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