1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "oops/objArrayKlass.hpp"
32 #include "opto/addnode.hpp"
33 #include "opto/arraycopynode.hpp"
34 #include "opto/c2compiler.hpp"
35 #include "opto/callGenerator.hpp"
36 #include "opto/castnode.hpp"
37 #include "opto/cfgnode.hpp"
38 #include "opto/convertnode.hpp"
39 #include "opto/countbitsnode.hpp"
40 #include "opto/intrinsicnode.hpp"
41 #include "opto/idealKit.hpp"
42 #include "opto/mathexactnode.hpp"
43 #include "opto/movenode.hpp"
44 #include "opto/mulnode.hpp"
45 #include "opto/narrowptrnode.hpp"
46 #include "opto/opaquenode.hpp"
47 #include "opto/parse.hpp"
48 #include "opto/runtime.hpp"
49 #include "opto/subnode.hpp"
50 #include "prims/nativeLookup.hpp"
51 #include "prims/unsafe.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #ifdef TRACE_HAVE_INTRINSICS
54 #include "trace/traceMacros.hpp"
55 #endif
56
57 class LibraryIntrinsic : public InlineCallGenerator {
58 // Extend the set of intrinsics known to the runtime:
59 public:
60 private:
61 bool _is_virtual;
62 bool _does_virtual_dispatch;
63 int8_t _predicates_count; // Intrinsic is predicated by several conditions
64 int8_t _last_predicate; // Last generated predicate
65 vmIntrinsics::ID _intrinsic_id;
66
67 public:
68 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
69 : InlineCallGenerator(m),
70 _is_virtual(is_virtual),
71 _does_virtual_dispatch(does_virtual_dispatch),
72 _predicates_count((int8_t)predicates_count),
73 _last_predicate((int8_t)-1),
74 _intrinsic_id(id)
75 {
76 }
77 virtual bool is_intrinsic() const { return true; }
78 virtual bool is_virtual() const { return _is_virtual; }
79 virtual bool is_predicated() const { return _predicates_count > 0; }
80 virtual int predicates_count() const { return _predicates_count; }
81 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
82 virtual JVMState* generate(JVMState* jvms);
83 virtual Node* generate_predicate(JVMState* jvms, int predicate);
84 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
85 };
86
87
88 // Local helper class for LibraryIntrinsic:
89 class LibraryCallKit : public GraphKit {
90 private:
91 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
92 Node* _result; // the result node, if any
93 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
94
95 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr = false);
96
97 public:
98 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
99 : GraphKit(jvms),
100 _intrinsic(intrinsic),
101 _result(NULL)
102 {
103 // Check if this is a root compile. In that case we don't have a caller.
104 if (!jvms->has_method()) {
105 _reexecute_sp = sp();
106 } else {
107 // Find out how many arguments the interpreter needs when deoptimizing
108 // and save the stack pointer value so it can used by uncommon_trap.
109 // We find the argument count by looking at the declared signature.
110 bool ignored_will_link;
111 ciSignature* declared_signature = NULL;
112 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
113 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
114 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
115 }
116 }
117
118 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
119
120 ciMethod* caller() const { return jvms()->method(); }
121 int bci() const { return jvms()->bci(); }
122 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
123 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
124 ciMethod* callee() const { return _intrinsic->method(); }
125
126 bool try_to_inline(int predicate);
127 Node* try_to_predicate(int predicate);
128
129 void push_result() {
130 // Push the result onto the stack.
131 if (!stopped() && result() != NULL) {
132 BasicType bt = result()->bottom_type()->basic_type();
133 push_node(bt, result());
134 }
135 }
136
137 private:
138 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
139 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
140 }
141
142 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
143 void set_result(RegionNode* region, PhiNode* value);
144 Node* result() { return _result; }
145
146 virtual int reexecute_sp() { return _reexecute_sp; }
147
148 // Helper functions to inline natives
149 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
150 Node* generate_slow_guard(Node* test, RegionNode* region);
151 Node* generate_fair_guard(Node* test, RegionNode* region);
152 Node* generate_negative_guard(Node* index, RegionNode* region,
153 // resulting CastII of index:
154 Node* *pos_index = NULL);
155 Node* generate_limit_guard(Node* offset, Node* subseq_length,
156 Node* array_length,
157 RegionNode* region);
158 void generate_string_range_check(Node* array, Node* offset,
159 Node* length, bool char_count);
160 Node* generate_current_thread(Node* &tls_output);
161 Node* load_mirror_from_klass(Node* klass);
162 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
163 RegionNode* region, int null_path,
164 int offset);
165 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
166 RegionNode* region, int null_path) {
167 int offset = java_lang_Class::klass_offset_in_bytes();
168 return load_klass_from_mirror_common(mirror, never_see_null,
169 region, null_path,
170 offset);
171 }
172 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
173 RegionNode* region, int null_path) {
174 int offset = java_lang_Class::array_klass_offset_in_bytes();
175 return load_klass_from_mirror_common(mirror, never_see_null,
176 region, null_path,
177 offset);
178 }
179 Node* generate_access_flags_guard(Node* kls,
180 int modifier_mask, int modifier_bits,
181 RegionNode* region);
182 Node* generate_interface_guard(Node* kls, RegionNode* region);
183 Node* generate_array_guard(Node* kls, RegionNode* region) {
184 return generate_array_guard_common(kls, region, false, false);
185 }
186 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
187 return generate_array_guard_common(kls, region, false, true);
188 }
189 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
190 return generate_array_guard_common(kls, region, true, false);
191 }
192 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
193 return generate_array_guard_common(kls, region, true, true);
194 }
195 Node* generate_array_guard_common(Node* kls, RegionNode* region,
196 bool obj_array, bool not_array);
197 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
198 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
199 bool is_virtual = false, bool is_static = false);
200 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
201 return generate_method_call(method_id, false, true);
202 }
203 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
204 return generate_method_call(method_id, true, false);
205 }
206 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
207 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
208
209 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
210 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
211 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
213 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
214 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
215 bool inline_string_indexOfChar();
216 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_toBytesU();
218 bool inline_string_getCharsU();
219 bool inline_string_copy(bool compress);
220 bool inline_string_char_access(bool is_store);
221 Node* round_double_node(Node* n);
222 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
223 bool inline_math_native(vmIntrinsics::ID id);
224 bool inline_trig(vmIntrinsics::ID id);
225 bool inline_math(vmIntrinsics::ID id);
226 template <typename OverflowOp>
227 bool inline_math_overflow(Node* arg1, Node* arg2);
228 void inline_math_mathExact(Node* math, Node* test);
229 bool inline_math_addExactI(bool is_increment);
230 bool inline_math_addExactL(bool is_increment);
231 bool inline_math_multiplyExactI();
232 bool inline_math_multiplyExactL();
233 bool inline_math_negateExactI();
234 bool inline_math_negateExactL();
235 bool inline_math_subtractExactI(bool is_decrement);
236 bool inline_math_subtractExactL(bool is_decrement);
237 bool inline_min_max(vmIntrinsics::ID id);
238 bool inline_notify(vmIntrinsics::ID id);
239 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
240 // This returns Type::AnyPtr, RawPtr, or OopPtr.
241 int classify_unsafe_addr(Node* &base, Node* &offset);
242 Node* make_unsafe_address(Node* base, Node* offset);
243 // Helper for inline_unsafe_access.
244 // Generates the guards that check whether the result of
245 // Unsafe.getObject should be recorded in an SATB log buffer.
246 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
247
248 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
249 bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
250 static bool klass_needs_init_guard(Node* kls);
251 bool inline_unsafe_allocate();
252 bool inline_unsafe_newArray(bool uninitialized);
253 bool inline_unsafe_copyMemory();
254 bool inline_native_currentThread();
255
256 bool inline_native_time_funcs(address method, const char* funcName);
257 bool inline_native_isInterrupted();
258 bool inline_native_Class_query(vmIntrinsics::ID id);
259 bool inline_native_subtype_check();
260 bool inline_native_getLength();
261 bool inline_array_copyOf(bool is_copyOfRange);
262 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
263 bool inline_objects_checkIndex();
264 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
265 bool inline_native_clone(bool is_virtual);
266 bool inline_native_Reflection_getCallerClass();
267 // Helper function for inlining native object hash method
268 bool inline_native_hashcode(bool is_virtual, bool is_static);
269 bool inline_native_getClass();
270
271 // Helper functions for inlining arraycopy
272 bool inline_arraycopy();
273 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
274 RegionNode* slow_region);
275 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
276 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
277
278 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
279 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
280 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
281 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
282 bool inline_unsafe_fence(vmIntrinsics::ID id);
283 bool inline_fp_conversions(vmIntrinsics::ID id);
284 bool inline_number_methods(vmIntrinsics::ID id);
285 bool inline_reference_get();
286 bool inline_Class_cast();
287 bool inline_aescrypt_Block(vmIntrinsics::ID id);
288 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
289 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
290 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
291 Node* inline_counterMode_AESCrypt_predicate();
292 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
293 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
294 bool inline_ghash_processBlocks();
295 bool inline_sha_implCompress(vmIntrinsics::ID id);
296 bool inline_digestBase_implCompressMB(int predicate);
297 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
298 bool long_state, address stubAddr, const char *stubName,
299 Node* src_start, Node* ofs, Node* limit);
300 Node* get_state_from_sha_object(Node *sha_object);
301 Node* get_state_from_sha5_object(Node *sha_object);
302 Node* inline_digestBase_implCompressMB_predicate(int predicate);
303 bool inline_encodeISOArray();
304 bool inline_updateCRC32();
305 bool inline_updateBytesCRC32();
306 bool inline_updateByteBufferCRC32();
307 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
308 bool inline_updateBytesCRC32C();
309 bool inline_updateDirectByteBufferCRC32C();
310 bool inline_updateBytesAdler32();
311 bool inline_updateByteBufferAdler32();
312 bool inline_multiplyToLen();
313 bool inline_hasNegatives();
314 bool inline_squareToLen();
315 bool inline_mulAdd();
316 bool inline_montgomeryMultiply();
317 bool inline_montgomerySquare();
318 bool inline_vectorizedMismatch();
319
320 bool inline_profileBoolean();
321 bool inline_isCompileConstant();
322
323 bool inline_deoptimize();
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_native_ptr = true;
479 const bool is_static = true;
480 const bool is_volatile = true;
481
482 if (!jvms()->has_method()) {
483 // Root JVMState has a null method.
484 assert(map()->memory()->Opcode() == Op_Parm, "");
485 // Insert the memory aliasing node
486 set_all_memory(reset_memory());
487 }
488 assert(merged_memory(), "");
489
490
491 switch (intrinsic_id()) {
492 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
493 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
494 case vmIntrinsics::_getClass: return inline_native_getClass();
495
496 case vmIntrinsics::_dsin:
497 case vmIntrinsics::_dcos:
498 case vmIntrinsics::_dtan:
499 case vmIntrinsics::_dabs:
500 case vmIntrinsics::_datan2:
501 case vmIntrinsics::_dsqrt:
502 case vmIntrinsics::_dexp:
503 case vmIntrinsics::_dlog:
504 case vmIntrinsics::_dlog10:
505 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
506
507 case vmIntrinsics::_min:
508 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
509
510 case vmIntrinsics::_notify:
511 case vmIntrinsics::_notifyAll:
512 if (InlineNotify) {
513 return inline_notify(intrinsic_id());
514 }
515 return false;
516
517 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
518 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
519 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
520 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
521 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
522 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
523 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
524 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
525 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
526 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
527 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
528 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
529
530 case vmIntrinsics::_arraycopy: return inline_arraycopy();
531
532 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
533 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
534 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
535 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
536
537 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
538 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
539 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
540 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
541 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
542 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
543 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
544
545 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
546 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
547
548 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
549 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
550 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
551 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
552
553 case vmIntrinsics::_compressStringC:
554 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
555 case vmIntrinsics::_inflateStringC:
556 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
557
558 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Relaxed, false);
559 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Relaxed, false);
560 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Relaxed, false);
561 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, false);
562 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, false);
563 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, false);
564 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, false);
565 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
566 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
567
568 case vmIntrinsics::_putObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Relaxed, false);
569 case vmIntrinsics::_putBoolean: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Relaxed, false);
570 case vmIntrinsics::_putByte: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Relaxed, false);
571 case vmIntrinsics::_putShort: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, false);
572 case vmIntrinsics::_putChar: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, false);
573 case vmIntrinsics::_putInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, false);
574 case vmIntrinsics::_putLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, false);
575 case vmIntrinsics::_putFloat: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Relaxed, false);
576 case vmIntrinsics::_putDouble: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
577
578 case vmIntrinsics::_getByte_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE, Relaxed, false);
579 case vmIntrinsics::_getShort_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT, Relaxed, false);
580 case vmIntrinsics::_getChar_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR, Relaxed, false);
581 case vmIntrinsics::_getInt_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_INT, Relaxed, false);
582 case vmIntrinsics::_getLong_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_LONG, Relaxed, false);
583 case vmIntrinsics::_getFloat_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
584 case vmIntrinsics::_getDouble_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
585 case vmIntrinsics::_getAddress_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS, Relaxed, false);
586
587 case vmIntrinsics::_putByte_raw: return inline_unsafe_access( is_native_ptr, is_store, T_BYTE, Relaxed, false);
588 case vmIntrinsics::_putShort_raw: return inline_unsafe_access( is_native_ptr, is_store, T_SHORT, Relaxed, false);
589 case vmIntrinsics::_putChar_raw: return inline_unsafe_access( is_native_ptr, is_store, T_CHAR, Relaxed, false);
590 case vmIntrinsics::_putInt_raw: return inline_unsafe_access( is_native_ptr, is_store, T_INT, Relaxed, false);
591 case vmIntrinsics::_putLong_raw: return inline_unsafe_access( is_native_ptr, is_store, T_LONG, Relaxed, false);
592 case vmIntrinsics::_putFloat_raw: return inline_unsafe_access( is_native_ptr, is_store, T_FLOAT, Relaxed, false);
593 case vmIntrinsics::_putDouble_raw: return inline_unsafe_access( is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
594 case vmIntrinsics::_putAddress_raw: return inline_unsafe_access( is_native_ptr, is_store, T_ADDRESS, Relaxed, false);
595
596 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Volatile, false);
597 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Volatile, false);
598 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Volatile, false);
599 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Volatile, false);
600 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Volatile, false);
601 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Volatile, false);
602 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Volatile, false);
603 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Volatile, false);
604 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Volatile, false);
605
606 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Volatile, false);
607 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Volatile, false);
608 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Volatile, false);
609 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Volatile, false);
610 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Volatile, false);
611 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Volatile, false);
612 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Volatile, false);
613 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Volatile, false);
614 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Volatile, false);
615
616 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, true);
617 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, true);
618 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, true);
619 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, true);
620
621 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, true);
622 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, true);
623 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, true);
624 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, true);
625
626 case vmIntrinsics::_putOrderedObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false);
627 case vmIntrinsics::_putOrderedInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false);
628 case vmIntrinsics::_putOrderedLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false);
629
630 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Acquire, false);
631 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Acquire, false);
632 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Acquire, false);
633 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Acquire, false);
634 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Acquire, false);
635 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Acquire, false);
636 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Acquire, false);
637 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Acquire, false);
638 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Acquire, false);
639
640 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false);
641 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Release, false);
642 case vmIntrinsics::_putByteRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Release, false);
643 case vmIntrinsics::_putShortRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Release, false);
644 case vmIntrinsics::_putCharRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Release, false);
645 case vmIntrinsics::_putIntRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false);
646 case vmIntrinsics::_putLongRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false);
647 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Release, false);
648 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Release, false);
649
650 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Opaque, false);
651 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Opaque, false);
652 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Opaque, false);
653 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Opaque, false);
654 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Opaque, false);
655 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Opaque, false);
656 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Opaque, false);
657 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Opaque, false);
658 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Opaque, false);
659
660 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Opaque, false);
661 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Opaque, false);
662 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Opaque, false);
663 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Opaque, false);
664 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Opaque, false);
665 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Opaque, false);
666 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Opaque, false);
667 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Opaque, false);
668 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Opaque, false);
669
670 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
671 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
672 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
673
674 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
675 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
676 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
677 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
678 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
679 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
680 case vmIntrinsics::_weakCompareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
681 case vmIntrinsics::_weakCompareAndSwapLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
682 case vmIntrinsics::_weakCompareAndSwapLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
683
684 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
685 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
686 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
687 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
688 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
689 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
690 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
691 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
692 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
693
694 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
695 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
696 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
697 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
698 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
699
700 case vmIntrinsics::_loadFence:
701 case vmIntrinsics::_storeFence:
702 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
703
704 case vmIntrinsics::_currentThread: return inline_native_currentThread();
705 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
706
707 #ifdef TRACE_HAVE_INTRINSICS
708 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
709 #endif
710 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
711 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
712 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
713 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
714 case vmIntrinsics::_getLength: return inline_native_getLength();
715 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
716 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
717 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
718 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
719 case vmIntrinsics::_Objects_checkIndex: return inline_objects_checkIndex();
720 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
721
722 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
723 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
724
725 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
726
727 case vmIntrinsics::_isInstance:
728 case vmIntrinsics::_getModifiers:
729 case vmIntrinsics::_isInterface:
730 case vmIntrinsics::_isArray:
731 case vmIntrinsics::_isPrimitive:
732 case vmIntrinsics::_getSuperclass:
733 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
734
735 case vmIntrinsics::_floatToRawIntBits:
736 case vmIntrinsics::_floatToIntBits:
737 case vmIntrinsics::_intBitsToFloat:
738 case vmIntrinsics::_doubleToRawLongBits:
739 case vmIntrinsics::_doubleToLongBits:
740 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
741
742 case vmIntrinsics::_numberOfLeadingZeros_i:
743 case vmIntrinsics::_numberOfLeadingZeros_l:
744 case vmIntrinsics::_numberOfTrailingZeros_i:
745 case vmIntrinsics::_numberOfTrailingZeros_l:
746 case vmIntrinsics::_bitCount_i:
747 case vmIntrinsics::_bitCount_l:
748 case vmIntrinsics::_reverseBytes_i:
749 case vmIntrinsics::_reverseBytes_l:
750 case vmIntrinsics::_reverseBytes_s:
751 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
752
753 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
754
755 case vmIntrinsics::_Reference_get: return inline_reference_get();
756
757 case vmIntrinsics::_Class_cast: return inline_Class_cast();
758
759 case vmIntrinsics::_aescrypt_encryptBlock:
760 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
761
762 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
763 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
764 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
765
766 case vmIntrinsics::_counterMode_AESCrypt:
767 return inline_counterMode_AESCrypt(intrinsic_id());
768
769 case vmIntrinsics::_sha_implCompress:
770 case vmIntrinsics::_sha2_implCompress:
771 case vmIntrinsics::_sha5_implCompress:
772 return inline_sha_implCompress(intrinsic_id());
773
774 case vmIntrinsics::_digestBase_implCompressMB:
775 return inline_digestBase_implCompressMB(predicate);
776
777 case vmIntrinsics::_multiplyToLen:
778 return inline_multiplyToLen();
779
780 case vmIntrinsics::_squareToLen:
781 return inline_squareToLen();
782
783 case vmIntrinsics::_mulAdd:
784 return inline_mulAdd();
785
786 case vmIntrinsics::_montgomeryMultiply:
787 return inline_montgomeryMultiply();
788 case vmIntrinsics::_montgomerySquare:
789 return inline_montgomerySquare();
790
791 case vmIntrinsics::_vectorizedMismatch:
792 return inline_vectorizedMismatch();
793
794 case vmIntrinsics::_ghash_processBlocks:
795 return inline_ghash_processBlocks();
796
797 case vmIntrinsics::_encodeISOArray:
798 case vmIntrinsics::_encodeByteISOArray:
799 return inline_encodeISOArray();
800
801 case vmIntrinsics::_updateCRC32:
802 return inline_updateCRC32();
803 case vmIntrinsics::_updateBytesCRC32:
804 return inline_updateBytesCRC32();
805 case vmIntrinsics::_updateByteBufferCRC32:
806 return inline_updateByteBufferCRC32();
807
808 case vmIntrinsics::_updateBytesCRC32C:
809 return inline_updateBytesCRC32C();
810 case vmIntrinsics::_updateDirectByteBufferCRC32C:
811 return inline_updateDirectByteBufferCRC32C();
812
813 case vmIntrinsics::_updateBytesAdler32:
814 return inline_updateBytesAdler32();
815 case vmIntrinsics::_updateByteBufferAdler32:
816 return inline_updateByteBufferAdler32();
817
818 case vmIntrinsics::_profileBoolean:
819 return inline_profileBoolean();
820 case vmIntrinsics::_isCompileConstant:
821 return inline_isCompileConstant();
822
823 case vmIntrinsics::_hasNegatives:
824 return inline_hasNegatives();
825
826 case vmIntrinsics::_deoptimize:
827 return inline_deoptimize();
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_objects_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 case vmIntrinsics::_dlog10: n = new Log10DNode(C, control(), arg); break;
1689 default: fatal_unexpected_iid(id); break;
1690 }
1691 set_result(_gvn.transform(n));
1692 return true;
1693 }
1694
1695 //------------------------------inline_trig----------------------------------
1696 // Inline sin/cos/tan instructions, if possible. If rounding is required, do
1697 // argument reduction which will turn into a fast/slow diamond.
1698 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1699 Node* arg = round_double_node(argument(0));
1700 Node* n = NULL;
1701
1702 switch (id) {
1703 case vmIntrinsics::_dtan: n = new TanDNode(C, control(), arg); break;
1704 default: fatal_unexpected_iid(id); break;
1705 }
1706 n = _gvn.transform(n);
1707
1708 // Rounding required? Check for argument reduction!
1709 if (Matcher::strict_fp_requires_explicit_rounding) {
1710 static const double pi_4 = 0.7853981633974483;
1711 static const double neg_pi_4 = -0.7853981633974483;
1712 // pi/2 in 80-bit extended precision
1713 // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
1714 // -pi/2 in 80-bit extended precision
1715 // 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};
1716 // Cutoff value for using this argument reduction technique
1717 //static const double pi_2_minus_epsilon = 1.564660403643354;
1718 //static const double neg_pi_2_plus_epsilon = -1.564660403643354;
1719
1720 // Pseudocode for sin:
1721 // if (x <= Math.PI / 4.0) {
1722 // if (x >= -Math.PI / 4.0) return fsin(x);
1723 // if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
1724 // } else {
1725 // if (x <= Math.PI / 2.0) return fcos(x - Math.PI / 2.0);
1726 // }
1727 // return StrictMath.sin(x);
1728
1729 // Pseudocode for cos:
1730 // if (x <= Math.PI / 4.0) {
1731 // if (x >= -Math.PI / 4.0) return fcos(x);
1732 // if (x >= -Math.PI / 2.0) return fsin(x + Math.PI / 2.0);
1733 // } else {
1734 // if (x <= Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
1735 // }
1736 // return StrictMath.cos(x);
1737
1738 // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
1739 // requires a special machine instruction to load it. Instead we'll try
1740 // the 'easy' case. If we really need the extra range +/- PI/2 we'll
1741 // probably do the math inside the SIN encoding.
