1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "oops/objArrayKlass.hpp"
32 #include "opto/addnode.hpp"
33 #include "opto/arraycopynode.hpp"
34 #include "opto/c2compiler.hpp"
35 #include "opto/callGenerator.hpp"
36 #include "opto/castnode.hpp"
37 #include "opto/cfgnode.hpp"
38 #include "opto/convertnode.hpp"
39 #include "opto/countbitsnode.hpp"
40 #include "opto/intrinsicnode.hpp"
41 #include "opto/idealKit.hpp"
42 #include "opto/mathexactnode.hpp"
43 #include "opto/movenode.hpp"
44 #include "opto/mulnode.hpp"
45 #include "opto/narrowptrnode.hpp"
46 #include "opto/opaquenode.hpp"
47 #include "opto/parse.hpp"
48 #include "opto/runtime.hpp"
49 #include "opto/subnode.hpp"
50 #include "prims/nativeLookup.hpp"
51 #include "runtime/sharedRuntime.hpp"
52 #include "trace/traceMacros.hpp"
53
54 class LibraryIntrinsic : public InlineCallGenerator {
55 // Extend the set of intrinsics known to the runtime:
56 public:
57 private:
58 bool _is_virtual;
59 bool _does_virtual_dispatch;
60 int8_t _predicates_count; // Intrinsic is predicated by several conditions
61 int8_t _last_predicate; // Last generated predicate
62 vmIntrinsics::ID _intrinsic_id;
63
64 public:
65 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
66 : InlineCallGenerator(m),
67 _is_virtual(is_virtual),
68 _does_virtual_dispatch(does_virtual_dispatch),
69 _predicates_count((int8_t)predicates_count),
70 _last_predicate((int8_t)-1),
71 _intrinsic_id(id)
72 {
73 }
74 virtual bool is_intrinsic() const { return true; }
75 virtual bool is_virtual() const { return _is_virtual; }
76 virtual bool is_predicated() const { return _predicates_count > 0; }
77 virtual int predicates_count() const { return _predicates_count; }
78 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
79 virtual JVMState* generate(JVMState* jvms);
80 virtual Node* generate_predicate(JVMState* jvms, int predicate);
81 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
82 };
83
84
85 // Local helper class for LibraryIntrinsic:
86 class LibraryCallKit : public GraphKit {
87 private:
88 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
89 Node* _result; // the result node, if any
90 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
91
92 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr = false);
93
94 public:
95 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
96 : GraphKit(jvms),
97 _intrinsic(intrinsic),
98 _result(NULL)
99 {
100 // Check if this is a root compile. In that case we don't have a caller.
101 if (!jvms->has_method()) {
102 _reexecute_sp = sp();
103 } else {
104 // Find out how many arguments the interpreter needs when deoptimizing
105 // and save the stack pointer value so it can used by uncommon_trap.
106 // We find the argument count by looking at the declared signature.
107 bool ignored_will_link;
108 ciSignature* declared_signature = NULL;
109 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
110 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
111 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
112 }
113 }
114
115 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
116
117 ciMethod* caller() const { return jvms()->method(); }
118 int bci() const { return jvms()->bci(); }
119 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
120 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
121 ciMethod* callee() const { return _intrinsic->method(); }
122
123 bool try_to_inline(int predicate);
124 Node* try_to_predicate(int predicate);
125
126 void push_result() {
127 // Push the result onto the stack.
128 if (!stopped() && result() != NULL) {
129 BasicType bt = result()->bottom_type()->basic_type();
130 push_node(bt, result());
131 }
132 }
133
134 private:
135 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
136 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
137 }
138
139 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
140 void set_result(RegionNode* region, PhiNode* value);
141 Node* result() { return _result; }
142
143 virtual int reexecute_sp() { return _reexecute_sp; }
144
145 // Helper functions to inline natives
146 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
147 Node* generate_slow_guard(Node* test, RegionNode* region);
148 Node* generate_fair_guard(Node* test, RegionNode* region);
149 Node* generate_negative_guard(Node* index, RegionNode* region,
150 // resulting CastII of index:
151 Node* *pos_index = NULL);
152 Node* generate_limit_guard(Node* offset, Node* subseq_length,
153 Node* array_length,
154 RegionNode* region);
155 void generate_string_range_check(Node* array, Node* offset,
156 Node* length, bool char_count);
157 Node* generate_current_thread(Node* &tls_output);
158 Node* load_mirror_from_klass(Node* klass);
159 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
160 RegionNode* region, int null_path,
161 int offset);
162 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
163 RegionNode* region, int null_path) {
164 int offset = java_lang_Class::klass_offset_in_bytes();
165 return load_klass_from_mirror_common(mirror, never_see_null,
166 region, null_path,
167 offset);
168 }
169 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
170 RegionNode* region, int null_path) {
171 int offset = java_lang_Class::array_klass_offset_in_bytes();
172 return load_klass_from_mirror_common(mirror, never_see_null,
173 region, null_path,
174 offset);
175 }
176 Node* generate_access_flags_guard(Node* kls,
177 int modifier_mask, int modifier_bits,
178 RegionNode* region);
179 Node* generate_interface_guard(Node* kls, RegionNode* region);
180 Node* generate_array_guard(Node* kls, RegionNode* region) {
181 return generate_array_guard_common(kls, region, false, false);
182 }
183 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
184 return generate_array_guard_common(kls, region, false, true);
185 }
186 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
187 return generate_array_guard_common(kls, region, true, false);
188 }
189 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
190 return generate_array_guard_common(kls, region, true, true);
191 }
192 Node* generate_array_guard_common(Node* kls, RegionNode* region,
193 bool obj_array, bool not_array);
194 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
195 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
196 bool is_virtual = false, bool is_static = false);
197 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
198 return generate_method_call(method_id, false, true);
199 }
200 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
201 return generate_method_call(method_id, true, false);
202 }
203 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
204 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
205
206 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
207 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
208 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
209 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
210 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
211 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_indexOfChar();
213 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
214 bool inline_string_toBytesU();
215 bool inline_string_getCharsU();
216 bool inline_string_copy(bool compress);
217 bool inline_string_char_access(bool is_store);
218 Node* round_double_node(Node* n);
219 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
220 bool inline_math_native(vmIntrinsics::ID id);
221 bool inline_trig(vmIntrinsics::ID id);
222 bool inline_math(vmIntrinsics::ID id);
223 template <typename OverflowOp>
224 bool inline_math_overflow(Node* arg1, Node* arg2);
225 void inline_math_mathExact(Node* math, Node* test);
226 bool inline_math_addExactI(bool is_increment);
227 bool inline_math_addExactL(bool is_increment);
228 bool inline_math_multiplyExactI();
229 bool inline_math_multiplyExactL();
230 bool inline_math_negateExactI();
231 bool inline_math_negateExactL();
232 bool inline_math_subtractExactI(bool is_decrement);
233 bool inline_math_subtractExactL(bool is_decrement);
234 bool inline_min_max(vmIntrinsics::ID id);
235 bool inline_notify(vmIntrinsics::ID id);
236 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
237 // This returns Type::AnyPtr, RawPtr, or OopPtr.
238 int classify_unsafe_addr(Node* &base, Node* &offset);
239 Node* make_unsafe_address(Node* base, Node* offset);
240 // Helper for inline_unsafe_access.
241 // Generates the guards that check whether the result of
242 // Unsafe.getObject should be recorded in an SATB log buffer.
243 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
244
245 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
246 bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
247 static bool klass_needs_init_guard(Node* kls);
248 bool inline_unsafe_allocate();
249 bool inline_unsafe_copyMemory();
250 bool inline_native_currentThread();
251 #ifdef TRACE_HAVE_INTRINSICS
252 bool inline_native_classID();
253 bool inline_native_threadID();
254 #endif
255 bool inline_native_time_funcs(address method, const char* funcName);
256 bool inline_native_isInterrupted();
257 bool inline_native_Class_query(vmIntrinsics::ID id);
258 bool inline_native_subtype_check();
259
260 bool inline_native_newArray();
261 bool inline_native_getLength();
262 bool inline_array_copyOf(bool is_copyOfRange);
263 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
264 bool inline_objects_checkIndex();
265 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
266 bool inline_native_clone(bool is_virtual);
267 bool inline_native_Reflection_getCallerClass();
268 // Helper function for inlining native object hash method
269 bool inline_native_hashcode(bool is_virtual, bool is_static);
270 bool inline_native_getClass();
271
272 // Helper functions for inlining arraycopy
273 bool inline_arraycopy();
274 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
275 RegionNode* slow_region);
276 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
277 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp);
278
279 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
280 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
281 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
282 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
283 bool inline_unsafe_fence(vmIntrinsics::ID id);
284 bool inline_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
321 bool inline_profileBoolean();
322 bool inline_isCompileConstant();
323
324 bool inline_deoptimize();
325 };
326
327 //---------------------------make_vm_intrinsic----------------------------
328 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
329 vmIntrinsics::ID id = m->intrinsic_id();
330 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
331
332 if (!m->is_loaded()) {
333 // Do not attempt to inline unloaded methods.
334 return NULL;
335 }
336
337 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
338 bool is_available = false;
339
340 {
341 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
342 // the compiler must transition to '_thread_in_vm' state because both
343 // methods access VM-internal data.
344 VM_ENTRY_MARK;
345 methodHandle mh(THREAD, m->get_Method());
346 is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
347 !C->directive()->is_intrinsic_disabled(mh) &&
348 !vmIntrinsics::is_disabled_by_flags(mh);
349
350 }
351
352 if (is_available) {
353 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
354 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
355 return new LibraryIntrinsic(m, is_virtual,
356 vmIntrinsics::predicates_needed(id),
357 vmIntrinsics::does_virtual_dispatch(id),
358 (vmIntrinsics::ID) id);
359 } else {
360 return NULL;
361 }
362 }
363
364 //----------------------register_library_intrinsics-----------------------
365 // Initialize this file's data structures, for each Compile instance.
366 void Compile::register_library_intrinsics() {
367 // Nothing to do here.
368 }
369
370 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
371 LibraryCallKit kit(jvms, this);
372 Compile* C = kit.C;
373 int nodes = C->unique();
374 #ifndef PRODUCT
375 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
376 char buf[1000];
377 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
378 tty->print_cr("Intrinsic %s", str);
379 }
380 #endif
381 ciMethod* callee = kit.callee();
382 const int bci = kit.bci();
383
384 // Try to inline the intrinsic.
385 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
386 kit.try_to_inline(_last_predicate)) {
387 if (C->print_intrinsics() || C->print_inlining()) {
388 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
389 }
390 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
391 if (C->log()) {
392 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
393 vmIntrinsics::name_at(intrinsic_id()),
394 (is_virtual() ? " virtual='1'" : ""),
395 C->unique() - nodes);
396 }
397 // Push the result from the inlined method onto the stack.
398 kit.push_result();
399 C->print_inlining_update(this);
400 return kit.transfer_exceptions_into_jvms();
401 }
402
403 // The intrinsic bailed out
404 if (C->print_intrinsics() || C->print_inlining()) {
405 if (jvms->has_method()) {
406 // Not a root compile.
407 const char* msg;
408 if (callee->intrinsic_candidate()) {
409 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
410 } else {
411 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
412 : "failed to inline (intrinsic), method not annotated";
413 }
414 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
415 } else {
416 // Root compile
417 tty->print("Did not generate intrinsic %s%s at bci:%d in",
418 vmIntrinsics::name_at(intrinsic_id()),
419 (is_virtual() ? " (virtual)" : ""), bci);
420 }
421 }
422 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
423 C->print_inlining_update(this);
424 return NULL;
425 }
426
427 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
428 LibraryCallKit kit(jvms, this);
429 Compile* C = kit.C;
430 int nodes = C->unique();
431 _last_predicate = predicate;
432 #ifndef PRODUCT
433 assert(is_predicated() && predicate < predicates_count(), "sanity");
434 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
435 char buf[1000];
436 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
437 tty->print_cr("Predicate for intrinsic %s", str);
438 }
439 #endif
440 ciMethod* callee = kit.callee();
441 const int bci = kit.bci();
442
443 Node* slow_ctl = kit.try_to_predicate(predicate);
444 if (!kit.failing()) {
445 if (C->print_intrinsics() || C->print_inlining()) {
446 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
447 }
448 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
449 if (C->log()) {
450 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
451 vmIntrinsics::name_at(intrinsic_id()),
452 (is_virtual() ? " virtual='1'" : ""),
453 C->unique() - nodes);
454 }
455 return slow_ctl; // Could be NULL if the check folds.
456 }
457
458 // The intrinsic bailed out
459 if (C->print_intrinsics() || C->print_inlining()) {
460 if (jvms->has_method()) {
461 // Not a root compile.
462 const char* msg = "failed to generate predicate for intrinsic";
463 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
464 } else {
465 // Root compile
466 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
467 vmIntrinsics::name_at(intrinsic_id()),
468 (is_virtual() ? " (virtual)" : ""), bci);
469 }
470 }
471 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
472 return NULL;
473 }
474
475 bool LibraryCallKit::try_to_inline(int predicate) {
476 // Handle symbolic names for otherwise undistinguished boolean switches:
477 const bool is_store = true;
478 const bool is_compress = true;
479 const bool is_native_ptr = true;
480 const bool is_static = true;
481 const bool is_volatile = true;
482
483 if (!jvms()->has_method()) {
484 // Root JVMState has a null method.
485 assert(map()->memory()->Opcode() == Op_Parm, "");
486 // Insert the memory aliasing node
487 set_all_memory(reset_memory());
488 }
489 assert(merged_memory(), "");
490
491
492 switch (intrinsic_id()) {
493 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
494 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
495 case vmIntrinsics::_getClass: return inline_native_getClass();
496
497 case vmIntrinsics::_dsin:
498 case vmIntrinsics::_dcos:
499 case vmIntrinsics::_dtan:
500 case vmIntrinsics::_dabs:
501 case vmIntrinsics::_datan2:
502 case vmIntrinsics::_dsqrt:
503 case vmIntrinsics::_dexp:
504 case vmIntrinsics::_dlog:
505 case vmIntrinsics::_dlog10:
506 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
507
508 case vmIntrinsics::_min:
509 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
510
511 case vmIntrinsics::_notify:
512 case vmIntrinsics::_notifyAll:
513 if (InlineNotify) {
514 return inline_notify(intrinsic_id());
515 }
516 return false;
517
518 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
519 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
520 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
521 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
522 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
523 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
524 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
525 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
526 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
527 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
528 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
529 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
530
531 case vmIntrinsics::_arraycopy: return inline_arraycopy();
532
533 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
534 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
535 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
536 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
537
538 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
539 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
540 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
541 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
542 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
543 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
544 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
545
546 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
547 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
548
549 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
550 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
551 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
552 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
553
554 case vmIntrinsics::_compressStringC:
555 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
556 case vmIntrinsics::_inflateStringC:
557 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
558
559 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Relaxed, false);
560 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Relaxed, false);
561 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Relaxed, false);
562 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, false);
563 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, false);
564 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, false);
565 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, false);
566 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
567 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
568
569 case vmIntrinsics::_putObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Relaxed, false);
570 case vmIntrinsics::_putBoolean: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Relaxed, false);
571 case vmIntrinsics::_putByte: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Relaxed, false);
572 case vmIntrinsics::_putShort: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, false);
573 case vmIntrinsics::_putChar: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, false);
574 case vmIntrinsics::_putInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, false);
575 case vmIntrinsics::_putLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, false);
576 case vmIntrinsics::_putFloat: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Relaxed, false);
577 case vmIntrinsics::_putDouble: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
578
579 case vmIntrinsics::_getByte_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE, Relaxed, false);
580 case vmIntrinsics::_getShort_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT, Relaxed, false);
581 case vmIntrinsics::_getChar_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR, Relaxed, false);
582 case vmIntrinsics::_getInt_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_INT, Relaxed, false);
583 case vmIntrinsics::_getLong_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_LONG, Relaxed, false);
584 case vmIntrinsics::_getFloat_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT, Relaxed, false);
585 case vmIntrinsics::_getDouble_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE, Relaxed, false);
586 case vmIntrinsics::_getAddress_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS, Relaxed, false);
587
588 case vmIntrinsics::_putByte_raw: return inline_unsafe_access( is_native_ptr, is_store, T_BYTE, Relaxed, false);
589 case vmIntrinsics::_putShort_raw: return inline_unsafe_access( is_native_ptr, is_store, T_SHORT, Relaxed, false);
590 case vmIntrinsics::_putChar_raw: return inline_unsafe_access( is_native_ptr, is_store, T_CHAR, Relaxed, false);
591 case vmIntrinsics::_putInt_raw: return inline_unsafe_access( is_native_ptr, is_store, T_INT, Relaxed, false);
592 case vmIntrinsics::_putLong_raw: return inline_unsafe_access( is_native_ptr, is_store, T_LONG, Relaxed, false);
593 case vmIntrinsics::_putFloat_raw: return inline_unsafe_access( is_native_ptr, is_store, T_FLOAT, Relaxed, false);
594 case vmIntrinsics::_putDouble_raw: return inline_unsafe_access( is_native_ptr, is_store, T_DOUBLE, Relaxed, false);
595 case vmIntrinsics::_putAddress_raw: return inline_unsafe_access( is_native_ptr, is_store, T_ADDRESS, Relaxed, false);
596
597 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Volatile, false);
598 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Volatile, false);
599 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Volatile, false);
600 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Volatile, false);
601 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Volatile, false);
602 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Volatile, false);
603 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Volatile, false);
604 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Volatile, false);
605 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Volatile, false);
606
607 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Volatile, false);
608 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Volatile, false);
609 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Volatile, false);
610 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Volatile, false);
611 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Volatile, false);
612 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Volatile, false);
613 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Volatile, false);
614 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Volatile, false);
615 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Volatile, false);
616
617 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, true);
618 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, true);
619 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, true);
620 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, true);
621
622 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, true);
623 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, true);
624 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, true);
625 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, true);
626
627 case vmIntrinsics::_putOrderedObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false);
628 case vmIntrinsics::_putOrderedInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false);
629 case vmIntrinsics::_putOrderedLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false);
630
631 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Acquire, false);
632 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Acquire, false);
633 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Acquire, false);
634 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Acquire, false);
635 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Acquire, false);
636 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Acquire, false);
637 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Acquire, false);
638 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Acquire, false);
639 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Acquire, false);
640
641 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false);
642 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Release, false);
643 case vmIntrinsics::_putByteRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Release, false);
644 case vmIntrinsics::_putShortRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Release, false);
645 case vmIntrinsics::_putCharRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Release, false);
646 case vmIntrinsics::_putIntRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false);
647 case vmIntrinsics::_putLongRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false);
648 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Release, false);
649 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Release, false);
650
651 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Opaque, false);
652 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Opaque, false);
653 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Opaque, false);
654 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Opaque, false);
655 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Opaque, false);
656 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Opaque, false);
657 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Opaque, false);
658 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Opaque, false);
659 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Opaque, false);
660
661 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Opaque, false);
662 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Opaque, false);
663 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Opaque, false);
664 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Opaque, false);
665 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Opaque, false);
666 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Opaque, false);
667 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Opaque, false);
668 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Opaque, false);
669 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Opaque, false);
670
671 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
672 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
673 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
674
675 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
676 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
677 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
678 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
679 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
680 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
681 case vmIntrinsics::_weakCompareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
682 case vmIntrinsics::_weakCompareAndSwapLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
683 case vmIntrinsics::_weakCompareAndSwapLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
684
685 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
686 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
687 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
688 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
689 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
690 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
691 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
692 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
693 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
694
695 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
696 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, 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::_currentThread: return inline_native_currentThread();
706 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
707
708 #ifdef TRACE_HAVE_INTRINSICS
709 case vmIntrinsics::_classID: return inline_native_classID();
710 case vmIntrinsics::_threadID: return inline_native_threadID();
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::_newArray: return inline_native_newArray();
718 case vmIntrinsics::_getLength: return inline_native_getLength();
719 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
720 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
721 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
722 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
723 case vmIntrinsics::_Objects_checkIndex: return inline_objects_checkIndex();
724 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
725
726 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
727
728 case vmIntrinsics::_isInstance:
729 case vmIntrinsics::_getModifiers:
730 case vmIntrinsics::_isInterface:
731 case vmIntrinsics::_isArray:
732 case vmIntrinsics::_isPrimitive:
733 case vmIntrinsics::_getSuperclass:
734 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
735
736 case vmIntrinsics::_floatToRawIntBits:
737 case vmIntrinsics::_floatToIntBits:
738 case vmIntrinsics::_intBitsToFloat:
739 case vmIntrinsics::_doubleToRawLongBits:
740 case vmIntrinsics::_doubleToLongBits:
741 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
742
743 case vmIntrinsics::_numberOfLeadingZeros_i:
744 case vmIntrinsics::_numberOfLeadingZeros_l:
745 case vmIntrinsics::_numberOfTrailingZeros_i:
746 case vmIntrinsics::_numberOfTrailingZeros_l:
747 case vmIntrinsics::_bitCount_i:
748 case vmIntrinsics::_bitCount_l:
749 case vmIntrinsics::_reverseBytes_i:
750 case vmIntrinsics::_reverseBytes_l:
751 case vmIntrinsics::_reverseBytes_s:
752 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
753
754 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
755
756 case vmIntrinsics::_Reference_get: return inline_reference_get();
757
758 case vmIntrinsics::_Class_cast: return inline_Class_cast();
759
760 case vmIntrinsics::_aescrypt_encryptBlock:
761 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
762
763 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
764 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
765 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
766
767 case vmIntrinsics::_counterMode_AESCrypt:
768 return inline_counterMode_AESCrypt(intrinsic_id());
769
770 case vmIntrinsics::_sha_implCompress:
771 case vmIntrinsics::_sha2_implCompress:
772 case vmIntrinsics::_sha5_implCompress:
773 return inline_sha_implCompress(intrinsic_id());
774
775 case vmIntrinsics::_digestBase_implCompressMB:
776 return inline_digestBase_implCompressMB(predicate);
777
778 case vmIntrinsics::_multiplyToLen:
779 return inline_multiplyToLen();
780
781 case vmIntrinsics::_squareToLen:
782 return inline_squareToLen();
783
784 case vmIntrinsics::_mulAdd:
785 return inline_mulAdd();
786
787 case vmIntrinsics::_montgomeryMultiply:
788 return inline_montgomeryMultiply();
789 case vmIntrinsics::_montgomerySquare:
790 return inline_montgomerySquare();
791
792 case vmIntrinsics::_vectorizedMismatch:
793 return inline_vectorizedMismatch();
794
795 case vmIntrinsics::_ghash_processBlocks:
796 return inline_ghash_processBlocks();
797
798 case vmIntrinsics::_encodeISOArray:
799 case vmIntrinsics::_encodeByteISOArray:
800 return inline_encodeISOArray();
801
802 case vmIntrinsics::_updateCRC32:
803 return inline_updateCRC32();
804 case vmIntrinsics::_updateBytesCRC32:
805 return inline_updateBytesCRC32();
806 case vmIntrinsics::_updateByteBufferCRC32:
807 return inline_updateByteBufferCRC32();
808
809 case vmIntrinsics::_updateBytesCRC32C:
810 return inline_updateBytesCRC32C();
811 case vmIntrinsics::_updateDirectByteBufferCRC32C:
812 return inline_updateDirectByteBufferCRC32C();
813
814 case vmIntrinsics::_updateBytesAdler32:
815 return inline_updateBytesAdler32();
816 case vmIntrinsics::_updateByteBufferAdler32:
817 return inline_updateByteBufferAdler32();
818
819 case vmIntrinsics::_profileBoolean:
820 return inline_profileBoolean();
821 case vmIntrinsics::_isCompileConstant:
822 return inline_isCompileConstant();
823
824 case vmIntrinsics::_hasNegatives:
825 return inline_hasNegatives();
826
827 case vmIntrinsics::_deoptimize:
828 return inline_deoptimize();
829
830 default:
831 // If you get here, it may be that someone has added a new intrinsic
832 // to the list in vmSymbols.hpp without implementing it here.
