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