1742
1743 // Make the merge point
1744 RegionNode* r = new RegionNode(3);
1745 Node* phi = new PhiNode(r, Type::DOUBLE);
1746
1747 // Flatten arg so we need only 1 test
1748 Node *abs = _gvn.transform(new AbsDNode(arg));
1749 // Node for PI/4 constant
1750 Node *pi4 = makecon(TypeD::make(pi_4));
1751 // Check PI/4 : abs(arg)
1752 Node *cmp = _gvn.transform(new CmpDNode(pi4,abs));
1753 // Check: If PI/4 < abs(arg) then go slow
1754 Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt ));
1755 // Branch either way
1756 IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1757 set_control(opt_iff(r,iff));
1758
1759 // Set fast path result
1760 phi->init_req(2, n);
1761
1762 // Slow path - non-blocking leaf call
1763 Node* call = NULL;
1764 switch (id) {
1765 case vmIntrinsics::_dtan:
1766 call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1767 CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
1768 "Tan", NULL, arg, top());
1769 break;
1770 }
1771 assert(control()->in(0) == call, "");
1772 Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1773 r->init_req(1, control());
1774 phi->init_req(1, slow_result);
1775
1776 // Post-merge
1777 set_control(_gvn.transform(r));
1778 record_for_igvn(r);
1779 n = _gvn.transform(phi);
1780
1781 C->set_has_split_ifs(true); // Has chance for split-if optimization
1782 }
1783 set_result(n);
1784 return true;
1785 }
1786
1787 //------------------------------runtime_math-----------------------------
1788 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1789 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1790 "must be (DD)D or (D)D type");
1791
1792 // Inputs
1793 Node* a = round_double_node(argument(0));
1794 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1795
1796 const TypePtr* no_memory_effects = NULL;
1797 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1798 no_memory_effects,
1799 a, top(), b, b ? top() : NULL);
1800 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1801 #ifdef ASSERT
1802 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1803 assert(value_top == top(), "second value must be top");
1804 #endif
1805
1806 set_result(value);
1807 return true;
1808 }
1809
1810 //------------------------------inline_math_native-----------------------------
1811 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1812 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1813 switch (id) {
1814 // These intrinsics are not properly supported on all hardware
1815 case vmIntrinsics::_dsin:
1816 return StubRoutines::dsin() != NULL ?
1817 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1818 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1819 case vmIntrinsics::_dcos:
1820 return StubRoutines::dcos() != NULL ?
1821 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1822 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1823 case vmIntrinsics::_dtan: return Matcher::has_match_rule(Op_TanD) ? inline_trig(id) :
1824 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1825
1826 case vmIntrinsics::_dlog:
1827 return StubRoutines::dlog() != NULL ?
1828 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1829 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1830 case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_math(id) :
1831 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1832
1833 // These intrinsics are supported on all hardware
1834 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1835 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1836
1837 case vmIntrinsics::_dexp:
1838 return StubRoutines::dexp() != NULL ?
1839 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1840 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1841 case vmIntrinsics::_dpow:
1842 return StubRoutines::dpow() != NULL ?
1843 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1844 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1845 #undef FN_PTR
1846
1847 // These intrinsics are not yet correctly implemented
1848 case vmIntrinsics::_datan2:
1849 return false;
1850
1851 default:
1852 fatal_unexpected_iid(id);
1853 return false;
1854 }
1855 }
1856
1857 static bool is_simple_name(Node* n) {
1858 return (n->req() == 1 // constant
1859 || (n->is_Type() && n->as_Type()->type()->singleton())
1860 || n->is_Proj() // parameter or return value
1861 || n->is_Phi() // local of some sort
1862 );
1863 }
1864
1865 //----------------------------inline_notify-----------------------------------*
1866 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1867 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1868 address func;
1869 if (id == vmIntrinsics::_notify) {
1870 func = OptoRuntime::monitor_notify_Java();
1871 } else {
1872 func = OptoRuntime::monitor_notifyAll_Java();
1873 }
1874 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1875 make_slow_call_ex(call, env()->Throwable_klass(), false);
1876 return true;
1877 }
1878
1879
1880 //----------------------------inline_min_max-----------------------------------
1881 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1882 set_result(generate_min_max(id, argument(0), argument(1)));
1883 return true;
1884 }
1885
1886 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1887 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1888 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1889 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1890 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1891
1892 {
1893 PreserveJVMState pjvms(this);
1894 PreserveReexecuteState preexecs(this);
1895 jvms()->set_should_reexecute(true);
1896
1897 set_control(slow_path);
1898 set_i_o(i_o());
1899
1900 uncommon_trap(Deoptimization::Reason_intrinsic,
1901 Deoptimization::Action_none);
1902 }
1903
1904 set_control(fast_path);
1905 set_result(math);
1906 }
1907
1908 template <typename OverflowOp>
1909 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1910 typedef typename OverflowOp::MathOp MathOp;
1911
1912 MathOp* mathOp = new MathOp(arg1, arg2);
1913 Node* operation = _gvn.transform( mathOp );
1914 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1915 inline_math_mathExact(operation, ofcheck);
1916 return true;
1917 }
1918
1919 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1920 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1921 }
1922
1923 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1924 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1925 }
1926
1927 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1928 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1929 }
1930
1931 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1932 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1933 }
1934
1935 bool LibraryCallKit::inline_math_negateExactI() {
1936 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1937 }
1938
1939 bool LibraryCallKit::inline_math_negateExactL() {
1940 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1941 }
1942
1943 bool LibraryCallKit::inline_math_multiplyExactI() {
1944 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1945 }
1946
1947 bool LibraryCallKit::inline_math_multiplyExactL() {
1948 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1949 }
1950
1951 Node*
1952 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1953 // These are the candidate return value:
1954 Node* xvalue = x0;
1955 Node* yvalue = y0;
1956
1957 if (xvalue == yvalue) {
1958 return xvalue;
1959 }
1960
1961 bool want_max = (id == vmIntrinsics::_max);
1962
1963 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1964 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1965 if (txvalue == NULL || tyvalue == NULL) return top();
1966 // This is not really necessary, but it is consistent with a
1967 // hypothetical MaxINode::Value method:
1968 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1969
1970 // %%% This folding logic should (ideally) be in a different place.
1971 // Some should be inside IfNode, and there to be a more reliable
1972 // transformation of ?: style patterns into cmoves. We also want
1973 // more powerful optimizations around cmove and min/max.
1974
1975 // Try to find a dominating comparison of these guys.
1976 // It can simplify the index computation for Arrays.copyOf
1977 // and similar uses of System.arraycopy.
1978 // First, compute the normalized version of CmpI(x, y).
1979 int cmp_op = Op_CmpI;
1980 Node* xkey = xvalue;
1981 Node* ykey = yvalue;
1982 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1983 if (ideal_cmpxy->is_Cmp()) {
1984 // E.g., if we have CmpI(length - offset, count),
1985 // it might idealize to CmpI(length, count + offset)
1986 cmp_op = ideal_cmpxy->Opcode();
1987 xkey = ideal_cmpxy->in(1);
1988 ykey = ideal_cmpxy->in(2);
1989 }
1990
1991 // Start by locating any relevant comparisons.
1992 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1993 Node* cmpxy = NULL;
1994 Node* cmpyx = NULL;
1995 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1996 Node* cmp = start_from->fast_out(k);
1997 if (cmp->outcnt() > 0 && // must have prior uses
1998 cmp->in(0) == NULL && // must be context-independent
1999 cmp->Opcode() == cmp_op) { // right kind of compare
2000 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2001 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2002 }
2003 }
2004
2005 const int NCMPS = 2;
2006 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2007 int cmpn;
2008 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2009 if (cmps[cmpn] != NULL) break; // find a result
2010 }
2011 if (cmpn < NCMPS) {
2012 // Look for a dominating test that tells us the min and max.
2013 int depth = 0; // Limit search depth for speed
2014 Node* dom = control();
2015 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2016 if (++depth >= 100) break;
2017 Node* ifproj = dom;
2018 if (!ifproj->is_Proj()) continue;
2019 Node* iff = ifproj->in(0);
2020 if (!iff->is_If()) continue;
2021 Node* bol = iff->in(1);
2022 if (!bol->is_Bool()) continue;
2023 Node* cmp = bol->in(1);
2024 if (cmp == NULL) continue;
2025 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2026 if (cmps[cmpn] == cmp) break;
2027 if (cmpn == NCMPS) continue;
2028 BoolTest::mask btest = bol->as_Bool()->_test._test;
2029 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2030 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2031 // At this point, we know that 'x btest y' is true.
2032 switch (btest) {
2033 case BoolTest::eq:
2034 // They are proven equal, so we can collapse the min/max.
2035 // Either value is the answer. Choose the simpler.
2036 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2037 return yvalue;
2038 return xvalue;
2039 case BoolTest::lt: // x < y
2040 case BoolTest::le: // x <= y
2041 return (want_max ? yvalue : xvalue);
2042 case BoolTest::gt: // x > y
2043 case BoolTest::ge: // x >= y
2044 return (want_max ? xvalue : yvalue);
2045 }
2046 }
2047 }
2048
2049 // We failed to find a dominating test.
2050 // Let's pick a test that might GVN with prior tests.
2051 Node* best_bol = NULL;
2052 BoolTest::mask best_btest = BoolTest::illegal;
2053 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2054 Node* cmp = cmps[cmpn];
2055 if (cmp == NULL) continue;
2056 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2057 Node* bol = cmp->fast_out(j);
2058 if (!bol->is_Bool()) continue;
2059 BoolTest::mask btest = bol->as_Bool()->_test._test;
2060 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2061 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2062 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2063 best_bol = bol->as_Bool();
2064 best_btest = btest;
2065 }
2066 }
2067 }
2068
2069 Node* answer_if_true = NULL;
2070 Node* answer_if_false = NULL;
2071 switch (best_btest) {
2072 default:
2073 if (cmpxy == NULL)
2074 cmpxy = ideal_cmpxy;
2075 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2076 // and fall through:
2077 case BoolTest::lt: // x < y
2078 case BoolTest::le: // x <= y
2079 answer_if_true = (want_max ? yvalue : xvalue);
2080 answer_if_false = (want_max ? xvalue : yvalue);
2081 break;
2082 case BoolTest::gt: // x > y
2083 case BoolTest::ge: // x >= y
2084 answer_if_true = (want_max ? xvalue : yvalue);
2085 answer_if_false = (want_max ? yvalue : xvalue);
2086 break;
2087 }
2088
2089 jint hi, lo;
2090 if (want_max) {
2091 // We can sharpen the minimum.
2092 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2093 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2094 } else {
2095 // We can sharpen the maximum.
2096 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2097 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2098 }
2099
2100 // Use a flow-free graph structure, to avoid creating excess control edges
2101 // which could hinder other optimizations.
2102 // Since Math.min/max is often used with arraycopy, we want
2103 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2104 Node* cmov = CMoveNode::make(NULL, best_bol,
2105 answer_if_false, answer_if_true,
2106 TypeInt::make(lo, hi, widen));
2107
2108 return _gvn.transform(cmov);
2109
2110 /*
2111 // This is not as desirable as it may seem, since Min and Max
2112 // nodes do not have a full set of optimizations.
2113 // And they would interfere, anyway, with 'if' optimizations
2114 // and with CMoveI canonical forms.
2115 switch (id) {
2116 case vmIntrinsics::_min:
2117 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2118 case vmIntrinsics::_max:
2119 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2120 default:
2121 ShouldNotReachHere();
2122 }
2123 */
2124 }
2125
2126 inline int
2127 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2128 const TypePtr* base_type = TypePtr::NULL_PTR;
2129 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2130 if (base_type == NULL) {
2131 // Unknown type.
2132 return Type::AnyPtr;
2133 } else if (base_type == TypePtr::NULL_PTR) {
2134 // Since this is a NULL+long form, we have to switch to a rawptr.
2135 base = _gvn.transform(new CastX2PNode(offset));
2136 offset = MakeConX(0);
2137 return Type::RawPtr;
2138 } else if (base_type->base() == Type::RawPtr) {
2139 return Type::RawPtr;
2140 } else if (base_type->isa_oopptr()) {
2141 // Base is never null => always a heap address.
2142 if (base_type->ptr() == TypePtr::NotNull) {
2143 return Type::OopPtr;
2144 }
2145 // Offset is small => always a heap address.
2146 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2147 if (offset_type != NULL &&
2148 base_type->offset() == 0 && // (should always be?)
2149 offset_type->_lo >= 0 &&
2150 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2151 return Type::OopPtr;
2152 }
2153 // Otherwise, it might either be oop+off or NULL+addr.
2154 return Type::AnyPtr;
2155 } else {
2156 // No information:
2157 return Type::AnyPtr;
2158 }
2159 }
2160
2161 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2162 int kind = classify_unsafe_addr(base, offset);
2163 if (kind == Type::RawPtr) {
2164 return basic_plus_adr(top(), base, offset);
2165 } else {
2166 return basic_plus_adr(base, offset);
2167 }
2168 }
2169
2170 //--------------------------inline_number_methods-----------------------------
2171 // inline int Integer.numberOfLeadingZeros(int)
2172 // inline int Long.numberOfLeadingZeros(long)
2173 //
2174 // inline int Integer.numberOfTrailingZeros(int)
2175 // inline int Long.numberOfTrailingZeros(long)
2176 //
2177 // inline int Integer.bitCount(int)
2178 // inline int Long.bitCount(long)
2179 //
2180 // inline char Character.reverseBytes(char)
2181 // inline short Short.reverseBytes(short)
2182 // inline int Integer.reverseBytes(int)
2183 // inline long Long.reverseBytes(long)
2184 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2185 Node* arg = argument(0);
2186 Node* n = NULL;
2187 switch (id) {
2188 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2189 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2190 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2191 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2192 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2193 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2194 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2195 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2196 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2197 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2198 default: fatal_unexpected_iid(id); break;
2199 }
2200 set_result(_gvn.transform(n));
2201 return true;
2202 }
2203
2204 //----------------------------inline_unsafe_access----------------------------
2205
2206 const static BasicType T_ADDRESS_HOLDER = T_LONG;
2207
2208 // Helper that guards and inserts a pre-barrier.
2209 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2210 Node* pre_val, bool need_mem_bar) {
2211 // We could be accessing the referent field of a reference object. If so, when G1
2212 // is enabled, we need to log the value in the referent field in an SATB buffer.
2213 // This routine performs some compile time filters and generates suitable
2214 // runtime filters that guard the pre-barrier code.
2215 // Also add memory barrier for non volatile load from the referent field
2216 // to prevent commoning of loads across safepoint.
2217 if (!UseG1GC && !need_mem_bar)
2218 return;
2219
2220 // Some compile time checks.
2221
2222 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2223 const TypeX* otype = offset->find_intptr_t_type();
2224 if (otype != NULL && otype->is_con() &&
2225 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2226 // Constant offset but not the reference_offset so just return
2227 return;
2228 }
2229
2230 // We only need to generate the runtime guards for instances.
2231 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2232 if (btype != NULL) {
2233 if (btype->isa_aryptr()) {
2234 // Array type so nothing to do
2235 return;
2236 }
2237
2238 const TypeInstPtr* itype = btype->isa_instptr();
2239 if (itype != NULL) {
2240 // Can the klass of base_oop be statically determined to be
2241 // _not_ a sub-class of Reference and _not_ Object?
2242 ciKlass* klass = itype->klass();
2243 if ( klass->is_loaded() &&
2244 !klass->is_subtype_of(env()->Reference_klass()) &&
2245 !env()->Object_klass()->is_subtype_of(klass)) {
2246 return;
2247 }
2248 }
2249 }
2250
2251 // The compile time filters did not reject base_oop/offset so
2252 // we need to generate the following runtime filters
2253 //
2254 // if (offset == java_lang_ref_Reference::_reference_offset) {
2255 // if (instance_of(base, java.lang.ref.Reference)) {
2256 // pre_barrier(_, pre_val, ...);
2257 // }
2258 // }
2259
2260 float likely = PROB_LIKELY( 0.999);
2261 float unlikely = PROB_UNLIKELY(0.999);
2262
2263 IdealKit ideal(this);
2264 #define __ ideal.
2265
2266 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2267
2268 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2269 // Update graphKit memory and control from IdealKit.
2270 sync_kit(ideal);
2271
2272 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2273 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2274
2275 // Update IdealKit memory and control from graphKit.
2276 __ sync_kit(this);
2277
2278 Node* one = __ ConI(1);
2279 // is_instof == 0 if base_oop == NULL
2280 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2281
2282 // Update graphKit from IdeakKit.
2283 sync_kit(ideal);
2284
2285 // Use the pre-barrier to record the value in the referent field
2286 pre_barrier(false /* do_load */,
2287 __ ctrl(),
2288 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2289 pre_val /* pre_val */,
2290 T_OBJECT);
2291 if (need_mem_bar) {
2292 // Add memory barrier to prevent commoning reads from this field
2293 // across safepoint since GC can change its value.
2294 insert_mem_bar(Op_MemBarCPUOrder);
2295 }
2296 // Update IdealKit from graphKit.
2297 __ sync_kit(this);
2298
2299 } __ end_if(); // _ref_type != ref_none
2300 } __ end_if(); // offset == referent_offset
2301
2302 // Final sync IdealKit and GraphKit.
2303 final_sync(ideal);
2304 #undef __
2305 }
2306
2307
2308 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
2309 // Attempt to infer a sharper value type from the offset and base type.
2310 ciKlass* sharpened_klass = NULL;
2311
2312 // See if it is an instance field, with an object type.
2313 if (alias_type->field() != NULL) {
2314 assert(!is_native_ptr, "native pointer op cannot use a java address");
2315 if (alias_type->field()->type()->is_klass()) {
2316 sharpened_klass = alias_type->field()->type()->as_klass();
2317 }
2318 }
2319
2320 // See if it is a narrow oop array.
2321 if (adr_type->isa_aryptr()) {
2322 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2323 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2324 if (elem_type != NULL) {
2325 sharpened_klass = elem_type->klass();
2326 }
2327 }
2328 }
2329
2330 // The sharpened class might be unloaded if there is no class loader
2331 // contraint in place.
2332 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2333 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2334
2335 #ifndef PRODUCT
2336 if (C->print_intrinsics() || C->print_inlining()) {
2337 tty->print(" from base type: "); adr_type->dump();
2338 tty->print(" sharpened value: "); tjp->dump();
2339 }
2340 #endif
2341 // Sharpen the value type.
2342 return tjp;
2343 }
2344 return NULL;
2345 }
2346
2347 bool LibraryCallKit::inline_unsafe_access(const bool is_native_ptr, bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2348 if (callee()->is_static()) return false; // caller must have the capability!
2349 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2350 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2351
2352 #ifndef PRODUCT
2353 {
2354 ResourceMark rm;
2355 // Check the signatures.
2356 ciSignature* sig = callee()->signature();
2357 #ifdef ASSERT
2358 if (!is_store) {
2359 // Object getObject(Object base, int/long offset), etc.
2360 BasicType rtype = sig->return_type()->basic_type();
2361 if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
2362 rtype = T_ADDRESS; // it is really a C void*
2363 assert(rtype == type, "getter must return the expected value");
2364 if (!is_native_ptr) {
2365 assert(sig->count() == 2, "oop getter has 2 arguments");
2366 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2367 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2368 } else {
2369 assert(sig->count() == 1, "native getter has 1 argument");
2370 assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
2371 }
2372 } else {
2373 // void putObject(Object base, int/long offset, Object x), etc.
2374 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2375 if (!is_native_ptr) {
2376 assert(sig->count() == 3, "oop putter has 3 arguments");
2377 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2378 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2379 } else {
2380 assert(sig->count() == 2, "native putter has 2 arguments");
2381 assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
2382 }
2383 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2384 if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
2385 vtype = T_ADDRESS; // it is really a C void*
2386 assert(vtype == type, "putter must accept the expected value");
2387 }
2388 #endif // ASSERT
2389 }
2390 #endif //PRODUCT
2391
2392 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2393
2394 Node* receiver = argument(0); // type: oop
2395
2396 // Build address expression.
2397 Node* adr;
2398 Node* heap_base_oop = top();
2399 Node* offset = top();
2400 Node* val;
2401
2402 if (!is_native_ptr) {
2403 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2404 Node* base = argument(1); // type: oop
2405 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2406 offset = argument(2); // type: long
2407 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2408 // to be plain byte offsets, which are also the same as those accepted
2409 // by oopDesc::field_base.
2410 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2411 "fieldOffset must be byte-scaled");
2412 // 32-bit machines ignore the high half!
2413 offset = ConvL2X(offset);
2414 adr = make_unsafe_address(base, offset);
2415 heap_base_oop = base;
2416 val = is_store ? argument(4) : NULL;
2417 } else {
2418 Node* ptr = argument(1); // type: long
2419 ptr = ConvL2X(ptr); // adjust Java long to machine word
2420 adr = make_unsafe_address(NULL, ptr);
2421 val = is_store ? argument(3) : NULL;
2422 }
2423
2424 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2425
2426 // First guess at the value type.
2427 const Type *value_type = Type::get_const_basic_type(type);
2428
2429 // Try to categorize the address. If it comes up as TypeJavaPtr::BOTTOM,
2430 // there was not enough information to nail it down.
2431 Compile::AliasType* alias_type = C->alias_type(adr_type);
2432 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2433
2434 // We will need memory barriers unless we can determine a unique
2435 // alias category for this reference. (Note: If for some reason
2436 // the barriers get omitted and the unsafe reference begins to "pollute"
2437 // the alias analysis of the rest of the graph, either Compile::can_alias
2438 // or Compile::must_alias will throw a diagnostic assert.)
2439 bool need_mem_bar;
2440 switch (kind) {
2441 case Relaxed:
2442 need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2443 break;
2444 case Opaque:
2445 // Opaque uses CPUOrder membars for protection against code movement.
2446 case Acquire:
2447 case Release:
2448 case Volatile:
2449 need_mem_bar = true;
2450 break;
2451 default:
2452 ShouldNotReachHere();
2453 }
2454
2455 // Some accesses require access atomicity for all types, notably longs and doubles.