833 #ifndef PRODUCT
834 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
835 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
836 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
837 }
838 #endif
839 return false;
840 }
841 }
842
843 Node* LibraryCallKit::try_to_predicate(int predicate) {
844 if (!jvms()->has_method()) {
845 // Root JVMState has a null method.
846 assert(map()->memory()->Opcode() == Op_Parm, "");
847 // Insert the memory aliasing node
848 set_all_memory(reset_memory());
849 }
850 assert(merged_memory(), "");
851
852 switch (intrinsic_id()) {
853 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
854 return inline_cipherBlockChaining_AESCrypt_predicate(false);
855 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
856 return inline_cipherBlockChaining_AESCrypt_predicate(true);
857 case vmIntrinsics::_counterMode_AESCrypt:
858 return inline_counterMode_AESCrypt_predicate();
859 case vmIntrinsics::_digestBase_implCompressMB:
860 return inline_digestBase_implCompressMB_predicate(predicate);
861
862 default:
863 // If you get here, it may be that someone has added a new intrinsic
864 // to the list in vmSymbols.hpp without implementing it here.
865 #ifndef PRODUCT
866 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
867 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
868 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
869 }
870 #endif
871 Node* slow_ctl = control();
872 set_control(top()); // No fast path instrinsic
873 return slow_ctl;
874 }
875 }
876
877 //------------------------------set_result-------------------------------
878 // Helper function for finishing intrinsics.
879 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
880 record_for_igvn(region);
881 set_control(_gvn.transform(region));
882 set_result( _gvn.transform(value));
883 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
884 }
885
886 //------------------------------generate_guard---------------------------
887 // Helper function for generating guarded fast-slow graph structures.
888 // The given 'test', if true, guards a slow path. If the test fails
889 // then a fast path can be taken. (We generally hope it fails.)
890 // In all cases, GraphKit::control() is updated to the fast path.
891 // The returned value represents the control for the slow path.
892 // The return value is never 'top'; it is either a valid control
893 // or NULL if it is obvious that the slow path can never be taken.
894 // Also, if region and the slow control are not NULL, the slow edge
895 // is appended to the region.
896 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
897 if (stopped()) {
898 // Already short circuited.
899 return NULL;
900 }
901
902 // Build an if node and its projections.
903 // If test is true we take the slow path, which we assume is uncommon.
904 if (_gvn.type(test) == TypeInt::ZERO) {
905 // The slow branch is never taken. No need to build this guard.
906 return NULL;
907 }
908
909 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
910
911 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
912 if (if_slow == top()) {
913 // The slow branch is never taken. No need to build this guard.
914 return NULL;
915 }
916
917 if (region != NULL)
918 region->add_req(if_slow);
919
920 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
921 set_control(if_fast);
922
923 return if_slow;
924 }
925
926 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
927 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
928 }
929 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
930 return generate_guard(test, region, PROB_FAIR);
931 }
932
933 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
934 Node* *pos_index) {
935 if (stopped())
936 return NULL; // already stopped
937 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
938 return NULL; // index is already adequately typed
939 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
940 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
941 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
942 if (is_neg != NULL && pos_index != NULL) {
943 // Emulate effect of Parse::adjust_map_after_if.
944 Node* ccast = new CastIINode(index, TypeInt::POS);
945 ccast->set_req(0, control());
946 (*pos_index) = _gvn.transform(ccast);
947 }
948 return is_neg;
949 }
950
951 // Make sure that 'position' is a valid limit index, in [0..length].
952 // There are two equivalent plans for checking this:
953 // A. (offset + copyLength) unsigned<= arrayLength
954 // B. offset <= (arrayLength - copyLength)
955 // We require that all of the values above, except for the sum and
956 // difference, are already known to be non-negative.
957 // Plan A is robust in the face of overflow, if offset and copyLength
958 // are both hugely positive.
959 //
960 // Plan B is less direct and intuitive, but it does not overflow at
961 // all, since the difference of two non-negatives is always
962 // representable. Whenever Java methods must perform the equivalent
963 // check they generally use Plan B instead of Plan A.
964 // For the moment we use Plan A.
965 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
966 Node* subseq_length,
967 Node* array_length,
968 RegionNode* region) {
969 if (stopped())
970 return NULL; // already stopped
971 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
972 if (zero_offset && subseq_length->eqv_uncast(array_length))
973 return NULL; // common case of whole-array copy
974 Node* last = subseq_length;
975 if (!zero_offset) // last += offset
976 last = _gvn.transform(new AddINode(last, offset));
977 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
978 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
979 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
980 return is_over;
981 }
982
983 // Emit range checks for the given String.value byte array
984 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
985 if (stopped()) {
986 return; // already stopped
987 }
988 RegionNode* bailout = new RegionNode(1);
989 record_for_igvn(bailout);
990 if (char_count) {
991 // Convert char count to byte count
992 count = _gvn.transform(new LShiftINode(count, intcon(1)));
993 }
994
995 // Offset and count must not be negative
996 generate_negative_guard(offset, bailout);
997 generate_negative_guard(count, bailout);
998 // Offset + count must not exceed length of array
999 generate_limit_guard(offset, count, load_array_length(array), bailout);
1000
1001 if (bailout->req() > 1) {
1002 PreserveJVMState pjvms(this);
1003 set_control(_gvn.transform(bailout));
1004 uncommon_trap(Deoptimization::Reason_intrinsic,
1005 Deoptimization::Action_maybe_recompile);
1006 }
1007 }
1008
1009 //--------------------------generate_current_thread--------------------
1010 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1011 ciKlass* thread_klass = env()->Thread_klass();
1012 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1013 Node* thread = _gvn.transform(new ThreadLocalNode());
1014 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1015 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1016 tls_output = thread;
1017 return threadObj;
1018 }
1019
1020
1021 //------------------------------make_string_method_node------------------------
1022 // Helper method for String intrinsic functions. This version is called with
1023 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1024 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1025 // containing the lengths of str1 and str2.
1026 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1027 Node* result = NULL;
1028 switch (opcode) {
1029 case Op_StrIndexOf:
1030 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1031 str1_start, cnt1, str2_start, cnt2, ae);
1032 break;
1033 case Op_StrComp:
1034 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1035 str1_start, cnt1, str2_start, cnt2, ae);
1036 break;
1037 case Op_StrEquals:
1038 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1039 // Use the constant length if there is one because optimized match rule may exist.
1040 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1041 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1042 break;
1043 default:
1044 ShouldNotReachHere();
1045 return NULL;
1046 }
1047
1048 // All these intrinsics have checks.
1049 C->set_has_split_ifs(true); // Has chance for split-if optimization
1050
1051 return _gvn.transform(result);
1052 }
1053
1054 //------------------------------inline_string_compareTo------------------------
1055 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1056 Node* arg1 = argument(0);
1057 Node* arg2 = argument(1);
1058
1059 // Get start addr and length of first argument
1060 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1061 Node* arg1_cnt = load_array_length(arg1);
1062
1063 // Get start addr and length of second argument
1064 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1065 Node* arg2_cnt = load_array_length(arg2);
1066
1067 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1068 set_result(result);
1069 return true;
1070 }
1071
1072 //------------------------------inline_string_equals------------------------
1073 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1074 Node* arg1 = argument(0);
1075 Node* arg2 = argument(1);
1076
1077 // paths (plus control) merge
1078 RegionNode* region = new RegionNode(3);
1079 Node* phi = new PhiNode(region, TypeInt::BOOL);
1080
1081 if (!stopped()) {
1082 // Get start addr and length of first argument
1083 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1084 Node* arg1_cnt = load_array_length(arg1);
1085
1086 // Get start addr and length of second argument
1087 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1088 Node* arg2_cnt = load_array_length(arg2);
1089
1090 // Check for arg1_cnt != arg2_cnt
1091 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1092 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1093 Node* if_ne = generate_slow_guard(bol, NULL);
1094 if (if_ne != NULL) {
1095 phi->init_req(2, intcon(0));
1096 region->init_req(2, if_ne);
1097 }
1098
1099 // Check for count == 0 is done by assembler code for StrEquals.
1100
1101 if (!stopped()) {
1102 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1103 phi->init_req(1, equals);
1104 region->init_req(1, control());
1105 }
1106 }
1107
1108 // post merge
1109 set_control(_gvn.transform(region));
1110 record_for_igvn(region);
1111
1112 set_result(_gvn.transform(phi));
1113 return true;
1114 }
1115
1116 //------------------------------inline_array_equals----------------------------
1117 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1118 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1119 Node* arg1 = argument(0);
1120 Node* arg2 = argument(1);
1121
1122 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1123 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1124 return true;
1125 }
1126
1127 //------------------------------inline_hasNegatives------------------------------
1128 bool LibraryCallKit::inline_hasNegatives() {
1129 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1130 return false;
1131 }
1132
1133 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1134 // no receiver since it is static method
1135 Node* ba = argument(0);
1136 Node* offset = argument(1);
1137 Node* len = argument(2);
1138
1139 // Range checks
1140 generate_string_range_check(ba, offset, len, false);
1141 if (stopped()) {
1142 return true;
1143 }
1144 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1145 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1146 set_result(_gvn.transform(result));
1147 return true;
1148 }
1149
1150 bool LibraryCallKit::inline_objects_checkIndex() {
1151 Node* index = argument(0);
1152 Node* length = argument(1);
1153 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1154 return false;
1155 }
1156
1157 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1158 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1159
1160 {
1161 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1162 uncommon_trap(Deoptimization::Reason_intrinsic,
1163 Deoptimization::Action_make_not_entrant);
1164 }
1165
1166 if (stopped()) {
1167 return false;
1168 }
1169
1170 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1171 BoolTest::mask btest = BoolTest::lt;
1172 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1173 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1174 _gvn.set_type(rc, rc->Value(&_gvn));
1175 if (!rc_bool->is_Con()) {
1176 record_for_igvn(rc);
1177 }
1178 set_control(_gvn.transform(new IfTrueNode(rc)));
1179 {
1180 PreserveJVMState pjvms(this);
1181 set_control(_gvn.transform(new IfFalseNode(rc)));
1182 uncommon_trap(Deoptimization::Reason_range_check,
1183 Deoptimization::Action_make_not_entrant);
1184 }
1185
1186 if (stopped()) {
1187 return false;
1188 }
1189
1190 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1191 result->set_req(0, control());
1192 result = _gvn.transform(result);
1193 set_result(result);
1194 replace_in_map(index, result);
1195 return true;
1196 }
1197
1198 //------------------------------inline_string_indexOf------------------------
1199 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1200 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1201 return false;
1202 }
1203 Node* src = argument(0);
1204 Node* tgt = argument(1);
1205
1206 // Make the merge point
1207 RegionNode* result_rgn = new RegionNode(4);
1208 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1209
1210 // Get start addr and length of source string
1211 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1212 Node* src_count = load_array_length(src);
1213
1214 // Get start addr and length of substring
1215 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1216 Node* tgt_count = load_array_length(tgt);
1217
1218 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1219 // Divide src size by 2 if String is UTF16 encoded
1220 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1221 }
1222 if (ae == StrIntrinsicNode::UU) {
1223 // Divide substring size by 2 if String is UTF16 encoded
1224 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1225 }
1226
1227 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1228 if (result != NULL) {
1229 result_phi->init_req(3, result);
1230 result_rgn->init_req(3, control());
1231 }
1232 set_control(_gvn.transform(result_rgn));
1233 record_for_igvn(result_rgn);
1234 set_result(_gvn.transform(result_phi));
1235
1236 return true;
1237 }
1238
1239 //-----------------------------inline_string_indexOf-----------------------
1240 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1241 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1242 return false;
1243 }
1244 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1245 return false;
1246 }
1247 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1248 Node* src = argument(0); // byte[]
1249 Node* src_count = argument(1); // char count
1250 Node* tgt = argument(2); // byte[]
1251 Node* tgt_count = argument(3); // char count
1252 Node* from_index = argument(4); // char index
1253
1254 // Multiply byte array index by 2 if String is UTF16 encoded
1255 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1256 src_count = _gvn.transform(new SubINode(src_count, from_index));
1257 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1258 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1259
1260 // Range checks
1261 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1262 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1263 if (stopped()) {
1264 return true;
1265 }
1266
1267 RegionNode* region = new RegionNode(5);
1268 Node* phi = new PhiNode(region, TypeInt::INT);
1269
1270 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1271 if (result != NULL) {
1272 // The result is index relative to from_index if substring was found, -1 otherwise.
1273 // Generate code which will fold into cmove.
1274 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1275 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1276
1277 Node* if_lt = generate_slow_guard(bol, NULL);
1278 if (if_lt != NULL) {
1279 // result == -1
1280 phi->init_req(3, result);
1281 region->init_req(3, if_lt);
1282 }
1283 if (!stopped()) {
1284 result = _gvn.transform(new AddINode(result, from_index));
1285 phi->init_req(4, result);
1286 region->init_req(4, control());
1287 }
1288 }
1289
1290 set_control(_gvn.transform(region));
1291 record_for_igvn(region);
1292 set_result(_gvn.transform(phi));
1293
1294 return true;
1295 }
1296
1297 // Create StrIndexOfNode with fast path checks
1298 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1299 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1300 // Check for substr count > string count
1301 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1302 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1303 Node* if_gt = generate_slow_guard(bol, NULL);
1304 if (if_gt != NULL) {
1305 phi->init_req(1, intcon(-1));
1306 region->init_req(1, if_gt);
1307 }
1308 if (!stopped()) {
1309 // Check for substr count == 0
1310 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1311 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1312 Node* if_zero = generate_slow_guard(bol, NULL);
1313 if (if_zero != NULL) {
1314 phi->init_req(2, intcon(0));
1315 region->init_req(2, if_zero);
1316 }
1317 }
1318 if (!stopped()) {
1319 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1320 }
1321 return NULL;
1322 }
1323
1324 //-----------------------------inline_string_indexOfChar-----------------------
1325 bool LibraryCallKit::inline_string_indexOfChar() {
1326 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1327 return false;
1328 }
1329 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1330 return false;
1331 }
1332 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1333 Node* src = argument(0); // byte[]
1334 Node* tgt = argument(1); // tgt is int ch
1335 Node* from_index = argument(2);
1336 Node* max = argument(3);
1337
1338 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1339 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1340 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1341
1342 // Range checks
1343 generate_string_range_check(src, src_offset, src_count, true);
1344 if (stopped()) {
1345 return true;
1346 }
1347
1348 RegionNode* region = new RegionNode(3);
1349 Node* phi = new PhiNode(region, TypeInt::INT);
1350
1351 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1352 C->set_has_split_ifs(true); // Has chance for split-if optimization
1353 _gvn.transform(result);
1354
1355 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1356 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1357
1358 Node* if_lt = generate_slow_guard(bol, NULL);
1359 if (if_lt != NULL) {
1360 // result == -1
1361 phi->init_req(2, result);
1362 region->init_req(2, if_lt);
1363 }
1364 if (!stopped()) {
1365 result = _gvn.transform(new AddINode(result, from_index));
1366 phi->init_req(1, result);
1367 region->init_req(1, control());
1368 }
1369 set_control(_gvn.transform(region));
1370 record_for_igvn(region);
1371 set_result(_gvn.transform(phi));
1372
1373 return true;
1374 }
1375 //---------------------------inline_string_copy---------------------
1376 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1377 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1378 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1379 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1380 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1381 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1382 bool LibraryCallKit::inline_string_copy(bool compress) {
1383 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1384 return false;
1385 }
1386 int nargs = 5; // 2 oops, 3 ints
1387 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1388
1389 Node* src = argument(0);
1390 Node* src_offset = argument(1);
1391 Node* dst = argument(2);
1392 Node* dst_offset = argument(3);
1393 Node* length = argument(4);
1394
1395 // Check for allocation before we add nodes that would confuse
1396 // tightly_coupled_allocation()
1397 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1398
1399 // Figure out the size and type of the elements we will be copying.
1400 const Type* src_type = src->Value(&_gvn);
1401 const Type* dst_type = dst->Value(&_gvn);
1402 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1403 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1404 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1405 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1406 "Unsupported array types for inline_string_copy");
1407
1408 // Range checks
1409 generate_string_range_check(src, src_offset, length, compress && src_elem == T_BYTE);
1410 generate_string_range_check(dst, dst_offset, length, !compress && dst_elem == T_BYTE);
1411 if (stopped()) {
1412 return true;
1413 }
1414
1415 // Convert char[] offsets to byte[] offsets
1416 if (compress && src_elem == T_BYTE) {
1417 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1418 } else if (!compress && dst_elem == T_BYTE) {
1419 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1420 }
1421
1422 Node* src_start = array_element_address(src, src_offset, src_elem);
1423 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1424 // 'src_start' points to src array + scaled offset
1425 // 'dst_start' points to dst array + scaled offset
1426 Node* count = NULL;
1427 if (compress) {
1428 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1429 } else {
1430 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1431 }
1432
1433 if (alloc != NULL) {
1434 if (alloc->maybe_set_complete(&_gvn)) {
1435 // "You break it, you buy it."
