1 /*
2 * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "opto/addnode.hpp"
34 #include "opto/arraycopynode.hpp"
35 #include "opto/c2compiler.hpp"
36 #include "opto/callGenerator.hpp"
37 #include "opto/castnode.hpp"
38 #include "opto/cfgnode.hpp"
39 #include "opto/convertnode.hpp"
40 #include "opto/countbitsnode.hpp"
41 #include "opto/intrinsicnode.hpp"
42 #include "opto/idealKit.hpp"
43 #include "opto/mathexactnode.hpp"
44 #include "opto/movenode.hpp"
45 #include "opto/mulnode.hpp"
46 #include "opto/narrowptrnode.hpp"
47 #include "opto/opaquenode.hpp"
48 #include "opto/parse.hpp"
49 #include "opto/runtime.hpp"
50 #include "opto/subnode.hpp"
51 #include "prims/nativeLookup.hpp"
52 #include "prims/unsafe.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #ifdef TRACE_HAVE_INTRINSICS
55 #include "trace/traceMacros.hpp"
56 #endif
57
58 class LibraryIntrinsic : public InlineCallGenerator {
59 // Extend the set of intrinsics known to the runtime:
60 public:
61 private:
62 bool _is_virtual;
63 bool _does_virtual_dispatch;
64 int8_t _predicates_count; // Intrinsic is predicated by several conditions
65 int8_t _last_predicate; // Last generated predicate
66 vmIntrinsics::ID _intrinsic_id;
67
68 public:
69 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
70 : InlineCallGenerator(m),
71 _is_virtual(is_virtual),
72 _does_virtual_dispatch(does_virtual_dispatch),
73 _predicates_count((int8_t)predicates_count),
74 _last_predicate((int8_t)-1),
75 _intrinsic_id(id)
76 {
77 }
78 virtual bool is_intrinsic() const { return true; }
79 virtual bool is_virtual() const { return _is_virtual; }
80 virtual bool is_predicated() const { return _predicates_count > 0; }
81 virtual int predicates_count() const { return _predicates_count; }
82 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
83 virtual JVMState* generate(JVMState* jvms);
84 virtual Node* generate_predicate(JVMState* jvms, int predicate);
85 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
86 };
87
88
89 // Local helper class for LibraryIntrinsic:
90 class LibraryCallKit : public GraphKit {
91 private:
92 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
93 Node* _result; // the result node, if any
94 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
95
96 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
97
98 public:
99 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
100 : GraphKit(jvms),
101 _intrinsic(intrinsic),
102 _result(NULL)
103 {
104 // Check if this is a root compile. In that case we don't have a caller.
105 if (!jvms->has_method()) {
106 _reexecute_sp = sp();
107 } else {
108 // Find out how many arguments the interpreter needs when deoptimizing
109 // and save the stack pointer value so it can used by uncommon_trap.
110 // We find the argument count by looking at the declared signature.
111 bool ignored_will_link;
112 ciSignature* declared_signature = NULL;
113 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
114 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
115 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
116 }
117 }
118
119 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
120
121 ciMethod* caller() const { return jvms()->method(); }
122 int bci() const { return jvms()->bci(); }
123 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
124 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
125 ciMethod* callee() const { return _intrinsic->method(); }
126
127 bool try_to_inline(int predicate);
128 Node* try_to_predicate(int predicate);
129
130 void push_result() {
131 // Push the result onto the stack.
132 if (!stopped() && result() != NULL) {
133 BasicType bt = result()->bottom_type()->basic_type();
134 push_node(bt, result());
135 }
136 }
137
138 private:
139 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
140 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
141 }
142
143 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
144 void set_result(RegionNode* region, PhiNode* value);
145 Node* result() { return _result; }
146
147 virtual int reexecute_sp() { return _reexecute_sp; }
148
149 // Helper functions to inline natives
150 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
151 Node* generate_slow_guard(Node* test, RegionNode* region);
152 Node* generate_fair_guard(Node* test, RegionNode* region);
153 Node* generate_negative_guard(Node* index, RegionNode* region,
154 // resulting CastII of index:
155 Node* *pos_index = NULL);
156 Node* generate_limit_guard(Node* offset, Node* subseq_length,
157 Node* array_length,
158 RegionNode* region);
159 void generate_string_range_check(Node* array, Node* offset,
160 Node* length, bool char_count);
161 Node* generate_current_thread(Node* &tls_output);
162 Node* load_mirror_from_klass(Node* klass);
163 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
164 RegionNode* region, int null_path,
165 int offset);
166 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
167 RegionNode* region, int null_path) {
168 int offset = java_lang_Class::klass_offset_in_bytes();
169 return load_klass_from_mirror_common(mirror, never_see_null,
170 region, null_path,
171 offset);
172 }
173 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
174 RegionNode* region, int null_path) {
175 int offset = java_lang_Class::array_klass_offset_in_bytes();
176 return load_klass_from_mirror_common(mirror, never_see_null,
177 region, null_path,
178 offset);
179 }
180 Node* generate_access_flags_guard(Node* kls,
181 int modifier_mask, int modifier_bits,
182 RegionNode* region);
183 Node* generate_interface_guard(Node* kls, RegionNode* region);
184 Node* generate_array_guard(Node* kls, RegionNode* region) {
185 return generate_array_guard_common(kls, region, false, false);
186 }
187 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
188 return generate_array_guard_common(kls, region, false, true);
189 }
190 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
191 return generate_array_guard_common(kls, region, true, false);
192 }
193 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
194 return generate_array_guard_common(kls, region, true, true);
195 }
196 Node* generate_array_guard_common(Node* kls, RegionNode* region,
197 bool obj_array, bool not_array);
198 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
199 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
200 bool is_virtual = false, bool is_static = false);
201 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
202 return generate_method_call(method_id, false, true);
203 }
204 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
205 return generate_method_call(method_id, true, false);
206 }
207 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
208 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
209
210 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
211 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
213 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
214 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
215 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOfChar();
217 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_toBytesU();
219 bool inline_string_getCharsU();
220 bool inline_string_copy(bool compress);
221 bool inline_string_char_access(bool is_store);
222 Node* round_double_node(Node* n);
223 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
224 bool inline_math_native(vmIntrinsics::ID id);
225 bool inline_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_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 #ifdef TRACE_HAVE_INTRINSICS
258 bool inline_native_classID();
259 bool inline_native_getBufferWriter();
260 #endif
261 bool inline_native_isInterrupted();
262 bool inline_native_Class_query(vmIntrinsics::ID id);
263 bool inline_native_subtype_check();
264 bool inline_native_getLength();
265 bool inline_array_copyOf(bool is_copyOfRange);
266 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
267 bool inline_preconditions_checkIndex();
268 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
269 bool inline_native_clone(bool is_virtual);
270 bool inline_native_Reflection_getCallerClass();
271 // Helper function for inlining native object hash method
272 bool inline_native_hashcode(bool is_virtual, bool is_static);
273 bool inline_native_getClass();
274
275 // Helper functions for inlining arraycopy
276 bool inline_arraycopy();
277 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
278 RegionNode* slow_region);
279 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
280 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
281 uint new_idx);
282
283 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
284 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
285 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
286 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
287 bool inline_unsafe_fence(vmIntrinsics::ID id);
288 bool inline_onspinwait();
289 bool inline_fp_conversions(vmIntrinsics::ID id);
290 bool inline_number_methods(vmIntrinsics::ID id);
291 bool inline_reference_get();
292 bool inline_Class_cast();
293 bool inline_aescrypt_Block(vmIntrinsics::ID id);
294 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
295 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
296 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
297 Node* inline_counterMode_AESCrypt_predicate();
298 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
299 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
300 bool inline_ghash_processBlocks();
301 bool inline_sha_implCompress(vmIntrinsics::ID id);
302 bool inline_digestBase_implCompressMB(int predicate);
303 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
304 bool long_state, address stubAddr, const char *stubName,
305 Node* src_start, Node* ofs, Node* limit);
306 Node* get_state_from_sha_object(Node *sha_object);
307 Node* get_state_from_sha5_object(Node *sha_object);
308 Node* inline_digestBase_implCompressMB_predicate(int predicate);
309 bool inline_encodeISOArray();
310 bool inline_updateCRC32();
311 bool inline_updateBytesCRC32();
312 bool inline_updateByteBufferCRC32();
313 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
314 bool inline_updateBytesCRC32C();
315 bool inline_updateDirectByteBufferCRC32C();
316 bool inline_updateBytesAdler32();
317 bool inline_updateByteBufferAdler32();
318 bool inline_multiplyToLen();
319 bool inline_hasNegatives();
320 bool inline_squareToLen();
321 bool inline_mulAdd();
322 bool inline_montgomeryMultiply();
323 bool inline_montgomerySquare();
324 bool inline_vectorizedMismatch();
325 bool inline_fma(vmIntrinsics::ID id);
326
327 bool inline_profileBoolean();
328 bool inline_isCompileConstant();
329 };
330
331 //---------------------------make_vm_intrinsic----------------------------
332 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
333 vmIntrinsics::ID id = m->intrinsic_id();
334 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
335
336 if (!m->is_loaded()) {
337 // Do not attempt to inline unloaded methods.
338 return NULL;
339 }
340
341 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
342 bool is_available = false;
343
344 {
345 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
346 // the compiler must transition to '_thread_in_vm' state because both
347 // methods access VM-internal data.
348 VM_ENTRY_MARK;
349 methodHandle mh(THREAD, m->get_Method());
350 is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
351 !C->directive()->is_intrinsic_disabled(mh) &&
352 !vmIntrinsics::is_disabled_by_flags(mh);
353
354 }
355
356 if (is_available) {
357 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
358 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
359 return new LibraryIntrinsic(m, is_virtual,
360 vmIntrinsics::predicates_needed(id),
361 vmIntrinsics::does_virtual_dispatch(id),
362 (vmIntrinsics::ID) id);
363 } else {
364 return NULL;
365 }
366 }
367
368 //----------------------register_library_intrinsics-----------------------
369 // Initialize this file's data structures, for each Compile instance.
370 void Compile::register_library_intrinsics() {
371 // Nothing to do here.
372 }
373
374 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
375 LibraryCallKit kit(jvms, this);
376 Compile* C = kit.C;
377 int nodes = C->unique();
378 #ifndef PRODUCT
379 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
380 char buf[1000];
381 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
382 tty->print_cr("Intrinsic %s", str);
383 }
384 #endif
385 ciMethod* callee = kit.callee();
386 const int bci = kit.bci();
387
388 // Try to inline the intrinsic.
389 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
390 kit.try_to_inline(_last_predicate)) {
391 if (C->print_intrinsics() || C->print_inlining()) {
392 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
393 }
394 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
395 if (C->log()) {
396 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
397 vmIntrinsics::name_at(intrinsic_id()),
398 (is_virtual() ? " virtual='1'" : ""),
399 C->unique() - nodes);
400 }
401 // Push the result from the inlined method onto the stack.
402 kit.push_result();
403 C->print_inlining_update(this);
404 return kit.transfer_exceptions_into_jvms();
405 }
406
407 // The intrinsic bailed out
408 if (C->print_intrinsics() || C->print_inlining()) {
409 if (jvms->has_method()) {
410 // Not a root compile.
411 const char* msg;
412 if (callee->intrinsic_candidate()) {
413 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
414 } else {
415 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
416 : "failed to inline (intrinsic), method not annotated";
417 }
418 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
419 } else {
420 // Root compile
421 tty->print("Did not generate intrinsic %s%s at bci:%d in",
422 vmIntrinsics::name_at(intrinsic_id()),
423 (is_virtual() ? " (virtual)" : ""), bci);
424 }
425 }
426 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
427 C->print_inlining_update(this);
428 return NULL;
429 }
430
431 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
432 LibraryCallKit kit(jvms, this);
433 Compile* C = kit.C;
434 int nodes = C->unique();
435 _last_predicate = predicate;
436 #ifndef PRODUCT
437 assert(is_predicated() && predicate < predicates_count(), "sanity");
438 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
439 char buf[1000];
440 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
441 tty->print_cr("Predicate for intrinsic %s", str);
442 }
443 #endif
444 ciMethod* callee = kit.callee();
445 const int bci = kit.bci();
446
447 Node* slow_ctl = kit.try_to_predicate(predicate);
448 if (!kit.failing()) {
449 if (C->print_intrinsics() || C->print_inlining()) {
450 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
451 }
452 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
453 if (C->log()) {
454 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
455 vmIntrinsics::name_at(intrinsic_id()),
456 (is_virtual() ? " virtual='1'" : ""),
457 C->unique() - nodes);
458 }
459 return slow_ctl; // Could be NULL if the check folds.
460 }
461
462 // The intrinsic bailed out
463 if (C->print_intrinsics() || C->print_inlining()) {
464 if (jvms->has_method()) {
465 // Not a root compile.
466 const char* msg = "failed to generate predicate for intrinsic";
467 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
468 } else {
469 // Root compile
470 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
471 vmIntrinsics::name_at(intrinsic_id()),
472 (is_virtual() ? " (virtual)" : ""), bci);
473 }
474 }
475 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
476 return NULL;
477 }
478
479 bool LibraryCallKit::try_to_inline(int predicate) {
480 // Handle symbolic names for otherwise undistinguished boolean switches:
481 const bool is_store = true;
482 const bool is_compress = true;
483 const bool is_static = true;
484 const bool is_volatile = true;
485
486 if (!jvms()->has_method()) {
487 // Root JVMState has a null method.
488 assert(map()->memory()->Opcode() == Op_Parm, "");
489 // Insert the memory aliasing node
490 set_all_memory(reset_memory());
491 }
492 assert(merged_memory(), "");
493
494
495 switch (intrinsic_id()) {
496 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
497 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
498 case vmIntrinsics::_getClass: return inline_native_getClass();
499
500 case vmIntrinsics::_dsin:
501 case vmIntrinsics::_dcos:
502 case vmIntrinsics::_dtan:
503 case vmIntrinsics::_dabs:
504 case vmIntrinsics::_datan2:
505 case vmIntrinsics::_dsqrt:
506 case vmIntrinsics::_dexp:
507 case vmIntrinsics::_dlog:
508 case vmIntrinsics::_dlog10:
509 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
510
511 case vmIntrinsics::_min:
512 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
513
514 case vmIntrinsics::_notify:
515 case vmIntrinsics::_notifyAll:
516 if (InlineNotify) {
517 return inline_notify(intrinsic_id());
518 }
519 return false;
520
521 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
522 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
523 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
524 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
525 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
526 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
527 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
528 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
529 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
530 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
531 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
532 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
533
534 case vmIntrinsics::_arraycopy: return inline_arraycopy();
535
536 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
537 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
538 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
539 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
540
541 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
542 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
543 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
544 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
545 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
546 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
547 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
548
549 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
550 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
551
552 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
553 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
554 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
555 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
556
557 case vmIntrinsics::_compressStringC:
558 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
559 case vmIntrinsics::_inflateStringC:
560 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
561
562 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
563 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
564 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
565 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
566 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
567 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
568 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
569 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
570 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
571
572 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
573 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
574 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
575 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
576 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
577 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
578 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
579 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
580 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
581
582 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
583 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
584 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
585 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
586 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
587 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
588 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
589 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
590 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
591
592 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
593 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
594 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
595 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
596 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
597 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
598 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
599 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
600 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
601
602 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
603 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
604 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
605 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
606
607 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
608 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
609 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
610 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
611
612 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
613 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
614 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
615 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
616 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
617 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
618 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
619 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
620 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
621
622 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
623 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
624 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
625 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
626 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
627 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
628 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
629 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
630 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
631
632 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
633 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
634 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
635 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
636 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
637 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
638 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
639 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
640 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
641
642 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
643 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
644 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
645 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
646 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
647 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
648 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
649 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
650 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
651
652 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
653 case vmIntrinsics::_compareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
654 case vmIntrinsics::_compareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
655 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
656 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
657
658 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
659 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
660 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
661 case vmIntrinsics::_weakCompareAndSwapObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
662 case vmIntrinsics::_weakCompareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
663 case vmIntrinsics::_weakCompareAndSwapByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
664 case vmIntrinsics::_weakCompareAndSwapByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
665 case vmIntrinsics::_weakCompareAndSwapByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
666 case vmIntrinsics::_weakCompareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
667 case vmIntrinsics::_weakCompareAndSwapShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
668 case vmIntrinsics::_weakCompareAndSwapShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
669 case vmIntrinsics::_weakCompareAndSwapShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
670 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
671 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
672 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
673 case vmIntrinsics::_weakCompareAndSwapIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
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 case vmIntrinsics::_weakCompareAndSwapLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
678
679 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
680 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
681 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
682 case vmIntrinsics::_compareAndExchangeByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
683 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
684 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
685 case vmIntrinsics::_compareAndExchangeShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
686 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
687 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
688 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
689 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
690 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
691 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
692 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
693 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
694
695 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
696 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
697 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
698 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
699
700 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
701 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
702 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
703 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
704 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
705
706 case vmIntrinsics::_loadFence:
707 case vmIntrinsics::_storeFence:
708 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
709
710 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
711
712 case vmIntrinsics::_currentThread: return inline_native_currentThread();
713 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
714
715 #ifdef TRACE_HAVE_INTRINSICS
716 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
717 case vmIntrinsics::_getClassId: return inline_native_classID();
718 case vmIntrinsics::_getBufferWriter: return inline_native_getBufferWriter();
719 #endif
720 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
721 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
722 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
723 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
724 case vmIntrinsics::_getLength: return inline_native_getLength();
725 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
726 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
727 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
728 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
729 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
730 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
731
732 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
733 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
734
735 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
736
737 case vmIntrinsics::_isInstance:
738 case vmIntrinsics::_getModifiers:
739 case vmIntrinsics::_isInterface:
740 case vmIntrinsics::_isArray:
741 case vmIntrinsics::_isPrimitive:
742 case vmIntrinsics::_getSuperclass:
743 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
744
745 case vmIntrinsics::_floatToRawIntBits:
746 case vmIntrinsics::_floatToIntBits:
747 case vmIntrinsics::_intBitsToFloat:
748 case vmIntrinsics::_doubleToRawLongBits:
749 case vmIntrinsics::_doubleToLongBits:
750 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
751
752 case vmIntrinsics::_numberOfLeadingZeros_i:
753 case vmIntrinsics::_numberOfLeadingZeros_l:
754 case vmIntrinsics::_numberOfTrailingZeros_i:
755 case vmIntrinsics::_numberOfTrailingZeros_l:
756 case vmIntrinsics::_bitCount_i:
757 case vmIntrinsics::_bitCount_l:
758 case vmIntrinsics::_reverseBytes_i:
759 case vmIntrinsics::_reverseBytes_l:
760 case vmIntrinsics::_reverseBytes_s:
761 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
762
763 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
764
765 case vmIntrinsics::_Reference_get: return inline_reference_get();
766
767 case vmIntrinsics::_Class_cast: return inline_Class_cast();
768
769 case vmIntrinsics::_aescrypt_encryptBlock:
770 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
771
772 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
773 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
774 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
775
776 case vmIntrinsics::_counterMode_AESCrypt:
777 return inline_counterMode_AESCrypt(intrinsic_id());
778
779 case vmIntrinsics::_sha_implCompress:
780 case vmIntrinsics::_sha2_implCompress:
781 case vmIntrinsics::_sha5_implCompress:
782 return inline_sha_implCompress(intrinsic_id());
783
784 case vmIntrinsics::_digestBase_implCompressMB:
785 return inline_digestBase_implCompressMB(predicate);
786
787 case vmIntrinsics::_multiplyToLen:
788 return inline_multiplyToLen();
789
790 case vmIntrinsics::_squareToLen:
791 return inline_squareToLen();
792
793 case vmIntrinsics::_mulAdd:
794 return inline_mulAdd();
795
796 case vmIntrinsics::_montgomeryMultiply:
797 return inline_montgomeryMultiply();
798 case vmIntrinsics::_montgomerySquare:
799 return inline_montgomerySquare();
800
801 case vmIntrinsics::_vectorizedMismatch:
802 return inline_vectorizedMismatch();
803
804 case vmIntrinsics::_ghash_processBlocks:
805 return inline_ghash_processBlocks();
806
807 case vmIntrinsics::_encodeISOArray:
808 case vmIntrinsics::_encodeByteISOArray:
809 return inline_encodeISOArray();
810
811 case vmIntrinsics::_updateCRC32:
812 return inline_updateCRC32();
813 case vmIntrinsics::_updateBytesCRC32:
814 return inline_updateBytesCRC32();
815 case vmIntrinsics::_updateByteBufferCRC32:
816 return inline_updateByteBufferCRC32();
817
818 case vmIntrinsics::_updateBytesCRC32C:
819 return inline_updateBytesCRC32C();
820 case vmIntrinsics::_updateDirectByteBufferCRC32C:
821 return inline_updateDirectByteBufferCRC32C();
822
823 case vmIntrinsics::_updateBytesAdler32:
824 return inline_updateBytesAdler32();
825 case vmIntrinsics::_updateByteBufferAdler32:
826 return inline_updateByteBufferAdler32();
827
828 case vmIntrinsics::_profileBoolean:
829 return inline_profileBoolean();
830 case vmIntrinsics::_isCompileConstant:
831 return inline_isCompileConstant();
832
833 case vmIntrinsics::_hasNegatives:
834 return inline_hasNegatives();
835
836 case vmIntrinsics::_fmaD:
837 case vmIntrinsics::_fmaF:
838 return inline_fma(intrinsic_id());
839
840 default:
841 // If you get here, it may be that someone has added a new intrinsic
842 // to the list in vmSymbols.hpp without implementing it here.
843 #ifndef PRODUCT
844 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
845 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
846 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
847 }
848 #endif
849 return false;
850 }
851 }
852
853 Node* LibraryCallKit::try_to_predicate(int predicate) {
854 if (!jvms()->has_method()) {
855 // Root JVMState has a null method.
856 assert(map()->memory()->Opcode() == Op_Parm, "");
857 // Insert the memory aliasing node
858 set_all_memory(reset_memory());
859 }
860 assert(merged_memory(), "");
861
862 switch (intrinsic_id()) {
863 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
864 return inline_cipherBlockChaining_AESCrypt_predicate(false);
865 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
866 return inline_cipherBlockChaining_AESCrypt_predicate(true);
867 case vmIntrinsics::_counterMode_AESCrypt:
868 return inline_counterMode_AESCrypt_predicate();
869 case vmIntrinsics::_digestBase_implCompressMB:
870 return inline_digestBase_implCompressMB_predicate(predicate);
871
872 default:
873 // If you get here, it may be that someone has added a new intrinsic
874 // to the list in vmSymbols.hpp without implementing it here.
875 #ifndef PRODUCT
876 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
877 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
878 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
879 }
880 #endif
881 Node* slow_ctl = control();
882 set_control(top()); // No fast path instrinsic
883 return slow_ctl;
884 }
885 }
886
887 //------------------------------set_result-------------------------------
888 // Helper function for finishing intrinsics.
889 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
890 record_for_igvn(region);
891 set_control(_gvn.transform(region));
892 set_result( _gvn.transform(value));
893 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
894 }
895
896 //------------------------------generate_guard---------------------------
897 // Helper function for generating guarded fast-slow graph structures.
898 // The given 'test', if true, guards a slow path. If the test fails
899 // then a fast path can be taken. (We generally hope it fails.)
900 // In all cases, GraphKit::control() is updated to the fast path.
901 // The returned value represents the control for the slow path.
902 // The return value is never 'top'; it is either a valid control
903 // or NULL if it is obvious that the slow path can never be taken.
904 // Also, if region and the slow control are not NULL, the slow edge
905 // is appended to the region.
906 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
907 if (stopped()) {
908 // Already short circuited.
909 return NULL;
910 }
911
912 // Build an if node and its projections.
913 // If test is true we take the slow path, which we assume is uncommon.
914 if (_gvn.type(test) == TypeInt::ZERO) {
915 // The slow branch is never taken. No need to build this guard.
916 return NULL;
917 }
918
919 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
920
921 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
922 if (if_slow == top()) {
923 // The slow branch is never taken. No need to build this guard.
924 return NULL;
925 }
926
927 if (region != NULL)
928 region->add_req(if_slow);
929
930 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
931 set_control(if_fast);
932
933 return if_slow;
934 }
935
936 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
937 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
938 }
939 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
940 return generate_guard(test, region, PROB_FAIR);
941 }
942
943 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
944 Node* *pos_index) {
945 if (stopped())
946 return NULL; // already stopped
947 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
948 return NULL; // index is already adequately typed
949 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
950 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
951 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
952 if (is_neg != NULL && pos_index != NULL) {
953 // Emulate effect of Parse::adjust_map_after_if.
954 Node* ccast = new CastIINode(index, TypeInt::POS);
955 ccast->set_req(0, control());
956 (*pos_index) = _gvn.transform(ccast);
957 }
958 return is_neg;
959 }
960
961 // Make sure that 'position' is a valid limit index, in [0..length].
962 // There are two equivalent plans for checking this:
963 // A. (offset + copyLength) unsigned<= arrayLength
964 // B. offset <= (arrayLength - copyLength)
965 // We require that all of the values above, except for the sum and
966 // difference, are already known to be non-negative.
967 // Plan A is robust in the face of overflow, if offset and copyLength
968 // are both hugely positive.
969 //
970 // Plan B is less direct and intuitive, but it does not overflow at
971 // all, since the difference of two non-negatives is always
972 // representable. Whenever Java methods must perform the equivalent
973 // check they generally use Plan B instead of Plan A.
974 // For the moment we use Plan A.