2456 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2457 bool requires_atomic_access = false;
2458 switch (kind) {
2459 case Relaxed:
2460 case Opaque:
2461 requires_atomic_access = AlwaysAtomicAccesses;
2462 break;
2463 case Acquire:
2464 case Release:
2465 case Volatile:
2466 requires_atomic_access = true;
2467 break;
2468 default:
2469 ShouldNotReachHere();
2470 }
2471
2472 // Figure out the memory ordering.
2473 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2474 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2475
2476 // If we are reading the value of the referent field of a Reference
2477 // object (either by using Unsafe directly or through reflection)
2478 // then, if G1 is enabled, we need to record the referent in an
2479 // SATB log buffer using the pre-barrier mechanism.
2480 // Also we need to add memory barrier to prevent commoning reads
2481 // from this field across safepoint since GC can change its value.
2482 bool need_read_barrier = !is_native_ptr && !is_store &&
2483 offset != top() && heap_base_oop != top();
2484
2485 if (!is_store && type == T_OBJECT) {
2486 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
2487 if (tjp != NULL) {
2488 value_type = tjp;
2489 }
2490 }
2491
2492 receiver = null_check(receiver);
2493 if (stopped()) {
2494 return true;
2495 }
2496 // Heap pointers get a null-check from the interpreter,
2497 // as a courtesy. However, this is not guaranteed by Unsafe,
2498 // and it is not possible to fully distinguish unintended nulls
2499 // from intended ones in this API.
2500
2501 // We need to emit leading and trailing CPU membars (see below) in
2502 // addition to memory membars for special access modes. This is a little
2503 // too strong, but avoids the need to insert per-alias-type
2504 // volatile membars (for stores; compare Parse::do_put_xxx), which
2505 // we cannot do effectively here because we probably only have a
2506 // rough approximation of type.
2507
2508 switch(kind) {
2509 case Relaxed:
2510 case Opaque:
2511 case Acquire:
2512 break;
2513 case Release:
2514 case Volatile:
2515 if (is_store) {
2516 insert_mem_bar(Op_MemBarRelease);
2517 } else {
2518 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2519 insert_mem_bar(Op_MemBarVolatile);
2520 }
2521 }
2522 break;
2523 default:
2524 ShouldNotReachHere();
2525 }
2526
2527 // Memory barrier to prevent normal and 'unsafe' accesses from
2528 // bypassing each other. Happens after null checks, so the
2529 // exception paths do not take memory state from the memory barrier,
2530 // so there's no problems making a strong assert about mixing users
2531 // of safe & unsafe memory.
2532 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2533
2534 assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2535 alias_type->field() != NULL || alias_type->element() != NULL, "field, array element or unknown");
2536 bool mismatched = false;
2537 if (alias_type->element() != NULL || alias_type->field() != NULL) {
2538 BasicType bt;
2539 if (alias_type->element() != NULL) {
2540 // Use address type to get the element type. Alias type doesn't provide
2541 // enough information (e.g., doesn't differentiate between byte[] and boolean[]).
2542 const Type* element = adr_type->is_aryptr()->elem();
2543 bt = element->isa_narrowoop() ? T_OBJECT : element->array_element_basic_type();
2544 } else {
2545 bt = alias_type->field()->layout_type();
2546 }
2547 if (bt == T_ARRAY) {
2548 // accessing an array field with getObject is not a mismatch
2549 bt = T_OBJECT;
2550 }
2551 if (bt != type) {
2552 mismatched = true;
2553 }
2554 }
2555 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2556
2557 if (!is_store) {
2558 Node* p = NULL;
2559 // Try to constant fold a load from a constant field
2560 ciField* field = alias_type->field();
2561 if (heap_base_oop != top() &&
2562 field != NULL && field->is_constant() && !mismatched) {
2563 // final or stable field
2564 const Type* con_type = Type::make_constant(alias_type->field(), heap_base_oop);
2565 if (con_type != NULL) {
2566 p = makecon(con_type);
2567 }
2568 }
2569 if (p == NULL) {
2570 // To be valid, unsafe loads may depend on other conditions than
2571 // the one that guards them: pin the Load node
2572 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2573 // load value
2574 switch (type) {
2575 case T_BOOLEAN:
2576 case T_CHAR:
2577 case T_BYTE:
2578 case T_SHORT:
2579 case T_INT:
2580 case T_LONG:
2581 case T_FLOAT:
2582 case T_DOUBLE:
2583 break;
2584 case T_OBJECT:
2585 if (need_read_barrier) {
2586 // We do not require a mem bar inside pre_barrier if need_mem_bar
2587 // is set: the barriers would be emitted by us.
2588 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2589 }
2590 break;
2591 case T_ADDRESS:
2592 // Cast to an int type.
2593 p = _gvn.transform(new CastP2XNode(NULL, p));
2594 p = ConvX2UL(p);
2595 break;
2596 default:
2597 fatal("unexpected type %d: %s", type, type2name(type));
2598 break;
2599 }
2600 }
2601 // The load node has the control of the preceding MemBarCPUOrder. All
2602 // following nodes will have the control of the MemBarCPUOrder inserted at
2603 // the end of this method. So, pushing the load onto the stack at a later
2604 // point is fine.
2605 set_result(p);
2606 } else {
2607 // place effect of store into memory
2608 switch (type) {
2609 case T_DOUBLE:
2610 val = dstore_rounding(val);
2611 break;
2612 case T_ADDRESS:
2613 // Repackage the long as a pointer.
2614 val = ConvL2X(val);
2615 val = _gvn.transform(new CastX2PNode(val));
2616 break;
2617 }
2618
2619 if (type != T_OBJECT) {
2620 (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2621 } else {
2622 // Possibly an oop being stored to Java heap or native memory
2623 if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2624 // oop to Java heap.
2625 (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2626 } else {
2627 // We can't tell at compile time if we are storing in the Java heap or outside
2628 // of it. So we need to emit code to conditionally do the proper type of
2629 // store.
2630
2631 IdealKit ideal(this);
2632 #define __ ideal.
2633 // QQQ who knows what probability is here??
2634 __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2635 // Sync IdealKit and graphKit.
2636 sync_kit(ideal);
2637 Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2638 // Update IdealKit memory.
2639 __ sync_kit(this);
2640 } __ else_(); {
2641 __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched);
2642 } __ end_if();
2643 // Final sync IdealKit and GraphKit.
2644 final_sync(ideal);
2645 #undef __
2646 }
2647 }
2648 }
2649
2650 switch(kind) {
2651 case Relaxed:
2652 case Opaque:
2653 case Release:
2654 break;
2655 case Acquire:
2656 case Volatile:
2657 if (!is_store) {
2658 insert_mem_bar(Op_MemBarAcquire);
2659 } else {
2660 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2661 insert_mem_bar(Op_MemBarVolatile);
2662 }
2663 }
2664 break;
2665 default:
2666 ShouldNotReachHere();
2667 }
2668
2669 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2670
2671 return true;
2672 }
2673
2674 //----------------------------inline_unsafe_load_store----------------------------
2675 // This method serves a couple of different customers (depending on LoadStoreKind):
2676 //
2677 // LS_cmp_swap:
2678 //
2679 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2680 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2681 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2682 //
2683 // LS_cmp_swap_weak:
2684 //
2685 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2686 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2687 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2688 //
2689 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2690 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2691 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2692 //
2693 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2694 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2695 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2696 //
2697 // LS_cmp_exchange:
2698 //
2699 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2700 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2701 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2702 //
2703 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2704 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2705 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2706 //
2707 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2708 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2709 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2710 //
2711 // LS_get_add:
2712 //
2713 // int getAndAddInt( Object o, long offset, int delta)
2714 // long getAndAddLong(Object o, long offset, long delta)
2715 //
2716 // LS_get_set:
2717 //
2718 // int getAndSet(Object o, long offset, int newValue)
2719 // long getAndSet(Object o, long offset, long newValue)
2720 // Object getAndSet(Object o, long offset, Object newValue)
2721 //
2722 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2723 // This basic scheme here is the same as inline_unsafe_access, but
2724 // differs in enough details that combining them would make the code
2725 // overly confusing. (This is a true fact! I originally combined
2726 // them, but even I was confused by it!) As much code/comments as
2727 // possible are retained from inline_unsafe_access though to make
2728 // the correspondences clearer. - dl
2729
2730 if (callee()->is_static()) return false; // caller must have the capability!
2731
2732 #ifndef PRODUCT
2733 BasicType rtype;
2734 {
2735 ResourceMark rm;
2736 // Check the signatures.
2737 ciSignature* sig = callee()->signature();
2738 rtype = sig->return_type()->basic_type();
2739 switch(kind) {
2740 case LS_get_add:
2741 case LS_get_set: {
2742 // Check the signatures.
2743 #ifdef ASSERT
2744 assert(rtype == type, "get and set must return the expected type");
2745 assert(sig->count() == 3, "get and set has 3 arguments");
2746 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2747 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2748 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2749 #endif // ASSERT
2750 break;
2751 }
2752 case LS_cmp_swap:
2753 case LS_cmp_swap_weak: {
2754 // Check the signatures.
2755 #ifdef ASSERT
2756 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2757 assert(sig->count() == 4, "CAS has 4 arguments");
2758 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2759 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2760 #endif // ASSERT
2761 break;
2762 }
2763 case LS_cmp_exchange: {
2764 // Check the signatures.
2765 #ifdef ASSERT
2766 assert(rtype == type, "CAS must return the expected type");
2767 assert(sig->count() == 4, "CAS has 4 arguments");
2768 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2769 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2770 #endif // ASSERT
2771 break;
2772 }
2773 default:
2774 ShouldNotReachHere();
2775 }
2776 }
2777 #endif //PRODUCT
2778
2779 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2780
2781 // Get arguments:
2782 Node* receiver = NULL;
2783 Node* base = NULL;
2784 Node* offset = NULL;
2785 Node* oldval = NULL;
2786 Node* newval = NULL;
2787 switch(kind) {
2788 case LS_cmp_swap:
2789 case LS_cmp_swap_weak:
2790 case LS_cmp_exchange: {
2791 const bool two_slot_type = type2size[type] == 2;
2792 receiver = argument(0); // type: oop
2793 base = argument(1); // type: oop
2794 offset = argument(2); // type: long
2795 oldval = argument(4); // type: oop, int, or long
2796 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2797 break;
2798 }
2799 case LS_get_add:
2800 case LS_get_set: {
2801 receiver = argument(0); // type: oop
2802 base = argument(1); // type: oop
2803 offset = argument(2); // type: long
2804 oldval = NULL;
2805 newval = argument(4); // type: oop, int, or long
2806 break;
2807 }
2808 default:
2809 ShouldNotReachHere();
2810 }
2811
2812 // Null check receiver.
2813 receiver = null_check(receiver);
2814 if (stopped()) {
2815 return true;
2816 }
2817
2818 // Build field offset expression.
2819 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2820 // to be plain byte offsets, which are also the same as those accepted
2821 // by oopDesc::field_base.
2822 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2823 // 32-bit machines ignore the high half of long offsets
2824 offset = ConvL2X(offset);
2825 Node* adr = make_unsafe_address(base, offset);
2826 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2827
2828 // For CAS, unlike inline_unsafe_access, there seems no point in
2829 // trying to refine types. Just use the coarse types here.
2830 const Type *value_type = Type::get_const_basic_type(type);
2831 Compile::AliasType* alias_type = C->alias_type(adr_type);
2832 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2833
2834 switch (kind) {
2835 case LS_get_set:
2836 case LS_cmp_exchange: {
2837 if (type == T_OBJECT) {
2838 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2839 if (tjp != NULL) {
2840 value_type = tjp;
2841 }
2842 }
2843 break;
2844 }
2845 case LS_cmp_swap:
2846 case LS_cmp_swap_weak:
2847 case LS_get_add:
2848 break;
2849 default:
2850 ShouldNotReachHere();
2851 }
2852
2853 int alias_idx = C->get_alias_index(adr_type);
2854
2855 // Memory-model-wise, a LoadStore acts like a little synchronized
2856 // block, so needs barriers on each side. These don't translate
2857 // into actual barriers on most machines, but we still need rest of
2858 // compiler to respect ordering.
2859
2860 switch (access_kind) {
2861 case Relaxed:
2862 case Acquire:
2863 break;
2864 case Release:
2865 case Volatile:
2866 insert_mem_bar(Op_MemBarRelease);
2867 break;
2868 default:
2869 ShouldNotReachHere();
2870 }
2871 insert_mem_bar(Op_MemBarCPUOrder);
2872
2873 // Figure out the memory ordering.
2874 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2875
2876 // 4984716: MemBars must be inserted before this
2877 // memory node in order to avoid a false
2878 // dependency which will confuse the scheduler.
2879 Node *mem = memory(alias_idx);
2880
2881 // For now, we handle only those cases that actually exist: ints,
2882 // longs, and Object. Adding others should be straightforward.
2883 Node* load_store = NULL;
2884 switch(type) {
2885 case T_INT:
2886 switch(kind) {
2887 case LS_get_add:
2888 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2889 break;
2890 case LS_get_set:
2891 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2892 break;
2893 case LS_cmp_swap_weak:
2894 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2895 break;
2896 case LS_cmp_swap:
2897 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2898 break;
2899 case LS_cmp_exchange:
2900 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2901 break;
2902 default:
2903 ShouldNotReachHere();
2904 }
2905 break;
2906 case T_LONG:
2907 switch(kind) {
2908 case LS_get_add:
2909 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2910 break;
2911 case LS_get_set:
2912 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2913 break;
2914 case LS_cmp_swap_weak:
2915 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2916 break;
2917 case LS_cmp_swap:
2918 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2919 break;
2920 case LS_cmp_exchange:
2921 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2922 break;
2923 default:
2924 ShouldNotReachHere();
2925 }
2926 break;
2927 case T_OBJECT:
2928 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2929 // could be delayed during Parse (for example, in adjust_map_after_if()).
2930 // Execute transformation here to avoid barrier generation in such case.
2931 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2932 newval = _gvn.makecon(TypePtr::NULL_PTR);
2933
2934 // Reference stores need a store barrier.
2935 switch(kind) {
2936 case LS_get_set: {
2937 // If pre-barrier must execute before the oop store, old value will require do_load here.
2938 if (!can_move_pre_barrier()) {
2939 pre_barrier(true /* do_load*/,
2940 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2941 NULL /* pre_val*/,
2942 T_OBJECT);
2943 } // Else move pre_barrier to use load_store value, see below.
2944 break;
2945 }
2946 case LS_cmp_swap_weak:
2947 case LS_cmp_swap:
2948 case LS_cmp_exchange: {
2949 // Same as for newval above:
2950 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2951 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2952 }
2953 // The only known value which might get overwritten is oldval.
2954 pre_barrier(false /* do_load */,
2955 control(), NULL, NULL, max_juint, NULL, NULL,
2956 oldval /* pre_val */,
2957 T_OBJECT);
2958 break;
2959 }
2960 default:
2961 ShouldNotReachHere();
2962 }
2963
2964 #ifdef _LP64
2965 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2966 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2967
2968 switch(kind) {
2969 case LS_get_set:
2970 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2971 break;
2972 case LS_cmp_swap_weak: {
2973 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2974 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2975 break;
2976 }
2977 case LS_cmp_swap: {
2978 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2979 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2980 break;
2981 }
2982 case LS_cmp_exchange: {
2983 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2984 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2985 break;
2986 }
2987 default:
2988 ShouldNotReachHere();
2989 }
2990 } else
2991 #endif
2992 switch (kind) {
2993 case LS_get_set:
2994 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2995 break;
2996 case LS_cmp_swap_weak:
2997 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2998 break;
2999 case LS_cmp_swap:
3000 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
3001 break;
3002 case LS_cmp_exchange:
3003 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
3004 break;
3005 default:
3006 ShouldNotReachHere();
3007 }
3008
3009 // Emit the post barrier only when the actual store happened. This makes sense
3010 // to check only for LS_cmp_* that can fail to set the value.
3011 // LS_cmp_exchange does not produce any branches by default, so there is no
3012 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
3013 // CompareAndExchange and move branches here, it would make sense to conditionalize
3014 // post_barriers for LS_cmp_exchange as well.
3015 //
3016 // CAS success path is marked more likely since we anticipate this is a performance
3017 // critical path, while CAS failure path can use the penalty for going through unlikely
3018 // path as backoff. Which is still better than doing a store barrier there.
3019 switch (kind) {
3020 case LS_get_set:
3021 case LS_cmp_exchange: {
3022 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3023 break;
3024 }
3025 case LS_cmp_swap_weak:
3026 case LS_cmp_swap: {
3027 IdealKit ideal(this);
3028 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3029 sync_kit(ideal);
3030 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3031 ideal.sync_kit(this);
3032 } ideal.end_if();
3033 final_sync(ideal);
3034 break;
3035 }
3036 default:
3037 ShouldNotReachHere();
3038 }
3039 break;
3040 default:
3041 fatal("unexpected type %d: %s", type, type2name(type));
3042 break;
3043 }
3044
3045 // SCMemProjNodes represent the memory state of a LoadStore. Their
3046 // main role is to prevent LoadStore nodes from being optimized away
3047 // when their results aren't used.
3048 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3049 set_memory(proj, alias_idx);
3050
3051 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3052 #ifdef _LP64
3053 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3054 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3055 }
3056 #endif
3057 if (can_move_pre_barrier()) {
3058 // Don't need to load pre_val. The old value is returned by load_store.
3059 // The pre_barrier can execute after the xchg as long as no safepoint
3060 // gets inserted between them.
3061 pre_barrier(false /* do_load */,
3062 control(), NULL, NULL, max_juint, NULL, NULL,
3063 load_store /* pre_val */,
3064 T_OBJECT);
3065 }
3066 }
3067
3068 // Add the trailing membar surrounding the access
3069 insert_mem_bar(Op_MemBarCPUOrder);
3070
3071 switch (access_kind) {
3072 case Relaxed:
3073 case Release:
3074 break; // do nothing
3075 case Acquire:
3076 case Volatile:
3077 insert_mem_bar(Op_MemBarAcquire);
3078 break;
3079 default:
3080 ShouldNotReachHere();
3081 }
3082
3083 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3084 set_result(load_store);
3085 return true;
3086 }
3087
3088 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3089 MemNode::MemOrd mo = MemNode::unset;
3090 switch(kind) {
3091 case Opaque:
3092 case Relaxed: mo = MemNode::unordered; break;
3093 case Acquire: mo = MemNode::acquire; break;
3094 case Release: mo = MemNode::release; break;
3095 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3096 default:
3097 ShouldNotReachHere();
3098 }
3099 guarantee(mo != MemNode::unset, "Should select memory ordering");
3100 return mo;
3101 }
3102
3103 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3104 MemNode::MemOrd mo = MemNode::unset;
3105 switch(kind) {
3106 case Opaque:
3107 case Relaxed: mo = MemNode::unordered; break;
3108 case Acquire: mo = MemNode::acquire; break;
3109 case Release: mo = MemNode::release; break;
3110 case Volatile: mo = MemNode::seqcst; break;
3111 default:
3112 ShouldNotReachHere();
3113 }
3114 guarantee(mo != MemNode::unset, "Should select memory ordering");
3115 return mo;
3116 }
3117
3118 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3119 // Regardless of form, don't allow previous ld/st to move down,
3120 // then issue acquire, release, or volatile mem_bar.
3121 insert_mem_bar(Op_MemBarCPUOrder);
3122 switch(id) {
3123 case vmIntrinsics::_loadFence:
3124 insert_mem_bar(Op_LoadFence);
3125 return true;
3126 case vmIntrinsics::_storeFence:
3127 insert_mem_bar(Op_StoreFence);
3128 return true;
3129 case vmIntrinsics::_fullFence:
3130 insert_mem_bar(Op_MemBarVolatile);
3131 return true;
3132 default:
3133 fatal_unexpected_iid(id);
3134 return false;
3135 }
3136 }
3137
3138 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3139 if (!kls->is_Con()) {
3140 return true;
3141 }
3142 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3143 if (klsptr == NULL) {
3144 return true;
3145 }
3146 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3147 // don't need a guard for a klass that is already initialized
3148 return !ik->is_initialized();
3149 }
3150
3151 //----------------------------inline_unsafe_allocate---------------------------
3152 // public native Object Unsafe.allocateInstance(Class<?> cls);
3153 bool LibraryCallKit::inline_unsafe_allocate() {
3154 if (callee()->is_static()) return false; // caller must have the capability!
3155
3156 null_check_receiver(); // null-check, then ignore
3157 Node* cls = null_check(argument(1));
3158 if (stopped()) return true;
3159
3160 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3161 kls = null_check(kls);
3162 if (stopped()) return true; // argument was like int.class
3163
3164 Node* test = NULL;
3165 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3166 // Note: The argument might still be an illegal value like
3167 // Serializable.class or Object[].class. The runtime will handle it.
3168 // But we must make an explicit check for initialization.
3169 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3170 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3171 // can generate code to load it as unsigned byte.
3172 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3173 Node* bits = intcon(InstanceKlass::fully_initialized);
3174 test = _gvn.transform(new SubINode(inst, bits));
3175 // The 'test' is non-zero if we need to take a slow path.
3176 }
3177
3178 Node* obj = new_instance(kls, test);
3179 set_result(obj);
3180 return true;
3181 }
3182
3183 //------------------------inline_native_time_funcs--------------
3184 // inline code for System.currentTimeMillis() and System.nanoTime()
3185 // these have the same type and signature
3186 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3187 const TypeFunc* tf = OptoRuntime::void_long_Type();
3188 const TypePtr* no_memory_effects = NULL;
3189 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3190 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3191 #ifdef ASSERT
3192 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3193 assert(value_top == top(), "second value must be top");
3194 #endif
3195 set_result(value);
3196 return true;
3197 }
3198
3199 //------------------------inline_native_currentThread------------------
3200 bool LibraryCallKit::inline_native_currentThread() {
3201 Node* junk = NULL;
3202 set_result(generate_current_thread(junk));
3203 return true;
3204 }
3205
3206 //------------------------inline_native_isInterrupted------------------
3207 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3208 bool LibraryCallKit::inline_native_isInterrupted() {
3209 // Add a fast path to t.isInterrupted(clear_int):
3210 // (t == Thread.current() &&
3211 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3212 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3213 // So, in the common case that the interrupt bit is false,
3214 // we avoid making a call into the VM. Even if the interrupt bit
3215 // is true, if the clear_int argument is false, we avoid the VM call.
3216 // However, if the receiver is not currentThread, we must call the VM,
3217 // because there must be some locking done around the operation.
3218
3219 // We only go to the fast case code if we pass two guards.
3220 // Paths which do not pass are accumulated in the slow_region.
3221
3222 enum {
3223 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3224 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3225 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3226 PATH_LIMIT
3227 };
3228
3229 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3230 // out of the function.