1436 InitializeNode* init = alloc->initialization();
1437 assert(init->is_complete(), "we just did this");
1438 init->set_complete_with_arraycopy();
1439 assert(dst->is_CheckCastPP(), "sanity");
1440 assert(dst->in(0)->in(0) == init, "dest pinned");
1441 }
1442 // Do not let stores that initialize this object be reordered with
1443 // a subsequent store that would make this object accessible by
1444 // other threads.
1445 // Record what AllocateNode this StoreStore protects so that
1446 // escape analysis can go from the MemBarStoreStoreNode to the
1447 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1448 // based on the escape status of the AllocateNode.
1449 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1450 }
1451 if (compress) {
1452 set_result(_gvn.transform(count));
1453 }
1454 return true;
1455 }
1456
1457 #ifdef _LP64
1458 #define XTOP ,top() /*additional argument*/
1459 #else //_LP64
1460 #define XTOP /*no additional argument*/
1461 #endif //_LP64
1462
1463 //------------------------inline_string_toBytesU--------------------------
1464 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1465 bool LibraryCallKit::inline_string_toBytesU() {
1466 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1467 return false;
1468 }
1469 // Get the arguments.
1470 Node* value = argument(0);
1471 Node* offset = argument(1);
1472 Node* length = argument(2);
1473
1474 Node* newcopy = NULL;
1475
1476 // Set the original stack and the reexecute bit for the interpreter to reexecute
1477 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1478 { PreserveReexecuteState preexecs(this);
1479 jvms()->set_should_reexecute(true);
1480
1481 // Check if a null path was taken unconditionally.
1482 value = null_check(value);
1483
1484 RegionNode* bailout = new RegionNode(1);
1485 record_for_igvn(bailout);
1486
1487 // Range checks
1488 generate_negative_guard(offset, bailout);
1489 generate_negative_guard(length, bailout);
1490 generate_limit_guard(offset, length, load_array_length(value), bailout);
1491 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1492 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1493
1494 if (bailout->req() > 1) {
1495 PreserveJVMState pjvms(this);
1496 set_control(_gvn.transform(bailout));
1497 uncommon_trap(Deoptimization::Reason_intrinsic,
1498 Deoptimization::Action_maybe_recompile);
1499 }
1500 if (stopped()) {
1501 return true;
1502 }
1503
1504 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1505 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1506 newcopy = new_array(klass_node, size, 0); // no arguments to push
1507 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1508
1509 // Calculate starting addresses.
1510 Node* src_start = array_element_address(value, offset, T_CHAR);
1511 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1512
1513 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1514 const TypeInt* toffset = gvn().type(offset)->is_int();
1515 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1516
1517 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1518 const char* copyfunc_name = "arraycopy";
1519 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1520 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1521 OptoRuntime::fast_arraycopy_Type(),
1522 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1523 src_start, dst_start, ConvI2X(length) XTOP);
1524 // Do not let reads from the cloned object float above the arraycopy.
1525 if (alloc != NULL) {
1526 if (alloc->maybe_set_complete(&_gvn)) {
1527 // "You break it, you buy it."
1528 InitializeNode* init = alloc->initialization();
1529 assert(init->is_complete(), "we just did this");
1530 init->set_complete_with_arraycopy();
1531 assert(newcopy->is_CheckCastPP(), "sanity");
1532 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1533 }
1534 // Do not let stores that initialize this object be reordered with
1535 // a subsequent store that would make this object accessible by
1536 // other threads.
1537 // Record what AllocateNode this StoreStore protects so that
1538 // escape analysis can go from the MemBarStoreStoreNode to the
1539 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1540 // based on the escape status of the AllocateNode.
1541 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1542 } else {
1543 insert_mem_bar(Op_MemBarCPUOrder);
1544 }
1545 } // original reexecute is set back here
1546
1547 C->set_has_split_ifs(true); // Has chance for split-if optimization
1548 if (!stopped()) {
1549 set_result(newcopy);
1550 }
1551 return true;
1552 }
1553
1554 //------------------------inline_string_getCharsU--------------------------
1555 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1556 bool LibraryCallKit::inline_string_getCharsU() {
1557 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1558 return false;
1559 }
1560
1561 // Get the arguments.
1562 Node* src = argument(0);
1563 Node* src_begin = argument(1);
1564 Node* src_end = argument(2); // exclusive offset (i < src_end)
1565 Node* dst = argument(3);
1566 Node* dst_begin = argument(4);
1567
1568 // Check for allocation before we add nodes that would confuse
1569 // tightly_coupled_allocation()
1570 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1571
1572 // Check if a null path was taken unconditionally.
1573 src = null_check(src);
1574 dst = null_check(dst);
1575 if (stopped()) {
1576 return true;
1577 }
1578
1579 // Get length and convert char[] offset to byte[] offset
1580 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1581 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1582
1583 // Range checks
1584 generate_string_range_check(src, src_begin, length, true);
1585 generate_string_range_check(dst, dst_begin, length, false);
1586 if (stopped()) {
1587 return true;
1588 }
1589
1590 if (!stopped()) {
1591 // Calculate starting addresses.
1592 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1593 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1594
1595 // Check if array addresses are aligned to HeapWordSize
1596 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1597 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1598 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1599 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1600
1601 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1602 const char* copyfunc_name = "arraycopy";
1603 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1604 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1605 OptoRuntime::fast_arraycopy_Type(),
1606 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1607 src_start, dst_start, ConvI2X(length) XTOP);
1608 // Do not let reads from the cloned object float above the arraycopy.
1609 if (alloc != NULL) {
1610 if (alloc->maybe_set_complete(&_gvn)) {
1611 // "You break it, you buy it."
1612 InitializeNode* init = alloc->initialization();
1613 assert(init->is_complete(), "we just did this");
1614 init->set_complete_with_arraycopy();
1615 assert(dst->is_CheckCastPP(), "sanity");
1616 assert(dst->in(0)->in(0) == init, "dest pinned");
1617 }
1618 // Do not let stores that initialize this object be reordered with
1619 // a subsequent store that would make this object accessible by
1620 // other threads.
1621 // Record what AllocateNode this StoreStore protects so that
1622 // escape analysis can go from the MemBarStoreStoreNode to the
1623 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1624 // based on the escape status of the AllocateNode.
1625 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1626 } else {
1627 insert_mem_bar(Op_MemBarCPUOrder);
1628 }
1629 }
1630
1631 C->set_has_split_ifs(true); // Has chance for split-if optimization
1632 return true;
1633 }
1634
1635 //----------------------inline_string_char_access----------------------------
1636 // Store/Load char to/from byte[] array.
1637 // static void StringUTF16.putChar(byte[] val, int index, int c)
1638 // static char StringUTF16.getChar(byte[] val, int index)
1639 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1640 Node* value = argument(0);
1641 Node* index = argument(1);
1642 Node* ch = is_store ? argument(2) : NULL;
1643
1644 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1645 // correctly requires matched array shapes.
1646 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1647 "sanity: byte[] and char[] bases agree");
1648 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1649 "sanity: byte[] and char[] scales agree");
1650
1651 // Bail when getChar over constants is requested: constant folding would
1652 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1653 // Java method would constant fold nicely instead.
1654 if (!is_store && value->is_Con() && index->is_Con()) {
1655 return false;
1656 }
1657
1658 Node* adr = array_element_address(value, index, T_CHAR);
1659 if (is_store) {
1660 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1661 false, false, true /* mismatched */);
1662 } else {
1663 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1664 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1665 set_result(ch);
1666 }
1667 return true;
1668 }
1669
1670 //--------------------------round_double_node--------------------------------
1671 // Round a double node if necessary.
1672 Node* LibraryCallKit::round_double_node(Node* n) {
1673 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1674 n = _gvn.transform(new RoundDoubleNode(0, n));
1675 return n;
1676 }
1677
1678 //------------------------------inline_math-----------------------------------
1679 // public static double Math.abs(double)
1680 // public static double Math.sqrt(double)
1681 // public static double Math.log(double)
1682 // public static double Math.log10(double)
1683 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1684 Node* arg = round_double_node(argument(0));
1685 Node* n = NULL;
1686 switch (id) {
1687 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1688 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1689 case vmIntrinsics::_dlog10: n = new Log10DNode(C, control(), arg); break;
1690 default: fatal_unexpected_iid(id); break;
1691 }
1692 set_result(_gvn.transform(n));
1693 return true;
1694 }
1695
1696 //------------------------------inline_trig----------------------------------
1697 // Inline sin/cos/tan instructions, if possible. If rounding is required, do
1698 // argument reduction which will turn into a fast/slow diamond.
1699 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1700 Node* arg = round_double_node(argument(0));
1701 Node* n = NULL;
1702
1703 switch (id) {
1704 case vmIntrinsics::_dtan: n = new TanDNode(C, control(), arg); break;
1705 default: fatal_unexpected_iid(id); break;
1706 }
1707 n = _gvn.transform(n);
1708
1709 // Rounding required? Check for argument reduction!
1710 if (Matcher::strict_fp_requires_explicit_rounding) {
1711 static const double pi_4 = 0.7853981633974483;
1712 static const double neg_pi_4 = -0.7853981633974483;
1713 // pi/2 in 80-bit extended precision
1714 // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
1715 // -pi/2 in 80-bit extended precision
1716 // static const unsigned char neg_pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0xbf,0x00,0x00,0x00,0x00,0x00,0x00};
1717 // Cutoff value for using this argument reduction technique
1718 //static const double pi_2_minus_epsilon = 1.564660403643354;
1719 //static const double neg_pi_2_plus_epsilon = -1.564660403643354;
1720
1721 // Pseudocode for sin:
1722 // if (x <= Math.PI / 4.0) {
1723 // if (x >= -Math.PI / 4.0) return fsin(x);
1724 // if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
1725 // } else {
1726 // if (x <= Math.PI / 2.0) return fcos(x - Math.PI / 2.0);
1727 // }
1728 // return StrictMath.sin(x);
1729
1730 // Pseudocode for cos:
1731 // if (x <= Math.PI / 4.0) {
1732 // if (x >= -Math.PI / 4.0) return fcos(x);
1733 // if (x >= -Math.PI / 2.0) return fsin(x + Math.PI / 2.0);
1734 // } else {
1735 // if (x <= Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
1736 // }
1737 // return StrictMath.cos(x);
1738
1739 // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
1740 // requires a special machine instruction to load it. Instead we'll try
1741 // the 'easy' case. If we really need the extra range +/- PI/2 we'll
1742 // probably do the math inside the SIN encoding.
1743
1744 // Make the merge point
1745 RegionNode* r = new RegionNode(3);
1746 Node* phi = new PhiNode(r, Type::DOUBLE);
1747
1748 // Flatten arg so we need only 1 test
1749 Node *abs = _gvn.transform(new AbsDNode(arg));
1750 // Node for PI/4 constant
1751 Node *pi4 = makecon(TypeD::make(pi_4));
1752 // Check PI/4 : abs(arg)
1753 Node *cmp = _gvn.transform(new CmpDNode(pi4,abs));
1754 // Check: If PI/4 < abs(arg) then go slow
1755 Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt ));
1756 // Branch either way
1757 IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1758 set_control(opt_iff(r,iff));
1759
1760 // Set fast path result
1761 phi->init_req(2, n);
1762
1763 // Slow path - non-blocking leaf call
1764 Node* call = NULL;
1765 switch (id) {
1766 case vmIntrinsics::_dtan:
1767 call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
1768 CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
1769 "Tan", NULL, arg, top());
1770 break;
1771 }
1772 assert(control()->in(0) == call, "");
1773 Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1774 r->init_req(1, control());
1775 phi->init_req(1, slow_result);
1776
1777 // Post-merge
1778 set_control(_gvn.transform(r));
1779 record_for_igvn(r);
1780 n = _gvn.transform(phi);
1781
1782 C->set_has_split_ifs(true); // Has chance for split-if optimization
1783 }
1784 set_result(n);
1785 return true;
1786 }
1787
1788 //------------------------------runtime_math-----------------------------
1789 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1790 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1791 "must be (DD)D or (D)D type");
1792
1793 // Inputs
1794 Node* a = round_double_node(argument(0));
1795 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1796
1797 const TypePtr* no_memory_effects = NULL;
1798 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1799 no_memory_effects,
1800 a, top(), b, b ? top() : NULL);
1801 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1802 #ifdef ASSERT
1803 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1804 assert(value_top == top(), "second value must be top");
1805 #endif
1806
1807 set_result(value);
1808 return true;
1809 }
1810
1811 //------------------------------inline_math_native-----------------------------
1812 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1813 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1814 switch (id) {
1815 // These intrinsics are not properly supported on all hardware
1816 case vmIntrinsics::_dsin:
1817 return StubRoutines::dsin() != NULL ?
1818 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1819 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1820 case vmIntrinsics::_dcos:
1821 return StubRoutines::dcos() != NULL ?
1822 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1823 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1824 case vmIntrinsics::_dtan: return Matcher::has_match_rule(Op_TanD) ? inline_trig(id) :
1825 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1826
1827 case vmIntrinsics::_dlog:
1828 return StubRoutines::dlog() != NULL ?
1829 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1830 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1831 case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_math(id) :
1832 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1833
1834 // These intrinsics are supported on all hardware
1835 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1836 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1837
1838 case vmIntrinsics::_dexp:
1839 return StubRoutines::dexp() != NULL ?
1840 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1841 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1842 case vmIntrinsics::_dpow:
1843 return StubRoutines::dpow() != NULL ?
1844 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1845 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1846 #undef FN_PTR
1847
1848 // These intrinsics are not yet correctly implemented
1849 case vmIntrinsics::_datan2:
1850 return false;
1851
1852 default:
1853 fatal_unexpected_iid(id);
1854 return false;
1855 }
1856 }
1857
1858 static bool is_simple_name(Node* n) {
1859 return (n->req() == 1 // constant
1860 || (n->is_Type() && n->as_Type()->type()->singleton())
1861 || n->is_Proj() // parameter or return value
1862 || n->is_Phi() // local of some sort
1863 );
1864 }
1865
1866 //----------------------------inline_notify-----------------------------------*
1867 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1868 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1869 address func;
1870 if (id == vmIntrinsics::_notify) {
1871 func = OptoRuntime::monitor_notify_Java();
1872 } else {
1873 func = OptoRuntime::monitor_notifyAll_Java();
1874 }
1875 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1876 make_slow_call_ex(call, env()->Throwable_klass(), false);
1877 return true;
1878 }
1879
1880
1881 //----------------------------inline_min_max-----------------------------------
1882 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1883 set_result(generate_min_max(id, argument(0), argument(1)));
1884 return true;
1885 }
1886
1887 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1888 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1889 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1890 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1891 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1892
1893 {
1894 PreserveJVMState pjvms(this);
1895 PreserveReexecuteState preexecs(this);
1896 jvms()->set_should_reexecute(true);
1897
1898 set_control(slow_path);
1899 set_i_o(i_o());
1900
1901 uncommon_trap(Deoptimization::Reason_intrinsic,
1902 Deoptimization::Action_none);
1903 }
1904
1905 set_control(fast_path);
1906 set_result(math);
1907 }
1908
1909 template <typename OverflowOp>
1910 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1911 typedef typename OverflowOp::MathOp MathOp;
1912
1913 MathOp* mathOp = new MathOp(arg1, arg2);
1914 Node* operation = _gvn.transform( mathOp );
1915 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1916 inline_math_mathExact(operation, ofcheck);
1917 return true;
1918 }
1919
1920 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1921 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1922 }
1923
1924 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1925 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1926 }
1927
1928 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1929 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1930 }
1931
1932 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1933 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1934 }
1935
1936 bool LibraryCallKit::inline_math_negateExactI() {
1937 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1938 }
1939
1940 bool LibraryCallKit::inline_math_negateExactL() {
1941 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1942 }
1943
1944 bool LibraryCallKit::inline_math_multiplyExactI() {
1945 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1946 }
1947
1948 bool LibraryCallKit::inline_math_multiplyExactL() {
1949 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1950 }
1951
1952 Node*
1953 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1954 // These are the candidate return value:
1955 Node* xvalue = x0;
1956 Node* yvalue = y0;
1957
1958 if (xvalue == yvalue) {
1959 return xvalue;
1960 }
1961
1962 bool want_max = (id == vmIntrinsics::_max);
1963
1964 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1965 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1966 if (txvalue == NULL || tyvalue == NULL) return top();
1967 // This is not really necessary, but it is consistent with a
1968 // hypothetical MaxINode::Value method:
1969 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1970
1971 // %%% This folding logic should (ideally) be in a different place.
1972 // Some should be inside IfNode, and there to be a more reliable
1973 // transformation of ?: style patterns into cmoves. We also want
1974 // more powerful optimizations around cmove and min/max.
1975
1976 // Try to find a dominating comparison of these guys.
1977 // It can simplify the index computation for Arrays.copyOf
1978 // and similar uses of System.arraycopy.
1979 // First, compute the normalized version of CmpI(x, y).
1980 int cmp_op = Op_CmpI;
1981 Node* xkey = xvalue;
1982 Node* ykey = yvalue;
1983 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1984 if (ideal_cmpxy->is_Cmp()) {
1985 // E.g., if we have CmpI(length - offset, count),
1986 // it might idealize to CmpI(length, count + offset)
1987 cmp_op = ideal_cmpxy->Opcode();
1988 xkey = ideal_cmpxy->in(1);
1989 ykey = ideal_cmpxy->in(2);
1990 }
1991
1992 // Start by locating any relevant comparisons.
1993 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1994 Node* cmpxy = NULL;
1995 Node* cmpyx = NULL;
1996 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1997 Node* cmp = start_from->fast_out(k);
1998 if (cmp->outcnt() > 0 && // must have prior uses
1999 cmp->in(0) == NULL && // must be context-independent
2000 cmp->Opcode() == cmp_op) { // right kind of compare
2001 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2002 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2003 }
2004 }
2005
2006 const int NCMPS = 2;
2007 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2008 int cmpn;
2009 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2010 if (cmps[cmpn] != NULL) break; // find a result
2011 }
2012 if (cmpn < NCMPS) {
2013 // Look for a dominating test that tells us the min and max.
2014 int depth = 0; // Limit search depth for speed
2015 Node* dom = control();
2016 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2017 if (++depth >= 100) break;
2018 Node* ifproj = dom;
2019 if (!ifproj->is_Proj()) continue;
2020 Node* iff = ifproj->in(0);
2021 if (!iff->is_If()) continue;
2022 Node* bol = iff->in(1);
2023 if (!bol->is_Bool()) continue;
2024 Node* cmp = bol->in(1);
2025 if (cmp == NULL) continue;
2026 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2027 if (cmps[cmpn] == cmp) break;
2028 if (cmpn == NCMPS) continue;
2029 BoolTest::mask btest = bol->as_Bool()->_test._test;
2030 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2031 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2032 // At this point, we know that 'x btest y' is true.
2033 switch (btest) {
2034 case BoolTest::eq:
2035 // They are proven equal, so we can collapse the min/max.
2036 // Either value is the answer. Choose the simpler.
2037 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2038 return yvalue;
2039 return xvalue;
2040 case BoolTest::lt: // x < y
2041 case BoolTest::le: // x <= y
2042 return (want_max ? yvalue : xvalue);
2043 case BoolTest::gt: // x > y
2044 case BoolTest::ge: // x >= y
2045 return (want_max ? xvalue : yvalue);
2046 }
2047 }
2048 }
2049
2050 // We failed to find a dominating test.
2051 // Let's pick a test that might GVN with prior tests.
2052 Node* best_bol = NULL;
2053 BoolTest::mask best_btest = BoolTest::illegal;
2054 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2055 Node* cmp = cmps[cmpn];
2056 if (cmp == NULL) continue;
2057 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2058 Node* bol = cmp->fast_out(j);
2059 if (!bol->is_Bool()) continue;
2060 BoolTest::mask btest = bol->as_Bool()->_test._test;
2061 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2062 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2063 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2064 best_bol = bol->as_Bool();
2065 best_btest = btest;
2066 }
2067 }
2068 }
2069
2070 Node* answer_if_true = NULL;
2071 Node* answer_if_false = NULL;
2072 switch (best_btest) {
2073 default:
2074 if (cmpxy == NULL)
2075 cmpxy = ideal_cmpxy;
2076 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2077 // and fall through:
2078 case BoolTest::lt: // x < y
2079 case BoolTest::le: // x <= y
2080 answer_if_true = (want_max ? yvalue : xvalue);
2081 answer_if_false = (want_max ? xvalue : yvalue);
2082 break;
2083 case BoolTest::gt: // x > y
2084 case BoolTest::ge: // x >= y
2085 answer_if_true = (want_max ? xvalue : yvalue);
2086 answer_if_false = (want_max ? yvalue : xvalue);
2087 break;
2088 }
2089
2090 jint hi, lo;
2091 if (want_max) {
2092 // We can sharpen the minimum.