975 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
976 Node* subseq_length,
977 Node* array_length,
978 RegionNode* region) {
979 if (stopped())
980 return NULL; // already stopped
981 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
982 if (zero_offset && subseq_length->eqv_uncast(array_length))
983 return NULL; // common case of whole-array copy
984 Node* last = subseq_length;
985 if (!zero_offset) // last += offset
986 last = _gvn.transform(new AddINode(last, offset));
987 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
988 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
989 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
990 return is_over;
991 }
992
993 // Emit range checks for the given String.value byte array
994 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
995 if (stopped()) {
996 return; // already stopped
997 }
998 RegionNode* bailout = new RegionNode(1);
999 record_for_igvn(bailout);
1000 if (char_count) {
1001 // Convert char count to byte count
1002 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1003 }
1004
1005 // Offset and count must not be negative
1006 generate_negative_guard(offset, bailout);
1007 generate_negative_guard(count, bailout);
1008 // Offset + count must not exceed length of array
1009 generate_limit_guard(offset, count, load_array_length(array), bailout);
1010
1011 if (bailout->req() > 1) {
1012 PreserveJVMState pjvms(this);
1013 set_control(_gvn.transform(bailout));
1014 uncommon_trap(Deoptimization::Reason_intrinsic,
1015 Deoptimization::Action_maybe_recompile);
1016 }
1017 }
1018
1019 //--------------------------generate_current_thread--------------------
1020 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1021 ciKlass* thread_klass = env()->Thread_klass();
1022 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1023 Node* thread = _gvn.transform(new ThreadLocalNode());
1024 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1025 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1026 tls_output = thread;
1027 return threadObj;
1028 }
1029
1030
1031 //------------------------------make_string_method_node------------------------
1032 // Helper method for String intrinsic functions. This version is called with
1033 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1034 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1035 // containing the lengths of str1 and str2.
1036 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1037 Node* result = NULL;
1038 switch (opcode) {
1039 case Op_StrIndexOf:
1040 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1041 str1_start, cnt1, str2_start, cnt2, ae);
1042 break;
1043 case Op_StrComp:
1044 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1045 str1_start, cnt1, str2_start, cnt2, ae);
1046 break;
1047 case Op_StrEquals:
1048 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1049 // Use the constant length if there is one because optimized match rule may exist.
1050 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1051 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1052 break;
1053 default:
1054 ShouldNotReachHere();
1055 return NULL;
1056 }
1057
1058 // All these intrinsics have checks.
1059 C->set_has_split_ifs(true); // Has chance for split-if optimization
1060
1061 return _gvn.transform(result);
1062 }
1063
1064 //------------------------------inline_string_compareTo------------------------
1065 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1066 Node* arg1 = argument(0);
1067 Node* arg2 = argument(1);
1068
1069 // Get start addr and length of first argument
1070 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1071 Node* arg1_cnt = load_array_length(arg1);
1072
1073 // Get start addr and length of second argument
1074 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1075 Node* arg2_cnt = load_array_length(arg2);
1076
1077 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1078 set_result(result);
1079 return true;
1080 }
1081
1082 //------------------------------inline_string_equals------------------------
1083 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1084 Node* arg1 = argument(0);
1085 Node* arg2 = argument(1);
1086
1087 // paths (plus control) merge
1088 RegionNode* region = new RegionNode(3);
1089 Node* phi = new PhiNode(region, TypeInt::BOOL);
1090
1091 if (!stopped()) {
1092 // Get start addr and length of first argument
1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1094 Node* arg1_cnt = load_array_length(arg1);
1095
1096 // Get start addr and length of second argument
1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1098 Node* arg2_cnt = load_array_length(arg2);
1099
1100 // Check for arg1_cnt != arg2_cnt
1101 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1102 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1103 Node* if_ne = generate_slow_guard(bol, NULL);
1104 if (if_ne != NULL) {
1105 phi->init_req(2, intcon(0));
1106 region->init_req(2, if_ne);
1107 }
1108
1109 // Check for count == 0 is done by assembler code for StrEquals.
1110
1111 if (!stopped()) {
1112 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1113 phi->init_req(1, equals);
1114 region->init_req(1, control());
1115 }
1116 }
1117
1118 // post merge
1119 set_control(_gvn.transform(region));
1120 record_for_igvn(region);
1121
1122 set_result(_gvn.transform(phi));
1123 return true;
1124 }
1125
1126 //------------------------------inline_array_equals----------------------------
1127 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1128 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1129 Node* arg1 = argument(0);
1130 Node* arg2 = argument(1);
1131
1132 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1133 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1134 return true;
1135 }
1136
1137 //------------------------------inline_hasNegatives------------------------------
1138 bool LibraryCallKit::inline_hasNegatives() {
1139 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1140 return false;
1141 }
1142
1143 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1144 // no receiver since it is static method
1145 Node* ba = argument(0);
1146 Node* offset = argument(1);
1147 Node* len = argument(2);
1148
1149 // Range checks
1150 generate_string_range_check(ba, offset, len, false);
1151 if (stopped()) {
1152 return true;
1153 }
1154 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1155 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1156 set_result(_gvn.transform(result));
1157 return true;
1158 }
1159
1160 bool LibraryCallKit::inline_preconditions_checkIndex() {
1161 Node* index = argument(0);
1162 Node* length = argument(1);
1163 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1164 return false;
1165 }
1166
1167 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1168 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1169
1170 {
1171 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1172 uncommon_trap(Deoptimization::Reason_intrinsic,
1173 Deoptimization::Action_make_not_entrant);
1174 }
1175
1176 if (stopped()) {
1177 return false;
1178 }
1179
1180 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1181 BoolTest::mask btest = BoolTest::lt;
1182 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1183 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1184 _gvn.set_type(rc, rc->Value(&_gvn));
1185 if (!rc_bool->is_Con()) {
1186 record_for_igvn(rc);
1187 }
1188 set_control(_gvn.transform(new IfTrueNode(rc)));
1189 {
1190 PreserveJVMState pjvms(this);
1191 set_control(_gvn.transform(new IfFalseNode(rc)));
1192 uncommon_trap(Deoptimization::Reason_range_check,
1193 Deoptimization::Action_make_not_entrant);
1194 }
1195
1196 if (stopped()) {
1197 return false;
1198 }
1199
1200 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1201 result->set_req(0, control());
1202 result = _gvn.transform(result);
1203 set_result(result);
1204 replace_in_map(index, result);
1205 return true;
1206 }
1207
1208 //------------------------------inline_string_indexOf------------------------
1209 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1210 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1211 return false;
1212 }
1213 Node* src = argument(0);
1214 Node* tgt = argument(1);
1215
1216 // Make the merge point
1217 RegionNode* result_rgn = new RegionNode(4);
1218 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1219
1220 // Get start addr and length of source string
1221 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1222 Node* src_count = load_array_length(src);
1223
1224 // Get start addr and length of substring
1225 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1226 Node* tgt_count = load_array_length(tgt);
1227
1228 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1229 // Divide src size by 2 if String is UTF16 encoded
1230 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1231 }
1232 if (ae == StrIntrinsicNode::UU) {
1233 // Divide substring size by 2 if String is UTF16 encoded
1234 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1235 }
1236
1237 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1238 if (result != NULL) {
1239 result_phi->init_req(3, result);
1240 result_rgn->init_req(3, control());
1241 }
1242 set_control(_gvn.transform(result_rgn));
1243 record_for_igvn(result_rgn);
1244 set_result(_gvn.transform(result_phi));
1245
1246 return true;
1247 }
1248
1249 //-----------------------------inline_string_indexOf-----------------------
1250 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1251 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1252 return false;
1253 }
1254 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1255 return false;
1256 }
1257 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1258 Node* src = argument(0); // byte[]
1259 Node* src_count = argument(1); // char count
1260 Node* tgt = argument(2); // byte[]
1261 Node* tgt_count = argument(3); // char count
1262 Node* from_index = argument(4); // char index
1263
1264 // Multiply byte array index by 2 if String is UTF16 encoded
1265 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1266 src_count = _gvn.transform(new SubINode(src_count, from_index));
1267 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1268 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1269
1270 // Range checks
1271 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1272 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1273 if (stopped()) {
1274 return true;
1275 }
1276
1277 RegionNode* region = new RegionNode(5);
1278 Node* phi = new PhiNode(region, TypeInt::INT);
1279
1280 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1281 if (result != NULL) {
1282 // The result is index relative to from_index if substring was found, -1 otherwise.
1283 // Generate code which will fold into cmove.
1284 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1285 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1286
1287 Node* if_lt = generate_slow_guard(bol, NULL);
1288 if (if_lt != NULL) {
1289 // result == -1
1290 phi->init_req(3, result);
1291 region->init_req(3, if_lt);
1292 }
1293 if (!stopped()) {
1294 result = _gvn.transform(new AddINode(result, from_index));
1295 phi->init_req(4, result);
1296 region->init_req(4, control());
1297 }
1298 }
1299
1300 set_control(_gvn.transform(region));
1301 record_for_igvn(region);
1302 set_result(_gvn.transform(phi));
1303
1304 return true;
1305 }
1306
1307 // Create StrIndexOfNode with fast path checks
1308 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1309 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1310 // Check for substr count > string count
1311 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1312 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1313 Node* if_gt = generate_slow_guard(bol, NULL);
1314 if (if_gt != NULL) {
1315 phi->init_req(1, intcon(-1));
1316 region->init_req(1, if_gt);
1317 }
1318 if (!stopped()) {
1319 // Check for substr count == 0
1320 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1321 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1322 Node* if_zero = generate_slow_guard(bol, NULL);
1323 if (if_zero != NULL) {
1324 phi->init_req(2, intcon(0));
1325 region->init_req(2, if_zero);
1326 }
1327 }
1328 if (!stopped()) {
1329 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1330 }
1331 return NULL;
1332 }
1333
1334 //-----------------------------inline_string_indexOfChar-----------------------
1335 bool LibraryCallKit::inline_string_indexOfChar() {
1336 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1337 return false;
1338 }
1339 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1340 return false;
1341 }
1342 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1343 Node* src = argument(0); // byte[]
1344 Node* tgt = argument(1); // tgt is int ch
1345 Node* from_index = argument(2);
1346 Node* max = argument(3);
1347
1348 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1349 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1350 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1351
1352 // Range checks
1353 generate_string_range_check(src, src_offset, src_count, true);
1354 if (stopped()) {
1355 return true;
1356 }
1357
1358 RegionNode* region = new RegionNode(3);
1359 Node* phi = new PhiNode(region, TypeInt::INT);
1360
1361 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1362 C->set_has_split_ifs(true); // Has chance for split-if optimization
1363 _gvn.transform(result);
1364
1365 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1366 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1367
1368 Node* if_lt = generate_slow_guard(bol, NULL);
1369 if (if_lt != NULL) {
1370 // result == -1
1371 phi->init_req(2, result);
1372 region->init_req(2, if_lt);
1373 }
1374 if (!stopped()) {
1375 result = _gvn.transform(new AddINode(result, from_index));
1376 phi->init_req(1, result);
1377 region->init_req(1, control());
1378 }
1379 set_control(_gvn.transform(region));
1380 record_for_igvn(region);
1381 set_result(_gvn.transform(phi));
1382
1383 return true;
1384 }
1385 //---------------------------inline_string_copy---------------------
1386 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1387 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1388 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1389 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1390 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1391 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1392 bool LibraryCallKit::inline_string_copy(bool compress) {
1393 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1394 return false;
1395 }
1396 int nargs = 5; // 2 oops, 3 ints
1397 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1398
1399 Node* src = argument(0);
1400 Node* src_offset = argument(1);
1401 Node* dst = argument(2);
1402 Node* dst_offset = argument(3);
1403 Node* length = argument(4);
1404
1405 // Check for allocation before we add nodes that would confuse
1406 // tightly_coupled_allocation()
1407 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1408
1409 // Figure out the size and type of the elements we will be copying.
1410 const Type* src_type = src->Value(&_gvn);
1411 const Type* dst_type = dst->Value(&_gvn);
1412 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1413 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1414 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1415 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1416 "Unsupported array types for inline_string_copy");
1417
1418 // Convert char[] offsets to byte[] offsets
1419 bool convert_src = (compress && src_elem == T_BYTE);
1420 bool convert_dst = (!compress && dst_elem == T_BYTE);
1421 if (convert_src) {
1422 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1423 } else if (convert_dst) {
1424 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1425 }
1426
1427 // Range checks
1428 generate_string_range_check(src, src_offset, length, convert_src);
1429 generate_string_range_check(dst, dst_offset, length, convert_dst);
1430 if (stopped()) {
1431 return true;
1432 }
1433
1434 Node* src_start = array_element_address(src, src_offset, src_elem);
1435 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1436 // 'src_start' points to src array + scaled offset
1437 // 'dst_start' points to dst array + scaled offset
1438 Node* count = NULL;
1439 if (compress) {
1440 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1441 } else {
1442 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1443 }
1444
1445 if (alloc != NULL) {
1446 if (alloc->maybe_set_complete(&_gvn)) {
1447 // "You break it, you buy it."
1448 InitializeNode* init = alloc->initialization();
1449 assert(init->is_complete(), "we just did this");
1450 init->set_complete_with_arraycopy();
1451 assert(dst->is_CheckCastPP(), "sanity");
1452 assert(dst->in(0)->in(0) == init, "dest pinned");
1453 }
1454 // Do not let stores that initialize this object be reordered with
1455 // a subsequent store that would make this object accessible by
1456 // other threads.
1457 // Record what AllocateNode this StoreStore protects so that
1458 // escape analysis can go from the MemBarStoreStoreNode to the
1459 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1460 // based on the escape status of the AllocateNode.
1461 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1462 }
1463 if (compress) {
1464 set_result(_gvn.transform(count));
1465 }
1466 return true;
1467 }
1468
1469 #ifdef _LP64
1470 #define XTOP ,top() /*additional argument*/
1471 #else //_LP64
1472 #define XTOP /*no additional argument*/
1473 #endif //_LP64
1474
1475 //------------------------inline_string_toBytesU--------------------------
1476 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1477 bool LibraryCallKit::inline_string_toBytesU() {
1478 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1479 return false;
1480 }
1481 // Get the arguments.
1482 Node* value = argument(0);
1483 Node* offset = argument(1);
1484 Node* length = argument(2);
1485
1486 Node* newcopy = NULL;
1487
1488 // Set the original stack and the reexecute bit for the interpreter to reexecute
1489 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1490 { PreserveReexecuteState preexecs(this);
1491 jvms()->set_should_reexecute(true);
1492
1493 // Check if a null path was taken unconditionally.
1494 value = null_check(value);
1495
1496 RegionNode* bailout = new RegionNode(1);
1497 record_for_igvn(bailout);
1498
1499 // Range checks
1500 generate_negative_guard(offset, bailout);
1501 generate_negative_guard(length, bailout);
1502 generate_limit_guard(offset, length, load_array_length(value), bailout);
1503 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1504 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1505
1506 if (bailout->req() > 1) {
1507 PreserveJVMState pjvms(this);
1508 set_control(_gvn.transform(bailout));
1509 uncommon_trap(Deoptimization::Reason_intrinsic,
1510 Deoptimization::Action_maybe_recompile);
1511 }
1512 if (stopped()) {
1513 return true;
1514 }
1515
1516 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1517 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1518 newcopy = new_array(klass_node, size, 0); // no arguments to push
1519 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1520
1521 // Calculate starting addresses.
1522 Node* src_start = array_element_address(value, offset, T_CHAR);
1523 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1524
1525 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1526 const TypeInt* toffset = gvn().type(offset)->is_int();
1527 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1528
1529 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1530 const char* copyfunc_name = "arraycopy";
1531 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1532 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1533 OptoRuntime::fast_arraycopy_Type(),
1534 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1535 src_start, dst_start, ConvI2X(length) XTOP);
1536 // Do not let reads from the cloned object float above the arraycopy.
1537 if (alloc != NULL) {
1538 if (alloc->maybe_set_complete(&_gvn)) {
1539 // "You break it, you buy it."
1540 InitializeNode* init = alloc->initialization();
1541 assert(init->is_complete(), "we just did this");
1542 init->set_complete_with_arraycopy();
1543 assert(newcopy->is_CheckCastPP(), "sanity");
1544 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1545 }
1546 // Do not let stores that initialize this object be reordered with
1547 // a subsequent store that would make this object accessible by
1548 // other threads.
1549 // Record what AllocateNode this StoreStore protects so that
1550 // escape analysis can go from the MemBarStoreStoreNode to the
1551 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1552 // based on the escape status of the AllocateNode.
1553 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1554 } else {
1555 insert_mem_bar(Op_MemBarCPUOrder);
1556 }
1557 } // original reexecute is set back here
1558
1559 C->set_has_split_ifs(true); // Has chance for split-if optimization
1560 if (!stopped()) {
1561 set_result(newcopy);
1562 }
1563 return true;
1564 }
1565
1566 //------------------------inline_string_getCharsU--------------------------
1567 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1568 bool LibraryCallKit::inline_string_getCharsU() {
1569 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1570 return false;
1571 }
1572
1573 // Get the arguments.
1574 Node* src = argument(0);
1575 Node* src_begin = argument(1);
1576 Node* src_end = argument(2); // exclusive offset (i < src_end)
1577 Node* dst = argument(3);
1578 Node* dst_begin = argument(4);
1579
1580 // Check for allocation before we add nodes that would confuse
1581 // tightly_coupled_allocation()
1582 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1583
1584 // Check if a null path was taken unconditionally.
1585 src = null_check(src);
1586 dst = null_check(dst);
1587 if (stopped()) {
1588 return true;
1589 }
1590
1591 // Get length and convert char[] offset to byte[] offset
1592 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1593 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1594
1595 // Range checks
1596 generate_string_range_check(src, src_begin, length, true);
1597 generate_string_range_check(dst, dst_begin, length, false);
1598 if (stopped()) {
1599 return true;
1600 }
1601
1602 if (!stopped()) {
1603 // Calculate starting addresses.
1604 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1605 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1606
1607 // Check if array addresses are aligned to HeapWordSize
1608 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1609 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1610 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1611 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1612
1613 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1614 const char* copyfunc_name = "arraycopy";
1615 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1616 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1617 OptoRuntime::fast_arraycopy_Type(),
1618 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1619 src_start, dst_start, ConvI2X(length) XTOP);
1620 // Do not let reads from the cloned object float above the arraycopy.
1621 if (alloc != NULL) {
1622 if (alloc->maybe_set_complete(&_gvn)) {
1623 // "You break it, you buy it."
1624 InitializeNode* init = alloc->initialization();
1625 assert(init->is_complete(), "we just did this");
1626 init->set_complete_with_arraycopy();
1627 assert(dst->is_CheckCastPP(), "sanity");
1628 assert(dst->in(0)->in(0) == init, "dest pinned");
1629 }
1630 // Do not let stores that initialize this object be reordered with
1631 // a subsequent store that would make this object accessible by
1632 // other threads.
1633 // Record what AllocateNode this StoreStore protects so that
1634 // escape analysis can go from the MemBarStoreStoreNode to the
1635 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1636 // based on the escape status of the AllocateNode.
1637 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1638 } else {
1639 insert_mem_bar(Op_MemBarCPUOrder);
1640 }
1641 }
1642
1643 C->set_has_split_ifs(true); // Has chance for split-if optimization
1644 return true;
1645 }
1646
1647 //----------------------inline_string_char_access----------------------------
1648 // Store/Load char to/from byte[] array.
1649 // static void StringUTF16.putChar(byte[] val, int index, int c)
1650 // static char StringUTF16.getChar(byte[] val, int index)
1651 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1652 Node* value = argument(0);
1653 Node* index = argument(1);
1654 Node* ch = is_store ? argument(2) : NULL;
1655
1656 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1657 // correctly requires matched array shapes.
1658 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1659 "sanity: byte[] and char[] bases agree");
1660 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1661 "sanity: byte[] and char[] scales agree");
1662
1663 // Bail when getChar over constants is requested: constant folding would
1664 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1665 // Java method would constant fold nicely instead.
1666 if (!is_store && value->is_Con() && index->is_Con()) {
1667 return false;
1668 }
1669
1670 Node* adr = array_element_address(value, index, T_CHAR);
1671 if (adr->is_top()) {
1672 return false;
1673 }
1674 if (is_store) {
1675 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1676 false, false, true /* mismatched */);
1677 } else {
1678 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1679 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1680 set_result(ch);
1681 }
1682 return true;
1683 }
1684
1685 //--------------------------round_double_node--------------------------------
1686 // Round a double node if necessary.
1687 Node* LibraryCallKit::round_double_node(Node* n) {
1688 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1689 n = _gvn.transform(new RoundDoubleNode(0, n));
1690 return n;
1691 }
1692
1693 //------------------------------inline_math-----------------------------------
1694 // public static double Math.abs(double)
1695 // public static double Math.sqrt(double)
1696 // public static double Math.log(double)
1697 // public static double Math.log10(double)
1698 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1699 Node* arg = round_double_node(argument(0));
1700 Node* n = NULL;
1701 switch (id) {
1702 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1703 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1704 default: fatal_unexpected_iid(id); break;
1705 }
1706 set_result(_gvn.transform(n));
1707 return true;
1708 }
1709
1710 //------------------------------runtime_math-----------------------------
1711 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1712 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1713 "must be (DD)D or (D)D type");
1714
1715 // Inputs
1716 Node* a = round_double_node(argument(0));
1717 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1718
1719 const TypePtr* no_memory_effects = NULL;
1720 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1721 no_memory_effects,
1722 a, top(), b, b ? top() : NULL);
1723 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1724 #ifdef ASSERT
1725 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1726 assert(value_top == top(), "second value must be top");
1727 #endif
1728
1729 set_result(value);
1730 return true;
1731 }
1732
1733 //------------------------------inline_math_native-----------------------------
1734 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1735 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1736 switch (id) {
1737 // These intrinsics are not properly supported on all hardware
1738 case vmIntrinsics::_dsin:
1739 return StubRoutines::dsin() != NULL ?
1740 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1741 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1742 case vmIntrinsics::_dcos:
1743 return StubRoutines::dcos() != NULL ?
1744 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1745 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1746 case vmIntrinsics::_dtan:
1747 return StubRoutines::dtan() != NULL ?
1748 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1749 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1750 case vmIntrinsics::_dlog:
1751 return StubRoutines::dlog() != NULL ?
1752 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1753 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1754 case vmIntrinsics::_dlog10:
1755 return StubRoutines::dlog10() != NULL ?
1756 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1757 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1758
1759 // These intrinsics are supported on all hardware
1760 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1761 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1762
1763 case vmIntrinsics::_dexp:
1764 return StubRoutines::dexp() != NULL ?
1765 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1766 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1767 case vmIntrinsics::_dpow:
1768 return StubRoutines::dpow() != NULL ?
1769 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1770 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1771 #undef FN_PTR
1772
1773 // These intrinsics are not yet correctly implemented
1774 case vmIntrinsics::_datan2:
1775 return false;
1776
1777 default:
1778 fatal_unexpected_iid(id);
1779 return false;
1780 }
1781 }
1782
1783 static bool is_simple_name(Node* n) {
1784 return (n->req() == 1 // constant
1785 || (n->is_Type() && n->as_Type()->type()->singleton())
1786 || n->is_Proj() // parameter or return value
1787 || n->is_Phi() // local of some sort
1788 );
1789 }
1790
1791 //----------------------------inline_notify-----------------------------------*
1792 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1793 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1794 address func;
1795 if (id == vmIntrinsics::_notify) {
1796 func = OptoRuntime::monitor_notify_Java();
1797 } else {
1798 func = OptoRuntime::monitor_notifyAll_Java();
1799 }
1800 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1801 make_slow_call_ex(call, env()->Throwable_klass(), false);
1802 return true;
1803 }
1804
1805
1806 //----------------------------inline_min_max-----------------------------------
1807 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1808 set_result(generate_min_max(id, argument(0), argument(1)));
1809 return true;
1810 }
1811
1812 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1813 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1814 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1815 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1816 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1817
1818 {
1819 PreserveJVMState pjvms(this);
1820 PreserveReexecuteState preexecs(this);
1821 jvms()->set_should_reexecute(true);
1822
1823 set_control(slow_path);
1824 set_i_o(i_o());
1825
1826 uncommon_trap(Deoptimization::Reason_intrinsic,
1827 Deoptimization::Action_none);
1828 }
1829
1830 set_control(fast_path);
1831 set_result(math);
1832 }
1833
1834 template <typename OverflowOp>
1835 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1836 typedef typename OverflowOp::MathOp MathOp;
1837
1838 MathOp* mathOp = new MathOp(arg1, arg2);
1839 Node* operation = _gvn.transform( mathOp );
1840 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1841 inline_math_mathExact(operation, ofcheck);
1842 return true;
1843 }
1844
1845 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1846 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1847 }
1848
1849 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1850 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1851 }
1852
1853 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1854 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1855 }
1856
1857 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1858 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1859 }
1860
1861 bool LibraryCallKit::inline_math_negateExactI() {
1862 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1863 }
1864
1865 bool LibraryCallKit::inline_math_negateExactL() {
1866 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1867 }
1868
1869 bool LibraryCallKit::inline_math_multiplyExactI() {
1870 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1871 }
1872
1873 bool LibraryCallKit::inline_math_multiplyExactL() {
1874 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1875 }
1876
1877 Node*
1878 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1879 // These are the candidate return value:
1880 Node* xvalue = x0;
1881 Node* yvalue = y0;
1882
1883 if (xvalue == yvalue) {
1884 return xvalue;
1885 }
1886
1887 bool want_max = (id == vmIntrinsics::_max);
1888
1889 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1890 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1891 if (txvalue == NULL || tyvalue == NULL) return top();
1892 // This is not really necessary, but it is consistent with a
1893 // hypothetical MaxINode::Value method:
1894 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1895
1896 // %%% This folding logic should (ideally) be in a different place.
1897 // Some should be inside IfNode, and there to be a more reliable
1898 // transformation of ?: style patterns into cmoves. We also want
1899 // more powerful optimizations around cmove and min/max.
1900
1901 // Try to find a dominating comparison of these guys.
1902 // It can simplify the index computation for Arrays.copyOf
1903 // and similar uses of System.arraycopy.
1904 // First, compute the normalized version of CmpI(x, y).