3231 insert_mem_bar(Op_MemBarCPUOrder);
3232
3233 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3234 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3235
3236 RegionNode* slow_region = new RegionNode(1);
3237 record_for_igvn(slow_region);
3238
3239 // (a) Receiving thread must be the current thread.
3240 Node* rec_thr = argument(0);
3241 Node* tls_ptr = NULL;
3242 Node* cur_thr = generate_current_thread(tls_ptr);
3243 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3244 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3245
3246 generate_slow_guard(bol_thr, slow_region);
3247
3248 // (b) Interrupt bit on TLS must be false.
3249 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3250 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3251 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3252
3253 // Set the control input on the field _interrupted read to prevent it floating up.
3254 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3255 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3256 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3257
3258 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3259
3260 // First fast path: if (!TLS._interrupted) return false;
3261 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3262 result_rgn->init_req(no_int_result_path, false_bit);
3263 result_val->init_req(no_int_result_path, intcon(0));
3264
3265 // drop through to next case
3266 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3267
3268 #ifndef TARGET_OS_FAMILY_windows
3269 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3270 Node* clr_arg = argument(1);
3271 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3272 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3273 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3274
3275 // Second fast path: ... else if (!clear_int) return true;
3276 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3277 result_rgn->init_req(no_clear_result_path, false_arg);
3278 result_val->init_req(no_clear_result_path, intcon(1));
3279
3280 // drop through to next case
3281 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3282 #else
3283 // To return true on Windows you must read the _interrupted field
3284 // and check the event state i.e. take the slow path.
3285 #endif // TARGET_OS_FAMILY_windows
3286
3287 // (d) Otherwise, go to the slow path.
3288 slow_region->add_req(control());
3289 set_control( _gvn.transform(slow_region));
3290
3291 if (stopped()) {
3292 // There is no slow path.
3293 result_rgn->init_req(slow_result_path, top());
3294 result_val->init_req(slow_result_path, top());
3295 } else {
3296 // non-virtual because it is a private non-static
3297 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3298
3299 Node* slow_val = set_results_for_java_call(slow_call);
3300 // this->control() comes from set_results_for_java_call
3301
3302 Node* fast_io = slow_call->in(TypeFunc::I_O);
3303 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3304
3305 // These two phis are pre-filled with copies of of the fast IO and Memory
3306 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3307 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3308
3309 result_rgn->init_req(slow_result_path, control());
3310 result_io ->init_req(slow_result_path, i_o());
3311 result_mem->init_req(slow_result_path, reset_memory());
3312 result_val->init_req(slow_result_path, slow_val);
3313
3314 set_all_memory(_gvn.transform(result_mem));
3315 set_i_o( _gvn.transform(result_io));
3316 }
3317
3318 C->set_has_split_ifs(true); // Has chance for split-if optimization
3319 set_result(result_rgn, result_val);
3320 return true;
3321 }
3322
3323 //---------------------------load_mirror_from_klass----------------------------
3324 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3325 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3326 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3327 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3328 }
3329
3330 //-----------------------load_klass_from_mirror_common-------------------------
3331 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3332 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3333 // and branch to the given path on the region.
3334 // If never_see_null, take an uncommon trap on null, so we can optimistically
3335 // compile for the non-null case.
3336 // If the region is NULL, force never_see_null = true.
3337 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3338 bool never_see_null,
3339 RegionNode* region,
3340 int null_path,
3341 int offset) {
3342 if (region == NULL) never_see_null = true;
3343 Node* p = basic_plus_adr(mirror, offset);
3344 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3345 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3346 Node* null_ctl = top();
3347 kls = null_check_oop(kls, &null_ctl, never_see_null);
3348 if (region != NULL) {
3349 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3350 region->init_req(null_path, null_ctl);
3351 } else {
3352 assert(null_ctl == top(), "no loose ends");
3353 }
3354 return kls;
3355 }
3356
3357 //--------------------(inline_native_Class_query helpers)---------------------
3358 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3359 // Fall through if (mods & mask) == bits, take the guard otherwise.
3360 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3361 // Branch around if the given klass has the given modifier bit set.
3362 // Like generate_guard, adds a new path onto the region.
3363 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3364 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3365 Node* mask = intcon(modifier_mask);
3366 Node* bits = intcon(modifier_bits);
3367 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3368 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3369 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3370 return generate_fair_guard(bol, region);
3371 }
3372 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3373 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3374 }
3375
3376 //-------------------------inline_native_Class_query-------------------
3377 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3378 const Type* return_type = TypeInt::BOOL;
3379 Node* prim_return_value = top(); // what happens if it's a primitive class?
3380 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3381 bool expect_prim = false; // most of these guys expect to work on refs
3382
3383 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3384
3385 Node* mirror = argument(0);
3386 Node* obj = top();
3387
3388 switch (id) {
3389 case vmIntrinsics::_isInstance:
3390 // nothing is an instance of a primitive type
3391 prim_return_value = intcon(0);
3392 obj = argument(1);
3393 break;
3394 case vmIntrinsics::_getModifiers:
3395 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3396 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3397 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3398 break;
3399 case vmIntrinsics::_isInterface:
3400 prim_return_value = intcon(0);
3401 break;
3402 case vmIntrinsics::_isArray:
3403 prim_return_value = intcon(0);
3404 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3405 break;
3406 case vmIntrinsics::_isPrimitive:
3407 prim_return_value = intcon(1);
3408 expect_prim = true; // obviously
3409 break;
3410 case vmIntrinsics::_getSuperclass:
3411 prim_return_value = null();
3412 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3413 break;
3414 case vmIntrinsics::_getClassAccessFlags:
3415 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3416 return_type = TypeInt::INT; // not bool! 6297094
3417 break;
3418 default:
3419 fatal_unexpected_iid(id);
3420 break;
3421 }
3422
3423 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3424 if (mirror_con == NULL) return false; // cannot happen?
3425
3426 #ifndef PRODUCT
3427 if (C->print_intrinsics() || C->print_inlining()) {
3428 ciType* k = mirror_con->java_mirror_type();
3429 if (k) {
3430 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3431 k->print_name();
3432 tty->cr();
3433 }
3434 }
3435 #endif
3436
3437 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3438 RegionNode* region = new RegionNode(PATH_LIMIT);
3439 record_for_igvn(region);
3440 PhiNode* phi = new PhiNode(region, return_type);
3441
3442 // The mirror will never be null of Reflection.getClassAccessFlags, however
3443 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3444 // if it is. See bug 4774291.
3445
3446 // For Reflection.getClassAccessFlags(), the null check occurs in
3447 // the wrong place; see inline_unsafe_access(), above, for a similar
3448 // situation.
3449 mirror = null_check(mirror);
3450 // If mirror or obj is dead, only null-path is taken.
3451 if (stopped()) return true;
3452
3453 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3454
3455 // Now load the mirror's klass metaobject, and null-check it.
3456 // Side-effects region with the control path if the klass is null.
3457 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3458 // If kls is null, we have a primitive mirror.
3459 phi->init_req(_prim_path, prim_return_value);
3460 if (stopped()) { set_result(region, phi); return true; }
3461 bool safe_for_replace = (region->in(_prim_path) == top());
3462
3463 Node* p; // handy temp
3464 Node* null_ctl;
3465
3466 // Now that we have the non-null klass, we can perform the real query.
3467 // For constant classes, the query will constant-fold in LoadNode::Value.
3468 Node* query_value = top();
3469 switch (id) {
3470 case vmIntrinsics::_isInstance:
3471 // nothing is an instance of a primitive type
3472 query_value = gen_instanceof(obj, kls, safe_for_replace);
3473 break;
3474
3475 case vmIntrinsics::_getModifiers:
3476 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3477 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3478 break;
3479
3480 case vmIntrinsics::_isInterface:
3481 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3482 if (generate_interface_guard(kls, region) != NULL)
3483 // A guard was added. If the guard is taken, it was an interface.
3484 phi->add_req(intcon(1));
3485 // If we fall through, it's a plain class.
3486 query_value = intcon(0);
3487 break;
3488
3489 case vmIntrinsics::_isArray:
3490 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3491 if (generate_array_guard(kls, region) != NULL)
3492 // A guard was added. If the guard is taken, it was an array.
3493 phi->add_req(intcon(1));
3494 // If we fall through, it's a plain class.
3495 query_value = intcon(0);
3496 break;
3497
3498 case vmIntrinsics::_isPrimitive:
3499 query_value = intcon(0); // "normal" path produces false
3500 break;
3501
3502 case vmIntrinsics::_getSuperclass:
3503 // The rules here are somewhat unfortunate, but we can still do better
3504 // with random logic than with a JNI call.
3505 // Interfaces store null or Object as _super, but must report null.
3506 // Arrays store an intermediate super as _super, but must report Object.
3507 // Other types can report the actual _super.
3508 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3509 if (generate_interface_guard(kls, region) != NULL)
3510 // A guard was added. If the guard is taken, it was an interface.
3511 phi->add_req(null());
3512 if (generate_array_guard(kls, region) != NULL)
3513 // A guard was added. If the guard is taken, it was an array.
3514 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3515 // If we fall through, it's a plain class. Get its _super.
3516 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3517 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3518 null_ctl = top();
3519 kls = null_check_oop(kls, &null_ctl);
3520 if (null_ctl != top()) {
3521 // If the guard is taken, Object.superClass is null (both klass and mirror).
3522 region->add_req(null_ctl);
3523 phi ->add_req(null());
3524 }
3525 if (!stopped()) {
3526 query_value = load_mirror_from_klass(kls);
3527 }
3528 break;
3529
3530 case vmIntrinsics::_getClassAccessFlags:
3531 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3532 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3533 break;
3534
3535 default:
3536 fatal_unexpected_iid(id);
3537 break;
3538 }
3539
3540 // Fall-through is the normal case of a query to a real class.
3541 phi->init_req(1, query_value);
3542 region->init_req(1, control());
3543
3544 C->set_has_split_ifs(true); // Has chance for split-if optimization
3545 set_result(region, phi);
3546 return true;
3547 }
3548
3549 //-------------------------inline_Class_cast-------------------
3550 bool LibraryCallKit::inline_Class_cast() {
3551 Node* mirror = argument(0); // Class
3552 Node* obj = argument(1);
3553 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3554 if (mirror_con == NULL) {
3555 return false; // dead path (mirror->is_top()).
3556 }
3557 if (obj == NULL || obj->is_top()) {
3558 return false; // dead path
3559 }
3560 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3561
3562 // First, see if Class.cast() can be folded statically.
3563 // java_mirror_type() returns non-null for compile-time Class constants.
3564 ciType* tm = mirror_con->java_mirror_type();
3565 if (tm != NULL && tm->is_klass() &&
3566 tp != NULL && tp->klass() != NULL) {
3567 if (!tp->klass()->is_loaded()) {
3568 // Don't use intrinsic when class is not loaded.
3569 return false;
3570 } else {
3571 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3572 if (static_res == Compile::SSC_always_true) {
3573 // isInstance() is true - fold the code.
3574 set_result(obj);
3575 return true;
3576 } else if (static_res == Compile::SSC_always_false) {
3577 // Don't use intrinsic, have to throw ClassCastException.
3578 // If the reference is null, the non-intrinsic bytecode will
3579 // be optimized appropriately.
3580 return false;
3581 }
3582 }
3583 }
3584
3585 // Bailout intrinsic and do normal inlining if exception path is frequent.
3586 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3587 return false;
3588 }
3589
3590 // Generate dynamic checks.
3591 // Class.cast() is java implementation of _checkcast bytecode.
3592 // Do checkcast (Parse::do_checkcast()) optimizations here.
3593
3594 mirror = null_check(mirror);
3595 // If mirror is dead, only null-path is taken.
3596 if (stopped()) {
3597 return true;
3598 }
3599
3600 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3601 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3602 RegionNode* region = new RegionNode(PATH_LIMIT);
3603 record_for_igvn(region);
3604
3605 // Now load the mirror's klass metaobject, and null-check it.
3606 // If kls is null, we have a primitive mirror and
3607 // nothing is an instance of a primitive type.
3608 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3609
3610 Node* res = top();
3611 if (!stopped()) {
3612 Node* bad_type_ctrl = top();
3613 // Do checkcast optimizations.
3614 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3615 region->init_req(_bad_type_path, bad_type_ctrl);
3616 }
3617 if (region->in(_prim_path) != top() ||
3618 region->in(_bad_type_path) != top()) {
3619 // Let Interpreter throw ClassCastException.
3620 PreserveJVMState pjvms(this);
3621 set_control(_gvn.transform(region));
3622 uncommon_trap(Deoptimization::Reason_intrinsic,
3623 Deoptimization::Action_maybe_recompile);
3624 }
3625 if (!stopped()) {
3626 set_result(res);
3627 }
3628 return true;
3629 }
3630
3631
3632 //--------------------------inline_native_subtype_check------------------------
3633 // This intrinsic takes the JNI calls out of the heart of
3634 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3635 bool LibraryCallKit::inline_native_subtype_check() {
3636 // Pull both arguments off the stack.
3637 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3638 args[0] = argument(0);
3639 args[1] = argument(1);
3640 Node* klasses[2]; // corresponding Klasses: superk, subk
3641 klasses[0] = klasses[1] = top();
3642
3643 enum {
3644 // A full decision tree on {superc is prim, subc is prim}:
3645 _prim_0_path = 1, // {P,N} => false
3646 // {P,P} & superc!=subc => false
3647 _prim_same_path, // {P,P} & superc==subc => true
3648 _prim_1_path, // {N,P} => false
3649 _ref_subtype_path, // {N,N} & subtype check wins => true
3650 _both_ref_path, // {N,N} & subtype check loses => false
3651 PATH_LIMIT
3652 };
3653
3654 RegionNode* region = new RegionNode(PATH_LIMIT);
3655 Node* phi = new PhiNode(region, TypeInt::BOOL);
3656 record_for_igvn(region);
3657
3658 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3659 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3660 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3661
3662 // First null-check both mirrors and load each mirror's klass metaobject.
3663 int which_arg;
3664 for (which_arg = 0; which_arg <= 1; which_arg++) {
3665 Node* arg = args[which_arg];
3666 arg = null_check(arg);
3667 if (stopped()) break;
3668 args[which_arg] = arg;
3669
3670 Node* p = basic_plus_adr(arg, class_klass_offset);
3671 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3672 klasses[which_arg] = _gvn.transform(kls);
3673 }
3674
3675 // Having loaded both klasses, test each for null.
3676 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3677 for (which_arg = 0; which_arg <= 1; which_arg++) {
3678 Node* kls = klasses[which_arg];
3679 Node* null_ctl = top();
3680 kls = null_check_oop(kls, &null_ctl, never_see_null);
3681 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3682 region->init_req(prim_path, null_ctl);
3683 if (stopped()) break;
3684 klasses[which_arg] = kls;
3685 }
3686
3687 if (!stopped()) {
3688 // now we have two reference types, in klasses[0..1]
3689 Node* subk = klasses[1]; // the argument to isAssignableFrom
3690 Node* superk = klasses[0]; // the receiver
3691 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3692 // now we have a successful reference subtype check
3693 region->set_req(_ref_subtype_path, control());
3694 }
3695
3696 // If both operands are primitive (both klasses null), then
3697 // we must return true when they are identical primitives.
3698 // It is convenient to test this after the first null klass check.
3699 set_control(region->in(_prim_0_path)); // go back to first null check
3700 if (!stopped()) {
3701 // Since superc is primitive, make a guard for the superc==subc case.
3702 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3703 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3704 generate_guard(bol_eq, region, PROB_FAIR);
3705 if (region->req() == PATH_LIMIT+1) {
3706 // A guard was added. If the added guard is taken, superc==subc.
3707 region->swap_edges(PATH_LIMIT, _prim_same_path);
3708 region->del_req(PATH_LIMIT);
3709 }
3710 region->set_req(_prim_0_path, control()); // Not equal after all.
3711 }
3712
3713 // these are the only paths that produce 'true':
3714 phi->set_req(_prim_same_path, intcon(1));
3715 phi->set_req(_ref_subtype_path, intcon(1));
3716
3717 // pull together the cases:
3718 assert(region->req() == PATH_LIMIT, "sane region");
3719 for (uint i = 1; i < region->req(); i++) {
3720 Node* ctl = region->in(i);
3721 if (ctl == NULL || ctl == top()) {
3722 region->set_req(i, top());
3723 phi ->set_req(i, top());
3724 } else if (phi->in(i) == NULL) {
3725 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3726 }
3727 }
3728
3729 set_control(_gvn.transform(region));
3730 set_result(_gvn.transform(phi));
3731 return true;
3732 }
3733
3734 //---------------------generate_array_guard_common------------------------
3735 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3736 bool obj_array, bool not_array) {
3737
3738 if (stopped()) {
3739 return NULL;
3740 }
3741
3742 // If obj_array/non_array==false/false:
3743 // Branch around if the given klass is in fact an array (either obj or prim).
3744 // If obj_array/non_array==false/true:
3745 // Branch around if the given klass is not an array klass of any kind.
3746 // If obj_array/non_array==true/true:
3747 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3748 // If obj_array/non_array==true/false:
3749 // Branch around if the kls is an oop array (Object[] or subtype)
3750 //
3751 // Like generate_guard, adds a new path onto the region.
3752 jint layout_con = 0;
3753 Node* layout_val = get_layout_helper(kls, layout_con);
3754 if (layout_val == NULL) {
3755 bool query = (obj_array
3756 ? Klass::layout_helper_is_objArray(layout_con)
3757 : Klass::layout_helper_is_array(layout_con));
3758 if (query == not_array) {
3759 return NULL; // never a branch
3760 } else { // always a branch
3761 Node* always_branch = control();
3762 if (region != NULL)
3763 region->add_req(always_branch);
3764 set_control(top());
3765 return always_branch;
3766 }
3767 }
3768 // Now test the correct condition.
3769 jint nval = (obj_array
3770 ? ((jint)Klass::_lh_array_tag_type_value
3771 << Klass::_lh_array_tag_shift)
3772 : Klass::_lh_neutral_value);
3773 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3774 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3775 // invert the test if we are looking for a non-array
3776 if (not_array) btest = BoolTest(btest).negate();
3777 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3778 return generate_fair_guard(bol, region);
3779 }
3780
3781
3782 //-----------------------inline_native_newArray--------------------------
3783 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3784 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3785 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3786 Node* mirror;
3787 Node* count_val;
3788 if (uninitialized) {
3789 mirror = argument(1);
3790 count_val = argument(2);
3791 } else {
3792 mirror = argument(0);
3793 count_val = argument(1);
3794 }
3795
3796 mirror = null_check(mirror);
3797 // If mirror or obj is dead, only null-path is taken.
3798 if (stopped()) return true;
3799
3800 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3801 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3802 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3803 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3804 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3805
3806 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3807 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3808 result_reg, _slow_path);
3809 Node* normal_ctl = control();
3810 Node* no_array_ctl = result_reg->in(_slow_path);
3811
3812 // Generate code for the slow case. We make a call to newArray().
3813 set_control(no_array_ctl);
3814 if (!stopped()) {
3815 // Either the input type is void.class, or else the
3816 // array klass has not yet been cached. Either the
3817 // ensuing call will throw an exception, or else it
3818 // will cache the array klass for next time.
3819 PreserveJVMState pjvms(this);
3820 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3821 Node* slow_result = set_results_for_java_call(slow_call);
3822 // this->control() comes from set_results_for_java_call
3823 result_reg->set_req(_slow_path, control());
3824 result_val->set_req(_slow_path, slow_result);
3825 result_io ->set_req(_slow_path, i_o());
3826 result_mem->set_req(_slow_path, reset_memory());
3827 }
3828
3829 set_control(normal_ctl);
3830 if (!stopped()) {
3831 // Normal case: The array type has been cached in the java.lang.Class.
3832 // The following call works fine even if the array type is polymorphic.
3833 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3834 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3835 result_reg->init_req(_normal_path, control());
3836 result_val->init_req(_normal_path, obj);
3837 result_io ->init_req(_normal_path, i_o());
3838 result_mem->init_req(_normal_path, reset_memory());
3839
3840 if (uninitialized) {
3841 // Mark the allocation so that zeroing is skipped
3842 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3843 alloc->maybe_set_complete(&_gvn);
3844 }
3845 }
3846
3847 // Return the combined state.
3848 set_i_o( _gvn.transform(result_io) );
3849 set_all_memory( _gvn.transform(result_mem));
3850
3851 C->set_has_split_ifs(true); // Has chance for split-if optimization
3852 set_result(result_reg, result_val);
3853 return true;
3854 }
3855
3856 //----------------------inline_native_getLength--------------------------
3857 // public static native int java.lang.reflect.Array.getLength(Object array);
3858 bool LibraryCallKit::inline_native_getLength() {
3859 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3860
3861 Node* array = null_check(argument(0));
3862 // If array is dead, only null-path is taken.
3863 if (stopped()) return true;
3864
3865 // Deoptimize if it is a non-array.
3866 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3867
3868 if (non_array != NULL) {
3869 PreserveJVMState pjvms(this);
3870 set_control(non_array);
3871 uncommon_trap(Deoptimization::Reason_intrinsic,
3872 Deoptimization::Action_maybe_recompile);
3873 }
3874
3875 // If control is dead, only non-array-path is taken.
3876 if (stopped()) return true;
3877
3878 // The works fine even if the array type is polymorphic.
3879 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3880 Node* result = load_array_length(array);
3881
3882 C->set_has_split_ifs(true); // Has chance for split-if optimization
3883 set_result(result);
3884 return true;
3885 }
3886
3887 //------------------------inline_array_copyOf----------------------------
3888 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3889 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3890 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3891 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3892
3893 // Get the arguments.
3894 Node* original = argument(0);
3895 Node* start = is_copyOfRange? argument(1): intcon(0);
3896 Node* end = is_copyOfRange? argument(2): argument(1);
3897 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3898
3899 Node* newcopy = NULL;
3900
3901 // Set the original stack and the reexecute bit for the interpreter to reexecute
3902 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3903 { PreserveReexecuteState preexecs(this);
3904 jvms()->set_should_reexecute(true);
3905
3906 array_type_mirror = null_check(array_type_mirror);
3907 original = null_check(original);
3908
3909 // Check if a null path was taken unconditionally.
3910 if (stopped()) return true;
3911
3912 Node* orig_length = load_array_length(original);
3913
3914 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3915 klass_node = null_check(klass_node);
3916
3917 RegionNode* bailout = new RegionNode(1);
3918 record_for_igvn(bailout);
3919
3920 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3921 // Bail out if that is so.
3922 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3923 if (not_objArray != NULL) {
3924 // Improve the klass node's type from the new optimistic assumption:
3925 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3926 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3927 Node* cast = new CastPPNode(klass_node, akls);
3928 cast->init_req(0, control());
3929 klass_node = _gvn.transform(cast);
3930 }
3931
3932 // Bail out if either start or end is negative.