2093 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2094 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2095 } else {
2096 // We can sharpen the maximum.
2097 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2098 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2099 }
2100
2101 // Use a flow-free graph structure, to avoid creating excess control edges
2102 // which could hinder other optimizations.
2103 // Since Math.min/max is often used with arraycopy, we want
2104 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2105 Node* cmov = CMoveNode::make(NULL, best_bol,
2106 answer_if_false, answer_if_true,
2107 TypeInt::make(lo, hi, widen));
2108
2109 return _gvn.transform(cmov);
2110
2111 /*
2112 // This is not as desirable as it may seem, since Min and Max
2113 // nodes do not have a full set of optimizations.
2114 // And they would interfere, anyway, with 'if' optimizations
2115 // and with CMoveI canonical forms.
2116 switch (id) {
2117 case vmIntrinsics::_min:
2118 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2119 case vmIntrinsics::_max:
2120 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2121 default:
2122 ShouldNotReachHere();
2123 }
2124 */
2125 }
2126
2127 inline int
2128 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2129 const TypePtr* base_type = TypePtr::NULL_PTR;
2130 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2131 if (base_type == NULL) {
2132 // Unknown type.
2133 return Type::AnyPtr;
2134 } else if (base_type == TypePtr::NULL_PTR) {
2135 // Since this is a NULL+long form, we have to switch to a rawptr.
2136 base = _gvn.transform(new CastX2PNode(offset));
2137 offset = MakeConX(0);
2138 return Type::RawPtr;
2139 } else if (base_type->base() == Type::RawPtr) {
2140 return Type::RawPtr;
2141 } else if (base_type->isa_oopptr()) {
2142 // Base is never null => always a heap address.
2143 if (base_type->ptr() == TypePtr::NotNull) {
2144 return Type::OopPtr;
2145 }
2146 // Offset is small => always a heap address.
2147 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2148 if (offset_type != NULL &&
2149 base_type->offset() == 0 && // (should always be?)
2150 offset_type->_lo >= 0 &&
2151 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2152 return Type::OopPtr;
2153 }
2154 // Otherwise, it might either be oop+off or NULL+addr.
2155 return Type::AnyPtr;
2156 } else {
2157 // No information:
2158 return Type::AnyPtr;
2159 }
2160 }
2161
2162 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2163 int kind = classify_unsafe_addr(base, offset);
2164 if (kind == Type::RawPtr) {
2165 return basic_plus_adr(top(), base, offset);
2166 } else {
2167 return basic_plus_adr(base, offset);
2168 }
2169 }
2170
2171 //--------------------------inline_number_methods-----------------------------
2172 // inline int Integer.numberOfLeadingZeros(int)
2173 // inline int Long.numberOfLeadingZeros(long)
2174 //
2175 // inline int Integer.numberOfTrailingZeros(int)
2176 // inline int Long.numberOfTrailingZeros(long)
2177 //
2178 // inline int Integer.bitCount(int)
2179 // inline int Long.bitCount(long)
2180 //
2181 // inline char Character.reverseBytes(char)
2182 // inline short Short.reverseBytes(short)
2183 // inline int Integer.reverseBytes(int)
2184 // inline long Long.reverseBytes(long)
2185 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2186 Node* arg = argument(0);
2187 Node* n = NULL;
2188 switch (id) {
2189 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2190 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2191 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2192 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2193 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2194 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2195 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2196 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2197 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2198 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2199 default: fatal_unexpected_iid(id); break;
2200 }
2201 set_result(_gvn.transform(n));
2202 return true;
2203 }
2204
2205 //----------------------------inline_unsafe_access----------------------------
2206
2207 const static BasicType T_ADDRESS_HOLDER = T_LONG;
2208
2209 // Helper that guards and inserts a pre-barrier.
2210 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2211 Node* pre_val, bool need_mem_bar) {
2212 // We could be accessing the referent field of a reference object. If so, when G1
2213 // is enabled, we need to log the value in the referent field in an SATB buffer.
2214 // This routine performs some compile time filters and generates suitable
2215 // runtime filters that guard the pre-barrier code.
2216 // Also add memory barrier for non volatile load from the referent field
2217 // to prevent commoning of loads across safepoint.
2218 if (!UseG1GC && !need_mem_bar)
2219 return;
2220
2221 // Some compile time checks.
2222
2223 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2224 const TypeX* otype = offset->find_intptr_t_type();
2225 if (otype != NULL && otype->is_con() &&
2226 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2227 // Constant offset but not the reference_offset so just return
2228 return;
2229 }
2230
2231 // We only need to generate the runtime guards for instances.
2232 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2233 if (btype != NULL) {
2234 if (btype->isa_aryptr()) {
2235 // Array type so nothing to do
2236 return;
2237 }
2238
2239 const TypeInstPtr* itype = btype->isa_instptr();
2240 if (itype != NULL) {
2241 // Can the klass of base_oop be statically determined to be
2242 // _not_ a sub-class of Reference and _not_ Object?
2243 ciKlass* klass = itype->klass();
2244 if ( klass->is_loaded() &&
2245 !klass->is_subtype_of(env()->Reference_klass()) &&
2246 !env()->Object_klass()->is_subtype_of(klass)) {
2247 return;
2248 }
2249 }
2250 }
2251
2252 // The compile time filters did not reject base_oop/offset so
2253 // we need to generate the following runtime filters
2254 //
2255 // if (offset == java_lang_ref_Reference::_reference_offset) {
2256 // if (instance_of(base, java.lang.ref.Reference)) {
2257 // pre_barrier(_, pre_val, ...);
2258 // }
2259 // }
2260
2261 float likely = PROB_LIKELY( 0.999);
2262 float unlikely = PROB_UNLIKELY(0.999);
2263
2264 IdealKit ideal(this);
2265 #define __ ideal.
2266
2267 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2268
2269 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2270 // Update graphKit memory and control from IdealKit.
2271 sync_kit(ideal);
2272
2273 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2274 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2275
2276 // Update IdealKit memory and control from graphKit.
2277 __ sync_kit(this);
2278
2279 Node* one = __ ConI(1);
2280 // is_instof == 0 if base_oop == NULL
2281 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2282
2283 // Update graphKit from IdeakKit.
2284 sync_kit(ideal);
2285
2286 // Use the pre-barrier to record the value in the referent field
2287 pre_barrier(false /* do_load */,
2288 __ ctrl(),
2289 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2290 pre_val /* pre_val */,
2291 T_OBJECT);
2292 if (need_mem_bar) {
2293 // Add memory barrier to prevent commoning reads from this field
2294 // across safepoint since GC can change its value.
2295 insert_mem_bar(Op_MemBarCPUOrder);
2296 }
2297 // Update IdealKit from graphKit.
2298 __ sync_kit(this);
2299
2300 } __ end_if(); // _ref_type != ref_none
2301 } __ end_if(); // offset == referent_offset
2302
2303 // Final sync IdealKit and GraphKit.
2304 final_sync(ideal);
2305 #undef __
2306 }
2307
2308
2309 // Interpret Unsafe.fieldOffset cookies correctly:
2310 extern jlong Unsafe_field_offset_to_byte_offset(jlong field_offset);
2311
2312 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
2313 // Attempt to infer a sharper value type from the offset and base type.
2314 ciKlass* sharpened_klass = NULL;
2315
2316 // See if it is an instance field, with an object type.
2317 if (alias_type->field() != NULL) {
2318 assert(!is_native_ptr, "native pointer op cannot use a java address");
2319 if (alias_type->field()->type()->is_klass()) {
2320 sharpened_klass = alias_type->field()->type()->as_klass();
2321 }
2322 }
2323
2324 // See if it is a narrow oop array.
2325 if (adr_type->isa_aryptr()) {
2326 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2327 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2328 if (elem_type != NULL) {
2329 sharpened_klass = elem_type->klass();
2330 }
2331 }
2332 }
2333
2334 // The sharpened class might be unloaded if there is no class loader
2335 // contraint in place.
2336 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2337 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2338
2339 #ifndef PRODUCT
2340 if (C->print_intrinsics() || C->print_inlining()) {
2341 tty->print(" from base type: "); adr_type->dump();
2342 tty->print(" sharpened value: "); tjp->dump();
2343 }
2344 #endif
2345 // Sharpen the value type.
2346 return tjp;
2347 }
2348 return NULL;
2349 }
2350
2351 bool LibraryCallKit::inline_unsafe_access(const bool is_native_ptr, bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2352 if (callee()->is_static()) return false; // caller must have the capability!
2353 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2354 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2355
2356 #ifndef PRODUCT
2357 {
2358 ResourceMark rm;
2359 // Check the signatures.
2360 ciSignature* sig = callee()->signature();
2361 #ifdef ASSERT
2362 if (!is_store) {
2363 // Object getObject(Object base, int/long offset), etc.
2364 BasicType rtype = sig->return_type()->basic_type();
2365 if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
2366 rtype = T_ADDRESS; // it is really a C void*
2367 assert(rtype == type, "getter must return the expected value");
2368 if (!is_native_ptr) {
2369 assert(sig->count() == 2, "oop getter has 2 arguments");
2370 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2371 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2372 } else {
2373 assert(sig->count() == 1, "native getter has 1 argument");
2374 assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
2375 }
2376 } else {
2377 // void putObject(Object base, int/long offset, Object x), etc.
2378 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2379 if (!is_native_ptr) {
2380 assert(sig->count() == 3, "oop putter has 3 arguments");
2381 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2382 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2383 } else {
2384 assert(sig->count() == 2, "native putter has 2 arguments");
2385 assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
2386 }
2387 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2388 if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
2389 vtype = T_ADDRESS; // it is really a C void*
2390 assert(vtype == type, "putter must accept the expected value");
2391 }
2392 #endif // ASSERT
2393 }
2394 #endif //PRODUCT
2395
2396 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2397
2398 Node* receiver = argument(0); // type: oop
2399
2400 // Build address expression.
2401 Node* adr;
2402 Node* heap_base_oop = top();
2403 Node* offset = top();
2404 Node* val;
2405
2406 if (!is_native_ptr) {
2407 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2408 Node* base = argument(1); // type: oop
2409 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2410 offset = argument(2); // type: long
2411 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2412 // to be plain byte offsets, which are also the same as those accepted
2413 // by oopDesc::field_base.
2414 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2415 "fieldOffset must be byte-scaled");
2416 // 32-bit machines ignore the high half!
2417 offset = ConvL2X(offset);
2418 adr = make_unsafe_address(base, offset);
2419 heap_base_oop = base;
2420 val = is_store ? argument(4) : NULL;
2421 } else {
2422 Node* ptr = argument(1); // type: long
2423 ptr = ConvL2X(ptr); // adjust Java long to machine word
2424 adr = make_unsafe_address(NULL, ptr);
2425 val = is_store ? argument(3) : NULL;
2426 }
2427
2428 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2429
2430 // First guess at the value type.
2431 const Type *value_type = Type::get_const_basic_type(type);
2432
2433 // Try to categorize the address. If it comes up as TypeJavaPtr::BOTTOM,
2434 // there was not enough information to nail it down.
2435 Compile::AliasType* alias_type = C->alias_type(adr_type);
2436 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2437
2438 // We will need memory barriers unless we can determine a unique
2439 // alias category for this reference. (Note: If for some reason
2440 // the barriers get omitted and the unsafe reference begins to "pollute"
2441 // the alias analysis of the rest of the graph, either Compile::can_alias
2442 // or Compile::must_alias will throw a diagnostic assert.)
2443 bool need_mem_bar;
2444 switch (kind) {
2445 case Relaxed:
2446 need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
2447 break;
2448 case Opaque:
2449 // Opaque uses CPUOrder membars for protection against code movement.
2450 case Acquire:
2451 case Release:
2452 case Volatile:
2453 need_mem_bar = true;
2454 break;
2455 default:
2456 ShouldNotReachHere();
2457 }
2458
2459 // Some accesses require access atomicity for all types, notably longs and doubles.
2460 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2461 bool requires_atomic_access = false;
2462 switch (kind) {
2463 case Relaxed:
2464 case Opaque:
2465 requires_atomic_access = AlwaysAtomicAccesses;
2466 break;
2467 case Acquire:
2468 case Release:
2469 case Volatile:
2470 requires_atomic_access = true;
2471 break;
2472 default:
2473 ShouldNotReachHere();
2474 }
2475
2476 // Figure out the memory ordering.
2477 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2478 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2479
2480 // If we are reading the value of the referent field of a Reference
2481 // object (either by using Unsafe directly or through reflection)
2482 // then, if G1 is enabled, we need to record the referent in an
2483 // SATB log buffer using the pre-barrier mechanism.
2484 // Also we need to add memory barrier to prevent commoning reads
2485 // from this field across safepoint since GC can change its value.
2486 bool need_read_barrier = !is_native_ptr && !is_store &&
2487 offset != top() && heap_base_oop != top();
2488
2489 if (!is_store && type == T_OBJECT) {
2490 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
2491 if (tjp != NULL) {
2492 value_type = tjp;
2493 }
2494 }
2495
2496 receiver = null_check(receiver);
2497 if (stopped()) {
2498 return true;
2499 }
2500 // Heap pointers get a null-check from the interpreter,
2501 // as a courtesy. However, this is not guaranteed by Unsafe,
2502 // and it is not possible to fully distinguish unintended nulls
2503 // from intended ones in this API.
2504
2505 // We need to emit leading and trailing CPU membars (see below) in
2506 // addition to memory membars for special access modes. This is a little
2507 // too strong, but avoids the need to insert per-alias-type
2508 // volatile membars (for stores; compare Parse::do_put_xxx), which
2509 // we cannot do effectively here because we probably only have a
2510 // rough approximation of type.
2511
2512 switch(kind) {
2513 case Relaxed:
2514 case Opaque:
2515 case Acquire:
2516 break;
2517 case Release:
2518 case Volatile:
2519 if (is_store) {
2520 insert_mem_bar(Op_MemBarRelease);
2521 } else {
2522 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2523 insert_mem_bar(Op_MemBarVolatile);
2524 }
2525 }
2526 break;
2527 default:
2528 ShouldNotReachHere();
2529 }
2530
2531 // Memory barrier to prevent normal and 'unsafe' accesses from
2532 // bypassing each other. Happens after null checks, so the
2533 // exception paths do not take memory state from the memory barrier,
2534 // so there's no problems making a strong assert about mixing users
2535 // of safe & unsafe memory.
2536 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2537
2538 assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM ||
2539 alias_type->field() != NULL || alias_type->element() != NULL, "field, array element or unknown");
2540 bool mismatched = false;
2541 if (alias_type->element() != NULL || alias_type->field() != NULL) {
2542 BasicType bt;
2543 if (alias_type->element() != NULL) {
2544 const Type* element = alias_type->element();
2545 bt = element->isa_narrowoop() ? T_OBJECT : element->array_element_basic_type();
2546 } else {
2547 bt = alias_type->field()->type()->basic_type();
2548 }
2549 if (bt == T_ARRAY) {
2550 // accessing an array field with getObject is not a mismatch
2551 bt = T_OBJECT;
2552 }
2553 if (bt != type) {
2554 mismatched = true;
2555 }
2556 }
2557 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2558
2559 if (!is_store) {
2560 Node* p = NULL;
2561 // Try to constant fold a load from a constant field
2562 ciField* field = alias_type->field();
2563 if (heap_base_oop != top() &&
2564 field != NULL && field->is_constant() && field->layout_type() == type) {
2565 // final or stable field
2566 const Type* con_type = Type::make_constant(alias_type->field(), heap_base_oop);
2567 if (con_type != NULL) {
2568 p = makecon(con_type);
2569 }
2570 }
2571 if (p == NULL) {
2572 // To be valid, unsafe loads may depend on other conditions than
2573 // the one that guards them: pin the Load node
2574 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2575 // load value
2576 switch (type) {
2577 case T_BOOLEAN:
2578 case T_CHAR:
2579 case T_BYTE:
2580 case T_SHORT:
2581 case T_INT:
2582 case T_LONG:
2583 case T_FLOAT:
2584 case T_DOUBLE:
2585 break;
2586 case T_OBJECT:
2587 if (need_read_barrier) {
2588 // We do not require a mem bar inside pre_barrier if need_mem_bar
2589 // is set: the barriers would be emitted by us.
2590 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2591 }
2592 break;
2593 case T_ADDRESS:
2594 // Cast to an int type.
2595 p = _gvn.transform(new CastP2XNode(NULL, p));
2596 p = ConvX2UL(p);
2597 break;
2598 default:
2599 fatal("unexpected type %d: %s", type, type2name(type));
2600 break;
2601 }
2602 }
2603 // The load node has the control of the preceding MemBarCPUOrder. All
2604 // following nodes will have the control of the MemBarCPUOrder inserted at
2605 // the end of this method. So, pushing the load onto the stack at a later
2606 // point is fine.
2607 set_result(p);
2608 } else {
2609 // place effect of store into memory
2610 switch (type) {
2611 case T_DOUBLE:
2612 val = dstore_rounding(val);
2613 break;
2614 case T_ADDRESS:
2615 // Repackage the long as a pointer.
2616 val = ConvL2X(val);
2617 val = _gvn.transform(new CastX2PNode(val));
2618 break;
2619 }
2620
2621 if (type != T_OBJECT) {
2622 (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2623 } else {
2624 // Possibly an oop being stored to Java heap or native memory
2625 if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
2626 // oop to Java heap.
2627 (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2628 } else {
2629 // We can't tell at compile time if we are storing in the Java heap or outside
2630 // of it. So we need to emit code to conditionally do the proper type of
2631 // store.
2632
2633 IdealKit ideal(this);
2634 #define __ ideal.
2635 // QQQ who knows what probability is here??
2636 __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
2637 // Sync IdealKit and graphKit.
2638 sync_kit(ideal);
2639 Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2640 // Update IdealKit memory.
2641 __ sync_kit(this);
2642 } __ else_(); {
2643 __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched);
2644 } __ end_if();
2645 // Final sync IdealKit and GraphKit.
2646 final_sync(ideal);
2647 #undef __
2648 }
2649 }
2650 }
2651
2652 switch(kind) {
2653 case Relaxed:
2654 case Opaque:
2655 case Release:
2656 break;
2657 case Acquire:
2658 case Volatile:
2659 if (!is_store) {
2660 insert_mem_bar(Op_MemBarAcquire);
2661 } else {
2662 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2663 insert_mem_bar(Op_MemBarVolatile);
2664 }
2665 }
2666 break;
2667 default:
2668 ShouldNotReachHere();
2669 }
2670
2671 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2672
2673 return true;
2674 }
2675
2676 //----------------------------inline_unsafe_load_store----------------------------
2677 // This method serves a couple of different customers (depending on LoadStoreKind):
2678 //
2679 // LS_cmp_swap:
2680 //
2681 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2682 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2683 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2684 //
2685 // LS_cmp_swap_weak:
2686 //
2687 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2688 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2689 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2690 //
2691 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2692 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2693 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2694 //
2695 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2696 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2697 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2698 //
2699 // LS_cmp_exchange:
2700 //
2701 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2702 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2703 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2704 //
2705 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2706 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2707 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2708 //
2709 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2710 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2711 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2712 //
2713 // LS_get_add:
2714 //
2715 // int getAndAddInt( Object o, long offset, int delta)
2716 // long getAndAddLong(Object o, long offset, long delta)
2717 //
2718 // LS_get_set:
2719 //
2720 // int getAndSet(Object o, long offset, int newValue)
2721 // long getAndSet(Object o, long offset, long newValue)
2722 // Object getAndSet(Object o, long offset, Object newValue)
2723 //
2724 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2725 // This basic scheme here is the same as inline_unsafe_access, but
2726 // differs in enough details that combining them would make the code
2727 // overly confusing. (This is a true fact! I originally combined
2728 // them, but even I was confused by it!) As much code/comments as
2729 // possible are retained from inline_unsafe_access though to make
2730 // the correspondences clearer. - dl
2731
2732 if (callee()->is_static()) return false; // caller must have the capability!
2733
2734 #ifndef PRODUCT
2735 BasicType rtype;
2736 {
2737 ResourceMark rm;
2738 // Check the signatures.
2739 ciSignature* sig = callee()->signature();
2740 rtype = sig->return_type()->basic_type();
2741 switch(kind) {
2742 case LS_get_add:
2743 case LS_get_set: {
2744 // Check the signatures.