1905 int cmp_op = Op_CmpI;
1906 Node* xkey = xvalue;
1907 Node* ykey = yvalue;
1908 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1909 if (ideal_cmpxy->is_Cmp()) {
1910 // E.g., if we have CmpI(length - offset, count),
1911 // it might idealize to CmpI(length, count + offset)
1912 cmp_op = ideal_cmpxy->Opcode();
1913 xkey = ideal_cmpxy->in(1);
1914 ykey = ideal_cmpxy->in(2);
1915 }
1916
1917 // Start by locating any relevant comparisons.
1918 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1919 Node* cmpxy = NULL;
1920 Node* cmpyx = NULL;
1921 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1922 Node* cmp = start_from->fast_out(k);
1923 if (cmp->outcnt() > 0 && // must have prior uses
1924 cmp->in(0) == NULL && // must be context-independent
1925 cmp->Opcode() == cmp_op) { // right kind of compare
1926 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1927 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1928 }
1929 }
1930
1931 const int NCMPS = 2;
1932 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1933 int cmpn;
1934 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1935 if (cmps[cmpn] != NULL) break; // find a result
1936 }
1937 if (cmpn < NCMPS) {
1938 // Look for a dominating test that tells us the min and max.
1939 int depth = 0; // Limit search depth for speed
1940 Node* dom = control();
1941 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1942 if (++depth >= 100) break;
1943 Node* ifproj = dom;
1944 if (!ifproj->is_Proj()) continue;
1945 Node* iff = ifproj->in(0);
1946 if (!iff->is_If()) continue;
1947 Node* bol = iff->in(1);
1948 if (!bol->is_Bool()) continue;
1949 Node* cmp = bol->in(1);
1950 if (cmp == NULL) continue;
1951 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1952 if (cmps[cmpn] == cmp) break;
1953 if (cmpn == NCMPS) continue;
1954 BoolTest::mask btest = bol->as_Bool()->_test._test;
1955 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1956 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1957 // At this point, we know that 'x btest y' is true.
1958 switch (btest) {
1959 case BoolTest::eq:
1960 // They are proven equal, so we can collapse the min/max.
1961 // Either value is the answer. Choose the simpler.
1962 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1963 return yvalue;
1964 return xvalue;
1965 case BoolTest::lt: // x < y
1966 case BoolTest::le: // x <= y
1967 return (want_max ? yvalue : xvalue);
1968 case BoolTest::gt: // x > y
1969 case BoolTest::ge: // x >= y
1970 return (want_max ? xvalue : yvalue);
1971 }
1972 }
1973 }
1974
1975 // We failed to find a dominating test.
1976 // Let's pick a test that might GVN with prior tests.
1977 Node* best_bol = NULL;
1978 BoolTest::mask best_btest = BoolTest::illegal;
1979 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1980 Node* cmp = cmps[cmpn];
1981 if (cmp == NULL) continue;
1982 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
1983 Node* bol = cmp->fast_out(j);
1984 if (!bol->is_Bool()) continue;
1985 BoolTest::mask btest = bol->as_Bool()->_test._test;
1986 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
1987 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1988 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
1989 best_bol = bol->as_Bool();
1990 best_btest = btest;
1991 }
1992 }
1993 }
1994
1995 Node* answer_if_true = NULL;
1996 Node* answer_if_false = NULL;
1997 switch (best_btest) {
1998 default:
1999 if (cmpxy == NULL)
2000 cmpxy = ideal_cmpxy;
2001 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2002 // and fall through:
2003 case BoolTest::lt: // x < y
2004 case BoolTest::le: // x <= y
2005 answer_if_true = (want_max ? yvalue : xvalue);
2006 answer_if_false = (want_max ? xvalue : yvalue);
2007 break;
2008 case BoolTest::gt: // x > y
2009 case BoolTest::ge: // x >= y
2010 answer_if_true = (want_max ? xvalue : yvalue);
2011 answer_if_false = (want_max ? yvalue : xvalue);
2012 break;
2013 }
2014
2015 jint hi, lo;
2016 if (want_max) {
2017 // We can sharpen the minimum.
2018 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2019 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2020 } else {
2021 // We can sharpen the maximum.
2022 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2023 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2024 }
2025
2026 // Use a flow-free graph structure, to avoid creating excess control edges
2027 // which could hinder other optimizations.
2028 // Since Math.min/max is often used with arraycopy, we want
2029 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2030 Node* cmov = CMoveNode::make(NULL, best_bol,
2031 answer_if_false, answer_if_true,
2032 TypeInt::make(lo, hi, widen));
2033
2034 return _gvn.transform(cmov);
2035
2036 /*
2037 // This is not as desirable as it may seem, since Min and Max
2038 // nodes do not have a full set of optimizations.
2039 // And they would interfere, anyway, with 'if' optimizations
2040 // and with CMoveI canonical forms.
2041 switch (id) {
2042 case vmIntrinsics::_min:
2043 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2044 case vmIntrinsics::_max:
2045 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2046 default:
2047 ShouldNotReachHere();
2048 }
2049 */
2050 }
2051
2052 inline int
2053 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2054 const TypePtr* base_type = TypePtr::NULL_PTR;
2055 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2056 if (base_type == NULL) {
2057 // Unknown type.
2058 return Type::AnyPtr;
2059 } else if (base_type == TypePtr::NULL_PTR) {
2060 // Since this is a NULL+long form, we have to switch to a rawptr.
2061 base = _gvn.transform(new CastX2PNode(offset));
2062 offset = MakeConX(0);
2063 return Type::RawPtr;
2064 } else if (base_type->base() == Type::RawPtr) {
2065 return Type::RawPtr;
2066 } else if (base_type->isa_oopptr()) {
2067 // Base is never null => always a heap address.
2068 if (base_type->ptr() == TypePtr::NotNull) {
2069 return Type::OopPtr;
2070 }
2071 // Offset is small => always a heap address.
2072 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2073 if (offset_type != NULL &&
2074 base_type->offset() == 0 && // (should always be?)
2075 offset_type->_lo >= 0 &&
2076 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2077 return Type::OopPtr;
2078 }
2079 // Otherwise, it might either be oop+off or NULL+addr.
2080 return Type::AnyPtr;
2081 } else {
2082 // No information:
2083 return Type::AnyPtr;
2084 }
2085 }
2086
2087 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2088 int kind = classify_unsafe_addr(base, offset);
2089 if (kind == Type::RawPtr) {
2090 return basic_plus_adr(top(), base, offset);
2091 } else {
2092 return basic_plus_adr(base, offset);
2093 }
2094 }
2095
2096 //--------------------------inline_number_methods-----------------------------
2097 // inline int Integer.numberOfLeadingZeros(int)
2098 // inline int Long.numberOfLeadingZeros(long)
2099 //
2100 // inline int Integer.numberOfTrailingZeros(int)
2101 // inline int Long.numberOfTrailingZeros(long)
2102 //
2103 // inline int Integer.bitCount(int)
2104 // inline int Long.bitCount(long)
2105 //
2106 // inline char Character.reverseBytes(char)
2107 // inline short Short.reverseBytes(short)
2108 // inline int Integer.reverseBytes(int)
2109 // inline long Long.reverseBytes(long)
2110 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2111 Node* arg = argument(0);
2112 Node* n = NULL;
2113 switch (id) {
2114 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2115 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2116 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2117 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2118 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2119 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2120 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2121 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2122 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2123 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2124 default: fatal_unexpected_iid(id); break;
2125 }
2126 set_result(_gvn.transform(n));
2127 return true;
2128 }
2129
2130 //----------------------------inline_unsafe_access----------------------------
2131
2132 // Helper that guards and inserts a pre-barrier.
2133 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2134 Node* pre_val, bool need_mem_bar) {
2135 // We could be accessing the referent field of a reference object. If so, when G1
2136 // is enabled, we need to log the value in the referent field in an SATB buffer.
2137 // This routine performs some compile time filters and generates suitable
2138 // runtime filters that guard the pre-barrier code.
2139 // Also add memory barrier for non volatile load from the referent field
2140 // to prevent commoning of loads across safepoint.
2141 if (!UseG1GC && !need_mem_bar)
2142 return;
2143
2144 // Some compile time checks.
2145
2146 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2147 const TypeX* otype = offset->find_intptr_t_type();
2148 if (otype != NULL && otype->is_con() &&
2149 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2150 // Constant offset but not the reference_offset so just return
2151 return;
2152 }
2153
2154 // We only need to generate the runtime guards for instances.
2155 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2156 if (btype != NULL) {
2157 if (btype->isa_aryptr()) {
2158 // Array type so nothing to do
2159 return;
2160 }
2161
2162 const TypeInstPtr* itype = btype->isa_instptr();
2163 if (itype != NULL) {
2164 // Can the klass of base_oop be statically determined to be
2165 // _not_ a sub-class of Reference and _not_ Object?
2166 ciKlass* klass = itype->klass();
2167 if ( klass->is_loaded() &&
2168 !klass->is_subtype_of(env()->Reference_klass()) &&
2169 !env()->Object_klass()->is_subtype_of(klass)) {
2170 return;
2171 }
2172 }
2173 }
2174
2175 // The compile time filters did not reject base_oop/offset so
2176 // we need to generate the following runtime filters
2177 //
2178 // if (offset == java_lang_ref_Reference::_reference_offset) {
2179 // if (instance_of(base, java.lang.ref.Reference)) {
2180 // pre_barrier(_, pre_val, ...);
2181 // }
2182 // }
2183
2184 float likely = PROB_LIKELY( 0.999);
2185 float unlikely = PROB_UNLIKELY(0.999);
2186
2187 IdealKit ideal(this);
2188 #define __ ideal.
2189
2190 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2191
2192 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2193 // Update graphKit memory and control from IdealKit.
2194 sync_kit(ideal);
2195
2196 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2197 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2198
2199 // Update IdealKit memory and control from graphKit.
2200 __ sync_kit(this);
2201
2202 Node* one = __ ConI(1);
2203 // is_instof == 0 if base_oop == NULL
2204 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2205
2206 // Update graphKit from IdeakKit.
2207 sync_kit(ideal);
2208
2209 // Use the pre-barrier to record the value in the referent field
2210 pre_barrier(false /* do_load */,
2211 __ ctrl(),
2212 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2213 pre_val /* pre_val */,
2214 T_OBJECT);
2215 if (need_mem_bar) {
2216 // Add memory barrier to prevent commoning reads from this field
2217 // across safepoint since GC can change its value.
2218 insert_mem_bar(Op_MemBarCPUOrder);
2219 }
2220 // Update IdealKit from graphKit.
2221 __ sync_kit(this);
2222
2223 } __ end_if(); // _ref_type != ref_none
2224 } __ end_if(); // offset == referent_offset
2225
2226 // Final sync IdealKit and GraphKit.
2227 final_sync(ideal);
2228 #undef __
2229 }
2230
2231
2232 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2233 // Attempt to infer a sharper value type from the offset and base type.
2234 ciKlass* sharpened_klass = NULL;
2235
2236 // See if it is an instance field, with an object type.
2237 if (alias_type->field() != NULL) {
2238 if (alias_type->field()->type()->is_klass()) {
2239 sharpened_klass = alias_type->field()->type()->as_klass();
2240 }
2241 }
2242
2243 // See if it is a narrow oop array.
2244 if (adr_type->isa_aryptr()) {
2245 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2246 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2247 if (elem_type != NULL) {
2248 sharpened_klass = elem_type->klass();
2249 }
2250 }
2251 }
2252
2253 // The sharpened class might be unloaded if there is no class loader
2254 // contraint in place.
2255 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2256 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2257
2258 #ifndef PRODUCT
2259 if (C->print_intrinsics() || C->print_inlining()) {
2260 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2261 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2262 }
2263 #endif
2264 // Sharpen the value type.
2265 return tjp;
2266 }
2267 return NULL;
2268 }
2269
2270 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2271 if (callee()->is_static()) return false; // caller must have the capability!
2272 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2273 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2274 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2275
2276 #ifndef PRODUCT
2277 {
2278 ResourceMark rm;
2279 // Check the signatures.
2280 ciSignature* sig = callee()->signature();
2281 #ifdef ASSERT
2282 if (!is_store) {
2283 // Object getObject(Object base, int/long offset), etc.
2284 BasicType rtype = sig->return_type()->basic_type();
2285 assert(rtype == type, "getter must return the expected value");
2286 assert(sig->count() == 2, "oop getter has 2 arguments");
2287 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2288 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2289 } else {
2290 // void putObject(Object base, int/long offset, Object x), etc.
2291 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2292 assert(sig->count() == 3, "oop putter has 3 arguments");
2293 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2294 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2295 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2296 assert(vtype == type, "putter must accept the expected value");
2297 }
2298 #endif // ASSERT
2299 }
2300 #endif //PRODUCT
2301
2302 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2303
2304 Node* receiver = argument(0); // type: oop
2305
2306 // Build address expression.
2307 Node* adr;
2308 Node* heap_base_oop = top();
2309 Node* offset = top();
2310 Node* val;
2311
2312 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2313 Node* base = argument(1); // type: oop
2314 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2315 offset = argument(2); // type: long
2316 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2317 // to be plain byte offsets, which are also the same as those accepted
2318 // by oopDesc::field_base.
2319 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2320 "fieldOffset must be byte-scaled");
2321 // 32-bit machines ignore the high half!
2322 offset = ConvL2X(offset);
2323 adr = make_unsafe_address(base, offset);
2324 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2325 heap_base_oop = base;
2326 } else if (type == T_OBJECT) {
2327 return false; // off-heap oop accesses are not supported
2328 }
2329
2330 // Can base be NULL? Otherwise, always on-heap access.
2331 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2332
2333 val = is_store ? argument(4) : NULL;
2334
2335 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2336
2337 // Try to categorize the address.
2338 Compile::AliasType* alias_type = C->alias_type(adr_type);
2339 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2340
2341 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2342 alias_type->adr_type() == TypeAryPtr::RANGE) {
2343 return false; // not supported
2344 }
2345
2346 bool mismatched = false;
2347 BasicType bt = alias_type->basic_type();
2348 if (bt != T_ILLEGAL) {
2349 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2350 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2351 // Alias type doesn't differentiate between byte[] and boolean[]).
2352 // Use address type to get the element type.
2353 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2354 }
2355 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2356 // accessing an array field with getObject is not a mismatch
2357 bt = T_OBJECT;
2358 }
2359 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2360 // Don't intrinsify mismatched object accesses
2361 return false;
2362 }
2363 mismatched = (bt != type);
2364 } else if (alias_type->adr_type()->isa_oopptr()) {
2365 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2366 }
2367
2368 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2369
2370 // First guess at the value type.
2371 const Type *value_type = Type::get_const_basic_type(type);
2372
2373 // We will need memory barriers unless we can determine a unique
2374 // alias category for this reference. (Note: If for some reason
2375 // the barriers get omitted and the unsafe reference begins to "pollute"
2376 // the alias analysis of the rest of the graph, either Compile::can_alias
2377 // or Compile::must_alias will throw a diagnostic assert.)
2378 bool need_mem_bar;
2379 switch (kind) {
2380 case Relaxed:
2381 need_mem_bar = mismatched || can_access_non_heap;
2382 break;
2383 case Opaque:
2384 // Opaque uses CPUOrder membars for protection against code movement.
2385 case Acquire:
2386 case Release:
2387 case Volatile:
2388 need_mem_bar = true;
2389 break;
2390 default:
2391 ShouldNotReachHere();
2392 }
2393
2394 // Some accesses require access atomicity for all types, notably longs and doubles.
2395 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2396 bool requires_atomic_access = false;
2397 switch (kind) {
2398 case Relaxed:
2399 requires_atomic_access = AlwaysAtomicAccesses;
2400 break;
2401 case Opaque:
2402 // Opaque accesses are atomic.
2403 case Acquire:
2404 case Release:
2405 case Volatile:
2406 requires_atomic_access = true;
2407 break;
2408 default:
2409 ShouldNotReachHere();
2410 }
2411
2412 // Figure out the memory ordering.
2413 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2414 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2415
2416 // If we are reading the value of the referent field of a Reference
2417 // object (either by using Unsafe directly or through reflection)
2418 // then, if G1 is enabled, we need to record the referent in an
2419 // SATB log buffer using the pre-barrier mechanism.
2420 // Also we need to add memory barrier to prevent commoning reads
2421 // from this field across safepoint since GC can change its value.
2422 bool need_read_barrier = !is_store &&
2423 offset != top() && heap_base_oop != top();
2424
2425 if (!is_store && type == T_OBJECT) {
2426 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2427 if (tjp != NULL) {
2428 value_type = tjp;
2429 }
2430 }
2431
2432 receiver = null_check(receiver);
2433 if (stopped()) {
2434 return true;
2435 }
2436 // Heap pointers get a null-check from the interpreter,
2437 // as a courtesy. However, this is not guaranteed by Unsafe,
2438 // and it is not possible to fully distinguish unintended nulls
2439 // from intended ones in this API.
2440
2441 // We need to emit leading and trailing CPU membars (see below) in
2442 // addition to memory membars for special access modes. This is a little
2443 // too strong, but avoids the need to insert per-alias-type
2444 // volatile membars (for stores; compare Parse::do_put_xxx), which
2445 // we cannot do effectively here because we probably only have a
2446 // rough approximation of type.
2447
2448 switch(kind) {
2449 case Relaxed:
2450 case Opaque:
2451 case Acquire:
2452 break;
2453 case Release:
2454 case Volatile:
2455 if (is_store) {
2456 insert_mem_bar(Op_MemBarRelease);
2457 } else {
2458 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2459 insert_mem_bar(Op_MemBarVolatile);
2460 }
2461 }
2462 break;
2463 default:
2464 ShouldNotReachHere();
2465 }
2466
2467 // Memory barrier to prevent normal and 'unsafe' accesses from
2468 // bypassing each other. Happens after null checks, so the
2469 // exception paths do not take memory state from the memory barrier,
2470 // so there's no problems making a strong assert about mixing users
2471 // of safe & unsafe memory.
2472 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2473
2474 if (!is_store) {
2475 Node* p = NULL;
2476 // Try to constant fold a load from a constant field
2477 ciField* field = alias_type->field();
2478 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2479 // final or stable field
2480 p = make_constant_from_field(field, heap_base_oop);
2481 }
2482 if (p == NULL) {
2483 // To be valid, unsafe loads may depend on other conditions than
2484 // the one that guards them: pin the Load node
2485 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2486 // load value
2487 switch (type) {
2488 case T_BOOLEAN:
2489 {
2490 // Normalize the value returned by getBoolean in the following cases
2491 if (mismatched ||
2492 heap_base_oop == top() || // - heap_base_oop is NULL or
2493 (can_access_non_heap && alias_type->field() == NULL) // - heap_base_oop is potentially NULL
2494 // and the unsafe access is made to large offset
2495 // (i.e., larger than the maximum offset necessary for any
2496 // field access)
2497 ) {
2498 IdealKit ideal = IdealKit(this);
2499 #define __ ideal.
2500 IdealVariable normalized_result(ideal);
2501 __ declarations_done();
2502 __ set(normalized_result, p);
2503 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2504 __ set(normalized_result, ideal.ConI(1));
2505 ideal.end_if();
2506 final_sync(ideal);
2507 p = __ value(normalized_result);
2508 #undef __
2509 }
2510 }
2511 case T_CHAR:
2512 case T_BYTE:
2513 case T_SHORT:
2514 case T_INT:
2515 case T_LONG:
2516 case T_FLOAT:
2517 case T_DOUBLE:
2518 break;
2519 case T_OBJECT:
2520 if (need_read_barrier) {
2521 // We do not require a mem bar inside pre_barrier if need_mem_bar
2522 // is set: the barriers would be emitted by us.
2523 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2524 }
2525 break;
2526 case T_ADDRESS:
2527 // Cast to an int type.
2528 p = _gvn.transform(new CastP2XNode(NULL, p));
2529 p = ConvX2UL(p);
2530 break;
2531 default:
2532 fatal("unexpected type %d: %s", type, type2name(type));
2533 break;
2534 }
2535 }
2536 // The load node has the control of the preceding MemBarCPUOrder. All
2537 // following nodes will have the control of the MemBarCPUOrder inserted at
2538 // the end of this method. So, pushing the load onto the stack at a later
2539 // point is fine.
2540 set_result(p);
2541 } else {
2542 // place effect of store into memory
2543 switch (type) {
2544 case T_DOUBLE:
2545 val = dstore_rounding(val);
2546 break;
2547 case T_ADDRESS:
2548 // Repackage the long as a pointer.
2549 val = ConvL2X(val);
2550 val = _gvn.transform(new CastX2PNode(val));
2551 break;
2552 }
2553
2554 if (type == T_OBJECT) {
2555 store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2556 } else {
2557 store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2558 }
2559 }
2560
2561 switch(kind) {
2562 case Relaxed:
2563 case Opaque:
2564 case Release:
2565 break;
2566 case Acquire:
2567 case Volatile:
2568 if (!is_store) {
2569 insert_mem_bar(Op_MemBarAcquire);
2570 } else {
2571 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2572 insert_mem_bar(Op_MemBarVolatile);
2573 }
2574 }
2575 break;
2576 default:
2577 ShouldNotReachHere();
2578 }
2579
2580 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2581
2582 return true;
2583 }
2584
2585 //----------------------------inline_unsafe_load_store----------------------------
2586 // This method serves a couple of different customers (depending on LoadStoreKind):
2587 //
2588 // LS_cmp_swap:
2589 //
2590 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2591 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2592 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2593 //
2594 // LS_cmp_swap_weak:
2595 //
2596 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2597 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2598 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2599 //
2600 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2601 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2602 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2603 //
2604 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2605 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2606 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2607 //
2608 // LS_cmp_exchange:
2609 //
2610 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2611 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2612 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2613 //
2614 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2616 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2617 //
2618 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2620 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2621 //
2622 // LS_get_add:
2623 //
2624 // int getAndAddInt( Object o, long offset, int delta)
2625 // long getAndAddLong(Object o, long offset, long delta)
2626 //
2627 // LS_get_set:
2628 //
2629 // int getAndSet(Object o, long offset, int newValue)
2630 // long getAndSet(Object o, long offset, long newValue)
2631 // Object getAndSet(Object o, long offset, Object newValue)
2632 //
2633 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2634 // This basic scheme here is the same as inline_unsafe_access, but
2635 // differs in enough details that combining them would make the code
2636 // overly confusing. (This is a true fact! I originally combined
2637 // them, but even I was confused by it!) As much code/comments as
2638 // possible are retained from inline_unsafe_access though to make
2639 // the correspondences clearer. - dl
2640
2641 if (callee()->is_static()) return false; // caller must have the capability!
2642
2643 #ifndef PRODUCT
2644 BasicType rtype;
2645 {
2646 ResourceMark rm;
2647 // Check the signatures.
2648 ciSignature* sig = callee()->signature();
2649 rtype = sig->return_type()->basic_type();
2650 switch(kind) {
2651 case LS_get_add:
2652 case LS_get_set: {
2653 // Check the signatures.
2654 #ifdef ASSERT
2655 assert(rtype == type, "get and set must return the expected type");
2656 assert(sig->count() == 3, "get and set has 3 arguments");
2657 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2658 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2659 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2660 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2661 #endif // ASSERT
2662 break;
2663 }
2664 case LS_cmp_swap:
2665 case LS_cmp_swap_weak: {
2666 // Check the signatures.
2667 #ifdef ASSERT
2668 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2669 assert(sig->count() == 4, "CAS has 4 arguments");
2670 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2671 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2672 #endif // ASSERT
2673 break;
2674 }
2675 case LS_cmp_exchange: {
2676 // Check the signatures.
2677 #ifdef ASSERT
2678 assert(rtype == type, "CAS must return the expected type");
2679 assert(sig->count() == 4, "CAS has 4 arguments");
2680 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2681 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2682 #endif // ASSERT
2683 break;
2684 }
2685 default:
2686 ShouldNotReachHere();
2687 }
2688 }
2689 #endif //PRODUCT
2690
2691 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2692
2693 // Get arguments:
2694 Node* receiver = NULL;
2695 Node* base = NULL;
2696 Node* offset = NULL;
2697 Node* oldval = NULL;
2698 Node* newval = NULL;
2699 switch(kind) {
2700 case LS_cmp_swap:
2701 case LS_cmp_swap_weak:
2702 case LS_cmp_exchange: {
2703 const bool two_slot_type = type2size[type] == 2;
2704 receiver = argument(0); // type: oop
2705 base = argument(1); // type: oop
2706 offset = argument(2); // type: long
2707 oldval = argument(4); // type: oop, int, or long
2708 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2709 break;
2710 }
2711 case LS_get_add:
2712 case LS_get_set: {
2713 receiver = argument(0); // type: oop
2714 base = argument(1); // type: oop
2715 offset = argument(2); // type: long
2716 oldval = NULL;
2717 newval = argument(4); // type: oop, int, or long
2718 break;
2719 }
2720 default:
2721 ShouldNotReachHere();
2722 }
2723
2724 // Build field offset expression.
2725 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2726 // to be plain byte offsets, which are also the same as those accepted
2727 // by oopDesc::field_base.
2728 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2729 // 32-bit machines ignore the high half of long offsets
2730 offset = ConvL2X(offset);
2731 Node* adr = make_unsafe_address(base, offset);
2732 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2733
2734 Compile::AliasType* alias_type = C->alias_type(adr_type);
2735 BasicType bt = alias_type->basic_type();
2736 if (bt != T_ILLEGAL &&
2737 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2738 // Don't intrinsify mismatched object accesses.
2739 return false;
2740 }
2741
2742 // For CAS, unlike inline_unsafe_access, there seems no point in
2743 // trying to refine types. Just use the coarse types here.
2744 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2745 const Type *value_type = Type::get_const_basic_type(type);
2746
2747 switch (kind) {
2748 case LS_get_set:
2749 case LS_cmp_exchange: {
2750 if (type == T_OBJECT) {
2751 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2752 if (tjp != NULL) {
2753 value_type = tjp;
2754 }
2755 }
2756 break;
2757 }
2758 case LS_cmp_swap:
2759 case LS_cmp_swap_weak:
2760 case LS_get_add:
2761 break;
2762 default:
2763 ShouldNotReachHere();
2764 }
2765
2766 // Null check receiver.
