, "sanity");
6685
6686 // The receiver was checked for NULL already.
6687 Node* digestBaseObj = argument(0);
6688
6689 // get DigestBase klass for instanceOf check
6690 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6691 assert(tinst != NULL, "digestBaseObj is null");
6692 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6693
6694 const char* klass_SHA_name = NULL;
6695 switch (predicate) {
6696 case 0:
6697 if (UseSHA1Intrinsics) {
6698 // we want to do an instanceof comparison against the SHA class
6699 klass_SHA_name = "sun/security/provider/SHA";
6700 }
6701 break;
6702 case 1:
6703 if (UseSHA256Intrinsics) {
6704 // we want to do an instanceof comparison against the SHA2 class
6705 klass_SHA_name = "sun/security/provider/SHA2";
6706 }
6707 break;
6708 case 2:
6709 if (UseSHA512Intrinsics) {
6710 // we want to do an instanceof comparison against the SHA5 class
6711 klass_SHA_name = "sun/security/provider/SHA5";
6712 }
6713 break;
6714 default:
6715 fatal("unknown SHA intrinsic predicate: %d", predicate);
6716 }
6717
6718 ciKlass* klass_SHA = NULL;
6719 if (klass_SHA_name != NULL) {
6720 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6721 }
6722 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6723 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6724 Node* ctrl = control();
6725 set_control(top()); // no intrinsic path
6726 return ctrl;
6727 }
6728 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6729
6730 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6731 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6732 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6733 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6734
6735 return instof_false; // even if it is NULL
6736 }
6737
6738 bool LibraryCallKit::inline_profileBoolean() {
6739 Node* counts = argument(1);
6740 const TypeAryPtr* ary = NULL;
6741 ciArray* aobj = NULL;
6742 if (counts->is_Con()
6743 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6744 && (aobj = ary->const_oop()->as_array()) != NULL
6745 && (aobj->length() == 2)) {
6746 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6747 jint false_cnt = aobj->element_value(0).as_int();
6748 jint true_cnt = aobj->element_value(1).as_int();
6749
6750 if (C->log() != NULL) {
6751 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6752 false_cnt, true_cnt);
6753 }
6754
6755 if (false_cnt + true_cnt == 0) {
6756 // According to profile, never executed.
6757 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6758 Deoptimization::Action_reinterpret);
6759 return true;
6760 }
6761
6762 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6763 // is a number of each value occurrences.
6764 Node* result = argument(0);
6765 if (false_cnt == 0 || true_cnt == 0) {
6766 // According to profile, one value has been never seen.
6767 int expected_val = (false_cnt == 0) ? 1 : 0;
6768
6769 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6770 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6771
6772 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6773 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6774 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6775
6776 { // Slow path: uncommon trap for never seen value and then reexecute
6777 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6778 // the value has been seen at least once.
6779 PreserveJVMState pjvms(this);
6780 PreserveReexecuteState preexecs(this);
6781 jvms()->set_should_reexecute(true);
6782
6783 set_control(slow_path);
6784 set_i_o(i_o());
6785
6786 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6787 Deoptimization::Action_reinterpret);
6788 }
6789 // The guard for never seen value enables sharpening of the result and
6790 // returning a constant. It allows to eliminate branches on the same value
6791 // later on.
6792 set_control(fast_path);
6793 result = intcon(expected_val);
6794 }
6795 // Stop profiling.
6796 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6797 // By replacing method body with profile data (represented as ProfileBooleanNode
6798 // on IR level) we effectively disable profiling.
6799 // It enables full speed execution once optimized code is generated.
6800 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6801 C->record_for_igvn(profile);
6802 set_result(profile);
6803 return true;
6804 } else {
6805 // Continue profiling.
6806 // Profile data isn't available at the moment. So, execute method's bytecode version.
6807 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6808 // is compiled and counters aren't available since corresponding MethodHandle
6809 // isn't a compile-time constant.
6810 return false;
6811 }
6812 }
6813
6814 bool LibraryCallKit::inline_isCompileConstant() {
6815 Node* n = argument(0);
6816 set_result(n->is_Con() ? intcon(1) : intcon(0));
6817 return true;
6818 }