2745 #ifdef ASSERT
2746 assert(rtype == type, "get and set must return the expected type");
2747 assert(sig->count() == 3, "get and set has 3 arguments");
2748 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2749 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2750 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2751 #endif // ASSERT
2752 break;
2753 }
2754 case LS_cmp_swap:
2755 case LS_cmp_swap_weak: {
2756 // Check the signatures.
2757 #ifdef ASSERT
2758 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2759 assert(sig->count() == 4, "CAS has 4 arguments");
2760 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2761 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2762 #endif // ASSERT
2763 break;
2764 }
2765 case LS_cmp_exchange: {
2766 // Check the signatures.
2767 #ifdef ASSERT
2768 assert(rtype == type, "CAS must return the expected type");
2769 assert(sig->count() == 4, "CAS has 4 arguments");
2770 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2771 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2772 #endif // ASSERT
2773 break;
2774 }
2775 default:
2776 ShouldNotReachHere();
2777 }
2778 }
2779 #endif //PRODUCT
2780
2781 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2782
2783 // Get arguments:
2784 Node* receiver = NULL;
2785 Node* base = NULL;
2786 Node* offset = NULL;
2787 Node* oldval = NULL;
2788 Node* newval = NULL;
2789 switch(kind) {
2790 case LS_cmp_swap:
2791 case LS_cmp_swap_weak:
2792 case LS_cmp_exchange: {
2793 const bool two_slot_type = type2size[type] == 2;
2794 receiver = argument(0); // type: oop
2795 base = argument(1); // type: oop
2796 offset = argument(2); // type: long
2797 oldval = argument(4); // type: oop, int, or long
2798 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2799 break;
2800 }
2801 case LS_get_add:
2802 case LS_get_set: {
2803 receiver = argument(0); // type: oop
2804 base = argument(1); // type: oop
2805 offset = argument(2); // type: long
2806 oldval = NULL;
2807 newval = argument(4); // type: oop, int, or long
2808 break;
2809 }
2810 default:
2811 ShouldNotReachHere();
2812 }
2813
2814 // Null check receiver.
2815 receiver = null_check(receiver);
2816 if (stopped()) {
2817 return true;
2818 }
2819
2820 // Build field offset expression.
2821 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2822 // to be plain byte offsets, which are also the same as those accepted
2823 // by oopDesc::field_base.
2824 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2825 // 32-bit machines ignore the high half of long offsets
2826 offset = ConvL2X(offset);
2827 Node* adr = make_unsafe_address(base, offset);
2828 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2829
2830 // For CAS, unlike inline_unsafe_access, there seems no point in
2831 // trying to refine types. Just use the coarse types here.
2832 const Type *value_type = Type::get_const_basic_type(type);
2833 Compile::AliasType* alias_type = C->alias_type(adr_type);
2834 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2835
2836 switch (kind) {
2837 case LS_get_set:
2838 case LS_cmp_exchange: {
2839 if (type == T_OBJECT) {
2840 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2841 if (tjp != NULL) {
2842 value_type = tjp;
2843 }
2844 }
2845 break;
2846 }
2847 case LS_cmp_swap:
2848 case LS_cmp_swap_weak:
2849 case LS_get_add:
2850 break;
2851 default:
2852 ShouldNotReachHere();
2853 }
2854
2855 int alias_idx = C->get_alias_index(adr_type);
2856
2857 // Memory-model-wise, a LoadStore acts like a little synchronized
2858 // block, so needs barriers on each side. These don't translate
2859 // into actual barriers on most machines, but we still need rest of
2860 // compiler to respect ordering.
2861
2862 switch (access_kind) {
2863 case Relaxed:
2864 case Acquire:
2865 break;
2866 case Release:
2867 case Volatile:
2868 insert_mem_bar(Op_MemBarRelease);
2869 break;
2870 default:
2871 ShouldNotReachHere();
2872 }
2873 insert_mem_bar(Op_MemBarCPUOrder);
2874
2875 // Figure out the memory ordering.
2876 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2877
2878 // 4984716: MemBars must be inserted before this
2879 // memory node in order to avoid a false
2880 // dependency which will confuse the scheduler.
2881 Node *mem = memory(alias_idx);
2882
2883 // For now, we handle only those cases that actually exist: ints,
2884 // longs, and Object. Adding others should be straightforward.
2885 Node* load_store = NULL;
2886 switch(type) {
2887 case T_INT:
2888 switch(kind) {
2889 case LS_get_add:
2890 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2891 break;
2892 case LS_get_set:
2893 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2894 break;
2895 case LS_cmp_swap_weak:
2896 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2897 break;
2898 case LS_cmp_swap:
2899 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2900 break;
2901 case LS_cmp_exchange:
2902 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2903 break;
2904 default:
2905 ShouldNotReachHere();
2906 }
2907 break;
2908 case T_LONG:
2909 switch(kind) {
2910 case LS_get_add:
2911 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2912 break;
2913 case LS_get_set:
2914 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2915 break;
2916 case LS_cmp_swap_weak:
2917 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2918 break;
2919 case LS_cmp_swap:
2920 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2921 break;
2922 case LS_cmp_exchange:
2923 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2924 break;
2925 default:
2926 ShouldNotReachHere();
2927 }
2928 break;
2929 case T_OBJECT:
2930 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2931 // could be delayed during Parse (for example, in adjust_map_after_if()).
2932 // Execute transformation here to avoid barrier generation in such case.
2933 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2934 newval = _gvn.makecon(TypePtr::NULL_PTR);
2935
2936 // Reference stores need a store barrier.
2937 switch(kind) {
2938 case LS_get_set: {
2939 // If pre-barrier must execute before the oop store, old value will require do_load here.
2940 if (!can_move_pre_barrier()) {
2941 pre_barrier(true /* do_load*/,
2942 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2943 NULL /* pre_val*/,
2944 T_OBJECT);
2945 } // Else move pre_barrier to use load_store value, see below.
2946 break;
2947 }
2948 case LS_cmp_swap_weak:
2949 case LS_cmp_swap:
2950 case LS_cmp_exchange: {
2951 // Same as for newval above:
2952 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2953 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2954 }
2955 // The only known value which might get overwritten is oldval.
2956 pre_barrier(false /* do_load */,
2957 control(), NULL, NULL, max_juint, NULL, NULL,
2958 oldval /* pre_val */,
2959 T_OBJECT);
2960 break;
2961 }
2962 default:
2963 ShouldNotReachHere();
2964 }
2965
2966 #ifdef _LP64
2967 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2968 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2969
2970 switch(kind) {
2971 case LS_get_set:
2972 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2973 break;
2974 case LS_cmp_swap_weak: {
2975 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2976 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2977 break;
2978 }
2979 case LS_cmp_swap: {
2980 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2981 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2982 break;
2983 }
2984 case LS_cmp_exchange: {
2985 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2986 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2987 break;
2988 }
2989 default:
2990 ShouldNotReachHere();
2991 }
2992 } else
2993 #endif
2994 switch (kind) {
2995 case LS_get_set:
2996 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2997 break;
2998 case LS_cmp_swap_weak:
2999 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
3000 break;
3001 case LS_cmp_swap:
3002 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
3003 break;
3004 case LS_cmp_exchange:
3005 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
3006 break;
3007 default:
3008 ShouldNotReachHere();
3009 }
3010
3011 // Emit the post barrier only when the actual store happened. This makes sense
3012 // to check only for LS_cmp_* that can fail to set the value.
3013 // LS_cmp_exchange does not produce any branches by default, so there is no
3014 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
3015 // CompareAndExchange and move branches here, it would make sense to conditionalize
3016 // post_barriers for LS_cmp_exchange as well.
3017 //
3018 // CAS success path is marked more likely since we anticipate this is a performance
3019 // critical path, while CAS failure path can use the penalty for going through unlikely
3020 // path as backoff. Which is still better than doing a store barrier there.
3021 switch (kind) {
3022 case LS_get_set:
3023 case LS_cmp_exchange: {
3024 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3025 break;
3026 }
3027 case LS_cmp_swap_weak:
3028 case LS_cmp_swap: {
3029 IdealKit ideal(this);
3030 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
3031 sync_kit(ideal);
3032 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
3033 ideal.sync_kit(this);
3034 } ideal.end_if();
3035 final_sync(ideal);
3036 break;
3037 }
3038 default:
3039 ShouldNotReachHere();
3040 }
3041 break;
3042 default:
3043 fatal("unexpected type %d: %s", type, type2name(type));
3044 break;
3045 }
3046
3047 // SCMemProjNodes represent the memory state of a LoadStore. Their
3048 // main role is to prevent LoadStore nodes from being optimized away
3049 // when their results aren't used.
3050 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3051 set_memory(proj, alias_idx);
3052
3053 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3054 #ifdef _LP64
3055 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3056 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3057 }
3058 #endif
3059 if (can_move_pre_barrier()) {
3060 // Don't need to load pre_val. The old value is returned by load_store.
3061 // The pre_barrier can execute after the xchg as long as no safepoint
3062 // gets inserted between them.
3063 pre_barrier(false /* do_load */,
3064 control(), NULL, NULL, max_juint, NULL, NULL,
3065 load_store /* pre_val */,
3066 T_OBJECT);
3067 }
3068 }
3069
3070 // Add the trailing membar surrounding the access
3071 insert_mem_bar(Op_MemBarCPUOrder);
3072
3073 switch (access_kind) {
3074 case Relaxed:
3075 case Release:
3076 break; // do nothing
3077 case Acquire:
3078 case Volatile:
3079 insert_mem_bar(Op_MemBarAcquire);
3080 break;
3081 default:
3082 ShouldNotReachHere();
3083 }
3084
3085 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3086 set_result(load_store);
3087 return true;
3088 }
3089
3090 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3091 MemNode::MemOrd mo = MemNode::unset;
3092 switch(kind) {
3093 case Opaque:
3094 case Relaxed: mo = MemNode::unordered; break;
3095 case Acquire: mo = MemNode::acquire; break;
3096 case Release: mo = MemNode::release; break;
3097 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3098 default:
3099 ShouldNotReachHere();
3100 }
3101 guarantee(mo != MemNode::unset, "Should select memory ordering");
3102 return mo;
3103 }
3104
3105 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3106 MemNode::MemOrd mo = MemNode::unset;
3107 switch(kind) {
3108 case Opaque:
3109 case Relaxed: mo = MemNode::unordered; break;
3110 case Acquire: mo = MemNode::acquire; break;
3111 case Release: mo = MemNode::release; break;
3112 case Volatile: mo = MemNode::seqcst; break;
3113 default:
3114 ShouldNotReachHere();
3115 }
3116 guarantee(mo != MemNode::unset, "Should select memory ordering");
3117 return mo;
3118 }
3119
3120 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3121 // Regardless of form, don't allow previous ld/st to move down,
3122 // then issue acquire, release, or volatile mem_bar.
3123 insert_mem_bar(Op_MemBarCPUOrder);
3124 switch(id) {
3125 case vmIntrinsics::_loadFence:
3126 insert_mem_bar(Op_LoadFence);
3127 return true;
3128 case vmIntrinsics::_storeFence:
3129 insert_mem_bar(Op_StoreFence);
3130 return true;
3131 case vmIntrinsics::_fullFence:
3132 insert_mem_bar(Op_MemBarVolatile);
3133 return true;
3134 default:
3135 fatal_unexpected_iid(id);
3136 return false;
3137 }
3138 }
3139
3140 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3141 if (!kls->is_Con()) {
3142 return true;
3143 }
3144 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3145 if (klsptr == NULL) {
3146 return true;
3147 }
3148 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3149 // don't need a guard for a klass that is already initialized
3150 return !ik->is_initialized();
3151 }
3152
3153 //----------------------------inline_unsafe_allocate---------------------------
3154 // public native Object Unsafe.allocateInstance(Class<?> cls);
3155 bool LibraryCallKit::inline_unsafe_allocate() {
3156 if (callee()->is_static()) return false; // caller must have the capability!
3157
3158 null_check_receiver(); // null-check, then ignore
3159 Node* cls = null_check(argument(1));
3160 if (stopped()) return true;
3161
3162 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3163 kls = null_check(kls);
3164 if (stopped()) return true; // argument was like int.class
3165
3166 Node* test = NULL;
3167 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3168 // Note: The argument might still be an illegal value like
3169 // Serializable.class or Object[].class. The runtime will handle it.
3170 // But we must make an explicit check for initialization.
3171 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3172 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3173 // can generate code to load it as unsigned byte.
3174 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3175 Node* bits = intcon(InstanceKlass::fully_initialized);
3176 test = _gvn.transform(new SubINode(inst, bits));
3177 // The 'test' is non-zero if we need to take a slow path.
3178 }
3179
3180 Node* obj = new_instance(kls, test);
3181 set_result(obj);
3182 return true;
3183 }
3184
3185 #ifdef TRACE_HAVE_INTRINSICS
3186 /*
3187 * oop -> myklass
3188 * myklass->trace_id |= USED
3189 * return myklass->trace_id & ~0x3
3190 */
3191 bool LibraryCallKit::inline_native_classID() {
3192 null_check_receiver(); // null-check, then ignore
3193 Node* cls = null_check(argument(1), T_OBJECT);
3194 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3195 kls = null_check(kls, T_OBJECT);
3196 ByteSize offset = TRACE_ID_OFFSET;
3197 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3198 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3199 Node* bits = longcon(~0x03l); // ignore bit 0 & 1
3200 Node* andl = _gvn.transform(new AndLNode(tvalue, bits));
3201 Node* clsused = longcon(0x01l); // set the class bit
3202 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3203
3204 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3205 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3206 set_result(andl);
3207 return true;
3208 }
3209
3210 bool LibraryCallKit::inline_native_threadID() {
3211 Node* tls_ptr = NULL;
3212 Node* cur_thr = generate_current_thread(tls_ptr);
3213 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3214 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3215 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::thread_id_offset()));
3216
3217 Node* threadid = NULL;
3218 size_t thread_id_size = OSThread::thread_id_size();
3219 if (thread_id_size == (size_t) BytesPerLong) {
3220 threadid = ConvL2I(make_load(control(), p, TypeLong::LONG, T_LONG, MemNode::unordered));
3221 } else if (thread_id_size == (size_t) BytesPerInt) {
3222 threadid = make_load(control(), p, TypeInt::INT, T_INT, MemNode::unordered);
3223 } else {
3224 ShouldNotReachHere();
3225 }
3226 set_result(threadid);
3227 return true;
3228 }
3229 #endif
3230
3231 //------------------------inline_native_time_funcs--------------
3232 // inline code for System.currentTimeMillis() and System.nanoTime()
3233 // these have the same type and signature
3234 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3235 const TypeFunc* tf = OptoRuntime::void_long_Type();
3236 const TypePtr* no_memory_effects = NULL;
3237 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3238 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3239 #ifdef ASSERT
3240 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3241 assert(value_top == top(), "second value must be top");
3242 #endif
3243 set_result(value);
3244 return true;
3245 }
3246
3247 //------------------------inline_native_currentThread------------------
3248 bool LibraryCallKit::inline_native_currentThread() {
3249 Node* junk = NULL;
3250 set_result(generate_current_thread(junk));
3251 return true;
3252 }
3253
3254 //------------------------inline_native_isInterrupted------------------
3255 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3256 bool LibraryCallKit::inline_native_isInterrupted() {
3257 // Add a fast path to t.isInterrupted(clear_int):
3258 // (t == Thread.current() &&
3259 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3260 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3261 // So, in the common case that the interrupt bit is false,
3262 // we avoid making a call into the VM. Even if the interrupt bit
3263 // is true, if the clear_int argument is false, we avoid the VM call.
3264 // However, if the receiver is not currentThread, we must call the VM,
3265 // because there must be some locking done around the operation.
3266
3267 // We only go to the fast case code if we pass two guards.
3268 // Paths which do not pass are accumulated in the slow_region.
3269
3270 enum {
3271 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3272 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3273 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3274 PATH_LIMIT
3275 };
3276
3277 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3278 // out of the function.
3279 insert_mem_bar(Op_MemBarCPUOrder);
3280
3281 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3282 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3283
3284 RegionNode* slow_region = new RegionNode(1);
3285 record_for_igvn(slow_region);
3286
3287 // (a) Receiving thread must be the current thread.
3288 Node* rec_thr = argument(0);
3289 Node* tls_ptr = NULL;
3290 Node* cur_thr = generate_current_thread(tls_ptr);
3291 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3292 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3293
3294 generate_slow_guard(bol_thr, slow_region);
3295
3296 // (b) Interrupt bit on TLS must be false.
3297 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3298 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3299 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3300
3301 // Set the control input on the field _interrupted read to prevent it floating up.
3302 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3303 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3304 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3305
3306 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3307
3308 // First fast path: if (!TLS._interrupted) return false;
3309 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3310 result_rgn->init_req(no_int_result_path, false_bit);
3311 result_val->init_req(no_int_result_path, intcon(0));
3312
3313 // drop through to next case
3314 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3315
3316 #ifndef TARGET_OS_FAMILY_windows
3317 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3318 Node* clr_arg = argument(1);
3319 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3320 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3321 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3322
3323 // Second fast path: ... else if (!clear_int) return true;
3324 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3325 result_rgn->init_req(no_clear_result_path, false_arg);
3326 result_val->init_req(no_clear_result_path, intcon(1));
3327
3328 // drop through to next case
3329 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3330 #else
3331 // To return true on Windows you must read the _interrupted field
3332 // and check the event state i.e. take the slow path.
3333 #endif // TARGET_OS_FAMILY_windows
3334
3335 // (d) Otherwise, go to the slow path.
3336 slow_region->add_req(control());
3337 set_control( _gvn.transform(slow_region));
3338
3339 if (stopped()) {
3340 // There is no slow path.
3341 result_rgn->init_req(slow_result_path, top());
3342 result_val->init_req(slow_result_path, top());
3343 } else {
3344 // non-virtual because it is a private non-static
3345 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3346
3347 Node* slow_val = set_results_for_java_call(slow_call);
3348 // this->control() comes from set_results_for_java_call
3349
3350 Node* fast_io = slow_call->in(TypeFunc::I_O);
3351 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3352
3353 // These two phis are pre-filled with copies of of the fast IO and Memory
3354 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3355 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3356
3357 result_rgn->init_req(slow_result_path, control());
3358 result_io ->init_req(slow_result_path, i_o());
3359 result_mem->init_req(slow_result_path, reset_memory());
3360 result_val->init_req(slow_result_path, slow_val);
3361
3362 set_all_memory(_gvn.transform(result_mem));
3363 set_i_o( _gvn.transform(result_io));
3364 }
3365
3366 C->set_has_split_ifs(true); // Has chance for split-if optimization
3367 set_result(result_rgn, result_val);
3368 return true;
3369 }
3370
3371 //---------------------------load_mirror_from_klass----------------------------
3372 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3373 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3374 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3375 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3376 }
3377
3378 //-----------------------load_klass_from_mirror_common-------------------------
3379 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3380 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3381 // and branch to the given path on the region.
3382 // If never_see_null, take an uncommon trap on null, so we can optimistically
3383 // compile for the non-null case.
3384 // If the region is NULL, force never_see_null = true.
3385 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3386 bool never_see_null,
3387 RegionNode* region,
3388 int null_path,
3389 int offset) {
3390 if (region == NULL) never_see_null = true;
3391 Node* p = basic_plus_adr(mirror, offset);
3392 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3393 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3394 Node* null_ctl = top();
3395 kls = null_check_oop(kls, &null_ctl, never_see_null);
3396 if (region != NULL) {
3397 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3398 region->init_req(null_path, null_ctl);
3399 } else {
3400 assert(null_ctl == top(), "no loose ends");
3401 }
3402 return kls;
3403 }
3404
3405 //--------------------(inline_native_Class_query helpers)---------------------
3406 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3407 // Fall through if (mods & mask) == bits, take the guard otherwise.
3408 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3409 // Branch around if the given klass has the given modifier bit set.
3410 // Like generate_guard, adds a new path onto the region.