2767 receiver = null_check(receiver);
2768 if (stopped()) {
2769 return true;
2770 }
2771
2772 int alias_idx = C->get_alias_index(adr_type);
2773
2774 // Memory-model-wise, a LoadStore acts like a little synchronized
2775 // block, so needs barriers on each side. These don't translate
2776 // into actual barriers on most machines, but we still need rest of
2777 // compiler to respect ordering.
2778
2779 switch (access_kind) {
2780 case Relaxed:
2781 case Acquire:
2782 break;
2783 case Release:
2784 insert_mem_bar(Op_MemBarRelease);
2785 break;
2786 case Volatile:
2787 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2788 insert_mem_bar(Op_MemBarVolatile);
2789 } else {
2790 insert_mem_bar(Op_MemBarRelease);
2791 }
2792 break;
2793 default:
2794 ShouldNotReachHere();
2795 }
2796 insert_mem_bar(Op_MemBarCPUOrder);
2797
2798 // Figure out the memory ordering.
2799 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2800
2801 // 4984716: MemBars must be inserted before this
2802 // memory node in order to avoid a false
2803 // dependency which will confuse the scheduler.
2804 Node *mem = memory(alias_idx);
2805
2806 // For now, we handle only those cases that actually exist: ints,
2807 // longs, and Object. Adding others should be straightforward.
2808 Node* load_store = NULL;
2809 switch(type) {
2810 case T_BYTE:
2811 switch(kind) {
2812 case LS_get_add:
2813 load_store = _gvn.transform(new GetAndAddBNode(control(), mem, adr, newval, adr_type));
2814 break;
2815 case LS_get_set:
2816 load_store = _gvn.transform(new GetAndSetBNode(control(), mem, adr, newval, adr_type));
2817 break;
2818 case LS_cmp_swap_weak:
2819 load_store = _gvn.transform(new WeakCompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2820 break;
2821 case LS_cmp_swap:
2822 load_store = _gvn.transform(new CompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2823 break;
2824 case LS_cmp_exchange:
2825 load_store = _gvn.transform(new CompareAndExchangeBNode(control(), mem, adr, newval, oldval, adr_type, mo));
2826 break;
2827 default:
2828 ShouldNotReachHere();
2829 }
2830 break;
2831 case T_SHORT:
2832 switch(kind) {
2833 case LS_get_add:
2834 load_store = _gvn.transform(new GetAndAddSNode(control(), mem, adr, newval, adr_type));
2835 break;
2836 case LS_get_set:
2837 load_store = _gvn.transform(new GetAndSetSNode(control(), mem, adr, newval, adr_type));
2838 break;
2839 case LS_cmp_swap_weak:
2840 load_store = _gvn.transform(new WeakCompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2841 break;
2842 case LS_cmp_swap:
2843 load_store = _gvn.transform(new CompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2844 break;
2845 case LS_cmp_exchange:
2846 load_store = _gvn.transform(new CompareAndExchangeSNode(control(), mem, adr, newval, oldval, adr_type, mo));
2847 break;
2848 default:
2849 ShouldNotReachHere();
2850 }
2851 break;
2852 case T_INT:
2853 switch(kind) {
2854 case LS_get_add:
2855 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2856 break;
2857 case LS_get_set:
2858 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2859 break;
2860 case LS_cmp_swap_weak:
2861 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2862 break;
2863 case LS_cmp_swap:
2864 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2865 break;
2866 case LS_cmp_exchange:
2867 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2868 break;
2869 default:
2870 ShouldNotReachHere();
2871 }
2872 break;
2873 case T_LONG:
2874 switch(kind) {
2875 case LS_get_add:
2876 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2877 break;
2878 case LS_get_set:
2879 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2880 break;
2881 case LS_cmp_swap_weak:
2882 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2883 break;
2884 case LS_cmp_swap:
2885 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2886 break;
2887 case LS_cmp_exchange:
2888 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2889 break;
2890 default:
2891 ShouldNotReachHere();
2892 }
2893 break;
2894 case T_OBJECT:
2895 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2896 // could be delayed during Parse (for example, in adjust_map_after_if()).
2897 // Execute transformation here to avoid barrier generation in such case.
2898 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2899 newval = _gvn.makecon(TypePtr::NULL_PTR);
2900
2901 // Reference stores need a store barrier.
2902 switch(kind) {
2903 case LS_get_set: {
2904 // If pre-barrier must execute before the oop store, old value will require do_load here.
2905 if (!can_move_pre_barrier()) {
2906 pre_barrier(true /* do_load*/,
2907 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2908 NULL /* pre_val*/,
2909 T_OBJECT);
2910 } // Else move pre_barrier to use load_store value, see below.
2911 break;
2912 }
2913 case LS_cmp_swap_weak:
2914 case LS_cmp_swap:
2915 case LS_cmp_exchange: {
2916 // Same as for newval above:
2917 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2918 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2919 }
2920 // The only known value which might get overwritten is oldval.
2921 pre_barrier(false /* do_load */,
2922 control(), NULL, NULL, max_juint, NULL, NULL,
2923 oldval /* pre_val */,
2924 T_OBJECT);
2925 break;
2926 }
2927 default:
2928 ShouldNotReachHere();
2929 }
2930
2931 #ifdef _LP64
2932 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2933 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2934
2935 switch(kind) {
2936 case LS_get_set:
2937 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2938 break;
2939 case LS_cmp_swap_weak: {
2940 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2941 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2942 break;
2943 }
2944 case LS_cmp_swap: {
2945 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2946 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2947 break;
2948 }
2949 case LS_cmp_exchange: {
2950 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2951 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2952 break;
2953 }
2954 default:
2955 ShouldNotReachHere();
2956 }
2957 } else
2958 #endif
2959 switch (kind) {
2960 case LS_get_set:
2961 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2962 break;
2963 case LS_cmp_swap_weak:
2964 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2965 break;
2966 case LS_cmp_swap:
2967 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2968 break;
2969 case LS_cmp_exchange:
2970 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
2971 break;
2972 default:
2973 ShouldNotReachHere();
2974 }
2975
2976 // Emit the post barrier only when the actual store happened. This makes sense
2977 // to check only for LS_cmp_* that can fail to set the value.
2978 // LS_cmp_exchange does not produce any branches by default, so there is no
2979 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
2980 // CompareAndExchange and move branches here, it would make sense to conditionalize
2981 // post_barriers for LS_cmp_exchange as well.
2982 //
2983 // CAS success path is marked more likely since we anticipate this is a performance
2984 // critical path, while CAS failure path can use the penalty for going through unlikely
2985 // path as backoff. Which is still better than doing a store barrier there.
2986 switch (kind) {
2987 case LS_get_set:
2988 case LS_cmp_exchange: {
2989 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2990 break;
2991 }
2992 case LS_cmp_swap_weak:
2993 case LS_cmp_swap: {
2994 IdealKit ideal(this);
2995 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
2996 sync_kit(ideal);
2997 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2998 ideal.sync_kit(this);
2999 } ideal.end_if();
3000 final_sync(ideal);
3001 break;
3002 }
3003 default:
3004 ShouldNotReachHere();
3005 }
3006 break;
3007 default:
3008 fatal("unexpected type %d: %s", type, type2name(type));
3009 break;
3010 }
3011
3012 // SCMemProjNodes represent the memory state of a LoadStore. Their
3013 // main role is to prevent LoadStore nodes from being optimized away
3014 // when their results aren't used.
3015 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3016 set_memory(proj, alias_idx);
3017
3018 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3019 #ifdef _LP64
3020 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3021 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3022 }
3023 #endif
3024 if (can_move_pre_barrier() && kind == LS_get_set) {
3025 // Don't need to load pre_val. The old value is returned by load_store.
3026 // The pre_barrier can execute after the xchg as long as no safepoint
3027 // gets inserted between them.
3028 pre_barrier(false /* do_load */,
3029 control(), NULL, NULL, max_juint, NULL, NULL,
3030 load_store /* pre_val */,
3031 T_OBJECT);
3032 }
3033 }
3034
3035 // Add the trailing membar surrounding the access
3036 insert_mem_bar(Op_MemBarCPUOrder);
3037
3038 switch (access_kind) {
3039 case Relaxed:
3040 case Release:
3041 break; // do nothing
3042 case Acquire:
3043 case Volatile:
3044 insert_mem_bar(Op_MemBarAcquire);
3045 // !support_IRIW_for_not_multiple_copy_atomic_cpu handled in platform code
3046 break;
3047 default:
3048 ShouldNotReachHere();
3049 }
3050
3051 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3052 set_result(load_store);
3053 return true;
3054 }
3055
3056 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3057 MemNode::MemOrd mo = MemNode::unset;
3058 switch(kind) {
3059 case Opaque:
3060 case Relaxed: mo = MemNode::unordered; break;
3061 case Acquire: mo = MemNode::acquire; break;
3062 case Release: mo = MemNode::release; break;
3063 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3064 default:
3065 ShouldNotReachHere();
3066 }
3067 guarantee(mo != MemNode::unset, "Should select memory ordering");
3068 return mo;
3069 }
3070
3071 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3072 MemNode::MemOrd mo = MemNode::unset;
3073 switch(kind) {
3074 case Opaque:
3075 case Relaxed: mo = MemNode::unordered; break;
3076 case Acquire: mo = MemNode::acquire; break;
3077 case Release: mo = MemNode::release; break;
3078 case Volatile: mo = MemNode::seqcst; break;
3079 default:
3080 ShouldNotReachHere();
3081 }
3082 guarantee(mo != MemNode::unset, "Should select memory ordering");
3083 return mo;
3084 }
3085
3086 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3087 // Regardless of form, don't allow previous ld/st to move down,
3088 // then issue acquire, release, or volatile mem_bar.
3089 insert_mem_bar(Op_MemBarCPUOrder);
3090 switch(id) {
3091 case vmIntrinsics::_loadFence:
3092 insert_mem_bar(Op_LoadFence);
3093 return true;
3094 case vmIntrinsics::_storeFence:
3095 insert_mem_bar(Op_StoreFence);
3096 return true;
3097 case vmIntrinsics::_fullFence:
3098 insert_mem_bar(Op_MemBarVolatile);
3099 return true;
3100 default:
3101 fatal_unexpected_iid(id);
3102 return false;
3103 }
3104 }
3105
3106 bool LibraryCallKit::inline_onspinwait() {
3107 insert_mem_bar(Op_OnSpinWait);
3108 return true;
3109 }
3110
3111 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3112 if (!kls->is_Con()) {
3113 return true;
3114 }
3115 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3116 if (klsptr == NULL) {
3117 return true;
3118 }
3119 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3120 // don't need a guard for a klass that is already initialized
3121 return !ik->is_initialized();
3122 }
3123
3124 //----------------------------inline_unsafe_allocate---------------------------
3125 // public native Object Unsafe.allocateInstance(Class<?> cls);
3126 bool LibraryCallKit::inline_unsafe_allocate() {
3127 if (callee()->is_static()) return false; // caller must have the capability!
3128
3129 null_check_receiver(); // null-check, then ignore
3130 Node* cls = null_check(argument(1));
3131 if (stopped()) return true;
3132
3133 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3134 kls = null_check(kls);
3135 if (stopped()) return true; // argument was like int.class
3136
3137 Node* test = NULL;
3138 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3139 // Note: The argument might still be an illegal value like
3140 // Serializable.class or Object[].class. The runtime will handle it.
3141 // But we must make an explicit check for initialization.
3142 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3143 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3144 // can generate code to load it as unsigned byte.
3145 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3146 Node* bits = intcon(InstanceKlass::fully_initialized);
3147 test = _gvn.transform(new SubINode(inst, bits));
3148 // The 'test' is non-zero if we need to take a slow path.
3149 }
3150
3151 Node* obj = new_instance(kls, test);
3152 set_result(obj);
3153 return true;
3154 }
3155
3156 //------------------------inline_native_time_funcs--------------
3157 // inline code for System.currentTimeMillis() and System.nanoTime()
3158 // these have the same type and signature
3159 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3160 const TypeFunc* tf = OptoRuntime::void_long_Type();
3161 const TypePtr* no_memory_effects = NULL;
3162 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3163 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3164 #ifdef ASSERT
3165 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3166 assert(value_top == top(), "second value must be top");
3167 #endif
3168 set_result(value);
3169 return true;
3170 }
3171
3172 #ifdef TRACE_HAVE_INTRINSICS
3173
3174 /*
3175 * oop -> myklass
3176 * myklass->trace_id |= USED
3177 * return myklass->trace_id & ~0x3
3178 */
3179 bool LibraryCallKit::inline_native_classID() {
3180 Node* cls = null_check(argument(0), T_OBJECT);
3181 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3182 kls = null_check(kls, T_OBJECT);
3183
3184 ByteSize offset = TRACE_KLASS_TRACE_ID_OFFSET;
3185 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3186 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3187
3188 Node* clsused = longcon(0x01l); // set the class bit
3189 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3190 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3191 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3192
3193 #ifdef TRACE_ID_META_BITS
3194 Node* mbits = longcon(~TRACE_ID_META_BITS);
3195 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
3196 #endif
3197 #ifdef TRACE_ID_CLASS_SHIFT
3198 Node* cbits = intcon(TRACE_ID_CLASS_SHIFT);
3199 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
3200 #endif
3201
3202 set_result(tvalue);
3203 return true;
3204
3205 }
3206
3207 bool LibraryCallKit::inline_native_getBufferWriter() {
3208 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3209
3210 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3211 in_bytes(TRACE_THREAD_DATA_WRITER_OFFSET)
3212 );
3213
3214 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3215
3216 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3217 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3218
3219 IfNode* iff_jobj_null =
3220 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3221
3222 enum { _normal_path = 1,
3223 _null_path = 2,
3224 PATH_LIMIT };
3225
3226 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3227 PhiNode* result_val = new PhiNode(result_rgn, TypePtr::BOTTOM);
3228
3229 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3230 result_rgn->init_req(_null_path, jobj_is_null);
3231 result_val->init_req(_null_path, null());
3232
3233 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3234 result_rgn->init_req(_normal_path, jobj_is_not_null);
3235
3236 Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
3237 result_val->init_req(_normal_path, res);
3238
3239 set_result(result_rgn, result_val);
3240
3241 return true;
3242 }
3243
3244 #endif
3245
3246 //------------------------inline_native_currentThread------------------
3247 bool LibraryCallKit::inline_native_currentThread() {
3248 Node* junk = NULL;
3249 set_result(generate_current_thread(junk));
3250 return true;
3251 }
3252
3253 //------------------------inline_native_isInterrupted------------------
3254 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3255 bool LibraryCallKit::inline_native_isInterrupted() {
3256 // Add a fast path to t.isInterrupted(clear_int):
3257 // (t == Thread.current() &&
3258 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3259 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3260 // So, in the common case that the interrupt bit is false,
3261 // we avoid making a call into the VM. Even if the interrupt bit
3262 // is true, if the clear_int argument is false, we avoid the VM call.
3263 // However, if the receiver is not currentThread, we must call the VM,
3264 // because there must be some locking done around the operation.
3265
3266 // We only go to the fast case code if we pass two guards.
3267 // Paths which do not pass are accumulated in the slow_region.
3268
3269 enum {
3270 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3271 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3272 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3273 PATH_LIMIT
3274 };
3275
3276 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3277 // out of the function.
3278 insert_mem_bar(Op_MemBarCPUOrder);
3279
3280 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3281 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3282
3283 RegionNode* slow_region = new RegionNode(1);
3284 record_for_igvn(slow_region);
3285
3286 // (a) Receiving thread must be the current thread.
3287 Node* rec_thr = argument(0);
3288 Node* tls_ptr = NULL;
3289 Node* cur_thr = generate_current_thread(tls_ptr);
3290 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3291 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3292
3293 generate_slow_guard(bol_thr, slow_region);
3294
3295 // (b) Interrupt bit on TLS must be false.
3296 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3297 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3298 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3299
3300 // Set the control input on the field _interrupted read to prevent it floating up.
3301 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3302 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3303 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3304
3305 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3306
3307 // First fast path: if (!TLS._interrupted) return false;
3308 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3309 result_rgn->init_req(no_int_result_path, false_bit);
3310 result_val->init_req(no_int_result_path, intcon(0));
3311
3312 // drop through to next case
3313 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3314
3315 #ifndef _WINDOWS
3316 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3317 Node* clr_arg = argument(1);
3318 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3319 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3320 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3321
3322 // Second fast path: ... else if (!clear_int) return true;
3323 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3324 result_rgn->init_req(no_clear_result_path, false_arg);
3325 result_val->init_req(no_clear_result_path, intcon(1));
3326
3327 // drop through to next case
3328 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3329 #else
3330 // To return true on Windows you must read the _interrupted field
3331 // and check the event state i.e. take the slow path.
3332 #endif // _WINDOWS
3333
3334 // (d) Otherwise, go to the slow path.
3335 slow_region->add_req(control());
3336 set_control( _gvn.transform(slow_region));
3337
3338 if (stopped()) {
3339 // There is no slow path.
3340 result_rgn->init_req(slow_result_path, top());
3341 result_val->init_req(slow_result_path, top());
3342 } else {
3343 // non-virtual because it is a private non-static
3344 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3345
3346 Node* slow_val = set_results_for_java_call(slow_call);
3347 // this->control() comes from set_results_for_java_call
3348
3349 Node* fast_io = slow_call->in(TypeFunc::I_O);
3350 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3351
3352 // These two phis are pre-filled with copies of of the fast IO and Memory
3353 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3354 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3355
3356 result_rgn->init_req(slow_result_path, control());
3357 result_io ->init_req(slow_result_path, i_o());
3358 result_mem->init_req(slow_result_path, reset_memory());
3359 result_val->init_req(slow_result_path, slow_val);
3360
3361 set_all_memory(_gvn.transform(result_mem));
3362 set_i_o( _gvn.transform(result_io));
3363 }
3364
3365 C->set_has_split_ifs(true); // Has chance for split-if optimization
3366 set_result(result_rgn, result_val);
3367 return true;
3368 }
3369
3370 //---------------------------load_mirror_from_klass----------------------------
3371 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3372 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3373 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3374 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3375 }
3376
3377 //-----------------------load_klass_from_mirror_common-------------------------
3378 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3379 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3380 // and branch to the given path on the region.
3381 // If never_see_null, take an uncommon trap on null, so we can optimistically
3382 // compile for the non-null case.
3383 // If the region is NULL, force never_see_null = true.
3384 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3385 bool never_see_null,
3386 RegionNode* region,
3387 int null_path,
3388 int offset) {
3389 if (region == NULL) never_see_null = true;
3390 Node* p = basic_plus_adr(mirror, offset);
3391 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3392 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3393 Node* null_ctl = top();
3394 kls = null_check_oop(kls, &null_ctl, never_see_null);
3395 if (region != NULL) {
3396 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3397 region->init_req(null_path, null_ctl);
3398 } else {
3399 assert(null_ctl == top(), "no loose ends");
3400 }
3401 return kls;
3402 }
3403
3404 //--------------------(inline_native_Class_query helpers)---------------------
3405 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3406 // Fall through if (mods & mask) == bits, take the guard otherwise.
3407 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3408 // Branch around if the given klass has the given modifier bit set.
3409 // Like generate_guard, adds a new path onto the region.
3410 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3411 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3412 Node* mask = intcon(modifier_mask);
3413 Node* bits = intcon(modifier_bits);
3414 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3415 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3416 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3417 return generate_fair_guard(bol, region);
3418 }
3419 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3420 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3421 }
3422
3423 //-------------------------inline_native_Class_query-------------------
3424 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3425 const Type* return_type = TypeInt::BOOL;
3426 Node* prim_return_value = top(); // what happens if it's a primitive class?
3427 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3428 bool expect_prim = false; // most of these guys expect to work on refs
3429
3430 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3431
3432 Node* mirror = argument(0);
3433 Node* obj = top();
3434
3435 switch (id) {
3436 case vmIntrinsics::_isInstance:
3437 // nothing is an instance of a primitive type
3438 prim_return_value = intcon(0);
3439 obj = argument(1);
3440 break;
3441 case vmIntrinsics::_getModifiers:
3442 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3443 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3444 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3445 break;
3446 case vmIntrinsics::_isInterface:
3447 prim_return_value = intcon(0);
3448 break;
3449 case vmIntrinsics::_isArray:
3450 prim_return_value = intcon(0);
3451 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3452 break;
3453 case vmIntrinsics::_isPrimitive:
3454 prim_return_value = intcon(1);
3455 expect_prim = true; // obviously
3456 break;
3457 case vmIntrinsics::_getSuperclass:
3458 prim_return_value = null();
3459 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3460 break;
3461 case vmIntrinsics::_getClassAccessFlags:
3462 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3463 return_type = TypeInt::INT; // not bool! 6297094
3464 break;
3465 default:
3466 fatal_unexpected_iid(id);
3467 break;
3468 }
3469
3470 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3471 if (mirror_con == NULL) return false; // cannot happen?
3472
3473 #ifndef PRODUCT
3474 if (C->print_intrinsics() || C->print_inlining()) {
3475 ciType* k = mirror_con->java_mirror_type();
3476 if (k) {
3477 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3478 k->print_name();
3479 tty->cr();
3480 }
3481 }
3482 #endif
3483
3484 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3485 RegionNode* region = new RegionNode(PATH_LIMIT);
3486 record_for_igvn(region);
3487 PhiNode* phi = new PhiNode(region, return_type);
3488
3489 // The mirror will never be null of Reflection.getClassAccessFlags, however
3490 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3491 // if it is. See bug 4774291.
3492
3493 // For Reflection.getClassAccessFlags(), the null check occurs in
3494 // the wrong place; see inline_unsafe_access(), above, for a similar
3495 // situation.
3496 mirror = null_check(mirror);
3497 // If mirror or obj is dead, only null-path is taken.
3498 if (stopped()) return true;
3499
3500 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3501
3502 // Now load the mirror's klass metaobject, and null-check it.
3503 // Side-effects region with the control path if the klass is null.
3504 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3505 // If kls is null, we have a primitive mirror.
3506 phi->init_req(_prim_path, prim_return_value);
3507 if (stopped()) { set_result(region, phi); return true; }
3508 bool safe_for_replace = (region->in(_prim_path) == top());
3509
3510 Node* p; // handy temp
3511 Node* null_ctl;
3512
3513 // Now that we have the non-null klass, we can perform the real query.
3514 // For constant classes, the query will constant-fold in LoadNode::Value.
3515 Node* query_value = top();
3516 switch (id) {
3517 case vmIntrinsics::_isInstance:
3518 // nothing is an instance of a primitive type
3519 query_value = gen_instanceof(obj, kls, safe_for_replace);
3520 break;
3521
3522 case vmIntrinsics::_getModifiers:
3523 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3524 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3525 break;
3526
3527 case vmIntrinsics::_isInterface:
3528 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3529 if (generate_interface_guard(kls, region) != NULL)
3530 // A guard was added. If the guard is taken, it was an interface.
3531 phi->add_req(intcon(1));
3532 // If we fall through, it's a plain class.
3533 query_value = intcon(0);
3534 break;
3535
3536 case vmIntrinsics::_isArray:
3537 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3538 if (generate_array_guard(kls, region) != NULL)
3539 // A guard was added. If the guard is taken, it was an array.
3540 phi->add_req(intcon(1));
3541 // If we fall through, it's a plain class.
3542 query_value = intcon(0);
3543 break;
3544
3545 case vmIntrinsics::_isPrimitive:
3546 query_value = intcon(0); // "normal" path produces false
3547 break;
3548
3549 case vmIntrinsics::_getSuperclass:
3550 // The rules here are somewhat unfortunate, but we can still do better
3551 // with random logic than with a JNI call.
3552 // Interfaces store null or Object as _super, but must report null.
3553 // Arrays store an intermediate super as _super, but must report Object.
3554 // Other types can report the actual _super.
3555 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3556 if (generate_interface_guard(kls, region) != NULL)
3557 // A guard was added. If the guard is taken, it was an interface.
3558 phi->add_req(null());
3559 if (generate_array_guard(kls, region) != NULL)
3560 // A guard was added. If the guard is taken, it was an array.
3561 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3562 // If we fall through, it's a plain class. Get its _super.
3563 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3564 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3565 null_ctl = top();
3566 kls = null_check_oop(kls, &null_ctl);
3567 if (null_ctl != top()) {
3568 // If the guard is taken, Object.superClass is null (both klass and mirror).
3569 region->add_req(null_ctl);
3570 phi ->add_req(null());
3571 }
3572 if (!stopped()) {
3573 query_value = load_mirror_from_klass(kls);
3574 }
3575 break;
3576
3577 case vmIntrinsics::_getClassAccessFlags:
3578 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3579 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3580 break;
3581
3582 default:
3583 fatal_unexpected_iid(id);
3584 break;
3585 }
3586
3587 // Fall-through is the normal case of a query to a real class.
3588 phi->init_req(1, query_value);
3589 region->init_req(1, control());
3590
3591 C->set_has_split_ifs(true); // Has chance for split-if optimization
3592 set_result(region, phi);
3593 return true;
3594 }
3595
3596 //-------------------------inline_Class_cast-------------------
3597 bool LibraryCallKit::inline_Class_cast() {
3598 Node* mirror = argument(0); // Class
3599 Node* obj = argument(1);
3600 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3601 if (mirror_con == NULL) {
3602 return false; // dead path (mirror->is_top()).
3603 }
3604 if (obj == NULL || obj->is_top()) {
3605 return false; // dead path
3606 }
3607 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3608
3609 // First, see if Class.cast() can be folded statically.
3610 // java_mirror_type() returns non-null for compile-time Class constants.
3611 ciType* tm = mirror_con->java_mirror_type();
3612 if (tm != NULL && tm->is_klass() &&
3613 tp != NULL && tp->klass() != NULL) {
3614 if (!tp->klass()->is_loaded()) {
3615 // Don't use intrinsic when class is not loaded.