3933 generate_negative_guard(start, bailout, &start);
3934 generate_negative_guard(end, bailout, &end);
3935
3936 Node* length = end;
3937 if (_gvn.type(start) != TypeInt::ZERO) {
3938 length = _gvn.transform(new SubINode(end, start));
3939 }
3940
3941 // Bail out if length is negative.
3942 // Without this the new_array would throw
3943 // NegativeArraySizeException but IllegalArgumentException is what
3944 // should be thrown
3945 generate_negative_guard(length, bailout, &length);
3946
3947 if (bailout->req() > 1) {
3948 PreserveJVMState pjvms(this);
3949 set_control(_gvn.transform(bailout));
3950 uncommon_trap(Deoptimization::Reason_intrinsic,
3951 Deoptimization::Action_maybe_recompile);
3952 }
3953
3954 if (!stopped()) {
3955 // How many elements will we copy from the original?
3956 // The answer is MinI(orig_length - start, length).
3957 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3958 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3959
3960 // Generate a direct call to the right arraycopy function(s).
3961 // We know the copy is disjoint but we might not know if the
3962 // oop stores need checking.
3963 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3964 // This will fail a store-check if x contains any non-nulls.
3965
3966 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3967 // loads/stores but it is legal only if we're sure the
3968 // Arrays.copyOf would succeed. So we need all input arguments
3969 // to the copyOf to be validated, including that the copy to the
3970 // new array won't trigger an ArrayStoreException. That subtype
3971 // check can be optimized if we know something on the type of
3972 // the input array from type speculation.
3973 if (_gvn.type(klass_node)->singleton()) {
3974 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3975 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3976
3977 int test = C->static_subtype_check(superk, subk);
3978 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3979 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3980 if (t_original->speculative_type() != NULL) {
3981 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3982 }
3983 }
3984 }
3985
3986 bool validated = false;
3987 // Reason_class_check rather than Reason_intrinsic because we
3988 // want to intrinsify even if this traps.
3989 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3990 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3991 klass_node);
3992
3993 if (not_subtype_ctrl != top()) {
3994 PreserveJVMState pjvms(this);
3995 set_control(not_subtype_ctrl);
3996 uncommon_trap(Deoptimization::Reason_class_check,
3997 Deoptimization::Action_make_not_entrant);
3998 assert(stopped(), "Should be stopped");
3999 }
4000 validated = true;
4001 }
4002
4003 if (!stopped()) {
4004 newcopy = new_array(klass_node, length, 0); // no arguments to push
4005
4006 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
4007 load_object_klass(original), klass_node);
4008 if (!is_copyOfRange) {
4009 ac->set_copyof(validated);
4010 } else {
4011 ac->set_copyofrange(validated);
4012 }
4013 Node* n = _gvn.transform(ac);
4014 if (n == ac) {
4015 ac->connect_outputs(this);
4016 } else {
4017 assert(validated, "shouldn't transform if all arguments not validated");
4018 set_all_memory(n);
4019 }
4020 }
4021 }
4022 } // original reexecute is set back here
4023
4024 C->set_has_split_ifs(true); // Has chance for split-if optimization
4025 if (!stopped()) {
4026 set_result(newcopy);
4027 }
4028 return true;
4029 }
4030
4031
4032 //----------------------generate_virtual_guard---------------------------
4033 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4034 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4035 RegionNode* slow_region) {
4036 ciMethod* method = callee();
4037 int vtable_index = method->vtable_index();
4038 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4039 "bad index %d", vtable_index);
4040 // Get the Method* out of the appropriate vtable entry.
4041 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4042 vtable_index*vtableEntry::size_in_bytes() +
4043 vtableEntry::method_offset_in_bytes();
4044 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4045 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4046
4047 // Compare the target method with the expected method (e.g., Object.hashCode).
4048 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4049
4050 Node* native_call = makecon(native_call_addr);
4051 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4052 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4053
4054 return generate_slow_guard(test_native, slow_region);
4055 }
4056
4057 //-----------------------generate_method_call----------------------------
4058 // Use generate_method_call to make a slow-call to the real
4059 // method if the fast path fails. An alternative would be to
4060 // use a stub like OptoRuntime::slow_arraycopy_Java.
4061 // This only works for expanding the current library call,
4062 // not another intrinsic. (E.g., don't use this for making an
4063 // arraycopy call inside of the copyOf intrinsic.)
4064 CallJavaNode*
4065 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4066 // When compiling the intrinsic method itself, do not use this technique.
4067 guarantee(callee() != C->method(), "cannot make slow-call to self");
4068
4069 ciMethod* method = callee();
4070 // ensure the JVMS we have will be correct for this call
4071 guarantee(method_id == method->intrinsic_id(), "must match");
4072
4073 const TypeFunc* tf = TypeFunc::make(method);
4074 CallJavaNode* slow_call;
4075 if (is_static) {
4076 assert(!is_virtual, "");
4077 slow_call = new CallStaticJavaNode(C, tf,
4078 SharedRuntime::get_resolve_static_call_stub(),
4079 method, bci());
4080 } else if (is_virtual) {
4081 null_check_receiver();
4082 int vtable_index = Method::invalid_vtable_index;
4083 if (UseInlineCaches) {
4084 // Suppress the vtable call
4085 } else {
4086 // hashCode and clone are not a miranda methods,
4087 // so the vtable index is fixed.
4088 // No need to use the linkResolver to get it.
4089 vtable_index = method->vtable_index();
4090 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4091 "bad index %d", vtable_index);
4092 }
4093 slow_call = new CallDynamicJavaNode(tf,
4094 SharedRuntime::get_resolve_virtual_call_stub(),
4095 method, vtable_index, bci());
4096 } else { // neither virtual nor static: opt_virtual
4097 null_check_receiver();
4098 slow_call = new CallStaticJavaNode(C, tf,
4099 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4100 method, bci());
4101 slow_call->set_optimized_virtual(true);
4102 }
4103 set_arguments_for_java_call(slow_call);
4104 set_edges_for_java_call(slow_call);
4105 return slow_call;
4106 }
4107
4108
4109 /**
4110 * Build special case code for calls to hashCode on an object. This call may
4111 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4112 * slightly different code.
4113 */
4114 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4115 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4116 assert(!(is_virtual && is_static), "either virtual, special, or static");
4117
4118 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4119
4120 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4121 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4122 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4123 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4124 Node* obj = NULL;
4125 if (!is_static) {
4126 // Check for hashing null object
4127 obj = null_check_receiver();
4128 if (stopped()) return true; // unconditionally null
4129 result_reg->init_req(_null_path, top());
4130 result_val->init_req(_null_path, top());
4131 } else {
4132 // Do a null check, and return zero if null.
4133 // System.identityHashCode(null) == 0
4134 obj = argument(0);
4135 Node* null_ctl = top();
4136 obj = null_check_oop(obj, &null_ctl);
4137 result_reg->init_req(_null_path, null_ctl);
4138 result_val->init_req(_null_path, _gvn.intcon(0));
4139 }
4140
4141 // Unconditionally null? Then return right away.
4142 if (stopped()) {
4143 set_control( result_reg->in(_null_path));
4144 if (!stopped())
4145 set_result(result_val->in(_null_path));
4146 return true;
4147 }
4148
4149 // We only go to the fast case code if we pass a number of guards. The
4150 // paths which do not pass are accumulated in the slow_region.
4151 RegionNode* slow_region = new RegionNode(1);
4152 record_for_igvn(slow_region);
4153
4154 // If this is a virtual call, we generate a funny guard. We pull out
4155 // the vtable entry corresponding to hashCode() from the target object.
4156 // If the target method which we are calling happens to be the native
4157 // Object hashCode() method, we pass the guard. We do not need this
4158 // guard for non-virtual calls -- the caller is known to be the native
4159 // Object hashCode().
4160 if (is_virtual) {
4161 // After null check, get the object's klass.
4162 Node* obj_klass = load_object_klass(obj);
4163 generate_virtual_guard(obj_klass, slow_region);
4164 }
4165
4166 // Get the header out of the object, use LoadMarkNode when available
4167 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4168 // The control of the load must be NULL. Otherwise, the load can move before
4169 // the null check after castPP removal.
4170 Node* no_ctrl = NULL;
4171 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4172
4173 // Test the header to see if it is unlocked.
4174 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4175 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4176 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4177 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4178 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4179
4180 generate_slow_guard(test_unlocked, slow_region);
4181
4182 // Get the hash value and check to see that it has been properly assigned.
4183 // We depend on hash_mask being at most 32 bits and avoid the use of
4184 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4185 // vm: see markOop.hpp.
4186 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4187 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4188 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4189 // This hack lets the hash bits live anywhere in the mark object now, as long
4190 // as the shift drops the relevant bits into the low 32 bits. Note that
4191 // Java spec says that HashCode is an int so there's no point in capturing
4192 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4193 hshifted_header = ConvX2I(hshifted_header);
4194 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4195
4196 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4197 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4198 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4199
4200 generate_slow_guard(test_assigned, slow_region);
4201
4202 Node* init_mem = reset_memory();
4203 // fill in the rest of the null path:
4204 result_io ->init_req(_null_path, i_o());
4205 result_mem->init_req(_null_path, init_mem);
4206
4207 result_val->init_req(_fast_path, hash_val);
4208 result_reg->init_req(_fast_path, control());
4209 result_io ->init_req(_fast_path, i_o());
4210 result_mem->init_req(_fast_path, init_mem);
4211
4212 // Generate code for the slow case. We make a call to hashCode().
4213 set_control(_gvn.transform(slow_region));
4214 if (!stopped()) {
4215 // No need for PreserveJVMState, because we're using up the present state.
4216 set_all_memory(init_mem);
4217 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4218 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4219 Node* slow_result = set_results_for_java_call(slow_call);
4220 // this->control() comes from set_results_for_java_call
4221 result_reg->init_req(_slow_path, control());
4222 result_val->init_req(_slow_path, slow_result);
4223 result_io ->set_req(_slow_path, i_o());
4224 result_mem ->set_req(_slow_path, reset_memory());
4225 }
4226
4227 // Return the combined state.
4228 set_i_o( _gvn.transform(result_io) );
4229 set_all_memory( _gvn.transform(result_mem));
4230
4231 set_result(result_reg, result_val);
4232 return true;
4233 }
4234
4235 //---------------------------inline_native_getClass----------------------------
4236 // public final native Class<?> java.lang.Object.getClass();
4237 //
4238 // Build special case code for calls to getClass on an object.
4239 bool LibraryCallKit::inline_native_getClass() {
4240 Node* obj = null_check_receiver();
4241 if (stopped()) return true;
4242 set_result(load_mirror_from_klass(load_object_klass(obj)));
4243 return true;
4244 }
4245
4246 //-----------------inline_native_Reflection_getCallerClass---------------------
4247 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4248 //
4249 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4250 //
4251 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4252 // in that it must skip particular security frames and checks for
4253 // caller sensitive methods.
4254 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4255 #ifndef PRODUCT
4256 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4257 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4258 }
4259 #endif
4260
4261 if (!jvms()->has_method()) {
4262 #ifndef PRODUCT
4263 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4264 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4265 }
4266 #endif
4267 return false;
4268 }
4269
4270 // Walk back up the JVM state to find the caller at the required
4271 // depth.
4272 JVMState* caller_jvms = jvms();
4273
4274 // Cf. JVM_GetCallerClass
4275 // NOTE: Start the loop at depth 1 because the current JVM state does
4276 // not include the Reflection.getCallerClass() frame.
4277 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4278 ciMethod* m = caller_jvms->method();
4279 switch (n) {
4280 case 0:
4281 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4282 break;
4283 case 1:
4284 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4285 if (!m->caller_sensitive()) {
4286 #ifndef PRODUCT
4287 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4288 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4289 }
4290 #endif
4291 return false; // bail-out; let JVM_GetCallerClass do the work
4292 }
4293 break;
4294 default:
4295 if (!m->is_ignored_by_security_stack_walk()) {
4296 // We have reached the desired frame; return the holder class.
4297 // Acquire method holder as java.lang.Class and push as constant.
4298 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4299 ciInstance* caller_mirror = caller_klass->java_mirror();
4300 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4301
4302 #ifndef PRODUCT
4303 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4304 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());
4305 tty->print_cr(" JVM state at this point:");
4306 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4307 ciMethod* m = jvms()->of_depth(i)->method();
4308 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4309 }
4310 }
4311 #endif
4312 return true;
4313 }
4314 break;
4315 }
4316 }
4317
4318 #ifndef PRODUCT
4319 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4320 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4321 tty->print_cr(" JVM state at this point:");
4322 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4323 ciMethod* m = jvms()->of_depth(i)->method();
4324 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4325 }
4326 }
4327 #endif
4328
4329 return false; // bail-out; let JVM_GetCallerClass do the work
4330 }
4331
4332 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4333 Node* arg = argument(0);
4334 Node* result = NULL;
4335
4336 switch (id) {
4337 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4338 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4339 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4340 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4341
4342 case vmIntrinsics::_doubleToLongBits: {
4343 // two paths (plus control) merge in a wood
4344 RegionNode *r = new RegionNode(3);
4345 Node *phi = new PhiNode(r, TypeLong::LONG);
4346
4347 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4348 // Build the boolean node
4349 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4350
4351 // Branch either way.
4352 // NaN case is less traveled, which makes all the difference.
4353 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4354 Node *opt_isnan = _gvn.transform(ifisnan);
4355 assert( opt_isnan->is_If(), "Expect an IfNode");
4356 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4357 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4358
4359 set_control(iftrue);
4360
4361 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4362 Node *slow_result = longcon(nan_bits); // return NaN
4363 phi->init_req(1, _gvn.transform( slow_result ));
4364 r->init_req(1, iftrue);
4365
4366 // Else fall through
4367 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4368 set_control(iffalse);
4369
4370 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4371 r->init_req(2, iffalse);
4372
4373 // Post merge
4374 set_control(_gvn.transform(r));
4375 record_for_igvn(r);
4376
4377 C->set_has_split_ifs(true); // Has chance for split-if optimization
4378 result = phi;
4379 assert(result->bottom_type()->isa_long(), "must be");
4380 break;
4381 }
4382
4383 case vmIntrinsics::_floatToIntBits: {
4384 // two paths (plus control) merge in a wood
4385 RegionNode *r = new RegionNode(3);
4386 Node *phi = new PhiNode(r, TypeInt::INT);
4387
4388 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4389 // Build the boolean node
4390 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4391
4392 // Branch either way.
4393 // NaN case is less traveled, which makes all the difference.
4394 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4395 Node *opt_isnan = _gvn.transform(ifisnan);
4396 assert( opt_isnan->is_If(), "Expect an IfNode");
4397 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4398 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4399
4400 set_control(iftrue);
4401
4402 static const jint nan_bits = 0x7fc00000;
4403 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4404 phi->init_req(1, _gvn.transform( slow_result ));
4405 r->init_req(1, iftrue);
4406
4407 // Else fall through
4408 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4409 set_control(iffalse);
4410
4411 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4412 r->init_req(2, iffalse);
4413
4414 // Post merge
4415 set_control(_gvn.transform(r));
4416 record_for_igvn(r);
4417
4418 C->set_has_split_ifs(true); // Has chance for split-if optimization
4419 result = phi;
4420 assert(result->bottom_type()->isa_int(), "must be");
4421 break;
4422 }
4423
4424 default:
4425 fatal_unexpected_iid(id);
4426 break;
4427 }
4428 set_result(_gvn.transform(result));
4429 return true;
4430 }
4431
4432 //----------------------inline_unsafe_copyMemory-------------------------
4433 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4434 bool LibraryCallKit::inline_unsafe_copyMemory() {
4435 if (callee()->is_static()) return false; // caller must have the capability!
4436 null_check_receiver(); // null-check receiver
4437 if (stopped()) return true;
4438
4439 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4440
4441 Node* src_ptr = argument(1); // type: oop
4442 Node* src_off = ConvL2X(argument(2)); // type: long
4443 Node* dst_ptr = argument(4); // type: oop
4444 Node* dst_off = ConvL2X(argument(5)); // type: long
4445 Node* size = ConvL2X(argument(7)); // type: long
4446
4447 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4448 "fieldOffset must be byte-scaled");
4449
4450 Node* src = make_unsafe_address(src_ptr, src_off);
4451 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4452
4453 // Conservatively insert a memory barrier on all memory slices.
4454 // Do not let writes of the copy source or destination float below the copy.
4455 insert_mem_bar(Op_MemBarCPUOrder);
4456
4457 // Call it. Note that the length argument is not scaled.
4458 make_runtime_call(RC_LEAF|RC_NO_FP,
4459 OptoRuntime::fast_arraycopy_Type(),
4460 StubRoutines::unsafe_arraycopy(),
4461 "unsafe_arraycopy",
4462 TypeRawPtr::BOTTOM,
4463 src, dst, size XTOP);
4464
4465 // Do not let reads of the copy destination float above the copy.
4466 insert_mem_bar(Op_MemBarCPUOrder);
4467
4468 return true;
4469 }
4470
4471 //------------------------clone_coping-----------------------------------
4472 // Helper function for inline_native_clone.
4473 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4474 assert(obj_size != NULL, "");
4475 Node* raw_obj = alloc_obj->in(1);
4476 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4477
4478 AllocateNode* alloc = NULL;
4479 if (ReduceBulkZeroing) {
4480 // We will be completely responsible for initializing this object -
4481 // mark Initialize node as complete.
4482 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4483 // The object was just allocated - there should be no any stores!
4484 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4485 // Mark as complete_with_arraycopy so that on AllocateNode
4486 // expansion, we know this AllocateNode is initialized by an array
4487 // copy and a StoreStore barrier exists after the array copy.
4488 alloc->initialization()->set_complete_with_arraycopy();
4489 }
4490
4491 // Copy the fastest available way.
4492 // TODO: generate fields copies for small objects instead.
4493 Node* src = obj;
4494 Node* dest = alloc_obj;
4495 Node* size = _gvn.transform(obj_size);
4496
4497 // Exclude the header but include array length to copy by 8 bytes words.
4498 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4499 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4500 instanceOopDesc::base_offset_in_bytes();
4501 // base_off:
4502 // 8 - 32-bit VM
4503 // 12 - 64-bit VM, compressed klass
4504 // 16 - 64-bit VM, normal klass
4505 if (base_off % BytesPerLong != 0) {
4506 assert(UseCompressedClassPointers, "");
4507 if (is_array) {
4508 // Exclude length to copy by 8 bytes words.
4509 base_off += sizeof(int);
4510 } else {
4511 // Include klass to copy by 8 bytes words.
4512 base_off = instanceOopDesc::klass_offset_in_bytes();
4513 }
4514 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4515 }
4516 src = basic_plus_adr(src, base_off);
4517 dest = basic_plus_adr(dest, base_off);
4518
4519 // Compute the length also, if needed:
4520 Node* countx = size;
4521 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4522 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4523
4524 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4525
4526 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4527 ac->set_clonebasic();
4528 Node* n = _gvn.transform(ac);
4529 if (n == ac) {
4530 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4531 } else {
4532 set_all_memory(n);
4533 }
4534
4535 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4536 if (card_mark) {
4537 assert(!is_array, "");
4538 // Put in store barrier for any and all oops we are sticking
4539 // into this object. (We could avoid this if we could prove
4540 // that the object type contains no oop fields at all.)
4541 Node* no_particular_value = NULL;
4542 Node* no_particular_field = NULL;
4543 int raw_adr_idx = Compile::AliasIdxRaw;
4544 post_barrier(control(),
4545 memory(raw_adr_type),
4546 alloc_obj,
4547 no_particular_field,
4548 raw_adr_idx,
4549 no_particular_value,
4550 T_OBJECT,
4551 false);
4552 }
4553
4554 // Do not let reads from the cloned object float above the arraycopy.
4555 if (alloc != NULL) {
4556 // Do not let stores that initialize this object be reordered with
4557 // a subsequent store that would make this object accessible by
4558 // other threads.
4559 // Record what AllocateNode this StoreStore protects so that
4560 // escape analysis can go from the MemBarStoreStoreNode to the
4561 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4562 // based on the escape status of the AllocateNode.
4563 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4564 } else {
4565 insert_mem_bar(Op_MemBarCPUOrder);
4566 }
4567 }
4568
4569 //------------------------inline_native_clone----------------------------
4570 // protected native Object java.lang.Object.clone();
4571 //
4572 // Here are the simple edge cases:
4573 // null receiver => normal trap
4574 // virtual and clone was overridden => slow path to out-of-line clone
4575 // not cloneable or finalizer => slow path to out-of-line Object.clone
4576 //
4577 // The general case has two steps, allocation and copying.
4578 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4579 //
4580 // Copying also has two cases, oop arrays and everything else.
4581 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4582 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4583 //
4584 // These steps fold up nicely if and when the cloned object's klass
4585 // can be sharply typed as an object array, a type array, or an instance.
4586 //
4587 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4588 PhiNode* result_val;
4589
4590 // Set the reexecute bit for the interpreter to reexecute
4591 // the bytecode that invokes Object.clone if deoptimization happens.
4592 { PreserveReexecuteState preexecs(this);
4593 jvms()->set_should_reexecute(true);
4594
4595 Node* obj = null_check_receiver();
4596 if (stopped()) return true;
4597
4598 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4599
4600 // If we are going to clone an instance, we need its exact type to
4601 // know the number and types of fields to convert the clone to
4602 // loads/stores. Maybe a speculative type can help us.
4603 if (!obj_type->klass_is_exact() &&
4604 obj_type->speculative_type() != NULL &&
4605 obj_type->speculative_type()->is_instance_klass()) {
4606 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4607 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4608 !spec_ik->has_injected_fields()) {
4609 ciKlass* k = obj_type->klass();
4610 if (!k->is_instance_klass() ||
4611 k->as_instance_klass()->is_interface() ||
4612 k->as_instance_klass()->has_subklass()) {
4613 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4614 }
4615 }
4616 }
4617
4618 Node* obj_klass = load_object_klass(obj);
4619 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4620 const TypeOopPtr* toop = ((tklass != NULL)
4621 ? tklass->as_instance_type()
4622 : TypeInstPtr::NOTNULL);
4623
4624 // Conservatively insert a memory barrier on all memory slices.
4625 // Do not let writes into the original float below the clone.
4626 insert_mem_bar(Op_MemBarCPUOrder);
4627
4628 // paths into result_reg:
4629 enum {
4630 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4631 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4632 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4633 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4634 PATH_LIMIT
4635 };
4636 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4637 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4638 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4639 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4640 record_for_igvn(result_reg);
4641
4642 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4643 int raw_adr_idx = Compile::AliasIdxRaw;
4644
4645 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4646 if (array_ctl != NULL) {
4647 // It's an array.