3411 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3412 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3413 Node* mask = intcon(modifier_mask);
3414 Node* bits = intcon(modifier_bits);
3415 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3416 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3417 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3418 return generate_fair_guard(bol, region);
3419 }
3420 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3421 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3422 }
3423
3424 //-------------------------inline_native_Class_query-------------------
3425 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3426 const Type* return_type = TypeInt::BOOL;
3427 Node* prim_return_value = top(); // what happens if it's a primitive class?
3428 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3429 bool expect_prim = false; // most of these guys expect to work on refs
3430
3431 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3432
3433 Node* mirror = argument(0);
3434 Node* obj = top();
3435
3436 switch (id) {
3437 case vmIntrinsics::_isInstance:
3438 // nothing is an instance of a primitive type
3439 prim_return_value = intcon(0);
3440 obj = argument(1);
3441 break;
3442 case vmIntrinsics::_getModifiers:
3443 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3444 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3445 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3446 break;
3447 case vmIntrinsics::_isInterface:
3448 prim_return_value = intcon(0);
3449 break;
3450 case vmIntrinsics::_isArray:
3451 prim_return_value = intcon(0);
3452 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3453 break;
3454 case vmIntrinsics::_isPrimitive:
3455 prim_return_value = intcon(1);
3456 expect_prim = true; // obviously
3457 break;
3458 case vmIntrinsics::_getSuperclass:
3459 prim_return_value = null();
3460 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3461 break;
3462 case vmIntrinsics::_getClassAccessFlags:
3463 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3464 return_type = TypeInt::INT; // not bool! 6297094
3465 break;
3466 default:
3467 fatal_unexpected_iid(id);
3468 break;
3469 }
3470
3471 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3472 if (mirror_con == NULL) return false; // cannot happen?
3473
3474 #ifndef PRODUCT
3475 if (C->print_intrinsics() || C->print_inlining()) {
3476 ciType* k = mirror_con->java_mirror_type();
3477 if (k) {
3478 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3479 k->print_name();
3480 tty->cr();
3481 }
3482 }
3483 #endif
3484
3485 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3486 RegionNode* region = new RegionNode(PATH_LIMIT);
3487 record_for_igvn(region);
3488 PhiNode* phi = new PhiNode(region, return_type);
3489
3490 // The mirror will never be null of Reflection.getClassAccessFlags, however
3491 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3492 // if it is. See bug 4774291.
3493
3494 // For Reflection.getClassAccessFlags(), the null check occurs in
3495 // the wrong place; see inline_unsafe_access(), above, for a similar
3496 // situation.
3497 mirror = null_check(mirror);
3498 // If mirror or obj is dead, only null-path is taken.
3499 if (stopped()) return true;
3500
3501 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3502
3503 // Now load the mirror's klass metaobject, and null-check it.
3504 // Side-effects region with the control path if the klass is null.
3505 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3506 // If kls is null, we have a primitive mirror.
3507 phi->init_req(_prim_path, prim_return_value);
3508 if (stopped()) { set_result(region, phi); return true; }
3509 bool safe_for_replace = (region->in(_prim_path) == top());
3510
3511 Node* p; // handy temp
3512 Node* null_ctl;
3513
3514 // Now that we have the non-null klass, we can perform the real query.
3515 // For constant classes, the query will constant-fold in LoadNode::Value.
3516 Node* query_value = top();
3517 switch (id) {
3518 case vmIntrinsics::_isInstance:
3519 // nothing is an instance of a primitive type
3520 query_value = gen_instanceof(obj, kls, safe_for_replace);
3521 break;
3522
3523 case vmIntrinsics::_getModifiers:
3524 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3525 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3526 break;
3527
3528 case vmIntrinsics::_isInterface:
3529 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3530 if (generate_interface_guard(kls, region) != NULL)
3531 // A guard was added. If the guard is taken, it was an interface.
3532 phi->add_req(intcon(1));
3533 // If we fall through, it's a plain class.
3534 query_value = intcon(0);
3535 break;
3536
3537 case vmIntrinsics::_isArray:
3538 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3539 if (generate_array_guard(kls, region) != NULL)
3540 // A guard was added. If the guard is taken, it was an array.
3541 phi->add_req(intcon(1));
3542 // If we fall through, it's a plain class.
3543 query_value = intcon(0);
3544 break;
3545
3546 case vmIntrinsics::_isPrimitive:
3547 query_value = intcon(0); // "normal" path produces false
3548 break;
3549
3550 case vmIntrinsics::_getSuperclass:
3551 // The rules here are somewhat unfortunate, but we can still do better
3552 // with random logic than with a JNI call.
3553 // Interfaces store null or Object as _super, but must report null.
3554 // Arrays store an intermediate super as _super, but must report Object.
3555 // Other types can report the actual _super.
3556 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3557 if (generate_interface_guard(kls, region) != NULL)
3558 // A guard was added. If the guard is taken, it was an interface.
3559 phi->add_req(null());
3560 if (generate_array_guard(kls, region) != NULL)
3561 // A guard was added. If the guard is taken, it was an array.
3562 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3563 // If we fall through, it's a plain class. Get its _super.
3564 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3565 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3566 null_ctl = top();
3567 kls = null_check_oop(kls, &null_ctl);
3568 if (null_ctl != top()) {
3569 // If the guard is taken, Object.superClass is null (both klass and mirror).
3570 region->add_req(null_ctl);
3571 phi ->add_req(null());
3572 }
3573 if (!stopped()) {
3574 query_value = load_mirror_from_klass(kls);
3575 }
3576 break;
3577
3578 case vmIntrinsics::_getClassAccessFlags:
3579 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3580 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3581 break;
3582
3583 default:
3584 fatal_unexpected_iid(id);
3585 break;
3586 }
3587
3588 // Fall-through is the normal case of a query to a real class.
3589 phi->init_req(1, query_value);
3590 region->init_req(1, control());
3591
3592 C->set_has_split_ifs(true); // Has chance for split-if optimization
3593 set_result(region, phi);
3594 return true;
3595 }
3596
3597 //-------------------------inline_Class_cast-------------------
3598 bool LibraryCallKit::inline_Class_cast() {
3599 Node* mirror = argument(0); // Class
3600 Node* obj = argument(1);
3601 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3602 if (mirror_con == NULL) {
3603 return false; // dead path (mirror->is_top()).
3604 }
3605 if (obj == NULL || obj->is_top()) {
3606 return false; // dead path
3607 }
3608 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3609
3610 // First, see if Class.cast() can be folded statically.
3611 // java_mirror_type() returns non-null for compile-time Class constants.
3612 ciType* tm = mirror_con->java_mirror_type();
3613 if (tm != NULL && tm->is_klass() &&
3614 tp != NULL && tp->klass() != NULL) {
3615 if (!tp->klass()->is_loaded()) {
3616 // Don't use intrinsic when class is not loaded.
3617 return false;
3618 } else {
3619 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3620 if (static_res == Compile::SSC_always_true) {
3621 // isInstance() is true - fold the code.
3622 set_result(obj);
3623 return true;
3624 } else if (static_res == Compile::SSC_always_false) {
3625 // Don't use intrinsic, have to throw ClassCastException.
3626 // If the reference is null, the non-intrinsic bytecode will
3627 // be optimized appropriately.
3628 return false;
3629 }
3630 }
3631 }
3632
3633 // Bailout intrinsic and do normal inlining if exception path is frequent.
3634 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3635 return false;
3636 }
3637
3638 // Generate dynamic checks.
3639 // Class.cast() is java implementation of _checkcast bytecode.
3640 // Do checkcast (Parse::do_checkcast()) optimizations here.
3641
3642 mirror = null_check(mirror);
3643 // If mirror is dead, only null-path is taken.
3644 if (stopped()) {
3645 return true;
3646 }
3647
3648 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3649 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3650 RegionNode* region = new RegionNode(PATH_LIMIT);
3651 record_for_igvn(region);
3652
3653 // Now load the mirror's klass metaobject, and null-check it.
3654 // If kls is null, we have a primitive mirror and
3655 // nothing is an instance of a primitive type.
3656 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3657
3658 Node* res = top();
3659 if (!stopped()) {
3660 Node* bad_type_ctrl = top();
3661 // Do checkcast optimizations.
3662 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3663 region->init_req(_bad_type_path, bad_type_ctrl);
3664 }
3665 if (region->in(_prim_path) != top() ||
3666 region->in(_bad_type_path) != top()) {
3667 // Let Interpreter throw ClassCastException.
3668 PreserveJVMState pjvms(this);
3669 set_control(_gvn.transform(region));
3670 uncommon_trap(Deoptimization::Reason_intrinsic,
3671 Deoptimization::Action_maybe_recompile);
3672 }
3673 if (!stopped()) {
3674 set_result(res);
3675 }
3676 return true;
3677 }
3678
3679
3680 //--------------------------inline_native_subtype_check------------------------
3681 // This intrinsic takes the JNI calls out of the heart of
3682 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3683 bool LibraryCallKit::inline_native_subtype_check() {
3684 // Pull both arguments off the stack.
3685 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3686 args[0] = argument(0);
3687 args[1] = argument(1);
3688 Node* klasses[2]; // corresponding Klasses: superk, subk
3689 klasses[0] = klasses[1] = top();
3690
3691 enum {
3692 // A full decision tree on {superc is prim, subc is prim}:
3693 _prim_0_path = 1, // {P,N} => false
3694 // {P,P} & superc!=subc => false
3695 _prim_same_path, // {P,P} & superc==subc => true
3696 _prim_1_path, // {N,P} => false
3697 _ref_subtype_path, // {N,N} & subtype check wins => true
3698 _both_ref_path, // {N,N} & subtype check loses => false
3699 PATH_LIMIT
3700 };
3701
3702 RegionNode* region = new RegionNode(PATH_LIMIT);
3703 Node* phi = new PhiNode(region, TypeInt::BOOL);
3704 record_for_igvn(region);
3705
3706 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3707 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3708 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3709
3710 // First null-check both mirrors and load each mirror's klass metaobject.
3711 int which_arg;
3712 for (which_arg = 0; which_arg <= 1; which_arg++) {
3713 Node* arg = args[which_arg];
3714 arg = null_check(arg);
3715 if (stopped()) break;
3716 args[which_arg] = arg;
3717
3718 Node* p = basic_plus_adr(arg, class_klass_offset);
3719 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3720 klasses[which_arg] = _gvn.transform(kls);
3721 }
3722
3723 // Having loaded both klasses, test each for null.
3724 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3725 for (which_arg = 0; which_arg <= 1; which_arg++) {
3726 Node* kls = klasses[which_arg];
3727 Node* null_ctl = top();
3728 kls = null_check_oop(kls, &null_ctl, never_see_null);
3729 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3730 region->init_req(prim_path, null_ctl);
3731 if (stopped()) break;
3732 klasses[which_arg] = kls;
3733 }
3734
3735 if (!stopped()) {
3736 // now we have two reference types, in klasses[0..1]
3737 Node* subk = klasses[1]; // the argument to isAssignableFrom
3738 Node* superk = klasses[0]; // the receiver
3739 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3740 // now we have a successful reference subtype check
3741 region->set_req(_ref_subtype_path, control());
3742 }
3743
3744 // If both operands are primitive (both klasses null), then
3745 // we must return true when they are identical primitives.
3746 // It is convenient to test this after the first null klass check.
3747 set_control(region->in(_prim_0_path)); // go back to first null check
3748 if (!stopped()) {
3749 // Since superc is primitive, make a guard for the superc==subc case.
3750 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3751 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3752 generate_guard(bol_eq, region, PROB_FAIR);
3753 if (region->req() == PATH_LIMIT+1) {
3754 // A guard was added. If the added guard is taken, superc==subc.
3755 region->swap_edges(PATH_LIMIT, _prim_same_path);
3756 region->del_req(PATH_LIMIT);
3757 }
3758 region->set_req(_prim_0_path, control()); // Not equal after all.
3759 }
3760
3761 // these are the only paths that produce 'true':
3762 phi->set_req(_prim_same_path, intcon(1));
3763 phi->set_req(_ref_subtype_path, intcon(1));
3764
3765 // pull together the cases:
3766 assert(region->req() == PATH_LIMIT, "sane region");
3767 for (uint i = 1; i < region->req(); i++) {
3768 Node* ctl = region->in(i);
3769 if (ctl == NULL || ctl == top()) {
3770 region->set_req(i, top());
3771 phi ->set_req(i, top());
3772 } else if (phi->in(i) == NULL) {
3773 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3774 }
3775 }
3776
3777 set_control(_gvn.transform(region));
3778 set_result(_gvn.transform(phi));
3779 return true;
3780 }
3781
3782 //---------------------generate_array_guard_common------------------------
3783 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3784 bool obj_array, bool not_array) {
3785
3786 if (stopped()) {
3787 return NULL;
3788 }
3789
3790 // If obj_array/non_array==false/false:
3791 // Branch around if the given klass is in fact an array (either obj or prim).
3792 // If obj_array/non_array==false/true:
3793 // Branch around if the given klass is not an array klass of any kind.
3794 // If obj_array/non_array==true/true:
3795 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3796 // If obj_array/non_array==true/false:
3797 // Branch around if the kls is an oop array (Object[] or subtype)
3798 //
3799 // Like generate_guard, adds a new path onto the region.
3800 jint layout_con = 0;
3801 Node* layout_val = get_layout_helper(kls, layout_con);
3802 if (layout_val == NULL) {
3803 bool query = (obj_array
3804 ? Klass::layout_helper_is_objArray(layout_con)
3805 : Klass::layout_helper_is_array(layout_con));
3806 if (query == not_array) {
3807 return NULL; // never a branch
3808 } else { // always a branch
3809 Node* always_branch = control();
3810 if (region != NULL)
3811 region->add_req(always_branch);
3812 set_control(top());
3813 return always_branch;
3814 }
3815 }
3816 // Now test the correct condition.
3817 jint nval = (obj_array
3818 ? ((jint)Klass::_lh_array_tag_type_value
3819 << Klass::_lh_array_tag_shift)
3820 : Klass::_lh_neutral_value);
3821 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3822 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3823 // invert the test if we are looking for a non-array
3824 if (not_array) btest = BoolTest(btest).negate();
3825 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3826 return generate_fair_guard(bol, region);
3827 }
3828
3829
3830 //-----------------------inline_native_newArray--------------------------
3831 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3832 bool LibraryCallKit::inline_native_newArray() {
3833 Node* mirror = argument(0);
3834 Node* count_val = argument(1);
3835
3836 mirror = null_check(mirror);
3837 // If mirror or obj is dead, only null-path is taken.
3838 if (stopped()) return true;
3839
3840 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3841 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3842 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3843 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3844 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3845
3846 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3847 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3848 result_reg, _slow_path);
3849 Node* normal_ctl = control();
3850 Node* no_array_ctl = result_reg->in(_slow_path);
3851
3852 // Generate code for the slow case. We make a call to newArray().
3853 set_control(no_array_ctl);
3854 if (!stopped()) {
3855 // Either the input type is void.class, or else the
3856 // array klass has not yet been cached. Either the
3857 // ensuing call will throw an exception, or else it
3858 // will cache the array klass for next time.
3859 PreserveJVMState pjvms(this);
3860 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3861 Node* slow_result = set_results_for_java_call(slow_call);
3862 // this->control() comes from set_results_for_java_call
3863 result_reg->set_req(_slow_path, control());
3864 result_val->set_req(_slow_path, slow_result);
3865 result_io ->set_req(_slow_path, i_o());
3866 result_mem->set_req(_slow_path, reset_memory());
3867 }
3868
3869 set_control(normal_ctl);
3870 if (!stopped()) {
3871 // Normal case: The array type has been cached in the java.lang.Class.
3872 // The following call works fine even if the array type is polymorphic.
3873 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3874 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3875 result_reg->init_req(_normal_path, control());
3876 result_val->init_req(_normal_path, obj);
3877 result_io ->init_req(_normal_path, i_o());
3878 result_mem->init_req(_normal_path, reset_memory());
3879 }
3880
3881 // Return the combined state.
3882 set_i_o( _gvn.transform(result_io) );
3883 set_all_memory( _gvn.transform(result_mem));
3884
3885 C->set_has_split_ifs(true); // Has chance for split-if optimization
3886 set_result(result_reg, result_val);
3887 return true;
3888 }
3889
3890 //----------------------inline_native_getLength--------------------------
3891 // public static native int java.lang.reflect.Array.getLength(Object array);
3892 bool LibraryCallKit::inline_native_getLength() {
3893 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3894
3895 Node* array = null_check(argument(0));
3896 // If array is dead, only null-path is taken.
3897 if (stopped()) return true;
3898
3899 // Deoptimize if it is a non-array.
3900 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3901
3902 if (non_array != NULL) {
3903 PreserveJVMState pjvms(this);
3904 set_control(non_array);
3905 uncommon_trap(Deoptimization::Reason_intrinsic,
3906 Deoptimization::Action_maybe_recompile);
3907 }
3908
3909 // If control is dead, only non-array-path is taken.
3910 if (stopped()) return true;
3911
3912 // The works fine even if the array type is polymorphic.
3913 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3914 Node* result = load_array_length(array);
3915
3916 C->set_has_split_ifs(true); // Has chance for split-if optimization
3917 set_result(result);
3918 return true;
3919 }
3920
3921 //------------------------inline_array_copyOf----------------------------
3922 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3923 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3924 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3925 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3926
3927 // Get the arguments.
3928 Node* original = argument(0);
3929 Node* start = is_copyOfRange? argument(1): intcon(0);
3930 Node* end = is_copyOfRange? argument(2): argument(1);
3931 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3932
3933 Node* newcopy = NULL;
3934
3935 // Set the original stack and the reexecute bit for the interpreter to reexecute
3936 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3937 { PreserveReexecuteState preexecs(this);
3938 jvms()->set_should_reexecute(true);
3939
3940 array_type_mirror = null_check(array_type_mirror);
3941 original = null_check(original);
3942
3943 // Check if a null path was taken unconditionally.
3944 if (stopped()) return true;
3945
3946 Node* orig_length = load_array_length(original);
3947
3948 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3949 klass_node = null_check(klass_node);
3950
3951 RegionNode* bailout = new RegionNode(1);
3952 record_for_igvn(bailout);
3953
3954 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3955 // Bail out if that is so.
3956 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3957 if (not_objArray != NULL) {
3958 // Improve the klass node's type from the new optimistic assumption:
3959 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3960 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3961 Node* cast = new CastPPNode(klass_node, akls);
3962 cast->init_req(0, control());
3963 klass_node = _gvn.transform(cast);
3964 }
3965
3966 // Bail out if either start or end is negative.
3967 generate_negative_guard(start, bailout, &start);
3968 generate_negative_guard(end, bailout, &end);
3969
3970 Node* length = end;
3971 if (_gvn.type(start) != TypeInt::ZERO) {
3972 length = _gvn.transform(new SubINode(end, start));
3973 }
3974
3975 // Bail out if length is negative.
3976 // Without this the new_array would throw
3977 // NegativeArraySizeException but IllegalArgumentException is what
3978 // should be thrown
3979 generate_negative_guard(length, bailout, &length);
3980
3981 if (bailout->req() > 1) {
3982 PreserveJVMState pjvms(this);
3983 set_control(_gvn.transform(bailout));
3984 uncommon_trap(Deoptimization::Reason_intrinsic,
3985 Deoptimization::Action_maybe_recompile);
3986 }
3987
3988 if (!stopped()) {
3989 // How many elements will we copy from the original?
3990 // The answer is MinI(orig_length - start, length).
3991 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3992 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3993
3994 // Generate a direct call to the right arraycopy function(s).
3995 // We know the copy is disjoint but we might not know if the
3996 // oop stores need checking.
3997 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3998 // This will fail a store-check if x contains any non-nulls.
3999
4000 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4001 // loads/stores but it is legal only if we're sure the
4002 // Arrays.copyOf would succeed. So we need all input arguments
4003 // to the copyOf to be validated, including that the copy to the
4004 // new array won't trigger an ArrayStoreException. That subtype
4005 // check can be optimized if we know something on the type of
4006 // the input array from type speculation.
4007 if (_gvn.type(klass_node)->singleton()) {
4008 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
4009 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4010
4011 int test = C->static_subtype_check(superk, subk);
4012 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4013 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4014 if (t_original->speculative_type() != NULL) {
4015 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4016 }
4017 }
4018 }
4019
4020 bool validated = false;
4021 // Reason_class_check rather than Reason_intrinsic because we
4022 // want to intrinsify even if this traps.
4023 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4024 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
4025 klass_node);
4026
4027 if (not_subtype_ctrl != top()) {
4028 PreserveJVMState pjvms(this);
4029 set_control(not_subtype_ctrl);
4030 uncommon_trap(Deoptimization::Reason_class_check,
4031 Deoptimization::Action_make_not_entrant);
4032 assert(stopped(), "Should be stopped");
4033 }
4034 validated = true;
4035 }
4036
4037 if (!stopped()) {
4038 newcopy = new_array(klass_node, length, 0); // no arguments to push
4039
4040 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true,
4041 load_object_klass(original), klass_node);
4042 if (!is_copyOfRange) {
4043 ac->set_copyof(validated);
4044 } else {
4045 ac->set_copyofrange(validated);
4046 }
4047 Node* n = _gvn.transform(ac);
4048 if (n == ac) {
4049 ac->connect_outputs(this);
4050 } else {
4051 assert(validated, "shouldn't transform if all arguments not validated");
4052 set_all_memory(n);
4053 }
4054 }
4055 }
4056 } // original reexecute is set back here
4057
4058 C->set_has_split_ifs(true); // Has chance for split-if optimization
4059 if (!stopped()) {
4060 set_result(newcopy);
4061 }
4062 return true;
4063 }
4064
4065
4066 //----------------------generate_virtual_guard---------------------------
4067 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4068 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4069 RegionNode* slow_region) {
4070 ciMethod* method = callee();
4071 int vtable_index = method->vtable_index();
4072 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4073 "bad index %d", vtable_index);
4074 // Get the Method* out of the appropriate vtable entry.