3616 return false;
3617 } else {
3618 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3619 if (static_res == Compile::SSC_always_true) {
3620 // isInstance() is true - fold the code.
3621 set_result(obj);
3622 return true;
3623 } else if (static_res == Compile::SSC_always_false) {
3624 // Don't use intrinsic, have to throw ClassCastException.
3625 // If the reference is null, the non-intrinsic bytecode will
3626 // be optimized appropriately.
3627 return false;
3628 }
3629 }
3630 }
3631
3632 // Bailout intrinsic and do normal inlining if exception path is frequent.
3633 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3634 return false;
3635 }
3636
3637 // Generate dynamic checks.
3638 // Class.cast() is java implementation of _checkcast bytecode.
3639 // Do checkcast (Parse::do_checkcast()) optimizations here.
3640
3641 mirror = null_check(mirror);
3642 // If mirror is dead, only null-path is taken.
3643 if (stopped()) {
3644 return true;
3645 }
3646
3647 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3648 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3649 RegionNode* region = new RegionNode(PATH_LIMIT);
3650 record_for_igvn(region);
3651
3652 // Now load the mirror's klass metaobject, and null-check it.
3653 // If kls is null, we have a primitive mirror and
3654 // nothing is an instance of a primitive type.
3655 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3656
3657 Node* res = top();
3658 if (!stopped()) {
3659 Node* bad_type_ctrl = top();
3660 // Do checkcast optimizations.
3661 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3662 region->init_req(_bad_type_path, bad_type_ctrl);
3663 }
3664 if (region->in(_prim_path) != top() ||
3665 region->in(_bad_type_path) != top()) {
3666 // Let Interpreter throw ClassCastException.
3667 PreserveJVMState pjvms(this);
3668 set_control(_gvn.transform(region));
3669 uncommon_trap(Deoptimization::Reason_intrinsic,
3670 Deoptimization::Action_maybe_recompile);
3671 }
3672 if (!stopped()) {
3673 set_result(res);
3674 }
3675 return true;
3676 }
3677
3678
3679 //--------------------------inline_native_subtype_check------------------------
3680 // This intrinsic takes the JNI calls out of the heart of
3681 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3682 bool LibraryCallKit::inline_native_subtype_check() {
3683 // Pull both arguments off the stack.
3684 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3685 args[0] = argument(0);
3686 args[1] = argument(1);
3687 Node* klasses[2]; // corresponding Klasses: superk, subk
3688 klasses[0] = klasses[1] = top();
3689
3690 enum {
3691 // A full decision tree on {superc is prim, subc is prim}:
3692 _prim_0_path = 1, // {P,N} => false
3693 // {P,P} & superc!=subc => false
3694 _prim_same_path, // {P,P} & superc==subc => true
3695 _prim_1_path, // {N,P} => false
3696 _ref_subtype_path, // {N,N} & subtype check wins => true
3697 _both_ref_path, // {N,N} & subtype check loses => false
3698 PATH_LIMIT
3699 };
3700
3701 RegionNode* region = new RegionNode(PATH_LIMIT);
3702 Node* phi = new PhiNode(region, TypeInt::BOOL);
3703 record_for_igvn(region);
3704
3705 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3706 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3707 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3708
3709 // First null-check both mirrors and load each mirror's klass metaobject.
3710 int which_arg;
3711 for (which_arg = 0; which_arg <= 1; which_arg++) {
3712 Node* arg = args[which_arg];
3713 arg = null_check(arg);
3714 if (stopped()) break;
3715 args[which_arg] = arg;
3716
3717 Node* p = basic_plus_adr(arg, class_klass_offset);
3718 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3719 klasses[which_arg] = _gvn.transform(kls);
3720 }
3721
3722 // Having loaded both klasses, test each for null.
3723 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3724 for (which_arg = 0; which_arg <= 1; which_arg++) {
3725 Node* kls = klasses[which_arg];
3726 Node* null_ctl = top();
3727 kls = null_check_oop(kls, &null_ctl, never_see_null);
3728 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3729 region->init_req(prim_path, null_ctl);
3730 if (stopped()) break;
3731 klasses[which_arg] = kls;
3732 }
3733
3734 if (!stopped()) {
3735 // now we have two reference types, in klasses[0..1]
3736 Node* subk = klasses[1]; // the argument to isAssignableFrom
3737 Node* superk = klasses[0]; // the receiver
3738 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3739 // now we have a successful reference subtype check
3740 region->set_req(_ref_subtype_path, control());
3741 }
3742
3743 // If both operands are primitive (both klasses null), then
3744 // we must return true when they are identical primitives.
3745 // It is convenient to test this after the first null klass check.
3746 set_control(region->in(_prim_0_path)); // go back to first null check
3747 if (!stopped()) {
3748 // Since superc is primitive, make a guard for the superc==subc case.
3749 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3750 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3751 generate_guard(bol_eq, region, PROB_FAIR);
3752 if (region->req() == PATH_LIMIT+1) {
3753 // A guard was added. If the added guard is taken, superc==subc.
3754 region->swap_edges(PATH_LIMIT, _prim_same_path);
3755 region->del_req(PATH_LIMIT);
3756 }
3757 region->set_req(_prim_0_path, control()); // Not equal after all.
3758 }
3759
3760 // these are the only paths that produce 'true':
3761 phi->set_req(_prim_same_path, intcon(1));
3762 phi->set_req(_ref_subtype_path, intcon(1));
3763
3764 // pull together the cases:
3765 assert(region->req() == PATH_LIMIT, "sane region");
3766 for (uint i = 1; i < region->req(); i++) {
3767 Node* ctl = region->in(i);
3768 if (ctl == NULL || ctl == top()) {
3769 region->set_req(i, top());
3770 phi ->set_req(i, top());
3771 } else if (phi->in(i) == NULL) {
3772 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3773 }
3774 }
3775
3776 set_control(_gvn.transform(region));
3777 set_result(_gvn.transform(phi));
3778 return true;
3779 }
3780
3781 //---------------------generate_array_guard_common------------------------
3782 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3783 bool obj_array, bool not_array) {
3784
3785 if (stopped()) {
3786 return NULL;
3787 }
3788
3789 // If obj_array/non_array==false/false:
3790 // Branch around if the given klass is in fact an array (either obj or prim).
3791 // If obj_array/non_array==false/true:
3792 // Branch around if the given klass is not an array klass of any kind.
3793 // If obj_array/non_array==true/true:
3794 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3795 // If obj_array/non_array==true/false:
3796 // Branch around if the kls is an oop array (Object[] or subtype)
3797 //
3798 // Like generate_guard, adds a new path onto the region.
3799 jint layout_con = 0;
3800 Node* layout_val = get_layout_helper(kls, layout_con);
3801 if (layout_val == NULL) {
3802 bool query = (obj_array
3803 ? Klass::layout_helper_is_objArray(layout_con)
3804 : Klass::layout_helper_is_array(layout_con));
3805 if (query == not_array) {
3806 return NULL; // never a branch
3807 } else { // always a branch
3808 Node* always_branch = control();
3809 if (region != NULL)
3810 region->add_req(always_branch);
3811 set_control(top());
3812 return always_branch;
3813 }
3814 }
3815 // Now test the correct condition.
3816 jint nval = (obj_array
3817 ? (jint)(Klass::_lh_array_tag_type_value
3818 << Klass::_lh_array_tag_shift)
3819 : Klass::_lh_neutral_value);
3820 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3821 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3822 // invert the test if we are looking for a non-array
3823 if (not_array) btest = BoolTest(btest).negate();
3824 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3825 return generate_fair_guard(bol, region);
3826 }
3827
3828
3829 //-----------------------inline_native_newArray--------------------------
3830 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3831 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3832 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3833 Node* mirror;
3834 Node* count_val;
3835 if (uninitialized) {
3836 mirror = argument(1);
3837 count_val = argument(2);
3838 } else {
3839 mirror = argument(0);
3840 count_val = argument(1);
3841 }
3842
3843 mirror = null_check(mirror);
3844 // If mirror or obj is dead, only null-path is taken.
3845 if (stopped()) return true;
3846
3847 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3848 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3849 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3850 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3851 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3852
3853 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3854 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3855 result_reg, _slow_path);
3856 Node* normal_ctl = control();
3857 Node* no_array_ctl = result_reg->in(_slow_path);
3858
3859 // Generate code for the slow case. We make a call to newArray().
3860 set_control(no_array_ctl);
3861 if (!stopped()) {
3862 // Either the input type is void.class, or else the
3863 // array klass has not yet been cached. Either the
3864 // ensuing call will throw an exception, or else it
3865 // will cache the array klass for next time.
3866 PreserveJVMState pjvms(this);
3867 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3868 Node* slow_result = set_results_for_java_call(slow_call);
3869 // this->control() comes from set_results_for_java_call
3870 result_reg->set_req(_slow_path, control());
3871 result_val->set_req(_slow_path, slow_result);
3872 result_io ->set_req(_slow_path, i_o());
3873 result_mem->set_req(_slow_path, reset_memory());
3874 }
3875
3876 set_control(normal_ctl);
3877 if (!stopped()) {
3878 // Normal case: The array type has been cached in the java.lang.Class.
3879 // The following call works fine even if the array type is polymorphic.
3880 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3881 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3882 result_reg->init_req(_normal_path, control());
3883 result_val->init_req(_normal_path, obj);
3884 result_io ->init_req(_normal_path, i_o());
3885 result_mem->init_req(_normal_path, reset_memory());
3886
3887 if (uninitialized) {
3888 // Mark the allocation so that zeroing is skipped
3889 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3890 alloc->maybe_set_complete(&_gvn);
3891 }
3892 }
3893
3894 // Return the combined state.
3895 set_i_o( _gvn.transform(result_io) );
3896 set_all_memory( _gvn.transform(result_mem));
3897
3898 C->set_has_split_ifs(true); // Has chance for split-if optimization
3899 set_result(result_reg, result_val);
3900 return true;
3901 }
3902
3903 //----------------------inline_native_getLength--------------------------
3904 // public static native int java.lang.reflect.Array.getLength(Object array);
3905 bool LibraryCallKit::inline_native_getLength() {
3906 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3907
3908 Node* array = null_check(argument(0));
3909 // If array is dead, only null-path is taken.
3910 if (stopped()) return true;
3911
3912 // Deoptimize if it is a non-array.
3913 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3914
3915 if (non_array != NULL) {
3916 PreserveJVMState pjvms(this);
3917 set_control(non_array);
3918 uncommon_trap(Deoptimization::Reason_intrinsic,
3919 Deoptimization::Action_maybe_recompile);
3920 }
3921
3922 // If control is dead, only non-array-path is taken.
3923 if (stopped()) return true;
3924
3925 // The works fine even if the array type is polymorphic.
3926 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3927 Node* result = load_array_length(array);
3928
3929 C->set_has_split_ifs(true); // Has chance for split-if optimization
3930 set_result(result);
3931 return true;
3932 }
3933
3934 //------------------------inline_array_copyOf----------------------------
3935 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3936 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3937 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3938 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3939
3940 // Get the arguments.
3941 Node* original = argument(0);
3942 Node* start = is_copyOfRange? argument(1): intcon(0);
3943 Node* end = is_copyOfRange? argument(2): argument(1);
3944 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3945
3946 Node* newcopy = NULL;
3947
3948 // Set the original stack and the reexecute bit for the interpreter to reexecute
3949 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3950 { PreserveReexecuteState preexecs(this);
3951 jvms()->set_should_reexecute(true);
3952
3953 array_type_mirror = null_check(array_type_mirror);
3954 original = null_check(original);
3955
3956 // Check if a null path was taken unconditionally.
3957 if (stopped()) return true;
3958
3959 Node* orig_length = load_array_length(original);
3960
3961 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3962 klass_node = null_check(klass_node);
3963
3964 RegionNode* bailout = new RegionNode(1);
3965 record_for_igvn(bailout);
3966
3967 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3968 // Bail out if that is so.
3969 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3970 if (not_objArray != NULL) {
3971 // Improve the klass node's type from the new optimistic assumption:
3972 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3973 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3974 Node* cast = new CastPPNode(klass_node, akls);
3975 cast->init_req(0, control());
3976 klass_node = _gvn.transform(cast);
3977 }
3978
3979 // Bail out if either start or end is negative.
3980 generate_negative_guard(start, bailout, &start);
3981 generate_negative_guard(end, bailout, &end);
3982
3983 Node* length = end;
3984 if (_gvn.type(start) != TypeInt::ZERO) {
3985 length = _gvn.transform(new SubINode(end, start));
3986 }
3987
3988 // Bail out if length is negative.
3989 // Without this the new_array would throw
3990 // NegativeArraySizeException but IllegalArgumentException is what
3991 // should be thrown
3992 generate_negative_guard(length, bailout, &length);
3993
3994 if (bailout->req() > 1) {
3995 PreserveJVMState pjvms(this);
3996 set_control(_gvn.transform(bailout));
3997 uncommon_trap(Deoptimization::Reason_intrinsic,
3998 Deoptimization::Action_maybe_recompile);
3999 }
4000
4001 if (!stopped()) {
4002 // How many elements will we copy from the original?
4003 // The answer is MinI(orig_length - start, length).
4004 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4005 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4006
4007 // Generate a direct call to the right arraycopy function(s).
4008 // We know the copy is disjoint but we might not know if the
4009 // oop stores need checking.
4010 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4011 // This will fail a store-check if x contains any non-nulls.
4012
4013 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4014 // loads/stores but it is legal only if we're sure the
4015 // Arrays.copyOf would succeed. So we need all input arguments
4016 // to the copyOf to be validated, including that the copy to the
4017 // new array won't trigger an ArrayStoreException. That subtype
4018 // check can be optimized if we know something on the type of
4019 // the input array from type speculation.
4020 if (_gvn.type(klass_node)->singleton()) {
4021 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
4022 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4023
4024 int test = C->static_subtype_check(superk, subk);
4025 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4026 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4027 if (t_original->speculative_type() != NULL) {
4028 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4029 }
4030 }
4031 }
4032
4033 bool validated = false;
4034 // Reason_class_check rather than Reason_intrinsic because we
4035 // want to intrinsify even if this traps.
4036 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4037 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
4038 klass_node);
4039
4040 if (not_subtype_ctrl != top()) {
4041 PreserveJVMState pjvms(this);
4042 set_control(not_subtype_ctrl);
4043 uncommon_trap(Deoptimization::Reason_class_check,
4044 Deoptimization::Action_make_not_entrant);
4045 assert(stopped(), "Should be stopped");
4046 }
4047 validated = true;
4048 }
4049
4050 if (!stopped()) {
4051 newcopy = new_array(klass_node, length, 0); // no arguments to push
4052
4053 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4054 load_object_klass(original), klass_node);
4055 if (!is_copyOfRange) {
4056 ac->set_copyof(validated);
4057 } else {
4058 ac->set_copyofrange(validated);
4059 }
4060 Node* n = _gvn.transform(ac);
4061 if (n == ac) {
4062 ac->connect_outputs(this);
4063 } else {
4064 assert(validated, "shouldn't transform if all arguments not validated");
4065 set_all_memory(n);
4066 }
4067 }
4068 }
4069 } // original reexecute is set back here
4070
4071 C->set_has_split_ifs(true); // Has chance for split-if optimization
4072 if (!stopped()) {
4073 set_result(newcopy);
4074 }
4075 return true;
4076 }
4077
4078
4079 //----------------------generate_virtual_guard---------------------------
4080 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4081 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4082 RegionNode* slow_region) {
4083 ciMethod* method = callee();
4084 int vtable_index = method->vtable_index();
4085 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4086 "bad index %d", vtable_index);
4087 // Get the Method* out of the appropriate vtable entry.
4088 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4089 vtable_index*vtableEntry::size_in_bytes() +
4090 vtableEntry::method_offset_in_bytes();
4091 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4092 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4093
4094 // Compare the target method with the expected method (e.g., Object.hashCode).
4095 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4096
4097 Node* native_call = makecon(native_call_addr);
4098 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4099 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4100
4101 return generate_slow_guard(test_native, slow_region);
4102 }
4103
4104 //-----------------------generate_method_call----------------------------
4105 // Use generate_method_call to make a slow-call to the real
4106 // method if the fast path fails. An alternative would be to
4107 // use a stub like OptoRuntime::slow_arraycopy_Java.
4108 // This only works for expanding the current library call,
4109 // not another intrinsic. (E.g., don't use this for making an
4110 // arraycopy call inside of the copyOf intrinsic.)
4111 CallJavaNode*
4112 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4113 // When compiling the intrinsic method itself, do not use this technique.
4114 guarantee(callee() != C->method(), "cannot make slow-call to self");
4115
4116 ciMethod* method = callee();
4117 // ensure the JVMS we have will be correct for this call
4118 guarantee(method_id == method->intrinsic_id(), "must match");
4119
4120 const TypeFunc* tf = TypeFunc::make(method);
4121 CallJavaNode* slow_call;
4122 if (is_static) {
4123 assert(!is_virtual, "");
4124 slow_call = new CallStaticJavaNode(C, tf,
4125 SharedRuntime::get_resolve_static_call_stub(),
4126 method, bci());
4127 } else if (is_virtual) {
4128 null_check_receiver();
4129 int vtable_index = Method::invalid_vtable_index;
4130 if (UseInlineCaches) {
4131 // Suppress the vtable call
4132 } else {
4133 // hashCode and clone are not a miranda methods,
4134 // so the vtable index is fixed.
4135 // No need to use the linkResolver to get it.
4136 vtable_index = method->vtable_index();
4137 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4138 "bad index %d", vtable_index);
4139 }
4140 slow_call = new CallDynamicJavaNode(tf,
4141 SharedRuntime::get_resolve_virtual_call_stub(),
4142 method, vtable_index, bci());
4143 } else { // neither virtual nor static: opt_virtual
4144 null_check_receiver();
4145 slow_call = new CallStaticJavaNode(C, tf,
4146 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4147 method, bci());
4148 slow_call->set_optimized_virtual(true);
4149 }
4150 set_arguments_for_java_call(slow_call);
4151 set_edges_for_java_call(slow_call);
4152 return slow_call;
4153 }
4154
4155
4156 /**
4157 * Build special case code for calls to hashCode on an object. This call may
4158 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4159 * slightly different code.
4160 */
4161 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4162 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4163 assert(!(is_virtual && is_static), "either virtual, special, or static");
4164
4165 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4166
4167 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4168 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4169 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4170 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4171 Node* obj = NULL;
4172 if (!is_static) {
4173 // Check for hashing null object
4174 obj = null_check_receiver();
4175 if (stopped()) return true; // unconditionally null
4176 result_reg->init_req(_null_path, top());
4177 result_val->init_req(_null_path, top());
4178 } else {
4179 // Do a null check, and return zero if null.
4180 // System.identityHashCode(null) == 0
4181 obj = argument(0);
4182 Node* null_ctl = top();
4183 obj = null_check_oop(obj, &null_ctl);
4184 result_reg->init_req(_null_path, null_ctl);
4185 result_val->init_req(_null_path, _gvn.intcon(0));
4186 }
4187
4188 // Unconditionally null? Then return right away.
4189 if (stopped()) {
4190 set_control( result_reg->in(_null_path));
4191 if (!stopped())
4192 set_result(result_val->in(_null_path));
4193 return true;
4194 }
4195
4196 // We only go to the fast case code if we pass a number of guards. The
4197 // paths which do not pass are accumulated in the slow_region.
4198 RegionNode* slow_region = new RegionNode(1);
4199 record_for_igvn(slow_region);
4200
4201 // If this is a virtual call, we generate a funny guard. We pull out
4202 // the vtable entry corresponding to hashCode() from the target object.
4203 // If the target method which we are calling happens to be the native
4204 // Object hashCode() method, we pass the guard. We do not need this
4205 // guard for non-virtual calls -- the caller is known to be the native
4206 // Object hashCode().
4207 if (is_virtual) {
4208 // After null check, get the object's klass.
4209 Node* obj_klass = load_object_klass(obj);
4210 generate_virtual_guard(obj_klass, slow_region);
4211 }
4212
4213 // Get the header out of the object, use LoadMarkNode when available
4214 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4215 // The control of the load must be NULL. Otherwise, the load can move before
4216 // the null check after castPP removal.
4217 Node* no_ctrl = NULL;
4218 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4219
4220 // Test the header to see if it is unlocked.
4221 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4222 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4223 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4224 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4225 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4226
4227 generate_slow_guard(test_unlocked, slow_region);
4228
4229 // Get the hash value and check to see that it has been properly assigned.
4230 // We depend on hash_mask being at most 32 bits and avoid the use of
4231 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4232 // vm: see markOop.hpp.
4233 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4234 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4235 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4236 // This hack lets the hash bits live anywhere in the mark object now, as long
4237 // as the shift drops the relevant bits into the low 32 bits. Note that
4238 // Java spec says that HashCode is an int so there's no point in capturing
4239 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4240 hshifted_header = ConvX2I(hshifted_header);
4241 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4242
4243 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4244 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4245 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4246
4247 generate_slow_guard(test_assigned, slow_region);
4248
4249 Node* init_mem = reset_memory();
4250 // fill in the rest of the null path:
4251 result_io ->init_req(_null_path, i_o());
4252 result_mem->init_req(_null_path, init_mem);
4253
4254 result_val->init_req(_fast_path, hash_val);
4255 result_reg->init_req(_fast_path, control());
4256 result_io ->init_req(_fast_path, i_o());
4257 result_mem->init_req(_fast_path, init_mem);
4258
4259 // Generate code for the slow case. We make a call to hashCode().
4260 set_control(_gvn.transform(slow_region));
4261 if (!stopped()) {
4262 // No need for PreserveJVMState, because we're using up the present state.
4263 set_all_memory(init_mem);
4264 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4265 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4266 Node* slow_result = set_results_for_java_call(slow_call);
4267 // this->control() comes from set_results_for_java_call
4268 result_reg->init_req(_slow_path, control());
4269 result_val->init_req(_slow_path, slow_result);
4270 result_io ->set_req(_slow_path, i_o());
4271 result_mem ->set_req(_slow_path, reset_memory());
4272 }
4273
4274 // Return the combined state.
4275 set_i_o( _gvn.transform(result_io) );
4276 set_all_memory( _gvn.transform(result_mem));
4277
4278 set_result(result_reg, result_val);
4279 return true;
4280 }
4281
4282 //---------------------------inline_native_getClass----------------------------
4283 // public final native Class<?> java.lang.Object.getClass();
4284 //
4285 // Build special case code for calls to getClass on an object.
4286 bool LibraryCallKit::inline_native_getClass() {
4287 Node* obj = null_check_receiver();
4288 if (stopped()) return true;
4289 set_result(load_mirror_from_klass(load_object_klass(obj)));
4290 return true;
4291 }
4292
4293 //-----------------inline_native_Reflection_getCallerClass---------------------
4294 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4295 //
4296 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4297 //
4298 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4299 // in that it must skip particular security frames and checks for
4300 // caller sensitive methods.
4301 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4302 #ifndef PRODUCT
4303 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4304 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4305 }
4306 #endif
4307
4308 if (!jvms()->has_method()) {
4309 #ifndef PRODUCT
4310 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4311 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4312 }
4313 #endif
4314 return false;
4315 }
4316
4317 // Walk back up the JVM state to find the caller at the required
4318 // depth.
4319 JVMState* caller_jvms = jvms();
4320
4321 // Cf. JVM_GetCallerClass
4322 // NOTE: Start the loop at depth 1 because the current JVM state does
4323 // not include the Reflection.getCallerClass() frame.
4324 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4325 ciMethod* m = caller_jvms->method();
4326 switch (n) {
4327 case 0:
4328 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4329 break;
4330 case 1:
4331 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4332 if (!m->caller_sensitive()) {
4333 #ifndef PRODUCT
4334 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4335 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4336 }
4337 #endif
4338 return false; // bail-out; let JVM_GetCallerClass do the work
4339 }
4340 break;
4341 default:
4342 if (!m->is_ignored_by_security_stack_walk()) {
4343 // We have reached the desired frame; return the holder class.
4344 // Acquire method holder as java.lang.Class and push as constant.
4345 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4346 ciInstance* caller_mirror = caller_klass->java_mirror();
4347 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4348
4349 #ifndef PRODUCT
4350 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4351 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());
4352 tty->print_cr(" JVM state at this point:");
4353 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4354 ciMethod* m = jvms()->of_depth(i)->method();
4355 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4356 }
4357 }
4358 #endif
4359 return true;
4360 }
4361 break;
4362 }
4363 }
4364
4365 #ifndef PRODUCT
4366 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4367 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4368 tty->print_cr(" JVM state at this point:");
4369 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4370 ciMethod* m = jvms()->of_depth(i)->method();
4371 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4372 }
4373 }
4374 #endif
4375
4376 return false; // bail-out; let JVM_GetCallerClass do the work
4377 }
4378
4379 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4380 Node* arg = argument(0);
4381 Node* result = NULL;
4382
4383 switch (id) {
4384 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4385 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4386 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4387 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4388
4389 case vmIntrinsics::_doubleToLongBits: {
4390 // two paths (plus control) merge in a wood
4391 RegionNode *r = new RegionNode(3);
4392 Node *phi = new PhiNode(r, TypeLong::LONG);
4393
4394 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4395 // Build the boolean node
4396 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4397
4398 // Branch either way.
4399 // NaN case is less traveled, which makes all the difference.
