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