4075 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4076 vtable_index*vtableEntry::size_in_bytes() +
4077 vtableEntry::method_offset_in_bytes();
4078 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4079 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4080
4081 // Compare the target method with the expected method (e.g., Object.hashCode).
4082 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4083
4084 Node* native_call = makecon(native_call_addr);
4085 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4086 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4087
4088 return generate_slow_guard(test_native, slow_region);
4089 }
4090
4091 //-----------------------generate_method_call----------------------------
4092 // Use generate_method_call to make a slow-call to the real
4093 // method if the fast path fails. An alternative would be to
4094 // use a stub like OptoRuntime::slow_arraycopy_Java.
4095 // This only works for expanding the current library call,
4096 // not another intrinsic. (E.g., don't use this for making an
4097 // arraycopy call inside of the copyOf intrinsic.)
4098 CallJavaNode*
4099 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4100 // When compiling the intrinsic method itself, do not use this technique.
4101 guarantee(callee() != C->method(), "cannot make slow-call to self");
4102
4103 ciMethod* method = callee();
4104 // ensure the JVMS we have will be correct for this call
4105 guarantee(method_id == method->intrinsic_id(), "must match");
4106
4107 const TypeFunc* tf = TypeFunc::make(method);
4108 CallJavaNode* slow_call;
4109 if (is_static) {
4110 assert(!is_virtual, "");
4111 slow_call = new CallStaticJavaNode(C, tf,
4112 SharedRuntime::get_resolve_static_call_stub(),
4113 method, bci());
4114 } else if (is_virtual) {
4115 null_check_receiver();
4116 int vtable_index = Method::invalid_vtable_index;
4117 if (UseInlineCaches) {
4118 // Suppress the vtable call
4119 } else {
4120 // hashCode and clone are not a miranda methods,
4121 // so the vtable index is fixed.
4122 // No need to use the linkResolver to get it.
4123 vtable_index = method->vtable_index();
4124 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4125 "bad index %d", vtable_index);
4126 }
4127 slow_call = new CallDynamicJavaNode(tf,
4128 SharedRuntime::get_resolve_virtual_call_stub(),
4129 method, vtable_index, bci());
4130 } else { // neither virtual nor static: opt_virtual
4131 null_check_receiver();
4132 slow_call = new CallStaticJavaNode(C, tf,
4133 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4134 method, bci());
4135 slow_call->set_optimized_virtual(true);
4136 }
4137 set_arguments_for_java_call(slow_call);
4138 set_edges_for_java_call(slow_call);
4139 return slow_call;
4140 }
4141
4142
4143 /**
4144 * Build special case code for calls to hashCode on an object. This call may
4145 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4146 * slightly different code.
4147 */
4148 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4149 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4150 assert(!(is_virtual && is_static), "either virtual, special, or static");
4151
4152 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4153
4154 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4155 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4156 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4157 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4158 Node* obj = NULL;
4159 if (!is_static) {
4160 // Check for hashing null object
4161 obj = null_check_receiver();
4162 if (stopped()) return true; // unconditionally null
4163 result_reg->init_req(_null_path, top());
4164 result_val->init_req(_null_path, top());
4165 } else {
4166 // Do a null check, and return zero if null.
4167 // System.identityHashCode(null) == 0
4168 obj = argument(0);
4169 Node* null_ctl = top();
4170 obj = null_check_oop(obj, &null_ctl);
4171 result_reg->init_req(_null_path, null_ctl);
4172 result_val->init_req(_null_path, _gvn.intcon(0));
4173 }
4174
4175 // Unconditionally null? Then return right away.
4176 if (stopped()) {
4177 set_control( result_reg->in(_null_path));
4178 if (!stopped())
4179 set_result(result_val->in(_null_path));
4180 return true;
4181 }
4182
4183 // We only go to the fast case code if we pass a number of guards. The
4184 // paths which do not pass are accumulated in the slow_region.
4185 RegionNode* slow_region = new RegionNode(1);
4186 record_for_igvn(slow_region);
4187
4188 // If this is a virtual call, we generate a funny guard. We pull out
4189 // the vtable entry corresponding to hashCode() from the target object.
4190 // If the target method which we are calling happens to be the native
4191 // Object hashCode() method, we pass the guard. We do not need this
4192 // guard for non-virtual calls -- the caller is known to be the native
4193 // Object hashCode().
4194 if (is_virtual) {
4195 // After null check, get the object's klass.
4196 Node* obj_klass = load_object_klass(obj);
4197 generate_virtual_guard(obj_klass, slow_region);
4198 }
4199
4200 // Get the header out of the object, use LoadMarkNode when available
4201 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4202 // The control of the load must be NULL. Otherwise, the load can move before
4203 // the null check after castPP removal.
4204 Node* no_ctrl = NULL;
4205 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4206
4207 // Test the header to see if it is unlocked.
4208 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4209 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4210 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4211 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4212 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4213
4214 generate_slow_guard(test_unlocked, slow_region);
4215
4216 // Get the hash value and check to see that it has been properly assigned.
4217 // We depend on hash_mask being at most 32 bits and avoid the use of
4218 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4219 // vm: see markOop.hpp.
4220 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4221 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4222 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4223 // This hack lets the hash bits live anywhere in the mark object now, as long
4224 // as the shift drops the relevant bits into the low 32 bits. Note that
4225 // Java spec says that HashCode is an int so there's no point in capturing
4226 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4227 hshifted_header = ConvX2I(hshifted_header);
4228 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4229
4230 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4231 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4232 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4233
4234 generate_slow_guard(test_assigned, slow_region);
4235
4236 Node* init_mem = reset_memory();
4237 // fill in the rest of the null path:
4238 result_io ->init_req(_null_path, i_o());
4239 result_mem->init_req(_null_path, init_mem);
4240
4241 result_val->init_req(_fast_path, hash_val);
4242 result_reg->init_req(_fast_path, control());
4243 result_io ->init_req(_fast_path, i_o());
4244 result_mem->init_req(_fast_path, init_mem);
4245
4246 // Generate code for the slow case. We make a call to hashCode().
4247 set_control(_gvn.transform(slow_region));
4248 if (!stopped()) {
4249 // No need for PreserveJVMState, because we're using up the present state.
4250 set_all_memory(init_mem);
4251 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4252 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4253 Node* slow_result = set_results_for_java_call(slow_call);
4254 // this->control() comes from set_results_for_java_call
4255 result_reg->init_req(_slow_path, control());
4256 result_val->init_req(_slow_path, slow_result);
4257 result_io ->set_req(_slow_path, i_o());
4258 result_mem ->set_req(_slow_path, reset_memory());
4259 }
4260
4261 // Return the combined state.
4262 set_i_o( _gvn.transform(result_io) );
4263 set_all_memory( _gvn.transform(result_mem));
4264
4265 set_result(result_reg, result_val);
4266 return true;
4267 }
4268
4269 //---------------------------inline_native_getClass----------------------------
4270 // public final native Class<?> java.lang.Object.getClass();
4271 //
4272 // Build special case code for calls to getClass on an object.
4273 bool LibraryCallKit::inline_native_getClass() {
4274 Node* obj = null_check_receiver();
4275 if (stopped()) return true;
4276 set_result(load_mirror_from_klass(load_object_klass(obj)));
4277 return true;
4278 }
4279
4280 //-----------------inline_native_Reflection_getCallerClass---------------------
4281 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4282 //
4283 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4284 //
4285 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4286 // in that it must skip particular security frames and checks for
4287 // caller sensitive methods.
4288 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4289 #ifndef PRODUCT
4290 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4291 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4292 }
4293 #endif
4294
4295 if (!jvms()->has_method()) {
4296 #ifndef PRODUCT
4297 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4298 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4299 }
4300 #endif
4301 return false;
4302 }
4303
4304 // Walk back up the JVM state to find the caller at the required
4305 // depth.
4306 JVMState* caller_jvms = jvms();
4307
4308 // Cf. JVM_GetCallerClass
4309 // NOTE: Start the loop at depth 1 because the current JVM state does
4310 // not include the Reflection.getCallerClass() frame.
4311 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4312 ciMethod* m = caller_jvms->method();
4313 switch (n) {
4314 case 0:
4315 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4316 break;
4317 case 1:
4318 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4319 if (!m->caller_sensitive()) {
4320 #ifndef PRODUCT
4321 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4322 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4323 }
4324 #endif
4325 return false; // bail-out; let JVM_GetCallerClass do the work
4326 }
4327 break;
4328 default:
4329 if (!m->is_ignored_by_security_stack_walk()) {
4330 // We have reached the desired frame; return the holder class.
4331 // Acquire method holder as java.lang.Class and push as constant.
4332 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4333 ciInstance* caller_mirror = caller_klass->java_mirror();
4334 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4335
4336 #ifndef PRODUCT
4337 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4338 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());
4339 tty->print_cr(" JVM state at this point:");
4340 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4341 ciMethod* m = jvms()->of_depth(i)->method();
4342 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4343 }
4344 }
4345 #endif
4346 return true;
4347 }
4348 break;
4349 }
4350 }
4351
4352 #ifndef PRODUCT
4353 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4354 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4355 tty->print_cr(" JVM state at this point:");
4356 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4357 ciMethod* m = jvms()->of_depth(i)->method();
4358 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4359 }
4360 }
4361 #endif
4362
4363 return false; // bail-out; let JVM_GetCallerClass do the work
4364 }
4365
4366 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4367 Node* arg = argument(0);
4368 Node* result = NULL;
4369
4370 switch (id) {
4371 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4372 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4373 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4374 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4375
4376 case vmIntrinsics::_doubleToLongBits: {
4377 // two paths (plus control) merge in a wood
4378 RegionNode *r = new RegionNode(3);
4379 Node *phi = new PhiNode(r, TypeLong::LONG);
4380
4381 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4382 // Build the boolean node
4383 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4384
4385 // Branch either way.
4386 // NaN case is less traveled, which makes all the difference.
4387 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4388 Node *opt_isnan = _gvn.transform(ifisnan);
4389 assert( opt_isnan->is_If(), "Expect an IfNode");
4390 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4391 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4392
4393 set_control(iftrue);
4394
4395 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4396 Node *slow_result = longcon(nan_bits); // return NaN
4397 phi->init_req(1, _gvn.transform( slow_result ));
4398 r->init_req(1, iftrue);
4399
4400 // Else fall through
4401 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4402 set_control(iffalse);
4403
4404 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4405 r->init_req(2, iffalse);
4406
4407 // Post merge
4408 set_control(_gvn.transform(r));
4409 record_for_igvn(r);
4410
4411 C->set_has_split_ifs(true); // Has chance for split-if optimization
4412 result = phi;
4413 assert(result->bottom_type()->isa_long(), "must be");
4414 break;
4415 }
4416
4417 case vmIntrinsics::_floatToIntBits: {
4418 // two paths (plus control) merge in a wood
4419 RegionNode *r = new RegionNode(3);
4420 Node *phi = new PhiNode(r, TypeInt::INT);
4421
4422 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4423 // Build the boolean node
4424 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4425
4426 // Branch either way.
4427 // NaN case is less traveled, which makes all the difference.
4428 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4429 Node *opt_isnan = _gvn.transform(ifisnan);
4430 assert( opt_isnan->is_If(), "Expect an IfNode");
4431 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4432 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4433
4434 set_control(iftrue);
4435
4436 static const jint nan_bits = 0x7fc00000;
4437 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4438 phi->init_req(1, _gvn.transform( slow_result ));
4439 r->init_req(1, iftrue);
4440
4441 // Else fall through
4442 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4443 set_control(iffalse);
4444
4445 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4446 r->init_req(2, iffalse);
4447
4448 // Post merge
4449 set_control(_gvn.transform(r));
4450 record_for_igvn(r);
4451
4452 C->set_has_split_ifs(true); // Has chance for split-if optimization
4453 result = phi;
4454 assert(result->bottom_type()->isa_int(), "must be");
4455 break;
4456 }
4457
4458 default:
4459 fatal_unexpected_iid(id);
4460 break;
4461 }
4462 set_result(_gvn.transform(result));
4463 return true;
4464 }
4465
4466 //----------------------inline_unsafe_copyMemory-------------------------
4467 // public native void Unsafe.copyMemory(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4468 bool LibraryCallKit::inline_unsafe_copyMemory() {
4469 if (callee()->is_static()) return false; // caller must have the capability!
4470 null_check_receiver(); // null-check receiver
4471 if (stopped()) return true;
4472
4473 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4474
4475 Node* src_ptr = argument(1); // type: oop
4476 Node* src_off = ConvL2X(argument(2)); // type: long
4477 Node* dst_ptr = argument(4); // type: oop
4478 Node* dst_off = ConvL2X(argument(5)); // type: long
4479 Node* size = ConvL2X(argument(7)); // type: long
4480
4481 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4482 "fieldOffset must be byte-scaled");
4483
4484 Node* src = make_unsafe_address(src_ptr, src_off);
4485 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4486
4487 // Conservatively insert a memory barrier on all memory slices.
4488 // Do not let writes of the copy source or destination float below the copy.
4489 insert_mem_bar(Op_MemBarCPUOrder);
4490
4491 // Call it. Note that the length argument is not scaled.
4492 make_runtime_call(RC_LEAF|RC_NO_FP,
4493 OptoRuntime::fast_arraycopy_Type(),
4494 StubRoutines::unsafe_arraycopy(),
4495 "unsafe_arraycopy",
4496 TypeRawPtr::BOTTOM,
4497 src, dst, size XTOP);
4498
4499 // Do not let reads of the copy destination float above the copy.
4500 insert_mem_bar(Op_MemBarCPUOrder);
4501
4502 return true;
4503 }
4504
4505 //------------------------clone_coping-----------------------------------
4506 // Helper function for inline_native_clone.
4507 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4508 assert(obj_size != NULL, "");
4509 Node* raw_obj = alloc_obj->in(1);
4510 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4511
4512 AllocateNode* alloc = NULL;
4513 if (ReduceBulkZeroing) {
4514 // We will be completely responsible for initializing this object -
4515 // mark Initialize node as complete.
4516 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4517 // The object was just allocated - there should be no any stores!
4518 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4519 // Mark as complete_with_arraycopy so that on AllocateNode
4520 // expansion, we know this AllocateNode is initialized by an array
4521 // copy and a StoreStore barrier exists after the array copy.
4522 alloc->initialization()->set_complete_with_arraycopy();
4523 }
4524
4525 // Copy the fastest available way.
4526 // TODO: generate fields copies for small objects instead.
4527 Node* src = obj;
4528 Node* dest = alloc_obj;
4529 Node* size = _gvn.transform(obj_size);
4530
4531 // Exclude the header but include array length to copy by 8 bytes words.
4532 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4533 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4534 instanceOopDesc::base_offset_in_bytes();
4535 // base_off:
4536 // 8 - 32-bit VM
4537 // 12 - 64-bit VM, compressed klass
4538 // 16 - 64-bit VM, normal klass
4539 if (base_off % BytesPerLong != 0) {
4540 assert(UseCompressedClassPointers, "");
4541 if (is_array) {
4542 // Exclude length to copy by 8 bytes words.
4543 base_off += sizeof(int);
4544 } else {
4545 // Include klass to copy by 8 bytes words.
4546 base_off = instanceOopDesc::klass_offset_in_bytes();
4547 }
4548 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4549 }
4550 src = basic_plus_adr(src, base_off);
4551 dest = basic_plus_adr(dest, base_off);
4552
4553 // Compute the length also, if needed:
4554 Node* countx = size;
4555 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4556 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4557
4558 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4559
4560 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false);
4561 ac->set_clonebasic();
4562 Node* n = _gvn.transform(ac);
4563 if (n == ac) {
4564 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4565 } else {
4566 set_all_memory(n);
4567 }
4568
4569 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4570 if (card_mark) {
4571 assert(!is_array, "");
4572 // Put in store barrier for any and all oops we are sticking
4573 // into this object. (We could avoid this if we could prove
4574 // that the object type contains no oop fields at all.)
4575 Node* no_particular_value = NULL;
4576 Node* no_particular_field = NULL;
4577 int raw_adr_idx = Compile::AliasIdxRaw;
4578 post_barrier(control(),
4579 memory(raw_adr_type),
4580 alloc_obj,
4581 no_particular_field,
4582 raw_adr_idx,
4583 no_particular_value,
4584 T_OBJECT,
4585 false);
4586 }
4587
4588 // Do not let reads from the cloned object float above the arraycopy.
4589 if (alloc != NULL) {
4590 // Do not let stores that initialize this object be reordered with
4591 // a subsequent store that would make this object accessible by
4592 // other threads.
4593 // Record what AllocateNode this StoreStore protects so that
4594 // escape analysis can go from the MemBarStoreStoreNode to the
4595 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4596 // based on the escape status of the AllocateNode.
4597 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4598 } else {
4599 insert_mem_bar(Op_MemBarCPUOrder);
4600 }
4601 }
4602
4603 //------------------------inline_native_clone----------------------------
4604 // protected native Object java.lang.Object.clone();
4605 //
4606 // Here are the simple edge cases:
4607 // null receiver => normal trap
4608 // virtual and clone was overridden => slow path to out-of-line clone
4609 // not cloneable or finalizer => slow path to out-of-line Object.clone
4610 //
4611 // The general case has two steps, allocation and copying.
4612 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4613 //
4614 // Copying also has two cases, oop arrays and everything else.
4615 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4616 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4617 //
4618 // These steps fold up nicely if and when the cloned object's klass
4619 // can be sharply typed as an object array, a type array, or an instance.
4620 //
4621 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4622 PhiNode* result_val;
4623
4624 // Set the reexecute bit for the interpreter to reexecute
4625 // the bytecode that invokes Object.clone if deoptimization happens.
4626 { PreserveReexecuteState preexecs(this);
4627 jvms()->set_should_reexecute(true);
4628
4629 Node* obj = null_check_receiver();
4630 if (stopped()) return true;
4631
4632 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4633
4634 // If we are going to clone an instance, we need its exact type to
4635 // know the number and types of fields to convert the clone to
4636 // loads/stores. Maybe a speculative type can help us.
4637 if (!obj_type->klass_is_exact() &&
4638 obj_type->speculative_type() != NULL &&
4639 obj_type->speculative_type()->is_instance_klass()) {
4640 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4641 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4642 !spec_ik->has_injected_fields()) {
4643 ciKlass* k = obj_type->klass();
4644 if (!k->is_instance_klass() ||
4645 k->as_instance_klass()->is_interface() ||
4646 k->as_instance_klass()->has_subklass()) {
4647 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4648 }
4649 }
4650 }
4651
4652 Node* obj_klass = load_object_klass(obj);
4653 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4654 const TypeOopPtr* toop = ((tklass != NULL)
4655 ? tklass->as_instance_type()
4656 : TypeInstPtr::NOTNULL);
4657
4658 // Conservatively insert a memory barrier on all memory slices.
4659 // Do not let writes into the original float below the clone.
4660 insert_mem_bar(Op_MemBarCPUOrder);
4661
4662 // paths into result_reg:
4663 enum {
4664 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4665 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4666 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4667 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4668 PATH_LIMIT
4669 };
4670 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4671 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4672 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4673 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4674 record_for_igvn(result_reg);
4675
4676 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4677 int raw_adr_idx = Compile::AliasIdxRaw;
4678
4679 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4680 if (array_ctl != NULL) {
4681 // It's an array.
4682 PreserveJVMState pjvms(this);
4683 set_control(array_ctl);
4684 Node* obj_length = load_array_length(obj);
4685 Node* obj_size = NULL;
4686 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4687
4688 if (!use_ReduceInitialCardMarks()) {
4689 // If it is an oop array, it requires very special treatment,
4690 // because card marking is required on each card of the array.
4691 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4692 if (is_obja != NULL) {
4693 PreserveJVMState pjvms2(this);
4694 set_control(is_obja);
4695 // Generate a direct call to the right arraycopy function(s).