4400 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4401 Node *opt_isnan = _gvn.transform(ifisnan);
4402 assert( opt_isnan->is_If(), "Expect an IfNode");
4403 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4404 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4405
4406 set_control(iftrue);
4407
4408 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4409 Node *slow_result = longcon(nan_bits); // return NaN
4410 phi->init_req(1, _gvn.transform( slow_result ));
4411 r->init_req(1, iftrue);
4412
4413 // Else fall through
4414 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4415 set_control(iffalse);
4416
4417 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4418 r->init_req(2, iffalse);
4419
4420 // Post merge
4421 set_control(_gvn.transform(r));
4422 record_for_igvn(r);
4423
4424 C->set_has_split_ifs(true); // Has chance for split-if optimization
4425 result = phi;
4426 assert(result->bottom_type()->isa_long(), "must be");
4427 break;
4428 }
4429
4430 case vmIntrinsics::_floatToIntBits: {
4431 // two paths (plus control) merge in a wood
4432 RegionNode *r = new RegionNode(3);
4433 Node *phi = new PhiNode(r, TypeInt::INT);
4434
4435 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4436 // Build the boolean node
4437 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4438
4439 // Branch either way.
4440 // NaN case is less traveled, which makes all the difference.
4441 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4442 Node *opt_isnan = _gvn.transform(ifisnan);
4443 assert( opt_isnan->is_If(), "Expect an IfNode");
4444 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4445 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4446
4447 set_control(iftrue);
4448
4449 static const jint nan_bits = 0x7fc00000;
4450 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4451 phi->init_req(1, _gvn.transform( slow_result ));
4452 r->init_req(1, iftrue);
4453
4454 // Else fall through
4455 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4456 set_control(iffalse);
4457
4458 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4459 r->init_req(2, iffalse);
4460
4461 // Post merge
4462 set_control(_gvn.transform(r));
4463 record_for_igvn(r);
4464
4465 C->set_has_split_ifs(true); // Has chance for split-if optimization
4466 result = phi;
4467 assert(result->bottom_type()->isa_int(), "must be");
4468 break;
4469 }
4470
4471 default:
4472 fatal_unexpected_iid(id);
4473 break;
4474 }
4475 set_result(_gvn.transform(result));
4476 return true;
4477 }
4478
4479 //----------------------inline_unsafe_copyMemory-------------------------
4480 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4481 bool LibraryCallKit::inline_unsafe_copyMemory() {
4482 if (callee()->is_static()) return false; // caller must have the capability!
4483 null_check_receiver(); // null-check receiver
4484 if (stopped()) return true;
4485
4486 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4487
4488 Node* src_ptr = argument(1); // type: oop
4489 Node* src_off = ConvL2X(argument(2)); // type: long
4490 Node* dst_ptr = argument(4); // type: oop
4491 Node* dst_off = ConvL2X(argument(5)); // type: long
4492 Node* size = ConvL2X(argument(7)); // type: long
4493
4494 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4495 "fieldOffset must be byte-scaled");
4496
4497 Node* src = make_unsafe_address(src_ptr, src_off);
4498 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4499
4500 // Conservatively insert a memory barrier on all memory slices.
4501 // Do not let writes of the copy source or destination float below the copy.
4502 insert_mem_bar(Op_MemBarCPUOrder);
4503
4504 // Call it. Note that the length argument is not scaled.
4505 make_runtime_call(RC_LEAF|RC_NO_FP,
4506 OptoRuntime::fast_arraycopy_Type(),
4507 StubRoutines::unsafe_arraycopy(),
4508 "unsafe_arraycopy",
4509 TypeRawPtr::BOTTOM,
4510 src, dst, size XTOP);
4511
4512 // Do not let reads of the copy destination float above the copy.
4513 insert_mem_bar(Op_MemBarCPUOrder);
4514
4515 return true;
4516 }
4517
4518 //------------------------clone_coping-----------------------------------
4519 // Helper function for inline_native_clone.
4520 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4521 assert(obj_size != NULL, "");
4522 Node* raw_obj = alloc_obj->in(1);
4523 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4524
4525 AllocateNode* alloc = NULL;
4526 if (ReduceBulkZeroing) {
4527 // We will be completely responsible for initializing this object -
4528 // mark Initialize node as complete.
4529 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4530 // The object was just allocated - there should be no any stores!
4531 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4532 // Mark as complete_with_arraycopy so that on AllocateNode
4533 // expansion, we know this AllocateNode is initialized by an array
4534 // copy and a StoreStore barrier exists after the array copy.
4535 alloc->initialization()->set_complete_with_arraycopy();
4536 }
4537
4538 // Copy the fastest available way.
4539 // TODO: generate fields copies for small objects instead.
4540 Node* src = obj;
4541 Node* dest = alloc_obj;
4542 Node* size = _gvn.transform(obj_size);
4543
4544 // Exclude the header but include array length to copy by 8 bytes words.
4545 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4546 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4547 instanceOopDesc::base_offset_in_bytes();
4548 // base_off:
4549 // 8 - 32-bit VM
4550 // 12 - 64-bit VM, compressed klass
4551 // 16 - 64-bit VM, normal klass
4552 if (base_off % BytesPerLong != 0) {
4553 assert(UseCompressedClassPointers, "");
4554 if (is_array) {
4555 // Exclude length to copy by 8 bytes words.
4556 base_off += sizeof(int);
4557 } else {
4558 // Include klass to copy by 8 bytes words.
4559 base_off = instanceOopDesc::klass_offset_in_bytes();
4560 }
4561 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4562 }
4563 src = basic_plus_adr(src, base_off);
4564 dest = basic_plus_adr(dest, base_off);
4565
4566 // Compute the length also, if needed:
4567 Node* countx = size;
4568 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4569 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4570
4571 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4572
4573 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false, false);
4574 ac->set_clonebasic();
4575 Node* n = _gvn.transform(ac);
4576 if (n == ac) {
4577 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4578 } else {
4579 set_all_memory(n);
4580 }
4581
4582 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4583 if (card_mark) {
4584 assert(!is_array, "");
4585 // Put in store barrier for any and all oops we are sticking
4586 // into this object. (We could avoid this if we could prove
4587 // that the object type contains no oop fields at all.)
4588 Node* no_particular_value = NULL;
4589 Node* no_particular_field = NULL;
4590 int raw_adr_idx = Compile::AliasIdxRaw;
4591 post_barrier(control(),
4592 memory(raw_adr_type),
4593 alloc_obj,
4594 no_particular_field,
4595 raw_adr_idx,
4596 no_particular_value,
4597 T_OBJECT,
4598 false);
4599 }
4600
4601 // Do not let reads from the cloned object float above the arraycopy.
4602 if (alloc != NULL) {
4603 // Do not let stores that initialize this object be reordered with
4604 // a subsequent store that would make this object accessible by
4605 // other threads.
4606 // Record what AllocateNode this StoreStore protects so that
4607 // escape analysis can go from the MemBarStoreStoreNode to the
4608 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4609 // based on the escape status of the AllocateNode.
4610 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4611 } else {
4612 insert_mem_bar(Op_MemBarCPUOrder);
4613 }
4614 }
4615
4616 //------------------------inline_native_clone----------------------------
4617 // protected native Object java.lang.Object.clone();
4618 //
4619 // Here are the simple edge cases:
4620 // null receiver => normal trap
4621 // virtual and clone was overridden => slow path to out-of-line clone
4622 // not cloneable or finalizer => slow path to out-of-line Object.clone
4623 //
4624 // The general case has two steps, allocation and copying.
4625 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4626 //
4627 // Copying also has two cases, oop arrays and everything else.
4628 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4629 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4630 //
4631 // These steps fold up nicely if and when the cloned object's klass
4632 // can be sharply typed as an object array, a type array, or an instance.
4633 //
4634 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4635 PhiNode* result_val;
4636
4637 // Set the reexecute bit for the interpreter to reexecute
4638 // the bytecode that invokes Object.clone if deoptimization happens.
4639 { PreserveReexecuteState preexecs(this);
4640 jvms()->set_should_reexecute(true);
4641
4642 Node* obj = null_check_receiver();
4643 if (stopped()) return true;
4644
4645 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4646
4647 // If we are going to clone an instance, we need its exact type to
4648 // know the number and types of fields to convert the clone to
4649 // loads/stores. Maybe a speculative type can help us.
4650 if (!obj_type->klass_is_exact() &&
4651 obj_type->speculative_type() != NULL &&
4652 obj_type->speculative_type()->is_instance_klass()) {
4653 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4654 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4655 !spec_ik->has_injected_fields()) {
4656 ciKlass* k = obj_type->klass();
4657 if (!k->is_instance_klass() ||
4658 k->as_instance_klass()->is_interface() ||
4659 k->as_instance_klass()->has_subklass()) {
4660 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4661 }
4662 }
4663 }
4664
4665 Node* obj_klass = load_object_klass(obj);
4666 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4667 const TypeOopPtr* toop = ((tklass != NULL)
4668 ? tklass->as_instance_type()
4669 : TypeInstPtr::NOTNULL);
4670
4671 // Conservatively insert a memory barrier on all memory slices.
4672 // Do not let writes into the original float below the clone.
4673 insert_mem_bar(Op_MemBarCPUOrder);
4674
4675 // paths into result_reg:
4676 enum {
4677 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4678 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4679 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4680 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4681 PATH_LIMIT
4682 };
4683 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4684 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4685 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4686 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4687 record_for_igvn(result_reg);
4688
4689 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4690 int raw_adr_idx = Compile::AliasIdxRaw;
4691
4692 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4693 if (array_ctl != NULL) {
4694 // It's an array.
4695 PreserveJVMState pjvms(this);
4696 set_control(array_ctl);
4697 Node* obj_length = load_array_length(obj);
4698 Node* obj_size = NULL;
4699 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4700
4701 if (!use_ReduceInitialCardMarks()) {
4702 // If it is an oop array, it requires very special treatment,
4703 // because card marking is required on each card of the array.
4704 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4705 if (is_obja != NULL) {
4706 PreserveJVMState pjvms2(this);
4707 set_control(is_obja);
4708 // Generate a direct call to the right arraycopy function(s).
4709 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4710 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4711 ac->set_cloneoop();
4712 Node* n = _gvn.transform(ac);
4713 assert(n == ac, "cannot disappear");
4714 ac->connect_outputs(this);
4715
4716 result_reg->init_req(_objArray_path, control());
4717 result_val->init_req(_objArray_path, alloc_obj);
4718 result_i_o ->set_req(_objArray_path, i_o());
4719 result_mem ->set_req(_objArray_path, reset_memory());
4720 }
4721 }
4722 // Otherwise, there are no card marks to worry about.
4723 // (We can dispense with card marks if we know the allocation
4724 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4725 // causes the non-eden paths to take compensating steps to
4726 // simulate a fresh allocation, so that no further
4727 // card marks are required in compiled code to initialize
4728 // the object.)
4729
4730 if (!stopped()) {
4731 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4732
4733 // Present the results of the copy.
4734 result_reg->init_req(_array_path, control());
4735 result_val->init_req(_array_path, alloc_obj);
4736 result_i_o ->set_req(_array_path, i_o());
4737 result_mem ->set_req(_array_path, reset_memory());
4738 }
4739 }
4740
4741 // We only go to the instance fast case code if we pass a number of guards.
4742 // The paths which do not pass are accumulated in the slow_region.
4743 RegionNode* slow_region = new RegionNode(1);
4744 record_for_igvn(slow_region);
4745 if (!stopped()) {
4746 // It's an instance (we did array above). Make the slow-path tests.
4747 // If this is a virtual call, we generate a funny guard. We grab
4748 // the vtable entry corresponding to clone() from the target object.
4749 // If the target method which we are calling happens to be the
4750 // Object clone() method, we pass the guard. We do not need this
4751 // guard for non-virtual calls; the caller is known to be the native
4752 // Object clone().
4753 if (is_virtual) {
4754 generate_virtual_guard(obj_klass, slow_region);
4755 }
4756
4757 // The object must be easily cloneable and must not have a finalizer.
4758 // Both of these conditions may be checked in a single test.
4759 // We could optimize the test further, but we don't care.
4760 generate_access_flags_guard(obj_klass,
4761 // Test both conditions:
4762 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4763 // Must be cloneable but not finalizer:
4764 JVM_ACC_IS_CLONEABLE_FAST,
4765 slow_region);
4766 }
4767
4768 if (!stopped()) {
4769 // It's an instance, and it passed the slow-path tests.
4770 PreserveJVMState pjvms(this);
4771 Node* obj_size = NULL;
4772 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4773 // is reexecuted if deoptimization occurs and there could be problems when merging
4774 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4775 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4776
4777 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4778
4779 // Present the results of the slow call.
4780 result_reg->init_req(_instance_path, control());
4781 result_val->init_req(_instance_path, alloc_obj);
4782 result_i_o ->set_req(_instance_path, i_o());
4783 result_mem ->set_req(_instance_path, reset_memory());
4784 }
4785
4786 // Generate code for the slow case. We make a call to clone().
4787 set_control(_gvn.transform(slow_region));
4788 if (!stopped()) {
4789 PreserveJVMState pjvms(this);
4790 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4791 Node* slow_result = set_results_for_java_call(slow_call);
4792 // this->control() comes from set_results_for_java_call
4793 result_reg->init_req(_slow_path, control());
4794 result_val->init_req(_slow_path, slow_result);
4795 result_i_o ->set_req(_slow_path, i_o());
4796 result_mem ->set_req(_slow_path, reset_memory());
4797 }
4798
4799 // Return the combined state.
4800 set_control( _gvn.transform(result_reg));
4801 set_i_o( _gvn.transform(result_i_o));
4802 set_all_memory( _gvn.transform(result_mem));
4803 } // original reexecute is set back here
4804
4805 set_result(_gvn.transform(result_val));
4806 return true;
4807 }
4808
4809 // If we have a tighly coupled allocation, the arraycopy may take care
4810 // of the array initialization. If one of the guards we insert between
4811 // the allocation and the arraycopy causes a deoptimization, an
4812 // unitialized array will escape the compiled method. To prevent that
4813 // we set the JVM state for uncommon traps between the allocation and
4814 // the arraycopy to the state before the allocation so, in case of
4815 // deoptimization, we'll reexecute the allocation and the
4816 // initialization.
4817 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4818 if (alloc != NULL) {
4819 ciMethod* trap_method = alloc->jvms()->method();
4820 int trap_bci = alloc->jvms()->bci();
4821
4822 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4823 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4824 // Make sure there's no store between the allocation and the
4825 // arraycopy otherwise visible side effects could be rexecuted
4826 // in case of deoptimization and cause incorrect execution.
4827 bool no_interfering_store = true;
4828 Node* mem = alloc->in(TypeFunc::Memory);
4829 if (mem->is_MergeMem()) {
4830 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4831 Node* n = mms.memory();
4832 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4833 assert(n->is_Store(), "what else?");
4834 no_interfering_store = false;
4835 break;
4836 }
4837 }
4838 } else {
4839 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4840 Node* n = mms.memory();
4841 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4842 assert(n->is_Store(), "what else?");
4843 no_interfering_store = false;
4844 break;
4845 }
4846 }
4847 }
4848
4849 if (no_interfering_store) {
4850 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4851 uint size = alloc->req();
4852 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4853 old_jvms->set_map(sfpt);
4854 for (uint i = 0; i < size; i++) {
4855 sfpt->init_req(i, alloc->in(i));
4856 }
4857 // re-push array length for deoptimization
4858 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4859 old_jvms->set_sp(old_jvms->sp()+1);
4860 old_jvms->set_monoff(old_jvms->monoff()+1);
4861 old_jvms->set_scloff(old_jvms->scloff()+1);
4862 old_jvms->set_endoff(old_jvms->endoff()+1);
4863 old_jvms->set_should_reexecute(true);
4864
4865 sfpt->set_i_o(map()->i_o());
4866 sfpt->set_memory(map()->memory());
4867 sfpt->set_control(map()->control());
4868
4869 JVMState* saved_jvms = jvms();
4870 saved_reexecute_sp = _reexecute_sp;
4871
4872 set_jvms(sfpt->jvms());
4873 _reexecute_sp = jvms()->sp();
4874
4875 return saved_jvms;
4876 }
4877 }
4878 }
4879 return NULL;
4880 }
4881
4882 // In case of a deoptimization, we restart execution at the
4883 // allocation, allocating a new array. We would leave an uninitialized
4884 // array in the heap that GCs wouldn't expect. Move the allocation
4885 // after the traps so we don't allocate the array if we
4886 // deoptimize. This is possible because tightly_coupled_allocation()
4887 // guarantees there's no observer of the allocated array at this point
4888 // and the control flow is simple enough.
4889 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4890 int saved_reexecute_sp, uint new_idx) {
4891 if (saved_jvms != NULL && !stopped()) {
4892 assert(alloc != NULL, "only with a tightly coupled allocation");
4893 // restore JVM state to the state at the arraycopy
4894 saved_jvms->map()->set_control(map()->control());
4895 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4896 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4897 // If we've improved the types of some nodes (null check) while
4898 // emitting the guards, propagate them to the current state
4899 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4900 set_jvms(saved_jvms);
4901 _reexecute_sp = saved_reexecute_sp;
4902
4903 // Remove the allocation from above the guards
4904 CallProjections callprojs;
4905 alloc->extract_projections(&callprojs, true);
4906 InitializeNode* init = alloc->initialization();
4907 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4908 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4909 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4910 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4911
4912 // move the allocation here (after the guards)
4913 _gvn.hash_delete(alloc);
4914 alloc->set_req(TypeFunc::Control, control());
4915 alloc->set_req(TypeFunc::I_O, i_o());
4916 Node *mem = reset_memory();
4917 set_all_memory(mem);
4918 alloc->set_req(TypeFunc::Memory, mem);
4919 set_control(init->proj_out(TypeFunc::Control));
4920 set_i_o(callprojs.fallthrough_ioproj);
4921
4922 // Update memory as done in GraphKit::set_output_for_allocation()
4923 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4924 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4925 if (ary_type->isa_aryptr() && length_type != NULL) {
4926 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4927 }
4928 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4929 int elemidx = C->get_alias_index(telemref);
4930 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4931 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4932
4933 Node* allocx = _gvn.transform(alloc);
4934 assert(allocx == alloc, "where has the allocation gone?");
4935 assert(dest->is_CheckCastPP(), "not an allocation result?");
4936
4937 _gvn.hash_delete(dest);
4938 dest->set_req(0, control());
4939 Node* destx = _gvn.transform(dest);
4940 assert(destx == dest, "where has the allocation result gone?");
4941 }
4942 }
4943
4944
4945 //------------------------------inline_arraycopy-----------------------
4946 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4947 // Object dest, int destPos,
4948 // int length);
4949 bool LibraryCallKit::inline_arraycopy() {
4950 // Get the arguments.
4951 Node* src = argument(0); // type: oop
4952 Node* src_offset = argument(1); // type: int
4953 Node* dest = argument(2); // type: oop
4954 Node* dest_offset = argument(3); // type: int
4955 Node* length = argument(4); // type: int
4956
4957 uint new_idx = C->unique();
4958
4959 // Check for allocation before we add nodes that would confuse
4960 // tightly_coupled_allocation()
4961 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4962
4963 int saved_reexecute_sp = -1;
4964 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4965 // See arraycopy_restore_alloc_state() comment
4966 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4967 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4968 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4969 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4970
4971 // The following tests must be performed
4972 // (1) src and dest are arrays.
4973 // (2) src and dest arrays must have elements of the same BasicType
4974 // (3) src and dest must not be null.
4975 // (4) src_offset must not be negative.
4976 // (5) dest_offset must not be negative.
4977 // (6) length must not be negative.
4978 // (7) src_offset + length must not exceed length of src.
4979 // (8) dest_offset + length must not exceed length of dest.
4980 // (9) each element of an oop array must be assignable
4981
4982 // (3) src and dest must not be null.
4983 // always do this here because we need the JVM state for uncommon traps
4984 Node* null_ctl = top();
4985 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4986 assert(null_ctl->is_top(), "no null control here");
4987 dest = null_check(dest, T_ARRAY);
4988
4989 if (!can_emit_guards) {
4990 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4991 // guards but the arraycopy node could still take advantage of a
4992 // tightly allocated allocation. tightly_coupled_allocation() is
4993 // called again to make sure it takes the null check above into
4994 // account: the null check is mandatory and if it caused an
4995 // uncommon trap to be emitted then the allocation can't be
4996 // considered tightly coupled in this context.
4997 alloc = tightly_coupled_allocation(dest, NULL);
4998 }
4999
5000 bool validated = false;
5001
5002 const Type* src_type = _gvn.type(src);
5003 const Type* dest_type = _gvn.type(dest);
5004 const TypeAryPtr* top_src = src_type->isa_aryptr();
5005 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5006
5007 // Do we have the type of src?
5008 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5009 // Do we have the type of dest?
5010 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5011 // Is the type for src from speculation?
5012 bool src_spec = false;
5013 // Is the type for dest from speculation?
5014 bool dest_spec = false;
5015
5016 if ((!has_src || !has_dest) && can_emit_guards) {
5017 // We don't have sufficient type information, let's see if
5018 // speculative types can help. We need to have types for both src
5019 // and dest so that it pays off.
5020
5021 // Do we already have or could we have type information for src
5022 bool could_have_src = has_src;
5023 // Do we already have or could we have type information for dest
5024 bool could_have_dest = has_dest;
5025
5026 ciKlass* src_k = NULL;
5027 if (!has_src) {
5028 src_k = src_type->speculative_type_not_null();
5029 if (src_k != NULL && src_k->is_array_klass()) {
5030 could_have_src = true;
5031 }
5032 }
5033
5034 ciKlass* dest_k = NULL;
5035 if (!has_dest) {
5036 dest_k = dest_type->speculative_type_not_null();
5037 if (dest_k != NULL && dest_k->is_array_klass()) {
5038 could_have_dest = true;
5039 }
5040 }
5041
5042 if (could_have_src && could_have_dest) {
5043 // This is going to pay off so emit the required guards
5044 if (!has_src) {
5045 src = maybe_cast_profiled_obj(src, src_k, true);
5046 src_type = _gvn.type(src);
5047 top_src = src_type->isa_aryptr();
5048 has_src = (top_src != NULL && top_src->klass() != NULL);
5049 src_spec = true;
5050 }
5051 if (!has_dest) {
5052 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5053 dest_type = _gvn.type(dest);
5054 top_dest = dest_type->isa_aryptr();
5055 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5056 dest_spec = true;
5057 }
5058 }
5059 }
5060
5061 if (has_src && has_dest && can_emit_guards) {
5062 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5063 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5064 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5065 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5066
5067 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5068 // If both arrays are object arrays then having the exact types
5069 // for both will remove the need for a subtype check at runtime
5070 // before the call and may make it possible to pick a faster copy
5071 // routine (without a subtype check on every element)
5072 // Do we have the exact type of src?
5073 bool could_have_src = src_spec;
5074 // Do we have the exact type of dest?
5075 bool could_have_dest = dest_spec;
5076 ciKlass* src_k = top_src->klass();
5077 ciKlass* dest_k = top_dest->klass();
5078 if (!src_spec) {
5079 src_k = src_type->speculative_type_not_null();
5080 if (src_k != NULL && src_k->is_array_klass()) {
5081 could_have_src = true;
5082 }
5083 }
5084 if (!dest_spec) {
5085 dest_k = dest_type->speculative_type_not_null();
5086 if (dest_k != NULL && dest_k->is_array_klass()) {
5087 could_have_dest = true;
5088 }
5089 }
5090 if (could_have_src && could_have_dest) {
5091 // If we can have both exact types, emit the missing guards
5092 if (could_have_src && !src_spec) {
5093 src = maybe_cast_profiled_obj(src, src_k, true);
5094 }
5095 if (could_have_dest && !dest_spec) {
5096 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5097 }
5098 }
5099 }
5100 }
5101
5102 ciMethod* trap_method = method();
5103 int trap_bci = bci();
5104 if (saved_jvms != NULL) {
5105 trap_method = alloc->jvms()->method();
5106 trap_bci = alloc->jvms()->bci();
5107 }
5108
5109 bool negative_length_guard_generated = false;
5110
5111 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5112 can_emit_guards &&
5113 !src->is_top() && !dest->is_top()) {
5114 // validate arguments: enables transformation the ArrayCopyNode
5115 validated = true;
5116
5117 RegionNode* slow_region = new RegionNode(1);
5118 record_for_igvn(slow_region);
5119
5120 // (1) src and dest are arrays.
5121 generate_non_array_guard(load_object_klass(src), slow_region);
5122 generate_non_array_guard(load_object_klass(dest), slow_region);
5123
5124 // (2) src and dest arrays must have elements of the same BasicType
5125 // done at macro expansion or at Ideal transformation time
5126
5127 // (4) src_offset must not be negative.
5128 generate_negative_guard(src_offset, slow_region);
5129
5130 // (5) dest_offset must not be negative.
5131 generate_negative_guard(dest_offset, slow_region);
5132
5133 // (7) src_offset + length must not exceed length of src.
5134 generate_limit_guard(src_offset, length,
5135 load_array_length(src),
5136 slow_region);
5137
5138 // (8) dest_offset + length must not exceed length of dest.
5139 generate_limit_guard(dest_offset, length,
5140 load_array_length(dest),
5141 slow_region);
5142
5143 // (6) length must not be negative.
5144 // This is also checked in generate_arraycopy() during macro expansion, but
5145 // we also have to check it here for the case where the ArrayCopyNode will
5146 // be eliminated by Escape Analysis.
5147 if (EliminateAllocations) {
5148 generate_negative_guard(length, slow_region);
5149 negative_length_guard_generated = true;
5150 }
5151
5152 // (9) each element of an oop array must be assignable
5153 Node* src_klass = load_object_klass(src);
5154 Node* dest_klass = load_object_klass(dest);
5155 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5156
5157 if (not_subtype_ctrl != top()) {
5158 PreserveJVMState pjvms(this);
5159 set_control(not_subtype_ctrl);
5160 uncommon_trap(Deoptimization::Reason_intrinsic,
5161 Deoptimization::Action_make_not_entrant);
5162 assert(stopped(), "Should be stopped");
5163 }
5164 {
5165 PreserveJVMState pjvms(this);
5166 set_control(_gvn.transform(slow_region));
5167 uncommon_trap(Deoptimization::Reason_intrinsic,
5168 Deoptimization::Action_make_not_entrant);
5169 assert(stopped(), "Should be stopped");
5170 }
5171 }
5172
5173 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5174
5175 if (stopped()) {
5176 return true;
5177 }
5178
5179 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5180 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5181 // so the compiler has a chance to eliminate them: during macro expansion,
5182 // we have to set their control (CastPP nodes are eliminated).