4648 PreserveJVMState pjvms(this);
4649 set_control(array_ctl);
4650 Node* obj_length = load_array_length(obj);
4651 Node* obj_size = NULL;
4652 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4653
4654 if (!use_ReduceInitialCardMarks()) {
4655 // If it is an oop array, it requires very special treatment,
4656 // because card marking is required on each card of the array.
4657 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4658 if (is_obja != NULL) {
4659 PreserveJVMState pjvms2(this);
4660 set_control(is_obja);
4661 // Generate a direct call to the right arraycopy function(s).
4662 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4663 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4664 ac->set_cloneoop();
4665 Node* n = _gvn.transform(ac);
4666 assert(n == ac, "cannot disappear");
4667 ac->connect_outputs(this);
4668
4669 result_reg->init_req(_objArray_path, control());
4670 result_val->init_req(_objArray_path, alloc_obj);
4671 result_i_o ->set_req(_objArray_path, i_o());
4672 result_mem ->set_req(_objArray_path, reset_memory());
4673 }
4674 }
4675 // Otherwise, there are no card marks to worry about.
4676 // (We can dispense with card marks if we know the allocation
4677 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4678 // causes the non-eden paths to take compensating steps to
4679 // simulate a fresh allocation, so that no further
4680 // card marks are required in compiled code to initialize
4681 // the object.)
4682
4683 if (!stopped()) {
4684 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4685
4686 // Present the results of the copy.
4687 result_reg->init_req(_array_path, control());
4688 result_val->init_req(_array_path, alloc_obj);
4689 result_i_o ->set_req(_array_path, i_o());
4690 result_mem ->set_req(_array_path, reset_memory());
4691 }
4692 }
4693
4694 // We only go to the instance fast case code if we pass a number of guards.
4695 // The paths which do not pass are accumulated in the slow_region.
4696 RegionNode* slow_region = new RegionNode(1);
4697 record_for_igvn(slow_region);
4698 if (!stopped()) {
4699 // It's an instance (we did array above). Make the slow-path tests.
4700 // If this is a virtual call, we generate a funny guard. We grab
4701 // the vtable entry corresponding to clone() from the target object.
4702 // If the target method which we are calling happens to be the
4703 // Object clone() method, we pass the guard. We do not need this
4704 // guard for non-virtual calls; the caller is known to be the native
4705 // Object clone().
4706 if (is_virtual) {
4707 generate_virtual_guard(obj_klass, slow_region);
4708 }
4709
4710 // The object must be easily cloneable and must not have a finalizer.
4711 // Both of these conditions may be checked in a single test.
4712 // We could optimize the test further, but we don't care.
4713 generate_access_flags_guard(obj_klass,
4714 // Test both conditions:
4715 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4716 // Must be cloneable but not finalizer:
4717 JVM_ACC_IS_CLONEABLE_FAST,
4718 slow_region);
4719 }
4720
4721 if (!stopped()) {
4722 // It's an instance, and it passed the slow-path tests.
4723 PreserveJVMState pjvms(this);
4724 Node* obj_size = NULL;
4725 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4726 // is reexecuted if deoptimization occurs and there could be problems when merging
4727 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4728 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4729
4730 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4731
4732 // Present the results of the slow call.
4733 result_reg->init_req(_instance_path, control());
4734 result_val->init_req(_instance_path, alloc_obj);
4735 result_i_o ->set_req(_instance_path, i_o());
4736 result_mem ->set_req(_instance_path, reset_memory());
4737 }
4738
4739 // Generate code for the slow case. We make a call to clone().
4740 set_control(_gvn.transform(slow_region));
4741 if (!stopped()) {
4742 PreserveJVMState pjvms(this);
4743 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4744 Node* slow_result = set_results_for_java_call(slow_call);
4745 // this->control() comes from set_results_for_java_call
4746 result_reg->init_req(_slow_path, control());
4747 result_val->init_req(_slow_path, slow_result);
4748 result_i_o ->set_req(_slow_path, i_o());
4749 result_mem ->set_req(_slow_path, reset_memory());
4750 }
4751
4752 // Return the combined state.
4753 set_control( _gvn.transform(result_reg));
4754 set_i_o( _gvn.transform(result_i_o));
4755 set_all_memory( _gvn.transform(result_mem));
4756 } // original reexecute is set back here
4757
4758 set_result(_gvn.transform(result_val));
4759 return true;
4760 }
4761
4762 // If we have a tighly coupled allocation, the arraycopy may take care
4763 // of the array initialization. If one of the guards we insert between
4764 // the allocation and the arraycopy causes a deoptimization, an
4765 // unitialized array will escape the compiled method. To prevent that
4766 // we set the JVM state for uncommon traps between the allocation and
4767 // the arraycopy to the state before the allocation so, in case of
4768 // deoptimization, we'll reexecute the allocation and the
4769 // initialization.
4770 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4771 if (alloc != NULL) {
4772 ciMethod* trap_method = alloc->jvms()->method();
4773 int trap_bci = alloc->jvms()->bci();
4774
4775 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4776 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4777 // Make sure there's no store between the allocation and the
4778 // arraycopy otherwise visible side effects could be rexecuted
4779 // in case of deoptimization and cause incorrect execution.
4780 bool no_interfering_store = true;
4781 Node* mem = alloc->in(TypeFunc::Memory);
4782 if (mem->is_MergeMem()) {
4783 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4784 Node* n = mms.memory();
4785 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4786 assert(n->is_Store(), "what else?");
4787 no_interfering_store = false;
4788 break;
4789 }
4790 }
4791 } else {
4792 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4793 Node* n = mms.memory();
4794 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4795 assert(n->is_Store(), "what else?");
4796 no_interfering_store = false;
4797 break;
4798 }
4799 }
4800 }
4801
4802 if (no_interfering_store) {
4803 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4804 uint size = alloc->req();
4805 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4806 old_jvms->set_map(sfpt);
4807 for (uint i = 0; i < size; i++) {
4808 sfpt->init_req(i, alloc->in(i));
4809 }
4810 // re-push array length for deoptimization
4811 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4812 old_jvms->set_sp(old_jvms->sp()+1);
4813 old_jvms->set_monoff(old_jvms->monoff()+1);
4814 old_jvms->set_scloff(old_jvms->scloff()+1);
4815 old_jvms->set_endoff(old_jvms->endoff()+1);
4816 old_jvms->set_should_reexecute(true);
4817
4818 sfpt->set_i_o(map()->i_o());
4819 sfpt->set_memory(map()->memory());
4820 sfpt->set_control(map()->control());
4821
4822 JVMState* saved_jvms = jvms();
4823 saved_reexecute_sp = _reexecute_sp;
4824
4825 set_jvms(sfpt->jvms());
4826 _reexecute_sp = jvms()->sp();
4827
4828 return saved_jvms;
4829 }
4830 }
4831 }
4832 return NULL;
4833 }
4834
4835 // In case of a deoptimization, we restart execution at the
4836 // allocation, allocating a new array. We would leave an uninitialized
4837 // array in the heap that GCs wouldn't expect. Move the allocation
4838 // after the traps so we don't allocate the array if we
4839 // deoptimize. This is possible because tightly_coupled_allocation()
4840 // guarantees there's no observer of the allocated array at this point
4841 // and the control flow is simple enough.
4842 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4843 if (saved_jvms != NULL && !stopped()) {
4844 assert(alloc != NULL, "only with a tightly coupled allocation");
4845 // restore JVM state to the state at the arraycopy
4846 saved_jvms->map()->set_control(map()->control());
4847 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4848 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4849 // If we've improved the types of some nodes (null check) while
4850 // emitting the guards, propagate them to the current state
4851 map()->replaced_nodes().apply(saved_jvms->map());
4852 set_jvms(saved_jvms);
4853 _reexecute_sp = saved_reexecute_sp;
4854
4855 // Remove the allocation from above the guards
4856 CallProjections callprojs;
4857 alloc->extract_projections(&callprojs, true);
4858 InitializeNode* init = alloc->initialization();
4859 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4860 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4861 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4862 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4863
4864 // move the allocation here (after the guards)
4865 _gvn.hash_delete(alloc);
4866 alloc->set_req(TypeFunc::Control, control());
4867 alloc->set_req(TypeFunc::I_O, i_o());
4868 Node *mem = reset_memory();
4869 set_all_memory(mem);
4870 alloc->set_req(TypeFunc::Memory, mem);
4871 set_control(init->proj_out(TypeFunc::Control));
4872 set_i_o(callprojs.fallthrough_ioproj);
4873
4874 // Update memory as done in GraphKit::set_output_for_allocation()
4875 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4876 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4877 if (ary_type->isa_aryptr() && length_type != NULL) {
4878 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4879 }
4880 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4881 int elemidx = C->get_alias_index(telemref);
4882 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4883 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4884
4885 Node* allocx = _gvn.transform(alloc);
4886 assert(allocx == alloc, "where has the allocation gone?");
4887 assert(dest->is_CheckCastPP(), "not an allocation result?");
4888
4889 _gvn.hash_delete(dest);
4890 dest->set_req(0, control());
4891 Node* destx = _gvn.transform(dest);
4892 assert(destx == dest, "where has the allocation result gone?");
4893 }
4894 }
4895
4896
4897 //------------------------------inline_arraycopy-----------------------
4898 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4899 // Object dest, int destPos,
4900 // int length);
4901 bool LibraryCallKit::inline_arraycopy() {
4902 // Get the arguments.
4903 Node* src = argument(0); // type: oop
4904 Node* src_offset = argument(1); // type: int
4905 Node* dest = argument(2); // type: oop
4906 Node* dest_offset = argument(3); // type: int
4907 Node* length = argument(4); // type: int
4908
4909
4910 // Check for allocation before we add nodes that would confuse
4911 // tightly_coupled_allocation()
4912 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4913
4914 int saved_reexecute_sp = -1;
4915 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4916 // See arraycopy_restore_alloc_state() comment
4917 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4918 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4919 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4920 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4921
4922 // The following tests must be performed
4923 // (1) src and dest are arrays.
4924 // (2) src and dest arrays must have elements of the same BasicType
4925 // (3) src and dest must not be null.
4926 // (4) src_offset must not be negative.
4927 // (5) dest_offset must not be negative.
4928 // (6) length must not be negative.
4929 // (7) src_offset + length must not exceed length of src.
4930 // (8) dest_offset + length must not exceed length of dest.
4931 // (9) each element of an oop array must be assignable
4932
4933 // (3) src and dest must not be null.
4934 // always do this here because we need the JVM state for uncommon traps
4935 Node* null_ctl = top();
4936 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4937 assert(null_ctl->is_top(), "no null control here");
4938 dest = null_check(dest, T_ARRAY);
4939
4940 if (!can_emit_guards) {
4941 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4942 // guards but the arraycopy node could still take advantage of a
4943 // tightly allocated allocation. tightly_coupled_allocation() is
4944 // called again to make sure it takes the null check above into
4945 // account: the null check is mandatory and if it caused an
4946 // uncommon trap to be emitted then the allocation can't be
4947 // considered tightly coupled in this context.
4948 alloc = tightly_coupled_allocation(dest, NULL);
4949 }
4950
4951 bool validated = false;
4952
4953 const Type* src_type = _gvn.type(src);
4954 const Type* dest_type = _gvn.type(dest);
4955 const TypeAryPtr* top_src = src_type->isa_aryptr();
4956 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4957
4958 // Do we have the type of src?
4959 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4960 // Do we have the type of dest?
4961 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4962 // Is the type for src from speculation?
4963 bool src_spec = false;
4964 // Is the type for dest from speculation?
4965 bool dest_spec = false;
4966
4967 if ((!has_src || !has_dest) && can_emit_guards) {
4968 // We don't have sufficient type information, let's see if
4969 // speculative types can help. We need to have types for both src
4970 // and dest so that it pays off.
4971
4972 // Do we already have or could we have type information for src
4973 bool could_have_src = has_src;
4974 // Do we already have or could we have type information for dest
4975 bool could_have_dest = has_dest;
4976
4977 ciKlass* src_k = NULL;
4978 if (!has_src) {
4979 src_k = src_type->speculative_type_not_null();
4980 if (src_k != NULL && src_k->is_array_klass()) {
4981 could_have_src = true;
4982 }
4983 }
4984
4985 ciKlass* dest_k = NULL;
4986 if (!has_dest) {
4987 dest_k = dest_type->speculative_type_not_null();
4988 if (dest_k != NULL && dest_k->is_array_klass()) {
4989 could_have_dest = true;
4990 }
4991 }
4992
4993 if (could_have_src && could_have_dest) {
4994 // This is going to pay off so emit the required guards
4995 if (!has_src) {
4996 src = maybe_cast_profiled_obj(src, src_k, true);
4997 src_type = _gvn.type(src);
4998 top_src = src_type->isa_aryptr();
4999 has_src = (top_src != NULL && top_src->klass() != NULL);
5000 src_spec = true;
5001 }
5002 if (!has_dest) {
5003 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5004 dest_type = _gvn.type(dest);
5005 top_dest = dest_type->isa_aryptr();
5006 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5007 dest_spec = true;
5008 }
5009 }
5010 }
5011
5012 if (has_src && has_dest && can_emit_guards) {
5013 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5014 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5015 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5016 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5017
5018 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5019 // If both arrays are object arrays then having the exact types
5020 // for both will remove the need for a subtype check at runtime
5021 // before the call and may make it possible to pick a faster copy
5022 // routine (without a subtype check on every element)
5023 // Do we have the exact type of src?
5024 bool could_have_src = src_spec;
5025 // Do we have the exact type of dest?
5026 bool could_have_dest = dest_spec;
5027 ciKlass* src_k = top_src->klass();
5028 ciKlass* dest_k = top_dest->klass();
5029 if (!src_spec) {
5030 src_k = src_type->speculative_type_not_null();
5031 if (src_k != NULL && src_k->is_array_klass()) {
5032 could_have_src = true;
5033 }
5034 }
5035 if (!dest_spec) {
5036 dest_k = dest_type->speculative_type_not_null();
5037 if (dest_k != NULL && dest_k->is_array_klass()) {
5038 could_have_dest = true;
5039 }
5040 }
5041 if (could_have_src && could_have_dest) {
5042 // If we can have both exact types, emit the missing guards
5043 if (could_have_src && !src_spec) {
5044 src = maybe_cast_profiled_obj(src, src_k, true);
5045 }
5046 if (could_have_dest && !dest_spec) {
5047 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5048 }
5049 }
5050 }
5051 }
5052
5053 ciMethod* trap_method = method();
5054 int trap_bci = bci();
5055 if (saved_jvms != NULL) {
5056 trap_method = alloc->jvms()->method();
5057 trap_bci = alloc->jvms()->bci();
5058 }
5059
5060 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5061 can_emit_guards &&
5062 !src->is_top() && !dest->is_top()) {
5063 // validate arguments: enables transformation the ArrayCopyNode
5064 validated = true;
5065
5066 RegionNode* slow_region = new RegionNode(1);
5067 record_for_igvn(slow_region);
5068
5069 // (1) src and dest are arrays.
5070 generate_non_array_guard(load_object_klass(src), slow_region);
5071 generate_non_array_guard(load_object_klass(dest), slow_region);
5072
5073 // (2) src and dest arrays must have elements of the same BasicType
5074 // done at macro expansion or at Ideal transformation time
5075
5076 // (4) src_offset must not be negative.
5077 generate_negative_guard(src_offset, slow_region);
5078
5079 // (5) dest_offset must not be negative.
5080 generate_negative_guard(dest_offset, slow_region);
5081
5082 // (7) src_offset + length must not exceed length of src.
5083 generate_limit_guard(src_offset, length,
5084 load_array_length(src),
5085 slow_region);
5086
5087 // (8) dest_offset + length must not exceed length of dest.
5088 generate_limit_guard(dest_offset, length,
5089 load_array_length(dest),
5090 slow_region);
5091
5092 // (9) each element of an oop array must be assignable
5093 Node* src_klass = load_object_klass(src);
5094 Node* dest_klass = load_object_klass(dest);
5095 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5096
5097 if (not_subtype_ctrl != top()) {
5098 PreserveJVMState pjvms(this);
5099 set_control(not_subtype_ctrl);
5100 uncommon_trap(Deoptimization::Reason_intrinsic,
5101 Deoptimization::Action_make_not_entrant);
5102 assert(stopped(), "Should be stopped");
5103 }
5104 {
5105 PreserveJVMState pjvms(this);
5106 set_control(_gvn.transform(slow_region));
5107 uncommon_trap(Deoptimization::Reason_intrinsic,
5108 Deoptimization::Action_make_not_entrant);
5109 assert(stopped(), "Should be stopped");
5110 }
5111 }
5112
5113 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5114
5115 if (stopped()) {
5116 return true;
5117 }
5118
5119 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5120 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5121 // so the compiler has a chance to eliminate them: during macro expansion,
5122 // we have to set their control (CastPP nodes are eliminated).
5123 load_object_klass(src), load_object_klass(dest),
5124 load_array_length(src), load_array_length(dest));
5125
5126 ac->set_arraycopy(validated);
5127
5128 Node* n = _gvn.transform(ac);
5129 if (n == ac) {
5130 ac->connect_outputs(this);
5131 } else {
5132 assert(validated, "shouldn't transform if all arguments not validated");
5133 set_all_memory(n);
5134 }
5135
5136 return true;
5137 }
5138
5139
5140 // Helper function which determines if an arraycopy immediately follows
5141 // an allocation, with no intervening tests or other escapes for the object.
5142 AllocateArrayNode*
5143 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5144 RegionNode* slow_region) {
5145 if (stopped()) return NULL; // no fast path
5146 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5147
5148 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5149 if (alloc == NULL) return NULL;
5150
5151 Node* rawmem = memory(Compile::AliasIdxRaw);
5152 // Is the allocation's memory state untouched?
5153 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5154 // Bail out if there have been raw-memory effects since the allocation.
5155 // (Example: There might have been a call or safepoint.)
5156 return NULL;
5157 }
5158 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5159 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5160 return NULL;
5161 }
5162
5163 // There must be no unexpected observers of this allocation.
5164 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5165 Node* obs = ptr->fast_out(i);
5166 if (obs != this->map()) {
5167 return NULL;
5168 }
5169 }
5170
5171 // This arraycopy must unconditionally follow the allocation of the ptr.
5172 Node* alloc_ctl = ptr->in(0);
5173 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5174
5175 Node* ctl = control();
5176 while (ctl != alloc_ctl) {
5177 // There may be guards which feed into the slow_region.
5178 // Any other control flow means that we might not get a chance
5179 // to finish initializing the allocated object.
5180 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5181 IfNode* iff = ctl->in(0)->as_If();
5182 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5183 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5184 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5185 ctl = iff->in(0); // This test feeds the known slow_region.
5186 continue;
5187 }
5188 // One more try: Various low-level checks bottom out in
5189 // uncommon traps. If the debug-info of the trap omits
5190 // any reference to the allocation, as we've already
5191 // observed, then there can be no objection to the trap.
5192 bool found_trap = false;
5193 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5194 Node* obs = not_ctl->fast_out(j);
5195 if (obs->in(0) == not_ctl && obs->is_Call() &&
5196 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5197 found_trap = true; break;
5198 }
5199 }
5200 if (found_trap) {
5201 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5202 continue;
5203 }
5204 }
5205 return NULL;
5206 }
5207
5208 // If we get this far, we have an allocation which immediately
5209 // precedes the arraycopy, and we can take over zeroing the new object.
5210 // The arraycopy will finish the initialization, and provide
5211 // a new control state to which we will anchor the destination pointer.
5212
5213 return alloc;
5214 }
5215
5216 //-------------inline_encodeISOArray-----------------------------------
5217 // encode char[] to byte[] in ISO_8859_1
5218 bool LibraryCallKit::inline_encodeISOArray() {
5219 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5220 // no receiver since it is static method
5221 Node *src = argument(0);
5222 Node *src_offset = argument(1);
5223 Node *dst = argument(2);
5224 Node *dst_offset = argument(3);
5225 Node *length = argument(4);
5226
5227 const Type* src_type = src->Value(&_gvn);
5228 const Type* dst_type = dst->Value(&_gvn);
5229 const TypeAryPtr* top_src = src_type->isa_aryptr();
5230 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5231 if (top_src == NULL || top_src->klass() == NULL ||
5232 top_dest == NULL || top_dest->klass() == NULL) {
5233 // failed array check
5234 return false;
5235 }
5236
5237 // Figure out the size and type of the elements we will be copying.
5238 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5239 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5240 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5241 return false;
5242 }
5243
5244 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5245 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5246 // 'src_start' points to src array + scaled offset
5247 // 'dst_start' points to dst array + scaled offset
5248
5249 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5250 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5251 enc = _gvn.transform(enc);
5252 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5253 set_memory(res_mem, mtype);
5254 set_result(enc);
5255 return true;
5256 }
5257
5258 //-------------inline_multiplyToLen-----------------------------------
5259 bool LibraryCallKit::inline_multiplyToLen() {
5260 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5261
5262 address stubAddr = StubRoutines::multiplyToLen();
5263 if (stubAddr == NULL) {
5264 return false; // Intrinsic's stub is not implemented on this platform
5265 }
5266 const char* stubName = "multiplyToLen";
5267
5268 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5269
5270 // no receiver because it is a static method
5271 Node* x = argument(0);
5272 Node* xlen = argument(1);
5273 Node* y = argument(2);
5274 Node* ylen = argument(3);
5275 Node* z = argument(4);
5276
5277 const Type* x_type = x->Value(&_gvn);
5278 const Type* y_type = y->Value(&_gvn);
5279 const TypeAryPtr* top_x = x_type->isa_aryptr();
5280 const TypeAryPtr* top_y = y_type->isa_aryptr();
5281 if (top_x == NULL || top_x->klass() == NULL ||
5282 top_y == NULL || top_y->klass() == NULL) {
5283 // failed array check
5284 return false;
5285 }
5286
5287 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5288 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5289 if (x_elem != T_INT || y_elem != T_INT) {
5290 return false;
5291 }
5292
5293 // Set the original stack and the reexecute bit for the interpreter to reexecute
5294 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5295 // on the return from z array allocation in runtime.
5296 { PreserveReexecuteState preexecs(this);
5297 jvms()->set_should_reexecute(true);
5298
5299 Node* x_start = array_element_address(x, intcon(0), x_elem);
5300 Node* y_start = array_element_address(y, intcon(0), y_elem);
5301 // 'x_start' points to x array + scaled xlen
5302 // 'y_start' points to y array + scaled ylen
5303
5304 // Allocate the result array
5305 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5306 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5307 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5308
5309 IdealKit ideal(this);
5310
5311 #define __ ideal.