5183 load_object_klass(src), load_object_klass(dest),
5184 load_array_length(src), load_array_length(dest));
5185
5186 ac->set_arraycopy(validated);
5187
5188 Node* n = _gvn.transform(ac);
5189 if (n == ac) {
5190 ac->connect_outputs(this);
5191 } else {
5192 assert(validated, "shouldn't transform if all arguments not validated");
5193 set_all_memory(n);
5194 }
5195
5196 return true;
5197 }
5198
5199
5200 // Helper function which determines if an arraycopy immediately follows
5201 // an allocation, with no intervening tests or other escapes for the object.
5202 AllocateArrayNode*
5203 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5204 RegionNode* slow_region) {
5205 if (stopped()) return NULL; // no fast path
5206 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5207
5208 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5209 if (alloc == NULL) return NULL;
5210
5211 Node* rawmem = memory(Compile::AliasIdxRaw);
5212 // Is the allocation's memory state untouched?
5213 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5214 // Bail out if there have been raw-memory effects since the allocation.
5215 // (Example: There might have been a call or safepoint.)
5216 return NULL;
5217 }
5218 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5219 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5220 return NULL;
5221 }
5222
5223 // There must be no unexpected observers of this allocation.
5224 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5225 Node* obs = ptr->fast_out(i);
5226 if (obs != this->map()) {
5227 return NULL;
5228 }
5229 }
5230
5231 // This arraycopy must unconditionally follow the allocation of the ptr.
5232 Node* alloc_ctl = ptr->in(0);
5233 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5234
5235 Node* ctl = control();
5236 while (ctl != alloc_ctl) {
5237 // There may be guards which feed into the slow_region.
5238 // Any other control flow means that we might not get a chance
5239 // to finish initializing the allocated object.
5240 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5241 IfNode* iff = ctl->in(0)->as_If();
5242 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5243 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5244 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5245 ctl = iff->in(0); // This test feeds the known slow_region.
5246 continue;
5247 }
5248 // One more try: Various low-level checks bottom out in
5249 // uncommon traps. If the debug-info of the trap omits
5250 // any reference to the allocation, as we've already
5251 // observed, then there can be no objection to the trap.
5252 bool found_trap = false;
5253 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5254 Node* obs = not_ctl->fast_out(j);
5255 if (obs->in(0) == not_ctl && obs->is_Call() &&
5256 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5257 found_trap = true; break;
5258 }
5259 }
5260 if (found_trap) {
5261 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5262 continue;
5263 }
5264 }
5265 return NULL;
5266 }
5267
5268 // If we get this far, we have an allocation which immediately
5269 // precedes the arraycopy, and we can take over zeroing the new object.
5270 // The arraycopy will finish the initialization, and provide
5271 // a new control state to which we will anchor the destination pointer.
5272
5273 return alloc;
5274 }
5275
5276 //-------------inline_encodeISOArray-----------------------------------
5277 // encode char[] to byte[] in ISO_8859_1
5278 bool LibraryCallKit::inline_encodeISOArray() {
5279 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5280 // no receiver since it is static method
5281 Node *src = argument(0);
5282 Node *src_offset = argument(1);
5283 Node *dst = argument(2);
5284 Node *dst_offset = argument(3);
5285 Node *length = argument(4);
5286
5287 const Type* src_type = src->Value(&_gvn);
5288 const Type* dst_type = dst->Value(&_gvn);
5289 const TypeAryPtr* top_src = src_type->isa_aryptr();
5290 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5291 if (top_src == NULL || top_src->klass() == NULL ||
5292 top_dest == NULL || top_dest->klass() == NULL) {
5293 // failed array check
5294 return false;
5295 }
5296
5297 // Figure out the size and type of the elements we will be copying.
5298 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5299 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5300 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5301 return false;
5302 }
5303
5304 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5305 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5306 // 'src_start' points to src array + scaled offset
5307 // 'dst_start' points to dst array + scaled offset
5308
5309 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5310 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5311 enc = _gvn.transform(enc);
5312 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5313 set_memory(res_mem, mtype);
5314 set_result(enc);
5315 return true;
5316 }
5317
5318 //-------------inline_multiplyToLen-----------------------------------
5319 bool LibraryCallKit::inline_multiplyToLen() {
5320 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5321
5322 address stubAddr = StubRoutines::multiplyToLen();
5323 if (stubAddr == NULL) {
5324 return false; // Intrinsic's stub is not implemented on this platform
5325 }
5326 const char* stubName = "multiplyToLen";
5327
5328 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5329
5330 // no receiver because it is a static method
5331 Node* x = argument(0);
5332 Node* xlen = argument(1);
5333 Node* y = argument(2);
5334 Node* ylen = argument(3);
5335 Node* z = argument(4);
5336
5337 const Type* x_type = x->Value(&_gvn);
5338 const Type* y_type = y->Value(&_gvn);
5339 const TypeAryPtr* top_x = x_type->isa_aryptr();
5340 const TypeAryPtr* top_y = y_type->isa_aryptr();
5341 if (top_x == NULL || top_x->klass() == NULL ||
5342 top_y == NULL || top_y->klass() == NULL) {
5343 // failed array check
5344 return false;
5345 }
5346
5347 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5348 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5349 if (x_elem != T_INT || y_elem != T_INT) {
5350 return false;
5351 }
5352
5353 // Set the original stack and the reexecute bit for the interpreter to reexecute
5354 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5355 // on the return from z array allocation in runtime.
5356 { PreserveReexecuteState preexecs(this);
5357 jvms()->set_should_reexecute(true);
5358
5359 Node* x_start = array_element_address(x, intcon(0), x_elem);
5360 Node* y_start = array_element_address(y, intcon(0), y_elem);
5361 // 'x_start' points to x array + scaled xlen
5362 // 'y_start' points to y array + scaled ylen
5363
5364 // Allocate the result array
5365 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5366 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5367 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5368
5369 IdealKit ideal(this);
5370
5371 #define __ ideal.
5372 Node* one = __ ConI(1);
5373 Node* zero = __ ConI(0);
5374 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5375 __ set(need_alloc, zero);
5376 __ set(z_alloc, z);
5377 __ if_then(z, BoolTest::eq, null()); {
5378 __ increment (need_alloc, one);
5379 } __ else_(); {
5380 // Update graphKit memory and control from IdealKit.
5381 sync_kit(ideal);
5382 Node* zlen_arg = load_array_length(z);
5383 // Update IdealKit memory and control from graphKit.
5384 __ sync_kit(this);
5385 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5386 __ increment (need_alloc, one);
5387 } __ end_if();
5388 } __ end_if();
5389
5390 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5391 // Update graphKit memory and control from IdealKit.
5392 sync_kit(ideal);
5393 Node * narr = new_array(klass_node, zlen, 1);
5394 // Update IdealKit memory and control from graphKit.
5395 __ sync_kit(this);
5396 __ set(z_alloc, narr);
5397 } __ end_if();
5398
5399 sync_kit(ideal);
5400 z = __ value(z_alloc);
5401 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5402 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5403 // Final sync IdealKit and GraphKit.
5404 final_sync(ideal);
5405 #undef __
5406
5407 Node* z_start = array_element_address(z, intcon(0), T_INT);
5408
5409 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5410 OptoRuntime::multiplyToLen_Type(),
5411 stubAddr, stubName, TypePtr::BOTTOM,
5412 x_start, xlen, y_start, ylen, z_start, zlen);
5413 } // original reexecute is set back here
5414
5415 C->set_has_split_ifs(true); // Has chance for split-if optimization
5416 set_result(z);
5417 return true;
5418 }
5419
5420 //-------------inline_squareToLen------------------------------------
5421 bool LibraryCallKit::inline_squareToLen() {
5422 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5423
5424 address stubAddr = StubRoutines::squareToLen();
5425 if (stubAddr == NULL) {
5426 return false; // Intrinsic's stub is not implemented on this platform
5427 }
5428 const char* stubName = "squareToLen";
5429
5430 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5431
5432 Node* x = argument(0);
5433 Node* len = argument(1);
5434 Node* z = argument(2);
5435 Node* zlen = argument(3);
5436
5437 const Type* x_type = x->Value(&_gvn);
5438 const Type* z_type = z->Value(&_gvn);
5439 const TypeAryPtr* top_x = x_type->isa_aryptr();
5440 const TypeAryPtr* top_z = z_type->isa_aryptr();
5441 if (top_x == NULL || top_x->klass() == NULL ||
5442 top_z == NULL || top_z->klass() == NULL) {
5443 // failed array check
5444 return false;
5445 }
5446
5447 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5448 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5449 if (x_elem != T_INT || z_elem != T_INT) {
5450 return false;
5451 }
5452
5453
5454 Node* x_start = array_element_address(x, intcon(0), x_elem);
5455 Node* z_start = array_element_address(z, intcon(0), z_elem);
5456
5457 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5458 OptoRuntime::squareToLen_Type(),
5459 stubAddr, stubName, TypePtr::BOTTOM,
5460 x_start, len, z_start, zlen);
5461
5462 set_result(z);
5463 return true;
5464 }
5465
5466 //-------------inline_mulAdd------------------------------------------
5467 bool LibraryCallKit::inline_mulAdd() {
5468 assert(UseMulAddIntrinsic, "not implemented on this platform");
5469
5470 address stubAddr = StubRoutines::mulAdd();
5471 if (stubAddr == NULL) {
5472 return false; // Intrinsic's stub is not implemented on this platform
5473 }
5474 const char* stubName = "mulAdd";
5475
5476 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5477
5478 Node* out = argument(0);
5479 Node* in = argument(1);
5480 Node* offset = argument(2);
5481 Node* len = argument(3);
5482 Node* k = argument(4);
5483
5484 const Type* out_type = out->Value(&_gvn);
5485 const Type* in_type = in->Value(&_gvn);
5486 const TypeAryPtr* top_out = out_type->isa_aryptr();
5487 const TypeAryPtr* top_in = in_type->isa_aryptr();
5488 if (top_out == NULL || top_out->klass() == NULL ||
5489 top_in == NULL || top_in->klass() == NULL) {
5490 // failed array check
5491 return false;
5492 }
5493
5494 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5495 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5496 if (out_elem != T_INT || in_elem != T_INT) {
5497 return false;
5498 }
5499
5500 Node* outlen = load_array_length(out);
5501 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5502 Node* out_start = array_element_address(out, intcon(0), out_elem);
5503 Node* in_start = array_element_address(in, intcon(0), in_elem);
5504
5505 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5506 OptoRuntime::mulAdd_Type(),
5507 stubAddr, stubName, TypePtr::BOTTOM,
5508 out_start,in_start, new_offset, len, k);
5509 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5510 set_result(result);
5511 return true;
5512 }
5513
5514 //-------------inline_montgomeryMultiply-----------------------------------
5515 bool LibraryCallKit::inline_montgomeryMultiply() {
5516 address stubAddr = StubRoutines::montgomeryMultiply();
5517 if (stubAddr == NULL) {
5518 return false; // Intrinsic's stub is not implemented on this platform
5519 }
5520
5521 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5522 const char* stubName = "montgomery_multiply";
5523
5524 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5525
5526 Node* a = argument(0);
5527 Node* b = argument(1);
5528 Node* n = argument(2);
5529 Node* len = argument(3);
5530 Node* inv = argument(4);
5531 Node* m = argument(6);
5532
5533 const Type* a_type = a->Value(&_gvn);
5534 const TypeAryPtr* top_a = a_type->isa_aryptr();
5535 const Type* b_type = b->Value(&_gvn);
5536 const TypeAryPtr* top_b = b_type->isa_aryptr();
5537 const Type* n_type = a->Value(&_gvn);
5538 const TypeAryPtr* top_n = n_type->isa_aryptr();
5539 const Type* m_type = a->Value(&_gvn);
5540 const TypeAryPtr* top_m = m_type->isa_aryptr();
5541 if (top_a == NULL || top_a->klass() == NULL ||
5542 top_b == NULL || top_b->klass() == NULL ||
5543 top_n == NULL || top_n->klass() == NULL ||
5544 top_m == NULL || top_m->klass() == NULL) {
5545 // failed array check
5546 return false;
5547 }
5548
5549 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5550 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5551 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5552 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5553 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5554 return false;
5555 }
5556
5557 // Make the call
5558 {
5559 Node* a_start = array_element_address(a, intcon(0), a_elem);
5560 Node* b_start = array_element_address(b, intcon(0), b_elem);
5561 Node* n_start = array_element_address(n, intcon(0), n_elem);
5562 Node* m_start = array_element_address(m, intcon(0), m_elem);
5563
5564 Node* call = make_runtime_call(RC_LEAF,
5565 OptoRuntime::montgomeryMultiply_Type(),
5566 stubAddr, stubName, TypePtr::BOTTOM,
5567 a_start, b_start, n_start, len, inv, top(),
5568 m_start);
5569 set_result(m);
5570 }
5571
5572 return true;
5573 }
5574
5575 bool LibraryCallKit::inline_montgomerySquare() {
5576 address stubAddr = StubRoutines::montgomerySquare();
5577 if (stubAddr == NULL) {
5578 return false; // Intrinsic's stub is not implemented on this platform
5579 }
5580
5581 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5582 const char* stubName = "montgomery_square";
5583
5584 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5585
5586 Node* a = argument(0);
5587 Node* n = argument(1);
5588 Node* len = argument(2);
5589 Node* inv = argument(3);
5590 Node* m = argument(5);
5591
5592 const Type* a_type = a->Value(&_gvn);
5593 const TypeAryPtr* top_a = a_type->isa_aryptr();
5594 const Type* n_type = a->Value(&_gvn);
5595 const TypeAryPtr* top_n = n_type->isa_aryptr();
5596 const Type* m_type = a->Value(&_gvn);
5597 const TypeAryPtr* top_m = m_type->isa_aryptr();
5598 if (top_a == NULL || top_a->klass() == NULL ||
5599 top_n == NULL || top_n->klass() == NULL ||
5600 top_m == NULL || top_m->klass() == NULL) {
5601 // failed array check
5602 return false;
5603 }
5604
5605 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5606 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5607 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5608 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5609 return false;
5610 }
5611
5612 // Make the call
5613 {
5614 Node* a_start = array_element_address(a, intcon(0), a_elem);
5615 Node* n_start = array_element_address(n, intcon(0), n_elem);
5616 Node* m_start = array_element_address(m, intcon(0), m_elem);
5617
5618 Node* call = make_runtime_call(RC_LEAF,
5619 OptoRuntime::montgomerySquare_Type(),
5620 stubAddr, stubName, TypePtr::BOTTOM,
5621 a_start, n_start, len, inv, top(),
5622 m_start);
5623 set_result(m);
5624 }
5625
5626 return true;
5627 }
5628
5629 //-------------inline_vectorizedMismatch------------------------------
5630 bool LibraryCallKit::inline_vectorizedMismatch() {
5631 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5632
5633 address stubAddr = StubRoutines::vectorizedMismatch();
5634 if (stubAddr == NULL) {
5635 return false; // Intrinsic's stub is not implemented on this platform
5636 }
5637 const char* stubName = "vectorizedMismatch";
5638 int size_l = callee()->signature()->size();
5639 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5640
5641 Node* obja = argument(0);
5642 Node* aoffset = argument(1);
5643 Node* objb = argument(3);
5644 Node* boffset = argument(4);
5645 Node* length = argument(6);
5646 Node* scale = argument(7);
5647
5648 const Type* a_type = obja->Value(&_gvn);
5649 const Type* b_type = objb->Value(&_gvn);
5650 const TypeAryPtr* top_a = a_type->isa_aryptr();
5651 const TypeAryPtr* top_b = b_type->isa_aryptr();
5652 if (top_a == NULL || top_a->klass() == NULL ||
5653 top_b == NULL || top_b->klass() == NULL) {
5654 // failed array check
5655 return false;
5656 }
5657
5658 Node* call;
5659 jvms()->set_should_reexecute(true);
5660
5661 Node* obja_adr = make_unsafe_address(obja, aoffset);
5662 Node* objb_adr = make_unsafe_address(objb, boffset);
5663
5664 call = make_runtime_call(RC_LEAF,
5665 OptoRuntime::vectorizedMismatch_Type(),
5666 stubAddr, stubName, TypePtr::BOTTOM,
5667 obja_adr, objb_adr, length, scale);
5668
5669 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5670 set_result(result);
5671 return true;
5672 }
5673
5674 /**
5675 * Calculate CRC32 for byte.
5676 * int java.util.zip.CRC32.update(int crc, int b)
5677 */
5678 bool LibraryCallKit::inline_updateCRC32() {
5679 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5680 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5681 // no receiver since it is static method
5682 Node* crc = argument(0); // type: int
5683 Node* b = argument(1); // type: int
5684
5685 /*
5686 * int c = ~ crc;
5687 * b = timesXtoThe32[(b ^ c) & 0xFF];
5688 * b = b ^ (c >>> 8);
5689 * crc = ~b;
5690 */
5691
5692 Node* M1 = intcon(-1);
5693 crc = _gvn.transform(new XorINode(crc, M1));
5694 Node* result = _gvn.transform(new XorINode(crc, b));
5695 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5696
5697 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5698 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5699 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5700 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5701
5702 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5703 result = _gvn.transform(new XorINode(crc, result));
5704 result = _gvn.transform(new XorINode(result, M1));
5705 set_result(result);
5706 return true;
5707 }
5708
5709 /**
5710 * Calculate CRC32 for byte[] array.
5711 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5712 */
5713 bool LibraryCallKit::inline_updateBytesCRC32() {
5714 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5715 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5716 // no receiver since it is static method
5717 Node* crc = argument(0); // type: int
5718 Node* src = argument(1); // type: oop
5719 Node* offset = argument(2); // type: int
5720 Node* length = argument(3); // type: int
5721
5722 const Type* src_type = src->Value(&_gvn);
5723 const TypeAryPtr* top_src = src_type->isa_aryptr();
5724 if (top_src == NULL || top_src->klass() == NULL) {
5725 // failed array check
5726 return false;
5727 }
5728
5729 // Figure out the size and type of the elements we will be copying.
5730 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5731 if (src_elem != T_BYTE) {
5732 return false;
5733 }
5734
5735 // 'src_start' points to src array + scaled offset
5736 Node* src_start = array_element_address(src, offset, src_elem);
5737
5738 // We assume that range check is done by caller.
5739 // TODO: generate range check (offset+length < src.length) in debug VM.
5740
5741 // Call the stub.
5742 address stubAddr = StubRoutines::updateBytesCRC32();
5743 const char *stubName = "updateBytesCRC32";
5744
5745 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5746 stubAddr, stubName, TypePtr::BOTTOM,
5747 crc, src_start, length);
5748 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5749 set_result(result);
5750 return true;
5751 }
5752
5753 /**
5754 * Calculate CRC32 for ByteBuffer.
5755 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5756 */
5757 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5758 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5759 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5760 // no receiver since it is static method
5761 Node* crc = argument(0); // type: int
5762 Node* src = argument(1); // type: long
5763 Node* offset = argument(3); // type: int
5764 Node* length = argument(4); // type: int
5765
5766 src = ConvL2X(src); // adjust Java long to machine word
5767 Node* base = _gvn.transform(new CastX2PNode(src));
5768 offset = ConvI2X(offset);
5769
5770 // 'src_start' points to src array + scaled offset
5771 Node* src_start = basic_plus_adr(top(), base, offset);
5772
5773 // Call the stub.
5774 address stubAddr = StubRoutines::updateBytesCRC32();
5775 const char *stubName = "updateBytesCRC32";
5776
5777 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5778 stubAddr, stubName, TypePtr::BOTTOM,
5779 crc, src_start, length);
5780 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5781 set_result(result);
5782 return true;
5783 }
5784
5785 //------------------------------get_table_from_crc32c_class-----------------------
5786 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5787 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5788 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5789
5790 return table;
5791 }
5792
5793 //------------------------------inline_updateBytesCRC32C-----------------------
5794 //
5795 // Calculate CRC32C for byte[] array.
5796 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5797 //
5798 bool LibraryCallKit::inline_updateBytesCRC32C() {
5799 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5800 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5801 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5802 // no receiver since it is a static method
5803 Node* crc = argument(0); // type: int
5804 Node* src = argument(1); // type: oop
5805 Node* offset = argument(2); // type: int
5806 Node* end = argument(3); // type: int
5807
5808 Node* length = _gvn.transform(new SubINode(end, offset));
5809
5810 const Type* src_type = src->Value(&_gvn);
5811 const TypeAryPtr* top_src = src_type->isa_aryptr();
5812 if (top_src == NULL || top_src->klass() == NULL) {
5813 // failed array check
5814 return false;
5815 }
5816
5817 // Figure out the size and type of the elements we will be copying.
5818 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5819 if (src_elem != T_BYTE) {
5820 return false;
5821 }
5822
5823 // 'src_start' points to src array + scaled offset
5824 Node* src_start = array_element_address(src, offset, src_elem);
5825
5826 // static final int[] byteTable in class CRC32C
5827 Node* table = get_table_from_crc32c_class(callee()->holder());
5828 Node* table_start = array_element_address(table, intcon(0), T_INT);
5829
5830 // We assume that range check is done by caller.
5831 // TODO: generate range check (offset+length < src.length) in debug VM.
5832
5833 // Call the stub.
5834 address stubAddr = StubRoutines::updateBytesCRC32C();
5835 const char *stubName = "updateBytesCRC32C";
5836
5837 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5838 stubAddr, stubName, TypePtr::BOTTOM,
5839 crc, src_start, length, table_start);
5840 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5841 set_result(result);
5842 return true;
5843 }
5844
5845 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5846 //
5847 // Calculate CRC32C for DirectByteBuffer.
5848 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5849 //
5850 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5851 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5852 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5853 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5854 // no receiver since it is a static method
5855 Node* crc = argument(0); // type: int
5856 Node* src = argument(1); // type: long
5857 Node* offset = argument(3); // type: int
5858 Node* end = argument(4); // type: int
5859
5860 Node* length = _gvn.transform(new SubINode(end, offset));
5861
5862 src = ConvL2X(src); // adjust Java long to machine word
5863 Node* base = _gvn.transform(new CastX2PNode(src));
5864 offset = ConvI2X(offset);
5865
5866 // 'src_start' points to src array + scaled offset
5867 Node* src_start = basic_plus_adr(top(), base, offset);
5868
5869 // static final int[] byteTable in class CRC32C
5870 Node* table = get_table_from_crc32c_class(callee()->holder());
5871 Node* table_start = array_element_address(table, intcon(0), T_INT);
5872
5873 // Call the stub.
5874 address stubAddr = StubRoutines::updateBytesCRC32C();
5875 const char *stubName = "updateBytesCRC32C";
5876
5877 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5878 stubAddr, stubName, TypePtr::BOTTOM,
5879 crc, src_start, length, table_start);
5880 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5881 set_result(result);
5882 return true;
5883 }
5884
5885 //------------------------------inline_updateBytesAdler32----------------------
5886 //
5887 // Calculate Adler32 checksum for byte[] array.
5888 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5889 //
5890 bool LibraryCallKit::inline_updateBytesAdler32() {
5891 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5892 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5893 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5894 // no receiver since it is static method
5895 Node* crc = argument(0); // type: int
5896 Node* src = argument(1); // type: oop
5897 Node* offset = argument(2); // type: int
5898 Node* length = argument(3); // type: int
5899
5900 const Type* src_type = src->Value(&_gvn);
5901 const TypeAryPtr* top_src = src_type->isa_aryptr();
5902 if (top_src == NULL || top_src->klass() == NULL) {
5903 // failed array check
5904 return false;
5905 }
5906
5907 // Figure out the size and type of the elements we will be copying.
5908 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5909 if (src_elem != T_BYTE) {
5910 return false;
5911 }
5912
5913 // 'src_start' points to src array + scaled offset
5914 Node* src_start = array_element_address(src, offset, src_elem);
5915
5916 // We assume that range check is done by caller.
5917 // TODO: generate range check (offset+length < src.length) in debug VM.
5918
5919 // Call the stub.
5920 address stubAddr = StubRoutines::updateBytesAdler32();
5921 const char *stubName = "updateBytesAdler32";
5922
5923 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5924 stubAddr, stubName, TypePtr::BOTTOM,
5925 crc, src_start, length);
5926 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5927 set_result(result);
5928 return true;
5929 }
5930
5931 //------------------------------inline_updateByteBufferAdler32---------------
5932 //
5933 // Calculate Adler32 checksum for DirectByteBuffer.
5934 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5935 //
5936 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5937 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5938 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5939 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5940 // no receiver since it is static method
5941 Node* crc = argument(0); // type: int
5942 Node* src = argument(1); // type: long
5943 Node* offset = argument(3); // type: int
5944 Node* length = argument(4); // type: int
5945
5946 src = ConvL2X(src); // adjust Java long to machine word
5947 Node* base = _gvn.transform(new CastX2PNode(src));
5948 offset = ConvI2X(offset);
5949
5950 // 'src_start' points to src array + scaled offset
5951 Node* src_start = basic_plus_adr(top(), base, offset);
5952
5953 // Call the stub.
5954 address stubAddr = StubRoutines::updateBytesAdler32();
5955 const char *stubName = "updateBytesAdler32";
5956
5957 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5958 stubAddr, stubName, TypePtr::BOTTOM,
5959 crc, src_start, length);
5960
5961 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5962 set_result(result);
5963 return true;
5964 }
5965
5966 //----------------------------inline_reference_get----------------------------
5967 // public T java.lang.ref.Reference.get();
5968 bool LibraryCallKit::inline_reference_get() {
5969 const int referent_offset = java_lang_ref_Reference::referent_offset;
5970 guarantee(referent_offset > 0, "should have already been set");
5971
5972 // Get the argument:
5973 Node* reference_obj = null_check_receiver();
5974 if (stopped()) return true;
5975
5976 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5977
5978 ciInstanceKlass* klass = env()->Object_klass();
5979 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5980
5981 Node* no_ctrl = NULL;
5982 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5983
5984 // Use the pre-barrier to record the value in the referent field
5985 pre_barrier(false /* do_load */,
5986 control(),
5987 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5988 result /* pre_val */,
5989 T_OBJECT);
5990
5991 // Add memory barrier to prevent commoning reads from this field
5992 // across safepoint since GC can change its value.