5312 Node* one = __ ConI(1);
5313 Node* zero = __ ConI(0);
5314 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5315 __ set(need_alloc, zero);
5316 __ set(z_alloc, z);
5317 __ if_then(z, BoolTest::eq, null()); {
5318 __ increment (need_alloc, one);
5319 } __ else_(); {
5320 // Update graphKit memory and control from IdealKit.
5321 sync_kit(ideal);
5322 Node* zlen_arg = load_array_length(z);
5323 // Update IdealKit memory and control from graphKit.
5324 __ sync_kit(this);
5325 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5326 __ increment (need_alloc, one);
5327 } __ end_if();
5328 } __ end_if();
5329
5330 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5331 // Update graphKit memory and control from IdealKit.
5332 sync_kit(ideal);
5333 Node * narr = new_array(klass_node, zlen, 1);
5334 // Update IdealKit memory and control from graphKit.
5335 __ sync_kit(this);
5336 __ set(z_alloc, narr);
5337 } __ end_if();
5338
5339 sync_kit(ideal);
5340 z = __ value(z_alloc);
5341 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5342 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5343 // Final sync IdealKit and GraphKit.
5344 final_sync(ideal);
5345 #undef __
5346
5347 Node* z_start = array_element_address(z, intcon(0), T_INT);
5348
5349 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5350 OptoRuntime::multiplyToLen_Type(),
5351 stubAddr, stubName, TypePtr::BOTTOM,
5352 x_start, xlen, y_start, ylen, z_start, zlen);
5353 } // original reexecute is set back here
5354
5355 C->set_has_split_ifs(true); // Has chance for split-if optimization
5356 set_result(z);
5357 return true;
5358 }
5359
5360 //-------------inline_squareToLen------------------------------------
5361 bool LibraryCallKit::inline_squareToLen() {
5362 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5363
5364 address stubAddr = StubRoutines::squareToLen();
5365 if (stubAddr == NULL) {
5366 return false; // Intrinsic's stub is not implemented on this platform
5367 }
5368 const char* stubName = "squareToLen";
5369
5370 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5371
5372 Node* x = argument(0);
5373 Node* len = argument(1);
5374 Node* z = argument(2);
5375 Node* zlen = argument(3);
5376
5377 const Type* x_type = x->Value(&_gvn);
5378 const Type* z_type = z->Value(&_gvn);
5379 const TypeAryPtr* top_x = x_type->isa_aryptr();
5380 const TypeAryPtr* top_z = z_type->isa_aryptr();
5381 if (top_x == NULL || top_x->klass() == NULL ||
5382 top_z == NULL || top_z->klass() == NULL) {
5383 // failed array check
5384 return false;
5385 }
5386
5387 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5388 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5389 if (x_elem != T_INT || z_elem != T_INT) {
5390 return false;
5391 }
5392
5393
5394 Node* x_start = array_element_address(x, intcon(0), x_elem);
5395 Node* z_start = array_element_address(z, intcon(0), z_elem);
5396
5397 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5398 OptoRuntime::squareToLen_Type(),
5399 stubAddr, stubName, TypePtr::BOTTOM,
5400 x_start, len, z_start, zlen);
5401
5402 set_result(z);
5403 return true;
5404 }
5405
5406 //-------------inline_mulAdd------------------------------------------
5407 bool LibraryCallKit::inline_mulAdd() {
5408 assert(UseMulAddIntrinsic, "not implemented on this platform");
5409
5410 address stubAddr = StubRoutines::mulAdd();
5411 if (stubAddr == NULL) {
5412 return false; // Intrinsic's stub is not implemented on this platform
5413 }
5414 const char* stubName = "mulAdd";
5415
5416 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5417
5418 Node* out = argument(0);
5419 Node* in = argument(1);
5420 Node* offset = argument(2);
5421 Node* len = argument(3);
5422 Node* k = argument(4);
5423
5424 const Type* out_type = out->Value(&_gvn);
5425 const Type* in_type = in->Value(&_gvn);
5426 const TypeAryPtr* top_out = out_type->isa_aryptr();
5427 const TypeAryPtr* top_in = in_type->isa_aryptr();
5428 if (top_out == NULL || top_out->klass() == NULL ||
5429 top_in == NULL || top_in->klass() == NULL) {
5430 // failed array check
5431 return false;
5432 }
5433
5434 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5435 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5436 if (out_elem != T_INT || in_elem != T_INT) {
5437 return false;
5438 }
5439
5440 Node* outlen = load_array_length(out);
5441 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5442 Node* out_start = array_element_address(out, intcon(0), out_elem);
5443 Node* in_start = array_element_address(in, intcon(0), in_elem);
5444
5445 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5446 OptoRuntime::mulAdd_Type(),
5447 stubAddr, stubName, TypePtr::BOTTOM,
5448 out_start,in_start, new_offset, len, k);
5449 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5450 set_result(result);
5451 return true;
5452 }
5453
5454 //-------------inline_montgomeryMultiply-----------------------------------
5455 bool LibraryCallKit::inline_montgomeryMultiply() {
5456 address stubAddr = StubRoutines::montgomeryMultiply();
5457 if (stubAddr == NULL) {
5458 return false; // Intrinsic's stub is not implemented on this platform
5459 }
5460
5461 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5462 const char* stubName = "montgomery_square";
5463
5464 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5465
5466 Node* a = argument(0);
5467 Node* b = argument(1);
5468 Node* n = argument(2);
5469 Node* len = argument(3);
5470 Node* inv = argument(4);
5471 Node* m = argument(6);
5472
5473 const Type* a_type = a->Value(&_gvn);
5474 const TypeAryPtr* top_a = a_type->isa_aryptr();
5475 const Type* b_type = b->Value(&_gvn);
5476 const TypeAryPtr* top_b = b_type->isa_aryptr();
5477 const Type* n_type = a->Value(&_gvn);
5478 const TypeAryPtr* top_n = n_type->isa_aryptr();
5479 const Type* m_type = a->Value(&_gvn);
5480 const TypeAryPtr* top_m = m_type->isa_aryptr();
5481 if (top_a == NULL || top_a->klass() == NULL ||
5482 top_b == NULL || top_b->klass() == NULL ||
5483 top_n == NULL || top_n->klass() == NULL ||
5484 top_m == NULL || top_m->klass() == NULL) {
5485 // failed array check
5486 return false;
5487 }
5488
5489 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5490 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5491 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5492 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5493 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5494 return false;
5495 }
5496
5497 // Make the call
5498 {
5499 Node* a_start = array_element_address(a, intcon(0), a_elem);
5500 Node* b_start = array_element_address(b, intcon(0), b_elem);
5501 Node* n_start = array_element_address(n, intcon(0), n_elem);
5502 Node* m_start = array_element_address(m, intcon(0), m_elem);
5503
5504 Node* call = make_runtime_call(RC_LEAF,
5505 OptoRuntime::montgomeryMultiply_Type(),
5506 stubAddr, stubName, TypePtr::BOTTOM,
5507 a_start, b_start, n_start, len, inv, top(),
5508 m_start);
5509 set_result(m);
5510 }
5511
5512 return true;
5513 }
5514
5515 bool LibraryCallKit::inline_montgomerySquare() {
5516 address stubAddr = StubRoutines::montgomerySquare();
5517 if (stubAddr == NULL) {
5518 return false; // Intrinsic's stub is not implemented on this platform
5519 }
5520
5521 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5522 const char* stubName = "montgomery_square";
5523
5524 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5525
5526 Node* a = argument(0);
5527 Node* n = argument(1);
5528 Node* len = argument(2);
5529 Node* inv = argument(3);
5530 Node* m = argument(5);
5531
5532 const Type* a_type = a->Value(&_gvn);
5533 const TypeAryPtr* top_a = a_type->isa_aryptr();
5534 const Type* n_type = a->Value(&_gvn);
5535 const TypeAryPtr* top_n = n_type->isa_aryptr();
5536 const Type* m_type = a->Value(&_gvn);
5537 const TypeAryPtr* top_m = m_type->isa_aryptr();
5538 if (top_a == NULL || top_a->klass() == NULL ||
5539 top_n == NULL || top_n->klass() == NULL ||
5540 top_m == NULL || top_m->klass() == NULL) {
5541 // failed array check
5542 return false;
5543 }
5544
5545 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5546 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5547 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5548 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5549 return false;
5550 }
5551
5552 // Make the call
5553 {
5554 Node* a_start = array_element_address(a, intcon(0), a_elem);
5555 Node* n_start = array_element_address(n, intcon(0), n_elem);
5556 Node* m_start = array_element_address(m, intcon(0), m_elem);
5557
5558 Node* call = make_runtime_call(RC_LEAF,
5559 OptoRuntime::montgomerySquare_Type(),
5560 stubAddr, stubName, TypePtr::BOTTOM,
5561 a_start, n_start, len, inv, top(),
5562 m_start);
5563 set_result(m);
5564 }
5565
5566 return true;
5567 }
5568
5569 //-------------inline_vectorizedMismatch------------------------------
5570 bool LibraryCallKit::inline_vectorizedMismatch() {
5571 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5572
5573 address stubAddr = StubRoutines::vectorizedMismatch();
5574 if (stubAddr == NULL) {
5575 return false; // Intrinsic's stub is not implemented on this platform
5576 }
5577 const char* stubName = "vectorizedMismatch";
5578 int size_l = callee()->signature()->size();
5579 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5580
5581 Node* obja = argument(0);
5582 Node* aoffset = argument(1);
5583 Node* objb = argument(3);
5584 Node* boffset = argument(4);
5585 Node* length = argument(6);
5586 Node* scale = argument(7);
5587
5588 const Type* a_type = obja->Value(&_gvn);
5589 const Type* b_type = objb->Value(&_gvn);
5590 const TypeAryPtr* top_a = a_type->isa_aryptr();
5591 const TypeAryPtr* top_b = b_type->isa_aryptr();
5592 if (top_a == NULL || top_a->klass() == NULL ||
5593 top_b == NULL || top_b->klass() == NULL) {
5594 // failed array check
5595 return false;
5596 }
5597
5598 Node* call;
5599 jvms()->set_should_reexecute(true);
5600
5601 Node* obja_adr = make_unsafe_address(obja, aoffset);
5602 Node* objb_adr = make_unsafe_address(objb, boffset);
5603
5604 call = make_runtime_call(RC_LEAF,
5605 OptoRuntime::vectorizedMismatch_Type(),
5606 stubAddr, stubName, TypePtr::BOTTOM,
5607 obja_adr, objb_adr, length, scale);
5608
5609 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5610 set_result(result);
5611 return true;
5612 }
5613
5614 /**
5615 * Calculate CRC32 for byte.
5616 * int java.util.zip.CRC32.update(int crc, int b)
5617 */
5618 bool LibraryCallKit::inline_updateCRC32() {
5619 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5620 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5621 // no receiver since it is static method
5622 Node* crc = argument(0); // type: int
5623 Node* b = argument(1); // type: int
5624
5625 /*
5626 * int c = ~ crc;
5627 * b = timesXtoThe32[(b ^ c) & 0xFF];
5628 * b = b ^ (c >>> 8);
5629 * crc = ~b;
5630 */
5631
5632 Node* M1 = intcon(-1);
5633 crc = _gvn.transform(new XorINode(crc, M1));
5634 Node* result = _gvn.transform(new XorINode(crc, b));
5635 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5636
5637 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5638 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5639 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5640 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5641
5642 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5643 result = _gvn.transform(new XorINode(crc, result));
5644 result = _gvn.transform(new XorINode(result, M1));
5645 set_result(result);
5646 return true;
5647 }
5648
5649 /**
5650 * Calculate CRC32 for byte[] array.
5651 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5652 */
5653 bool LibraryCallKit::inline_updateBytesCRC32() {
5654 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5655 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5656 // no receiver since it is static method
5657 Node* crc = argument(0); // type: int
5658 Node* src = argument(1); // type: oop
5659 Node* offset = argument(2); // type: int
5660 Node* length = argument(3); // type: int
5661
5662 const Type* src_type = src->Value(&_gvn);
5663 const TypeAryPtr* top_src = src_type->isa_aryptr();
5664 if (top_src == NULL || top_src->klass() == NULL) {
5665 // failed array check
5666 return false;
5667 }
5668
5669 // Figure out the size and type of the elements we will be copying.
5670 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5671 if (src_elem != T_BYTE) {
5672 return false;
5673 }
5674
5675 // 'src_start' points to src array + scaled offset
5676 Node* src_start = array_element_address(src, offset, src_elem);
5677
5678 // We assume that range check is done by caller.
5679 // TODO: generate range check (offset+length < src.length) in debug VM.
5680
5681 // Call the stub.
5682 address stubAddr = StubRoutines::updateBytesCRC32();
5683 const char *stubName = "updateBytesCRC32";
5684
5685 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5686 stubAddr, stubName, TypePtr::BOTTOM,
5687 crc, src_start, length);
5688 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5689 set_result(result);
5690 return true;
5691 }
5692
5693 /**
5694 * Calculate CRC32 for ByteBuffer.
5695 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5696 */
5697 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5698 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5699 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5700 // no receiver since it is static method
5701 Node* crc = argument(0); // type: int
5702 Node* src = argument(1); // type: long
5703 Node* offset = argument(3); // type: int
5704 Node* length = argument(4); // type: int
5705
5706 src = ConvL2X(src); // adjust Java long to machine word
5707 Node* base = _gvn.transform(new CastX2PNode(src));
5708 offset = ConvI2X(offset);
5709
5710 // 'src_start' points to src array + scaled offset
5711 Node* src_start = basic_plus_adr(top(), base, offset);
5712
5713 // Call the stub.
5714 address stubAddr = StubRoutines::updateBytesCRC32();
5715 const char *stubName = "updateBytesCRC32";
5716
5717 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5718 stubAddr, stubName, TypePtr::BOTTOM,
5719 crc, src_start, length);
5720 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5721 set_result(result);
5722 return true;
5723 }
5724
5725 //------------------------------get_table_from_crc32c_class-----------------------
5726 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5727 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5728 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5729
5730 return table;
5731 }
5732
5733 //------------------------------inline_updateBytesCRC32C-----------------------
5734 //
5735 // Calculate CRC32C for byte[] array.
5736 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5737 //
5738 bool LibraryCallKit::inline_updateBytesCRC32C() {
5739 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5740 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5741 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5742 // no receiver since it is a static method
5743 Node* crc = argument(0); // type: int
5744 Node* src = argument(1); // type: oop
5745 Node* offset = argument(2); // type: int
5746 Node* end = argument(3); // type: int
5747
5748 Node* length = _gvn.transform(new SubINode(end, offset));
5749
5750 const Type* src_type = src->Value(&_gvn);
5751 const TypeAryPtr* top_src = src_type->isa_aryptr();
5752 if (top_src == NULL || top_src->klass() == NULL) {
5753 // failed array check
5754 return false;
5755 }
5756
5757 // Figure out the size and type of the elements we will be copying.
5758 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5759 if (src_elem != T_BYTE) {
5760 return false;
5761 }
5762
5763 // 'src_start' points to src array + scaled offset
5764 Node* src_start = array_element_address(src, offset, src_elem);
5765
5766 // static final int[] byteTable in class CRC32C
5767 Node* table = get_table_from_crc32c_class(callee()->holder());
5768 Node* table_start = array_element_address(table, intcon(0), T_INT);
5769
5770 // We assume that range check is done by caller.
5771 // TODO: generate range check (offset+length < src.length) in debug VM.
5772
5773 // Call the stub.
5774 address stubAddr = StubRoutines::updateBytesCRC32C();
5775 const char *stubName = "updateBytesCRC32C";
5776
5777 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5778 stubAddr, stubName, TypePtr::BOTTOM,
5779 crc, src_start, length, table_start);
5780 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5781 set_result(result);
5782 return true;
5783 }
5784
5785 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5786 //
5787 // Calculate CRC32C for DirectByteBuffer.
5788 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5789 //
5790 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5791 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5792 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5793 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5794 // no receiver since it is a static method
5795 Node* crc = argument(0); // type: int
5796 Node* src = argument(1); // type: long
5797 Node* offset = argument(3); // type: int
5798 Node* end = argument(4); // type: int
5799
5800 Node* length = _gvn.transform(new SubINode(end, offset));
5801
5802 src = ConvL2X(src); // adjust Java long to machine word
5803 Node* base = _gvn.transform(new CastX2PNode(src));
5804 offset = ConvI2X(offset);
5805
5806 // 'src_start' points to src array + scaled offset
5807 Node* src_start = basic_plus_adr(top(), base, offset);
5808
5809 // static final int[] byteTable in class CRC32C
5810 Node* table = get_table_from_crc32c_class(callee()->holder());
5811 Node* table_start = array_element_address(table, intcon(0), T_INT);
5812
5813 // Call the stub.
5814 address stubAddr = StubRoutines::updateBytesCRC32C();
5815 const char *stubName = "updateBytesCRC32C";
5816
5817 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5818 stubAddr, stubName, TypePtr::BOTTOM,
5819 crc, src_start, length, table_start);
5820 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5821 set_result(result);
5822 return true;
5823 }
5824
5825 //------------------------------inline_updateBytesAdler32----------------------
5826 //
5827 // Calculate Adler32 checksum for byte[] array.
5828 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5829 //
5830 bool LibraryCallKit::inline_updateBytesAdler32() {
5831 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5832 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5833 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5834 // no receiver since it is static method
5835 Node* crc = argument(0); // type: int
5836 Node* src = argument(1); // type: oop
5837 Node* offset = argument(2); // type: int
5838 Node* length = argument(3); // type: int
5839
5840 const Type* src_type = src->Value(&_gvn);
5841 const TypeAryPtr* top_src = src_type->isa_aryptr();
5842 if (top_src == NULL || top_src->klass() == NULL) {
5843 // failed array check
5844 return false;
5845 }
5846
5847 // Figure out the size and type of the elements we will be copying.
5848 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5849 if (src_elem != T_BYTE) {
5850 return false;
5851 }
5852
5853 // 'src_start' points to src array + scaled offset
5854 Node* src_start = array_element_address(src, offset, src_elem);
5855
5856 // We assume that range check is done by caller.
5857 // TODO: generate range check (offset+length < src.length) in debug VM.
5858
5859 // Call the stub.
5860 address stubAddr = StubRoutines::updateBytesAdler32();
5861 const char *stubName = "updateBytesAdler32";
5862
5863 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5864 stubAddr, stubName, TypePtr::BOTTOM,
5865 crc, src_start, length);
5866 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5867 set_result(result);
5868 return true;
5869 }
5870
5871 //------------------------------inline_updateByteBufferAdler32---------------
5872 //
5873 // Calculate Adler32 checksum for DirectByteBuffer.
5874 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5875 //
5876 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5877 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5878 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5879 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5880 // no receiver since it is static method
5881 Node* crc = argument(0); // type: int
5882 Node* src = argument(1); // type: long
5883 Node* offset = argument(3); // type: int
5884 Node* length = argument(4); // type: int
5885
5886 src = ConvL2X(src); // adjust Java long to machine word
5887 Node* base = _gvn.transform(new CastX2PNode(src));
5888 offset = ConvI2X(offset);
5889
5890 // 'src_start' points to src array + scaled offset
5891 Node* src_start = basic_plus_adr(top(), base, offset);
5892
5893 // Call the stub.
5894 address stubAddr = StubRoutines::updateBytesAdler32();
5895 const char *stubName = "updateBytesAdler32";
5896
5897 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5898 stubAddr, stubName, TypePtr::BOTTOM,
5899 crc, src_start, length);
5900
5901 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5902 set_result(result);
5903 return true;
5904 }
5905
5906 //----------------------------inline_reference_get----------------------------
5907 // public T java.lang.ref.Reference.get();
5908 bool LibraryCallKit::inline_reference_get() {
5909 const int referent_offset = java_lang_ref_Reference::referent_offset;
5910 guarantee(referent_offset > 0, "should have already been set");
5911
5912 // Get the argument:
5913 Node* reference_obj = null_check_receiver();
5914 if (stopped()) return true;
5915
5916 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5917
5918 ciInstanceKlass* klass = env()->Object_klass();
5919 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5920
5921 Node* no_ctrl = NULL;
5922 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5923
5924 // Use the pre-barrier to record the value in the referent field
5925 pre_barrier(false /* do_load */,
5926 control(),
5927 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5928 result /* pre_val */,
5929 T_OBJECT);
5930
5931 // Add memory barrier to prevent commoning reads from this field
5932 // across safepoint since GC can change its value.
5933 insert_mem_bar(Op_MemBarCPUOrder);
5934
5935 set_result(result);
5936 return true;
5937 }
5938
5939
5940 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5941 bool is_exact=true, bool is_static=false,
5942 ciInstanceKlass * fromKls=NULL) {
5943 if (fromKls == NULL) {
5944 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5945 assert(tinst != NULL, "obj is null");
5946 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5947 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5948 fromKls = tinst->klass()->as_instance_klass();
5949 } else {
5950 assert(is_static, "only for static field access");
5951 }
5952 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5953 ciSymbol::make(fieldTypeString),
5954 is_static);
5955
5956 assert (field != NULL, "undefined field");
5957 if (field == NULL) return (Node *) NULL;
5958
5959 if (is_static) {
5960 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5961 fromObj = makecon(tip);
5962 }
5963
5964 // Next code copied from Parse::do_get_xxx():
5965
5966 // Compute address and memory type.
5967 int offset = field->offset_in_bytes();
5968 bool is_vol = field->is_volatile();
5969 ciType* field_klass = field->type();
5970 assert(field_klass->is_loaded(), "should be loaded");
5971 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5972 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5973 BasicType bt = field->layout_type();
5974
5975 // Build the resultant type of the load
5976 const Type *type;
5977 if (bt == T_OBJECT) {
5978 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5979 } else {
5980 type = Type::get_const_basic_type(bt);
5981 }
5982
5983 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
5984 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
5985 }
5986 // Build the load.
5987 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
5988 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
5989 // If reference is volatile, prevent following memory ops from
5990 // floating up past the volatile read. Also prevents commoning
5991 // another volatile read.
5992 if (is_vol) {
5993 // Memory barrier includes bogus read of value to force load BEFORE membar
5994 insert_mem_bar(Op_MemBarAcquire, loadedField);
5995 }
5996 return loadedField;
5997 }
5998
5999 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6000 bool is_exact = true, bool is_static = false,
6001 ciInstanceKlass * fromKls = NULL) {
6002 if (fromKls == NULL) {
6003 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6004 assert(tinst != NULL, "obj is null");
6005 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6006 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6007 fromKls = tinst->klass()->as_instance_klass();
6008 }
6009 else {
6010 assert(is_static, "only for static field access");
6011 }
6012 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6013 ciSymbol::make(fieldTypeString),
6014 is_static);
6015
6016 assert(field != NULL, "undefined field");
6017 assert(!field->is_volatile(), "not defined for volatile fields");
6018
6019 if (is_static) {
6020 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6021 fromObj = makecon(tip);
6022 }
6023
6024 // Next code copied from Parse::do_get_xxx():
6025
6026 // Compute address and memory type.
6027 int offset = field->offset_in_bytes();
6028 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6029
6030 return adr;
6031 }
6032
6033 //------------------------------inline_aescrypt_Block-----------------------
6034 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6035 address stubAddr = NULL;
6036 const char *stubName;
6037 assert(UseAES, "need AES instruction support");
6038
6039 switch(id) {
6040 case vmIntrinsics::_aescrypt_encryptBlock:
6041 stubAddr = StubRoutines::aescrypt_encryptBlock();
6042 stubName = "aescrypt_encryptBlock";
6043 break;
6044 case vmIntrinsics::_aescrypt_decryptBlock:
6045 stubAddr = StubRoutines::aescrypt_decryptBlock();
6046 stubName = "aescrypt_decryptBlock";
6047 break;
6048 }
6049 if (stubAddr == NULL) return false;
6050
6051 Node* aescrypt_object = argument(0);
6052 Node* src = argument(1);
6053 Node* src_offset = argument(2);
6054 Node* dest = argument(3);
6055 Node* dest_offset = argument(4);
6056
6057 // (1) src and dest are arrays.
6058 const Type* src_type = src->Value(&_gvn);
6059 const Type* dest_type = dest->Value(&_gvn);
6060 const TypeAryPtr* top_src = src_type->isa_aryptr();
6061 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6062 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6063
6064 // for the quick and dirty code we will skip all the checks.
6065 // we are just trying to get the call to be generated.
6066 Node* src_start = src;
6067 Node* dest_start = dest;
6068 if (src_offset != NULL || dest_offset != NULL) {
6069 assert(src_offset != NULL && dest_offset != NULL, "");
6070 src_start = array_element_address(src, src_offset, T_BYTE);
6071 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6072 }
6073
6074 // now need to get the start of its expanded key array
6075 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6076 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6077 if (k_start == NULL) return false;
6078
6079 if (Matcher::pass_original_key_for_aes()) {
6080 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6081 // compatibility issues between Java key expansion and SPARC crypto instructions
6082 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6083 if (original_k_start == NULL) return false;
6084
6085 // Call the stub.
6086 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6087 stubAddr, stubName, TypePtr::BOTTOM,
6088 src_start, dest_start, k_start, original_k_start);
6089 } else {
6090 // Call the stub.
6091 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6092 stubAddr, stubName, TypePtr::BOTTOM,
6093 src_start, dest_start, k_start);
6094 }
6095
6096 return true;
6097 }
6098
6099 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6100 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6101 address stubAddr = NULL;
6102 const char *stubName = NULL;
6103
6104 assert(UseAES, "need AES instruction support");
6105
6106 switch(id) {
6107 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6108 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6109 stubName = "cipherBlockChaining_encryptAESCrypt";
6110 break;
6111 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6112 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6113 stubName = "cipherBlockChaining_decryptAESCrypt";
6114 break;
6115 }
6116 if (stubAddr == NULL) return false;
6117
6118 Node* cipherBlockChaining_object = argument(0);
6119 Node* src = argument(1);
6120 Node* src_offset = argument(2);
6121 Node* len = argument(3);
6122 Node* dest = argument(4);
6123 Node* dest_offset = argument(5);
6124
6125 // (1) src and dest are arrays.
6126 const Type* src_type = src->Value(&_gvn);
6127 const Type* dest_type = dest->Value(&_gvn);
6128 const TypeAryPtr* top_src = src_type->isa_aryptr();
6129 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6130 assert (top_src != NULL && top_src->klass() != NULL
6131 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6132
6133 // checks are the responsibility of the caller
6134 Node* src_start = src;
6135 Node* dest_start = dest;
6136 if (src_offset != NULL || dest_offset != NULL) {
6137 assert(src_offset != NULL && dest_offset != NULL, "");
6138 src_start = array_element_address(src, src_offset, T_BYTE);
6139 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6140 }
6141
6142 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6143 // (because of the predicated logic executed earlier).
6144 // so we cast it here safely.
6145 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6146
6147 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6148 if (embeddedCipherObj == NULL) return false;
6149
6150 // cast it to what we know it will be at runtime
6151 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6152 assert(tinst != NULL, "CBC obj is null");
6153 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6154 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6155 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6156
6157 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6158 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6159 const TypeOopPtr* xtype = aklass->as_instance_type();
6160 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6161 aescrypt_object = _gvn.transform(aescrypt_object);
6162
6163 // we need to get the start of the aescrypt_object's expanded key array
6164 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6165 if (k_start == NULL) return false;
6166
6167 // similarly, get the start address of the r vector
6168 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6169 if (objRvec == NULL) return false;
6170 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6171
6172 Node* cbcCrypt;
6173 if (Matcher::pass_original_key_for_aes()) {
6174 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6175 // compatibility issues between Java key expansion and SPARC crypto instructions
6176 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6177 if (original_k_start == NULL) return false;
6178
6179 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6180 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6181 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6182 stubAddr, stubName, TypePtr::BOTTOM,
6183 src_start, dest_start, k_start, r_start, len, original_k_start);
6184 } else {
6185 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6186 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6187 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6188 stubAddr, stubName, TypePtr::BOTTOM,
6189 src_start, dest_start, k_start, r_start, len);
6190 }
6191
6192 // return cipher length (int)
6193 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6194 set_result(retvalue);
6195 return true;
6196 }
6197
6198 //------------------------------inline_counterMode_AESCrypt-----------------------
6199 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6200 assert(UseAES, "need AES instruction support");
6201 if (!UseAESCTRIntrinsics) return false;
6202
6203 address stubAddr = NULL;
6204 const char *stubName = NULL;
6205 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6206 stubAddr = StubRoutines::counterMode_AESCrypt();
6207 stubName = "counterMode_AESCrypt";
6208 }
6209 if (stubAddr == NULL) return false;
6210
6211 Node* counterMode_object = argument(0);
6212 Node* src = argument(1);
6213 Node* src_offset = argument(2);
6214 Node* len = argument(3);
6215 Node* dest = argument(4);
6216 Node* dest_offset = argument(5);
6217
6218 // (1) src and dest are arrays.
6219 const Type* src_type = src->Value(&_gvn);
6220 const Type* dest_type = dest->Value(&_gvn);
6221 const TypeAryPtr* top_src = src_type->isa_aryptr();
6222 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6223 assert(top_src != NULL && top_src->klass() != NULL &&
6224 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6225
6226 // checks are the responsibility of the caller
6227 Node* src_start = src;
6228 Node* dest_start = dest;
6229 if (src_offset != NULL || dest_offset != NULL) {
6230 assert(src_offset != NULL && dest_offset != NULL, "");
6231 src_start = array_element_address(src, src_offset, T_BYTE);
6232 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6233 }
6234
6235 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6236 // (because of the predicated logic executed earlier).
6237 // so we cast it here safely.
6238 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6239 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6240 if (embeddedCipherObj == NULL) return false;
6241 // cast it to what we know it will be at runtime
6242 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6243 assert(tinst != NULL, "CTR obj is null");
6244 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6245 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6246 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6247 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6248 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6249 const TypeOopPtr* xtype = aklass->as_instance_type();
6250 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6251 aescrypt_object = _gvn.transform(aescrypt_object);
6252 // we need to get the start of the aescrypt_object's expanded key array
6253 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6254 if (k_start == NULL) return false;
6255 // similarly, get the start address of the r vector
6256 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6257 if (obj_counter == NULL) return false;
6258 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6259
6260 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6261 if (saved_encCounter == NULL) return false;
6262 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6263 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6264
6265 Node* ctrCrypt;
6266 if (Matcher::pass_original_key_for_aes()) {
6267 // no SPARC version for AES/CTR intrinsics now.
6268 return false;
6269 }
6270 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6271 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6272 OptoRuntime::counterMode_aescrypt_Type(),
6273 stubAddr, stubName, TypePtr::BOTTOM,
6274 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6275
6276 // return cipher length (int)
6277 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6278 set_result(retvalue);
6279 return true;
6280 }
6281
6282 //------------------------------get_key_start_from_aescrypt_object-----------------------
6283 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6284 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6285 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6286 if (objAESCryptKey == NULL) return (Node *) NULL;
6287
6288 // now have the array, need to get the start address of the K array
6289 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6290 return k_start;
6291 }
6292
6293 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6294 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6295 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6296 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6297 if (objAESCryptKey == NULL) return (Node *) NULL;
6298
6299 // now have the array, need to get the start address of the lastKey array
6300 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6301 return original_k_start;
6302 }
6303
6304 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6305 // Return node representing slow path of predicate check.
6306 // the pseudo code we want to emulate with this predicate is:
6307 // for encryption:
6308 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6309 // for decryption:
6310 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6311 // note cipher==plain is more conservative than the original java code but that's OK
6312 //
6313 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6314 // The receiver was checked for NULL already.
6315 Node* objCBC = argument(0);
6316
6317 // Load embeddedCipher field of CipherBlockChaining object.
6318 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6319
6320 // get AESCrypt klass for instanceOf check
6321 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6322 // will have same classloader as CipherBlockChaining object
6323 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6324 assert(tinst != NULL, "CBCobj is null");
6325 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6326
6327 // we want to do an instanceof comparison against the AESCrypt class
6328 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6329 if (!klass_AESCrypt->is_loaded()) {
6330 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6331 Node* ctrl = control();
6332 set_control(top()); // no regular fast path
6333 return ctrl;
6334 }
6335 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6336
6337 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6338 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6339 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6340
6341 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6342
6343 // for encryption, we are done
6344 if (!decrypting)
6345 return instof_false; // even if it is NULL
6346
6347 // for decryption, we need to add a further check to avoid
6348 // taking the intrinsic path when cipher and plain are the same
6349 // see the original java code for why.
6350 RegionNode* region = new RegionNode(3);
6351 region->init_req(1, instof_false);
6352 Node* src = argument(1);
6353 Node* dest = argument(4);
6354 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6355 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6356 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6357 region->init_req(2, src_dest_conjoint);
6358
6359 record_for_igvn(region);
6360 return _gvn.transform(region);
6361 }
6362
6363 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6364 // Return node representing slow path of predicate check.
6365 // the pseudo code we want to emulate with this predicate is:
6366 // for encryption:
6367 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6368 // for decryption:
6369 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6370 // note cipher==plain is more conservative than the original java code but that's OK
6371 //
6372
6373 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6374 // The receiver was checked for NULL already.
6375 Node* objCTR = argument(0);
6376
6377 // Load embeddedCipher field of CipherBlockChaining object.
6378 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6379
6380 // get AESCrypt klass for instanceOf check
6381 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6382 // will have same classloader as CipherBlockChaining object
6383 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6384 assert(tinst != NULL, "CTRobj is null");
6385 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6386
6387 // we want to do an instanceof comparison against the AESCrypt class
6388 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6389 if (!klass_AESCrypt->is_loaded()) {
6390 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6391 Node* ctrl = control();
6392 set_control(top()); // no regular fast path
6393 return ctrl;
6394 }
6395
6396 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6397 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6398 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6399 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6400 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6401
6402 return instof_false; // even if it is NULL
6403 }
6404
6405 //------------------------------inline_ghash_processBlocks
6406 bool LibraryCallKit::inline_ghash_processBlocks() {
6407 address stubAddr;
6408 const char *stubName;
6409 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6410
6411 stubAddr = StubRoutines::ghash_processBlocks();
6412 stubName = "ghash_processBlocks";
6413
6414 Node* data = argument(0);
6415 Node* offset = argument(1);
6416 Node* len = argument(2);
6417 Node* state = argument(3);
6418 Node* subkeyH = argument(4);
6419
6420 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6421 assert(state_start, "state is NULL");
6422 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6423 assert(subkeyH_start, "subkeyH is NULL");
6424 Node* data_start = array_element_address(data, offset, T_BYTE);
6425 assert(data_start, "data is NULL");
6426
6427 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6428 OptoRuntime::ghash_processBlocks_Type(),
6429 stubAddr, stubName, TypePtr::BOTTOM,
6430 state_start, subkeyH_start, data_start, len);
6431 return true;
6432 }
6433
6434 //------------------------------inline_sha_implCompress-----------------------
6435 //
6436 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6437 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6438 //
6439 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6440 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6441 //
6442 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6443 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6444 //
6445 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6446 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6447
6448 Node* sha_obj = argument(0);
6449 Node* src = argument(1); // type oop
6450 Node* ofs = argument(2); // type int
6451
6452 const Type* src_type = src->Value(&_gvn);
6453 const TypeAryPtr* top_src = src_type->isa_aryptr();
6454 if (top_src == NULL || top_src->klass() == NULL) {
6455 // failed array check
6456 return false;
6457 }
6458 // Figure out the size and type of the elements we will be copying.
6459 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6460 if (src_elem != T_BYTE) {
6461 return false;
6462 }
6463 // 'src_start' points to src array + offset
6464 Node* src_start = array_element_address(src, ofs, src_elem);
6465 Node* state = NULL;
6466 address stubAddr;
6467 const char *stubName;
6468
6469 switch(id) {
6470 case vmIntrinsics::_sha_implCompress:
6471 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6472 state = get_state_from_sha_object(sha_obj);
6473 stubAddr = StubRoutines::sha1_implCompress();
6474 stubName = "sha1_implCompress";
6475 break;
6476 case vmIntrinsics::_sha2_implCompress:
6477 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6478 state = get_state_from_sha_object(sha_obj);
6479 stubAddr = StubRoutines::sha256_implCompress();
6480 stubName = "sha256_implCompress";
6481 break;
6482 case vmIntrinsics::_sha5_implCompress:
6483 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6484 state = get_state_from_sha5_object(sha_obj);
6485 stubAddr = StubRoutines::sha512_implCompress();
6486 stubName = "sha512_implCompress";
6487 break;
6488 default:
6489 fatal_unexpected_iid(id);
6490 return false;
6491 }
6492 if (state == NULL) return false;
6493
6494 // Call the stub.
6495 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6496 stubAddr, stubName, TypePtr::BOTTOM,
6497 src_start, state);
6498
6499 return true;
6500 }
6501
6502 //------------------------------inline_digestBase_implCompressMB-----------------------
6503 //
6504 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6505 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6506 //
6507 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6508 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6509 "need SHA1/SHA256/SHA512 instruction support");
6510 assert((uint)predicate < 3, "sanity");
6511 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6512
6513 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6514 Node* src = argument(1); // byte[] array
6515 Node* ofs = argument(2); // type int
6516 Node* limit = argument(3); // type int
6517
6518 const Type* src_type = src->Value(&_gvn);
6519 const TypeAryPtr* top_src = src_type->isa_aryptr();
6520 if (top_src == NULL || top_src->klass() == NULL) {
6521 // failed array check
6522 return false;
6523 }
6524 // Figure out the size and type of the elements we will be copying.
6525 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6526 if (src_elem != T_BYTE) {
6527 return false;
6528 }
6529 // 'src_start' points to src array + offset
6530 Node* src_start = array_element_address(src, ofs, src_elem);
6531
6532 const char* klass_SHA_name = NULL;
6533 const char* stub_name = NULL;
6534 address stub_addr = NULL;
6535 bool long_state = false;
6536
6537 switch (predicate) {
6538 case 0:
6539 if (UseSHA1Intrinsics) {
6540 klass_SHA_name = "sun/security/provider/SHA";
6541 stub_name = "sha1_implCompressMB";
6542 stub_addr = StubRoutines::sha1_implCompressMB();
6543 }
6544 break;
6545 case 1:
6546 if (UseSHA256Intrinsics) {
6547 klass_SHA_name = "sun/security/provider/SHA2";
6548 stub_name = "sha256_implCompressMB";
6549 stub_addr = StubRoutines::sha256_implCompressMB();
6550 }
6551 break;
6552 case 2:
6553 if (UseSHA512Intrinsics) {
6554 klass_SHA_name = "sun/security/provider/SHA5";
6555 stub_name = "sha512_implCompressMB";
6556 stub_addr = StubRoutines::sha512_implCompressMB();
6557 long_state = true;
6558 }
6559 break;
6560 default:
6561 fatal("unknown SHA intrinsic predicate: %d", predicate);
6562 }
6563 if (klass_SHA_name != NULL) {
6564 // get DigestBase klass to lookup for SHA klass
6565 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6566 assert(tinst != NULL, "digestBase_obj is not instance???");
6567 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6568
6569 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6570 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6571 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6572 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6573 }
6574 return false;
6575 }
6576 //------------------------------inline_sha_implCompressMB-----------------------
6577 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6578 bool long_state, address stubAddr, const char *stubName,
6579 Node* src_start, Node* ofs, Node* limit) {
6580 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6581 const TypeOopPtr* xtype = aklass->as_instance_type();
6582 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6583 sha_obj = _gvn.transform(sha_obj);
6584
6585 Node* state;
6586 if (long_state) {
6587 state = get_state_from_sha5_object(sha_obj);
6588 } else {
6589 state = get_state_from_sha_object(sha_obj);
6590 }
6591 if (state == NULL) return false;
6592
6593 // Call the stub.
6594 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6595 OptoRuntime::digestBase_implCompressMB_Type(),
6596 stubAddr, stubName, TypePtr::BOTTOM,
6597 src_start, state, ofs, limit);
6598 // return ofs (int)
6599 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6600 set_result(result);
6601
6602 return true;
6603 }
6604
6605 //------------------------------get_state_from_sha_object-----------------------
6606 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6607 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6608 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6609 if (sha_state == NULL) return (Node *) NULL;
6610
6611 // now have the array, need to get the start address of the state array
6612 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6613 return state;
6614 }
6615
6616 //------------------------------get_state_from_sha5_object-----------------------
6617 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6618 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6619 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6620 if (sha_state == NULL) return (Node *) NULL;
6621
6622 // now have the array, need to get the start address of the state array
6623 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6624 return state;
6625 }
6626
6627 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6628 // Return node representing slow path of predicate check.
6629 // the pseudo code we want to emulate with this predicate is:
6630 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6631 //
6632 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6633 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6634 "need SHA1/SHA256/SHA512 instruction support");
6635 assert((uint)predicate < 3, "sanity");
6636
6637 // The receiver was checked for NULL already.
6638 Node* digestBaseObj = argument(0);
6639
6640 // get DigestBase klass for instanceOf check
6641 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6642 assert(tinst != NULL, "digestBaseObj is null");
6643 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6644
6645 const char* klass_SHA_name = NULL;
6646 switch (predicate) {
6647 case 0:
6648 if (UseSHA1Intrinsics) {
6649 // we want to do an instanceof comparison against the SHA class
6650 klass_SHA_name = "sun/security/provider/SHA";
6651 }
6652 break;
6653 case 1:
6654 if (UseSHA256Intrinsics) {
6655 // we want to do an instanceof comparison against the SHA2 class
6656 klass_SHA_name = "sun/security/provider/SHA2";
6657 }
6658 break;
6659 case 2:
6660 if (UseSHA512Intrinsics) {
6661 // we want to do an instanceof comparison against the SHA5 class
6662 klass_SHA_name = "sun/security/provider/SHA5";
6663 }
6664 break;
6665 default:
6666 fatal("unknown SHA intrinsic predicate: %d", predicate);
6667 }
6668
6669 ciKlass* klass_SHA = NULL;
6670 if (klass_SHA_name != NULL) {
6671 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6672 }
6673 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6674 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6675 Node* ctrl = control();
6676 set_control(top()); // no intrinsic path
6677 return ctrl;
6678 }
6679 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6680
6681 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6682 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6683 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6684 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6685
6686 return instof_false; // even if it is NULL
6687 }
6688
6689 bool LibraryCallKit::inline_profileBoolean() {
6690 Node* counts = argument(1);
6691 const TypeAryPtr* ary = NULL;
6692 ciArray* aobj = NULL;
6693 if (counts->is_Con()
6694 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6695 && (aobj = ary->const_oop()->as_array()) != NULL
6696 && (aobj->length() == 2)) {
6697 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6698 jint false_cnt = aobj->element_value(0).as_int();
6699 jint true_cnt = aobj->element_value(1).as_int();
6700
6701 if (C->log() != NULL) {
6702 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6703 false_cnt, true_cnt);
6704 }
6705
6706 if (false_cnt + true_cnt == 0) {
6707 // According to profile, never executed.
6708 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6709 Deoptimization::Action_reinterpret);
6710 return true;
6711 }
6712
6713 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6714 // is a number of each value occurrences.
6715 Node* result = argument(0);
6716 if (false_cnt == 0 || true_cnt == 0) {
6717 // According to profile, one value has been never seen.
6718 int expected_val = (false_cnt == 0) ? 1 : 0;
6719
6720 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6721 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6722
6723 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6724 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6725 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6726
6727 { // Slow path: uncommon trap for never seen value and then reexecute
6728 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6729 // the value has been seen at least once.
6730 PreserveJVMState pjvms(this);
6731 PreserveReexecuteState preexecs(this);
6732 jvms()->set_should_reexecute(true);
6733
6734 set_control(slow_path);
6735 set_i_o(i_o());
6736
6737 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6738 Deoptimization::Action_reinterpret);
6739 }
6740 // The guard for never seen value enables sharpening of the result and
6741 // returning a constant. It allows to eliminate branches on the same value
6742 // later on.
6743 set_control(fast_path);
6744 result = intcon(expected_val);
6745 }
6746 // Stop profiling.
6747 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6748 // By replacing method body with profile data (represented as ProfileBooleanNode
6749 // on IR level) we effectively disable profiling.
6750 // It enables full speed execution once optimized code is generated.
6751 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6752 C->record_for_igvn(profile);
6753 set_result(profile);
6754 return true;
6755 } else {
6756 // Continue profiling.
6757 // Profile data isn't available at the moment. So, execute method's bytecode version.
6758 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6759 // is compiled and counters aren't available since corresponding MethodHandle
6760 // isn't a compile-time constant.
6761 return false;
6762 }
6763 }
6764
6765 bool LibraryCallKit::inline_isCompileConstant() {
6766 Node* n = argument(0);
6767 set_result(n->is_Con() ? intcon(1) : intcon(0));
6768 return true;
6769 }
6770
6771 bool LibraryCallKit::inline_deoptimize() {
6772 assert(WhiteBoxAPI, "");
6773 PreserveReexecuteState preexecs(this);
6774 jvms()->set_should_reexecute(false);
6775 uncommon_trap(Deoptimization::Reason_intrinsic,
6776 Deoptimization::Action_none);
6777 return true;
6778 }