5993 insert_mem_bar(Op_MemBarCPUOrder);
5994
5995 set_result(result);
5996 return true;
5997 }
5998
5999
6000 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6001 bool is_exact=true, bool is_static=false,
6002 ciInstanceKlass * fromKls=NULL) {
6003 if (fromKls == NULL) {
6004 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6005 assert(tinst != NULL, "obj is null");
6006 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6007 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6008 fromKls = tinst->klass()->as_instance_klass();
6009 } else {
6010 assert(is_static, "only for static field access");
6011 }
6012 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6013 ciSymbol::make(fieldTypeString),
6014 is_static);
6015
6016 assert (field != NULL, "undefined field");
6017 if (field == NULL) return (Node *) NULL;
6018
6019 if (is_static) {
6020 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6021 fromObj = makecon(tip);
6022 }
6023
6024 // Next code copied from Parse::do_get_xxx():
6025
6026 // Compute address and memory type.
6027 int offset = field->offset_in_bytes();
6028 bool is_vol = field->is_volatile();
6029 ciType* field_klass = field->type();
6030 assert(field_klass->is_loaded(), "should be loaded");
6031 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6032 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6033 BasicType bt = field->layout_type();
6034
6035 // Build the resultant type of the load
6036 const Type *type;
6037 if (bt == T_OBJECT) {
6038 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6039 } else {
6040 type = Type::get_const_basic_type(bt);
6041 }
6042
6043 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
6044 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
6045 }
6046 // Build the load.
6047 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
6048 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
6049 // If reference is volatile, prevent following memory ops from
6050 // floating up past the volatile read. Also prevents commoning
6051 // another volatile read.
6052 if (is_vol) {
6053 // Memory barrier includes bogus read of value to force load BEFORE membar
6054 insert_mem_bar(Op_MemBarAcquire, loadedField);
6055 }
6056 return loadedField;
6057 }
6058
6059 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6060 bool is_exact = true, bool is_static = false,
6061 ciInstanceKlass * fromKls = NULL) {
6062 if (fromKls == NULL) {
6063 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6064 assert(tinst != NULL, "obj is null");
6065 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6066 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6067 fromKls = tinst->klass()->as_instance_klass();
6068 }
6069 else {
6070 assert(is_static, "only for static field access");
6071 }
6072 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6073 ciSymbol::make(fieldTypeString),
6074 is_static);
6075
6076 assert(field != NULL, "undefined field");
6077 assert(!field->is_volatile(), "not defined for volatile fields");
6078
6079 if (is_static) {
6080 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6081 fromObj = makecon(tip);
6082 }
6083
6084 // Next code copied from Parse::do_get_xxx():
6085
6086 // Compute address and memory type.
6087 int offset = field->offset_in_bytes();
6088 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6089
6090 return adr;
6091 }
6092
6093 //------------------------------inline_aescrypt_Block-----------------------
6094 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6095 address stubAddr = NULL;
6096 const char *stubName;
6097 assert(UseAES, "need AES instruction support");
6098
6099 switch(id) {
6100 case vmIntrinsics::_aescrypt_encryptBlock:
6101 stubAddr = StubRoutines::aescrypt_encryptBlock();
6102 stubName = "aescrypt_encryptBlock";
6103 break;
6104 case vmIntrinsics::_aescrypt_decryptBlock:
6105 stubAddr = StubRoutines::aescrypt_decryptBlock();
6106 stubName = "aescrypt_decryptBlock";
6107 break;
6108 }
6109 if (stubAddr == NULL) return false;
6110
6111 Node* aescrypt_object = argument(0);
6112 Node* src = argument(1);
6113 Node* src_offset = argument(2);
6114 Node* dest = argument(3);
6115 Node* dest_offset = argument(4);
6116
6117 // (1) src and dest are arrays.
6118 const Type* src_type = src->Value(&_gvn);
6119 const Type* dest_type = dest->Value(&_gvn);
6120 const TypeAryPtr* top_src = src_type->isa_aryptr();
6121 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6122 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6123
6124 // for the quick and dirty code we will skip all the checks.
6125 // we are just trying to get the call to be generated.
6126 Node* src_start = src;
6127 Node* dest_start = dest;
6128 if (src_offset != NULL || dest_offset != NULL) {
6129 assert(src_offset != NULL && dest_offset != NULL, "");
6130 src_start = array_element_address(src, src_offset, T_BYTE);
6131 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6132 }
6133
6134 // now need to get the start of its expanded key array
6135 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6136 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6137 if (k_start == NULL) return false;
6138
6139 if (Matcher::pass_original_key_for_aes()) {
6140 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6141 // compatibility issues between Java key expansion and SPARC crypto instructions
6142 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6143 if (original_k_start == NULL) return false;
6144
6145 // Call the stub.
6146 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6147 stubAddr, stubName, TypePtr::BOTTOM,
6148 src_start, dest_start, k_start, original_k_start);
6149 } else {
6150 // Call the stub.
6151 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6152 stubAddr, stubName, TypePtr::BOTTOM,
6153 src_start, dest_start, k_start);
6154 }
6155
6156 return true;
6157 }
6158
6159 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6160 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6161 address stubAddr = NULL;
6162 const char *stubName = NULL;
6163
6164 assert(UseAES, "need AES instruction support");
6165
6166 switch(id) {
6167 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6168 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6169 stubName = "cipherBlockChaining_encryptAESCrypt";
6170 break;
6171 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6172 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6173 stubName = "cipherBlockChaining_decryptAESCrypt";
6174 break;
6175 }
6176 if (stubAddr == NULL) return false;
6177
6178 Node* cipherBlockChaining_object = argument(0);
6179 Node* src = argument(1);
6180 Node* src_offset = argument(2);
6181 Node* len = argument(3);
6182 Node* dest = argument(4);
6183 Node* dest_offset = argument(5);
6184
6185 // (1) src and dest are arrays.
6186 const Type* src_type = src->Value(&_gvn);
6187 const Type* dest_type = dest->Value(&_gvn);
6188 const TypeAryPtr* top_src = src_type->isa_aryptr();
6189 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6190 assert (top_src != NULL && top_src->klass() != NULL
6191 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6192
6193 // checks are the responsibility of the caller
6194 Node* src_start = src;
6195 Node* dest_start = dest;
6196 if (src_offset != NULL || dest_offset != NULL) {
6197 assert(src_offset != NULL && dest_offset != NULL, "");
6198 src_start = array_element_address(src, src_offset, T_BYTE);
6199 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6200 }
6201
6202 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6203 // (because of the predicated logic executed earlier).
6204 // so we cast it here safely.
6205 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6206
6207 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6208 if (embeddedCipherObj == NULL) return false;
6209
6210 // cast it to what we know it will be at runtime
6211 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6212 assert(tinst != NULL, "CBC obj is null");
6213 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6214 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6215 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6216
6217 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6218 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6219 const TypeOopPtr* xtype = aklass->as_instance_type();
6220 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6221 aescrypt_object = _gvn.transform(aescrypt_object);
6222
6223 // we need to get the start of the aescrypt_object's expanded key array
6224 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6225 if (k_start == NULL) return false;
6226
6227 // similarly, get the start address of the r vector
6228 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6229 if (objRvec == NULL) return false;
6230 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6231
6232 Node* cbcCrypt;
6233 if (Matcher::pass_original_key_for_aes()) {
6234 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6235 // compatibility issues between Java key expansion and SPARC crypto instructions
6236 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6237 if (original_k_start == NULL) return false;
6238
6239 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6240 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6241 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6242 stubAddr, stubName, TypePtr::BOTTOM,
6243 src_start, dest_start, k_start, r_start, len, original_k_start);
6244 } else {
6245 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6246 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6247 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6248 stubAddr, stubName, TypePtr::BOTTOM,
6249 src_start, dest_start, k_start, r_start, len);
6250 }
6251
6252 // return cipher length (int)
6253 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6254 set_result(retvalue);
6255 return true;
6256 }
6257
6258 //------------------------------inline_counterMode_AESCrypt-----------------------
6259 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6260 assert(UseAES, "need AES instruction support");
6261 if (!UseAESCTRIntrinsics) return false;
6262
6263 address stubAddr = NULL;
6264 const char *stubName = NULL;
6265 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6266 stubAddr = StubRoutines::counterMode_AESCrypt();
6267 stubName = "counterMode_AESCrypt";
6268 }
6269 if (stubAddr == NULL) return false;
6270
6271 Node* counterMode_object = argument(0);
6272 Node* src = argument(1);
6273 Node* src_offset = argument(2);
6274 Node* len = argument(3);
6275 Node* dest = argument(4);
6276 Node* dest_offset = argument(5);
6277
6278 // (1) src and dest are arrays.
6279 const Type* src_type = src->Value(&_gvn);
6280 const Type* dest_type = dest->Value(&_gvn);
6281 const TypeAryPtr* top_src = src_type->isa_aryptr();
6282 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6283 assert(top_src != NULL && top_src->klass() != NULL &&
6284 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6285
6286 // checks are the responsibility of the caller
6287 Node* src_start = src;
6288 Node* dest_start = dest;
6289 if (src_offset != NULL || dest_offset != NULL) {
6290 assert(src_offset != NULL && dest_offset != NULL, "");
6291 src_start = array_element_address(src, src_offset, T_BYTE);
6292 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6293 }
6294
6295 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6296 // (because of the predicated logic executed earlier).
6297 // so we cast it here safely.
6298 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6299 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6300 if (embeddedCipherObj == NULL) return false;
6301 // cast it to what we know it will be at runtime
6302 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6303 assert(tinst != NULL, "CTR obj is null");
6304 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6305 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6306 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6307 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6308 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6309 const TypeOopPtr* xtype = aklass->as_instance_type();
6310 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6311 aescrypt_object = _gvn.transform(aescrypt_object);
6312 // we need to get the start of the aescrypt_object's expanded key array
6313 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6314 if (k_start == NULL) return false;
6315 // similarly, get the start address of the r vector
6316 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6317 if (obj_counter == NULL) return false;
6318 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6319
6320 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6321 if (saved_encCounter == NULL) return false;
6322 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6323 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6324
6325 Node* ctrCrypt;
6326 if (Matcher::pass_original_key_for_aes()) {
6327 // no SPARC version for AES/CTR intrinsics now.
6328 return false;
6329 }
6330 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6331 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6332 OptoRuntime::counterMode_aescrypt_Type(),
6333 stubAddr, stubName, TypePtr::BOTTOM,
6334 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6335
6336 // return cipher length (int)
6337 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6338 set_result(retvalue);
6339 return true;
6340 }
6341
6342 //------------------------------get_key_start_from_aescrypt_object-----------------------
6343 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6344 #if defined(PPC64) || defined(S390)
6345 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6346 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6347 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6348 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6349 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6350 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6351 if (objSessionK == NULL) {
6352 return (Node *) NULL;
6353 }
6354 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6355 #else
6356 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6357 #endif // PPC64
6358 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6359 if (objAESCryptKey == NULL) return (Node *) NULL;
6360
6361 // now have the array, need to get the start address of the K array
6362 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6363 return k_start;
6364 }
6365
6366 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6367 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6368 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6369 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6370 if (objAESCryptKey == NULL) return (Node *) NULL;
6371
6372 // now have the array, need to get the start address of the lastKey array
6373 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6374 return original_k_start;
6375 }
6376
6377 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6378 // Return node representing slow path of predicate check.
6379 // the pseudo code we want to emulate with this predicate is:
6380 // for encryption:
6381 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6382 // for decryption:
6383 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6384 // note cipher==plain is more conservative than the original java code but that's OK
6385 //
6386 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6387 // The receiver was checked for NULL already.
6388 Node* objCBC = argument(0);
6389
6390 // Load embeddedCipher field of CipherBlockChaining object.
6391 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6392
6393 // get AESCrypt klass for instanceOf check
6394 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6395 // will have same classloader as CipherBlockChaining object
6396 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6397 assert(tinst != NULL, "CBCobj is null");
6398 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6399
6400 // we want to do an instanceof comparison against the AESCrypt class
6401 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6402 if (!klass_AESCrypt->is_loaded()) {
6403 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6404 Node* ctrl = control();
6405 set_control(top()); // no regular fast path
6406 return ctrl;
6407 }
6408 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6409
6410 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6411 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6412 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6413
6414 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6415
6416 // for encryption, we are done
6417 if (!decrypting)
6418 return instof_false; // even if it is NULL
6419
6420 // for decryption, we need to add a further check to avoid
6421 // taking the intrinsic path when cipher and plain are the same
6422 // see the original java code for why.
6423 RegionNode* region = new RegionNode(3);
6424 region->init_req(1, instof_false);
6425 Node* src = argument(1);
6426 Node* dest = argument(4);
6427 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6428 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6429 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6430 region->init_req(2, src_dest_conjoint);
6431
6432 record_for_igvn(region);
6433 return _gvn.transform(region);
6434 }
6435
6436 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6437 // Return node representing slow path of predicate check.
6438 // the pseudo code we want to emulate with this predicate is:
6439 // for encryption:
6440 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6441 // for decryption:
6442 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6443 // note cipher==plain is more conservative than the original java code but that's OK
6444 //
6445
6446 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6447 // The receiver was checked for NULL already.
6448 Node* objCTR = argument(0);
6449
6450 // Load embeddedCipher field of CipherBlockChaining object.
6451 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6452
6453 // get AESCrypt klass for instanceOf check
6454 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6455 // will have same classloader as CipherBlockChaining object
6456 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6457 assert(tinst != NULL, "CTRobj is null");
6458 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6459
6460 // we want to do an instanceof comparison against the AESCrypt class
6461 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6462 if (!klass_AESCrypt->is_loaded()) {
6463 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6464 Node* ctrl = control();
6465 set_control(top()); // no regular fast path
6466 return ctrl;
6467 }
6468
6469 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6470 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6471 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6472 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6473 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6474
6475 return instof_false; // even if it is NULL
6476 }
6477
6478 //------------------------------inline_ghash_processBlocks
6479 bool LibraryCallKit::inline_ghash_processBlocks() {
6480 address stubAddr;
6481 const char *stubName;
6482 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6483
6484 stubAddr = StubRoutines::ghash_processBlocks();
6485 stubName = "ghash_processBlocks";
6486
6487 Node* data = argument(0);
6488 Node* offset = argument(1);
6489 Node* len = argument(2);
6490 Node* state = argument(3);
6491 Node* subkeyH = argument(4);
6492
6493 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6494 assert(state_start, "state is NULL");
6495 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6496 assert(subkeyH_start, "subkeyH is NULL");
6497 Node* data_start = array_element_address(data, offset, T_BYTE);
6498 assert(data_start, "data is NULL");
6499
6500 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6501 OptoRuntime::ghash_processBlocks_Type(),
6502 stubAddr, stubName, TypePtr::BOTTOM,
6503 state_start, subkeyH_start, data_start, len);
6504 return true;
6505 }
6506
6507 //------------------------------inline_sha_implCompress-----------------------
6508 //
6509 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6510 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6511 //
6512 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6513 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6514 //
6515 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6516 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6517 //
6518 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6519 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6520
6521 Node* sha_obj = argument(0);
6522 Node* src = argument(1); // type oop
6523 Node* ofs = argument(2); // type int
6524
6525 const Type* src_type = src->Value(&_gvn);
6526 const TypeAryPtr* top_src = src_type->isa_aryptr();
6527 if (top_src == NULL || top_src->klass() == NULL) {
6528 // failed array check
6529 return false;
6530 }
6531 // Figure out the size and type of the elements we will be copying.
6532 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6533 if (src_elem != T_BYTE) {
6534 return false;
6535 }
6536 // 'src_start' points to src array + offset
6537 Node* src_start = array_element_address(src, ofs, src_elem);
6538 Node* state = NULL;
6539 address stubAddr;
6540 const char *stubName;
6541
6542 switch(id) {
6543 case vmIntrinsics::_sha_implCompress:
6544 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6545 state = get_state_from_sha_object(sha_obj);
6546 stubAddr = StubRoutines::sha1_implCompress();
6547 stubName = "sha1_implCompress";
6548 break;
6549 case vmIntrinsics::_sha2_implCompress:
6550 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6551 state = get_state_from_sha_object(sha_obj);
6552 stubAddr = StubRoutines::sha256_implCompress();
6553 stubName = "sha256_implCompress";
6554 break;
6555 case vmIntrinsics::_sha5_implCompress:
6556 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6557 state = get_state_from_sha5_object(sha_obj);
6558 stubAddr = StubRoutines::sha512_implCompress();
6559 stubName = "sha512_implCompress";
6560 break;
6561 default:
6562 fatal_unexpected_iid(id);
6563 return false;
6564 }
6565 if (state == NULL) return false;
6566
6567 // Call the stub.
6568 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6569 stubAddr, stubName, TypePtr::BOTTOM,
6570 src_start, state);
6571
6572 return true;
6573 }
6574
6575 //------------------------------inline_digestBase_implCompressMB-----------------------
6576 //
6577 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6578 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6579 //
6580 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6581 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6582 "need SHA1/SHA256/SHA512 instruction support");
6583 assert((uint)predicate < 3, "sanity");
6584 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6585
6586 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6587 Node* src = argument(1); // byte[] array
6588 Node* ofs = argument(2); // type int
6589 Node* limit = argument(3); // type int
6590
6591 const Type* src_type = src->Value(&_gvn);
6592 const TypeAryPtr* top_src = src_type->isa_aryptr();
6593 if (top_src == NULL || top_src->klass() == NULL) {
6594 // failed array check
6595 return false;
6596 }
6597 // Figure out the size and type of the elements we will be copying.
6598 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6599 if (src_elem != T_BYTE) {
6600 return false;
6601 }
6602 // 'src_start' points to src array + offset
6603 Node* src_start = array_element_address(src, ofs, src_elem);
6604
6605 const char* klass_SHA_name = NULL;
6606 const char* stub_name = NULL;
6607 address stub_addr = NULL;
6608 bool long_state = false;
6609
6610 switch (predicate) {
6611 case 0:
6612 if (UseSHA1Intrinsics) {
6613 klass_SHA_name = "sun/security/provider/SHA";
6614 stub_name = "sha1_implCompressMB";
6615 stub_addr = StubRoutines::sha1_implCompressMB();
6616 }
6617 break;
6618 case 1:
6619 if (UseSHA256Intrinsics) {
6620 klass_SHA_name = "sun/security/provider/SHA2";
6621 stub_name = "sha256_implCompressMB";
6622 stub_addr = StubRoutines::sha256_implCompressMB();
6623 }
6624 break;
6625 case 2:
6626 if (UseSHA512Intrinsics) {
6627 klass_SHA_name = "sun/security/provider/SHA5";
6628 stub_name = "sha512_implCompressMB";
6629 stub_addr = StubRoutines::sha512_implCompressMB();
6630 long_state = true;
6631 }
6632 break;
6633 default:
6634 fatal("unknown SHA intrinsic predicate: %d", predicate);
6635 }
6636 if (klass_SHA_name != NULL) {
6637 // get DigestBase klass to lookup for SHA klass
6638 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6639 assert(tinst != NULL, "digestBase_obj is not instance???");
6640 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6641
6642 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6643 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6644 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6645 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6646 }
6647 return false;
6648 }
6649 //------------------------------inline_sha_implCompressMB-----------------------
6650 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6651 bool long_state, address stubAddr, const char *stubName,
6652 Node* src_start, Node* ofs, Node* limit) {
6653 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6654 const TypeOopPtr* xtype = aklass->as_instance_type();
6655 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6656 sha_obj = _gvn.transform(sha_obj);
6657
6658 Node* state;
6659 if (long_state) {
6660 state = get_state_from_sha5_object(sha_obj);
6661 } else {
6662 state = get_state_from_sha_object(sha_obj);
6663 }
6664 if (state == NULL) return false;
6665
6666 // Call the stub.
6667 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6668 OptoRuntime::digestBase_implCompressMB_Type(),
6669 stubAddr, stubName, TypePtr::BOTTOM,
6670 src_start, state, ofs, limit);
6671 // return ofs (int)
6672 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6673 set_result(result);
6674
6675 return true;
6676 }
6677
6678 //------------------------------get_state_from_sha_object-----------------------
6679 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6680 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6681 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6682 if (sha_state == NULL) return (Node *) NULL;
6683
6684 // now have the array, need to get the start address of the state array
6685 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6686 return state;
6687 }
6688
6689 //------------------------------get_state_from_sha5_object-----------------------
6690 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6691 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6692 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6693 if (sha_state == NULL) return (Node *) NULL;
6694
6695 // now have the array, need to get the start address of the state array
6696 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6697 return state;
6698 }
6699
6700 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6701 // Return node representing slow path of predicate check.
6702 // the pseudo code we want to emulate with this predicate is:
6703 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6704 //
6705 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6706 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6707 "need SHA1/SHA256/SHA512 instruction support");
6708 assert((uint)predicate < 3, "sanity");
6709
6710 // The receiver was checked for NULL already.
6711 Node* digestBaseObj = argument(0);
6712
6713 // get DigestBase klass for instanceOf check
6714 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6715 assert(tinst != NULL, "digestBaseObj is null");
6716 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6717
6718 const char* klass_SHA_name = NULL;
6719 switch (predicate) {
6720 case 0:
6721 if (UseSHA1Intrinsics) {
6722 // we want to do an instanceof comparison against the SHA class
6723 klass_SHA_name = "sun/security/provider/SHA";
6724 }
6725 break;
6726 case 1:
6727 if (UseSHA256Intrinsics) {
6728 // we want to do an instanceof comparison against the SHA2 class
6729 klass_SHA_name = "sun/security/provider/SHA2";
6730 }
6731 break;
6732 case 2:
6733 if (UseSHA512Intrinsics) {
6734 // we want to do an instanceof comparison against the SHA5 class
6735 klass_SHA_name = "sun/security/provider/SHA5";
6736 }
6737 break;
6738 default:
6739 fatal("unknown SHA intrinsic predicate: %d", predicate);
6740 }
6741
6742 ciKlass* klass_SHA = NULL;
6743 if (klass_SHA_name != NULL) {
6744 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6745 }
6746 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6747 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6748 Node* ctrl = control();
6749 set_control(top()); // no intrinsic path
6750 return ctrl;
6751 }
6752 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6753
6754 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6755 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6756 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6757 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6758
6759 return instof_false; // even if it is NULL
6760 }
6761
6762 //-------------inline_fma-----------------------------------
6763 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6764 Node *a = NULL;
6765 Node *b = NULL;
6766 Node *c = NULL;
6767 Node* result = NULL;
6768 switch (id) {
6769 case vmIntrinsics::_fmaD:
6770 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6771 // no receiver since it is static method
6772 a = round_double_node(argument(0));
6773 b = round_double_node(argument(2));
6774 c = round_double_node(argument(4));
6775 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6776 break;
6777 case vmIntrinsics::_fmaF:
6778 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6779 a = argument(0);
6780 b = argument(1);
6781 c = argument(2);
6782 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6783 break;
6784 default:
6785 fatal_unexpected_iid(id); break;
6786 }
6787 set_result(result);
6788 return true;
6789 }
6790
6791 bool LibraryCallKit::inline_profileBoolean() {
6792 Node* counts = argument(1);
6793 const TypeAryPtr* ary = NULL;
6794 ciArray* aobj = NULL;
6795 if (counts->is_Con()
6796 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6797 && (aobj = ary->const_oop()->as_array()) != NULL
6798 && (aobj->length() == 2)) {
6799 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6800 jint false_cnt = aobj->element_value(0).as_int();
6801 jint true_cnt = aobj->element_value(1).as_int();
6802
6803 if (C->log() != NULL) {
6804 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6805 false_cnt, true_cnt);
6806 }
6807
6808 if (false_cnt + true_cnt == 0) {
6809 // According to profile, never executed.
6810 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6811 Deoptimization::Action_reinterpret);
6812 return true;
6813 }
6814
6815 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6816 // is a number of each value occurrences.
6817 Node* result = argument(0);
6818 if (false_cnt == 0 || true_cnt == 0) {
6819 // According to profile, one value has been never seen.
6820 int expected_val = (false_cnt == 0) ? 1 : 0;
6821
6822 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6823 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6824
6825 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6826 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6827 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6828
6829 { // Slow path: uncommon trap for never seen value and then reexecute
6830 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6831 // the value has been seen at least once.
6832 PreserveJVMState pjvms(this);
6833 PreserveReexecuteState preexecs(this);
6834 jvms()->set_should_reexecute(true);
6835
6836 set_control(slow_path);
6837 set_i_o(i_o());
6838
6839 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6840 Deoptimization::Action_reinterpret);
6841 }
6842 // The guard for never seen value enables sharpening of the result and
6843 // returning a constant. It allows to eliminate branches on the same value
6844 // later on.
6845 set_control(fast_path);
6846 result = intcon(expected_val);
6847 }
6848 // Stop profiling.
6849 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6850 // By replacing method body with profile data (represented as ProfileBooleanNode
6851 // on IR level) we effectively disable profiling.
6852 // It enables full speed execution once optimized code is generated.
6853 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6854 C->record_for_igvn(profile);
6855 set_result(profile);
6856 return true;
6857 } else {
6858 // Continue profiling.
6859 // Profile data isn't available at the moment. So, execute method's bytecode version.
6860 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6861 // is compiled and counters aren't available since corresponding MethodHandle
6862 // isn't a compile-time constant.
6863 return false;
6864 }
6865 }
6866
6867 bool LibraryCallKit::inline_isCompileConstant() {
6868 Node* n = argument(0);
6869 set_result(n->is_Con() ? intcon(1) : intcon(0));
6870 return true;
6871 }