4696 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4697 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL);
4698 ac->set_cloneoop();
4699 Node* n = _gvn.transform(ac);
4700 assert(n == ac, "cannot disappear");
4701 ac->connect_outputs(this);
4702
4703 result_reg->init_req(_objArray_path, control());
4704 result_val->init_req(_objArray_path, alloc_obj);
4705 result_i_o ->set_req(_objArray_path, i_o());
4706 result_mem ->set_req(_objArray_path, reset_memory());
4707 }
4708 }
4709 // Otherwise, there are no card marks to worry about.
4710 // (We can dispense with card marks if we know the allocation
4711 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4712 // causes the non-eden paths to take compensating steps to
4713 // simulate a fresh allocation, so that no further
4714 // card marks are required in compiled code to initialize
4715 // the object.)
4716
4717 if (!stopped()) {
4718 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4719
4720 // Present the results of the copy.
4721 result_reg->init_req(_array_path, control());
4722 result_val->init_req(_array_path, alloc_obj);
4723 result_i_o ->set_req(_array_path, i_o());
4724 result_mem ->set_req(_array_path, reset_memory());
4725 }
4726 }
4727
4728 // We only go to the instance fast case code if we pass a number of guards.
4729 // The paths which do not pass are accumulated in the slow_region.
4730 RegionNode* slow_region = new RegionNode(1);
4731 record_for_igvn(slow_region);
4732 if (!stopped()) {
4733 // It's an instance (we did array above). Make the slow-path tests.
4734 // If this is a virtual call, we generate a funny guard. We grab
4735 // the vtable entry corresponding to clone() from the target object.
4736 // If the target method which we are calling happens to be the
4737 // Object clone() method, we pass the guard. We do not need this
4738 // guard for non-virtual calls; the caller is known to be the native
4739 // Object clone().
4740 if (is_virtual) {
4741 generate_virtual_guard(obj_klass, slow_region);
4742 }
4743
4744 // The object must be easily cloneable and must not have a finalizer.
4745 // Both of these conditions may be checked in a single test.
4746 // We could optimize the test further, but we don't care.
4747 generate_access_flags_guard(obj_klass,
4748 // Test both conditions:
4749 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4750 // Must be cloneable but not finalizer:
4751 JVM_ACC_IS_CLONEABLE_FAST,
4752 slow_region);
4753 }
4754
4755 if (!stopped()) {
4756 // It's an instance, and it passed the slow-path tests.
4757 PreserveJVMState pjvms(this);
4758 Node* obj_size = NULL;
4759 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4760 // is reexecuted if deoptimization occurs and there could be problems when merging
4761 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4762 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4763
4764 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4765
4766 // Present the results of the slow call.
4767 result_reg->init_req(_instance_path, control());
4768 result_val->init_req(_instance_path, alloc_obj);
4769 result_i_o ->set_req(_instance_path, i_o());
4770 result_mem ->set_req(_instance_path, reset_memory());
4771 }
4772
4773 // Generate code for the slow case. We make a call to clone().
4774 set_control(_gvn.transform(slow_region));
4775 if (!stopped()) {
4776 PreserveJVMState pjvms(this);
4777 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4778 Node* slow_result = set_results_for_java_call(slow_call);
4779 // this->control() comes from set_results_for_java_call
4780 result_reg->init_req(_slow_path, control());
4781 result_val->init_req(_slow_path, slow_result);
4782 result_i_o ->set_req(_slow_path, i_o());
4783 result_mem ->set_req(_slow_path, reset_memory());
4784 }
4785
4786 // Return the combined state.
4787 set_control( _gvn.transform(result_reg));
4788 set_i_o( _gvn.transform(result_i_o));
4789 set_all_memory( _gvn.transform(result_mem));
4790 } // original reexecute is set back here
4791
4792 set_result(_gvn.transform(result_val));
4793 return true;
4794 }
4795
4796 // If we have a tighly coupled allocation, the arraycopy may take care
4797 // of the array initialization. If one of the guards we insert between
4798 // the allocation and the arraycopy causes a deoptimization, an
4799 // unitialized array will escape the compiled method. To prevent that
4800 // we set the JVM state for uncommon traps between the allocation and
4801 // the arraycopy to the state before the allocation so, in case of
4802 // deoptimization, we'll reexecute the allocation and the
4803 // initialization.
4804 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4805 if (alloc != NULL) {
4806 ciMethod* trap_method = alloc->jvms()->method();
4807 int trap_bci = alloc->jvms()->bci();
4808
4809 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4810 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4811 // Make sure there's no store between the allocation and the
4812 // arraycopy otherwise visible side effects could be rexecuted
4813 // in case of deoptimization and cause incorrect execution.
4814 bool no_interfering_store = true;
4815 Node* mem = alloc->in(TypeFunc::Memory);
4816 if (mem->is_MergeMem()) {
4817 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4818 Node* n = mms.memory();
4819 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4820 assert(n->is_Store(), "what else?");
4821 no_interfering_store = false;
4822 break;
4823 }
4824 }
4825 } else {
4826 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4827 Node* n = mms.memory();
4828 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4829 assert(n->is_Store(), "what else?");
4830 no_interfering_store = false;
4831 break;
4832 }
4833 }
4834 }
4835
4836 if (no_interfering_store) {
4837 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4838 uint size = alloc->req();
4839 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4840 old_jvms->set_map(sfpt);
4841 for (uint i = 0; i < size; i++) {
4842 sfpt->init_req(i, alloc->in(i));
4843 }
4844 // re-push array length for deoptimization
4845 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4846 old_jvms->set_sp(old_jvms->sp()+1);
4847 old_jvms->set_monoff(old_jvms->monoff()+1);
4848 old_jvms->set_scloff(old_jvms->scloff()+1);
4849 old_jvms->set_endoff(old_jvms->endoff()+1);
4850 old_jvms->set_should_reexecute(true);
4851
4852 sfpt->set_i_o(map()->i_o());
4853 sfpt->set_memory(map()->memory());
4854 sfpt->set_control(map()->control());
4855
4856 JVMState* saved_jvms = jvms();
4857 saved_reexecute_sp = _reexecute_sp;
4858
4859 set_jvms(sfpt->jvms());
4860 _reexecute_sp = jvms()->sp();
4861
4862 return saved_jvms;
4863 }
4864 }
4865 }
4866 return NULL;
4867 }
4868
4869 // In case of a deoptimization, we restart execution at the
4870 // allocation, allocating a new array. We would leave an uninitialized
4871 // array in the heap that GCs wouldn't expect. Move the allocation
4872 // after the traps so we don't allocate the array if we
4873 // deoptimize. This is possible because tightly_coupled_allocation()
4874 // guarantees there's no observer of the allocated array at this point
4875 // and the control flow is simple enough.
4876 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) {
4877 if (saved_jvms != NULL && !stopped()) {
4878 assert(alloc != NULL, "only with a tightly coupled allocation");
4879 // restore JVM state to the state at the arraycopy
4880 saved_jvms->map()->set_control(map()->control());
4881 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4882 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4883 // If we've improved the types of some nodes (null check) while
4884 // emitting the guards, propagate them to the current state
4885 map()->replaced_nodes().apply(saved_jvms->map());
4886 set_jvms(saved_jvms);
4887 _reexecute_sp = saved_reexecute_sp;
4888
4889 // Remove the allocation from above the guards
4890 CallProjections callprojs;
4891 alloc->extract_projections(&callprojs, true);
4892 InitializeNode* init = alloc->initialization();
4893 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4894 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4895 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4896 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4897
4898 // move the allocation here (after the guards)
4899 _gvn.hash_delete(alloc);
4900 alloc->set_req(TypeFunc::Control, control());
4901 alloc->set_req(TypeFunc::I_O, i_o());
4902 Node *mem = reset_memory();
4903 set_all_memory(mem);
4904 alloc->set_req(TypeFunc::Memory, mem);
4905 set_control(init->proj_out(TypeFunc::Control));
4906 set_i_o(callprojs.fallthrough_ioproj);
4907
4908 // Update memory as done in GraphKit::set_output_for_allocation()
4909 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4910 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4911 if (ary_type->isa_aryptr() && length_type != NULL) {
4912 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4913 }
4914 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4915 int elemidx = C->get_alias_index(telemref);
4916 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4917 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4918
4919 Node* allocx = _gvn.transform(alloc);
4920 assert(allocx == alloc, "where has the allocation gone?");
4921 assert(dest->is_CheckCastPP(), "not an allocation result?");
4922
4923 _gvn.hash_delete(dest);
4924 dest->set_req(0, control());
4925 Node* destx = _gvn.transform(dest);
4926 assert(destx == dest, "where has the allocation result gone?");
4927 }
4928 }
4929
4930
4931 //------------------------------inline_arraycopy-----------------------
4932 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4933 // Object dest, int destPos,
4934 // int length);
4935 bool LibraryCallKit::inline_arraycopy() {
4936 // Get the arguments.
4937 Node* src = argument(0); // type: oop
4938 Node* src_offset = argument(1); // type: int
4939 Node* dest = argument(2); // type: oop
4940 Node* dest_offset = argument(3); // type: int
4941 Node* length = argument(4); // type: int
4942
4943
4944 // Check for allocation before we add nodes that would confuse
4945 // tightly_coupled_allocation()
4946 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4947
4948 int saved_reexecute_sp = -1;
4949 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4950 // See arraycopy_restore_alloc_state() comment
4951 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4952 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4953 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4954 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4955
4956 // The following tests must be performed
4957 // (1) src and dest are arrays.
4958 // (2) src and dest arrays must have elements of the same BasicType
4959 // (3) src and dest must not be null.
4960 // (4) src_offset must not be negative.
4961 // (5) dest_offset must not be negative.
4962 // (6) length must not be negative.
4963 // (7) src_offset + length must not exceed length of src.
4964 // (8) dest_offset + length must not exceed length of dest.
4965 // (9) each element of an oop array must be assignable
4966
4967 // (3) src and dest must not be null.
4968 // always do this here because we need the JVM state for uncommon traps
4969 Node* null_ctl = top();
4970 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4971 assert(null_ctl->is_top(), "no null control here");
4972 dest = null_check(dest, T_ARRAY);
4973
4974 if (!can_emit_guards) {
4975 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4976 // guards but the arraycopy node could still take advantage of a
4977 // tightly allocated allocation. tightly_coupled_allocation() is
4978 // called again to make sure it takes the null check above into
4979 // account: the null check is mandatory and if it caused an
4980 // uncommon trap to be emitted then the allocation can't be
4981 // considered tightly coupled in this context.
4982 alloc = tightly_coupled_allocation(dest, NULL);
4983 }
4984
4985 bool validated = false;
4986
4987 const Type* src_type = _gvn.type(src);
4988 const Type* dest_type = _gvn.type(dest);
4989 const TypeAryPtr* top_src = src_type->isa_aryptr();
4990 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4991
4992 // Do we have the type of src?
4993 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4994 // Do we have the type of dest?
4995 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4996 // Is the type for src from speculation?
4997 bool src_spec = false;
4998 // Is the type for dest from speculation?
4999 bool dest_spec = false;
5000
5001 if ((!has_src || !has_dest) && can_emit_guards) {
5002 // We don't have sufficient type information, let's see if
5003 // speculative types can help. We need to have types for both src
5004 // and dest so that it pays off.
5005
5006 // Do we already have or could we have type information for src
5007 bool could_have_src = has_src;
5008 // Do we already have or could we have type information for dest
5009 bool could_have_dest = has_dest;
5010
5011 ciKlass* src_k = NULL;
5012 if (!has_src) {
5013 src_k = src_type->speculative_type_not_null();
5014 if (src_k != NULL && src_k->is_array_klass()) {
5015 could_have_src = true;
5016 }
5017 }
5018
5019 ciKlass* dest_k = NULL;
5020 if (!has_dest) {
5021 dest_k = dest_type->speculative_type_not_null();
5022 if (dest_k != NULL && dest_k->is_array_klass()) {
5023 could_have_dest = true;
5024 }
5025 }
5026
5027 if (could_have_src && could_have_dest) {
5028 // This is going to pay off so emit the required guards
5029 if (!has_src) {
5030 src = maybe_cast_profiled_obj(src, src_k, true);
5031 src_type = _gvn.type(src);
5032 top_src = src_type->isa_aryptr();
5033 has_src = (top_src != NULL && top_src->klass() != NULL);
5034 src_spec = true;
5035 }
5036 if (!has_dest) {
5037 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5038 dest_type = _gvn.type(dest);
5039 top_dest = dest_type->isa_aryptr();
5040 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5041 dest_spec = true;
5042 }
5043 }
5044 }
5045
5046 if (has_src && has_dest && can_emit_guards) {
5047 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5048 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5049 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5050 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5051
5052 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5053 // If both arrays are object arrays then having the exact types
5054 // for both will remove the need for a subtype check at runtime
5055 // before the call and may make it possible to pick a faster copy
5056 // routine (without a subtype check on every element)
5057 // Do we have the exact type of src?
5058 bool could_have_src = src_spec;
5059 // Do we have the exact type of dest?
5060 bool could_have_dest = dest_spec;
5061 ciKlass* src_k = top_src->klass();
5062 ciKlass* dest_k = top_dest->klass();
5063 if (!src_spec) {
5064 src_k = src_type->speculative_type_not_null();
5065 if (src_k != NULL && src_k->is_array_klass()) {
5066 could_have_src = true;
5067 }
5068 }
5069 if (!dest_spec) {
5070 dest_k = dest_type->speculative_type_not_null();
5071 if (dest_k != NULL && dest_k->is_array_klass()) {
5072 could_have_dest = true;
5073 }
5074 }
5075 if (could_have_src && could_have_dest) {
5076 // If we can have both exact types, emit the missing guards
5077 if (could_have_src && !src_spec) {
5078 src = maybe_cast_profiled_obj(src, src_k, true);
5079 }
5080 if (could_have_dest && !dest_spec) {
5081 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5082 }
5083 }
5084 }
5085 }
5086
5087 ciMethod* trap_method = method();
5088 int trap_bci = bci();
5089 if (saved_jvms != NULL) {
5090 trap_method = alloc->jvms()->method();
5091 trap_bci = alloc->jvms()->bci();
5092 }
5093
5094 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5095 can_emit_guards &&
5096 !src->is_top() && !dest->is_top()) {
5097 // validate arguments: enables transformation the ArrayCopyNode
5098 validated = true;
5099
5100 RegionNode* slow_region = new RegionNode(1);
5101 record_for_igvn(slow_region);
5102
5103 // (1) src and dest are arrays.
5104 generate_non_array_guard(load_object_klass(src), slow_region);
5105 generate_non_array_guard(load_object_klass(dest), slow_region);
5106
5107 // (2) src and dest arrays must have elements of the same BasicType
5108 // done at macro expansion or at Ideal transformation time
5109
5110 // (4) src_offset must not be negative.
5111 generate_negative_guard(src_offset, slow_region);
5112
5113 // (5) dest_offset must not be negative.
5114 generate_negative_guard(dest_offset, slow_region);
5115
5116 // (7) src_offset + length must not exceed length of src.
5117 generate_limit_guard(src_offset, length,
5118 load_array_length(src),
5119 slow_region);
5120
5121 // (8) dest_offset + length must not exceed length of dest.
5122 generate_limit_guard(dest_offset, length,
5123 load_array_length(dest),
5124 slow_region);
5125
5126 // (9) each element of an oop array must be assignable
5127 Node* src_klass = load_object_klass(src);
5128 Node* dest_klass = load_object_klass(dest);
5129 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5130
5131 if (not_subtype_ctrl != top()) {
5132 PreserveJVMState pjvms(this);
5133 set_control(not_subtype_ctrl);
5134 uncommon_trap(Deoptimization::Reason_intrinsic,
5135 Deoptimization::Action_make_not_entrant);
5136 assert(stopped(), "Should be stopped");
5137 }
5138 {
5139 PreserveJVMState pjvms(this);
5140 set_control(_gvn.transform(slow_region));
5141 uncommon_trap(Deoptimization::Reason_intrinsic,
5142 Deoptimization::Action_make_not_entrant);
5143 assert(stopped(), "Should be stopped");
5144 }
5145 }
5146
5147 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp);
5148
5149 if (stopped()) {
5150 return true;
5151 }
5152
5153 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL,
5154 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5155 // so the compiler has a chance to eliminate them: during macro expansion,
5156 // we have to set their control (CastPP nodes are eliminated).
5157 load_object_klass(src), load_object_klass(dest),
5158 load_array_length(src), load_array_length(dest));
5159
5160 ac->set_arraycopy(validated);
5161
5162 Node* n = _gvn.transform(ac);
5163 if (n == ac) {
5164 ac->connect_outputs(this);
5165 } else {
5166 assert(validated, "shouldn't transform if all arguments not validated");
5167 set_all_memory(n);
5168 }
5169
5170 return true;
5171 }
5172
5173
5174 // Helper function which determines if an arraycopy immediately follows
5175 // an allocation, with no intervening tests or other escapes for the object.
5176 AllocateArrayNode*
5177 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5178 RegionNode* slow_region) {
5179 if (stopped()) return NULL; // no fast path
5180 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5181
5182 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5183 if (alloc == NULL) return NULL;
5184
5185 Node* rawmem = memory(Compile::AliasIdxRaw);
5186 // Is the allocation's memory state untouched?
5187 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5188 // Bail out if there have been raw-memory effects since the allocation.
5189 // (Example: There might have been a call or safepoint.)
5190 return NULL;
5191 }
5192 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5193 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5194 return NULL;
5195 }
5196
5197 // There must be no unexpected observers of this allocation.
5198 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5199 Node* obs = ptr->fast_out(i);
5200 if (obs != this->map()) {
5201 return NULL;
5202 }
5203 }
5204
5205 // This arraycopy must unconditionally follow the allocation of the ptr.
5206 Node* alloc_ctl = ptr->in(0);
5207 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5208
5209 Node* ctl = control();
5210 while (ctl != alloc_ctl) {
5211 // There may be guards which feed into the slow_region.
5212 // Any other control flow means that we might not get a chance
5213 // to finish initializing the allocated object.
5214 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5215 IfNode* iff = ctl->in(0)->as_If();
5216 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5217 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5218 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5219 ctl = iff->in(0); // This test feeds the known slow_region.
5220 continue;
5221 }
5222 // One more try: Various low-level checks bottom out in
5223 // uncommon traps. If the debug-info of the trap omits
5224 // any reference to the allocation, as we've already
5225 // observed, then there can be no objection to the trap.
5226 bool found_trap = false;
5227 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5228 Node* obs = not_ctl->fast_out(j);
5229 if (obs->in(0) == not_ctl && obs->is_Call() &&
5230 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5231 found_trap = true; break;
5232 }
5233 }
5234 if (found_trap) {
5235 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5236 continue;
5237 }
5238 }
5239 return NULL;
5240 }
5241
5242 // If we get this far, we have an allocation which immediately
5243 // precedes the arraycopy, and we can take over zeroing the new object.
5244 // The arraycopy will finish the initialization, and provide
5245 // a new control state to which we will anchor the destination pointer.
5246
5247 return alloc;
5248 }
5249
5250 //-------------inline_encodeISOArray-----------------------------------
5251 // encode char[] to byte[] in ISO_8859_1
5252 bool LibraryCallKit::inline_encodeISOArray() {
5253 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5254 // no receiver since it is static method
5255 Node *src = argument(0);
5256 Node *src_offset = argument(1);
5257 Node *dst = argument(2);
5258 Node *dst_offset = argument(3);
5259 Node *length = argument(4);
5260
5261 const Type* src_type = src->Value(&_gvn);
5262 const Type* dst_type = dst->Value(&_gvn);
5263 const TypeAryPtr* top_src = src_type->isa_aryptr();
5264 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5265 if (top_src == NULL || top_src->klass() == NULL ||
5266 top_dest == NULL || top_dest->klass() == NULL) {
5267 // failed array check
5268 return false;
5269 }
5270
5271 // Figure out the size and type of the elements we will be copying.
5272 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5273 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5274 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5275 return false;
5276 }
5277
5278 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5279 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5280 // 'src_start' points to src array + scaled offset
5281 // 'dst_start' points to dst array + scaled offset
5282
5283 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5284 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5285 enc = _gvn.transform(enc);
5286 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5287 set_memory(res_mem, mtype);
5288 set_result(enc);
5289 return true;
5290 }
5291
5292 //-------------inline_multiplyToLen-----------------------------------
5293 bool LibraryCallKit::inline_multiplyToLen() {
5294 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5295
5296 address stubAddr = StubRoutines::multiplyToLen();
5297 if (stubAddr == NULL) {
5298 return false; // Intrinsic's stub is not implemented on this platform
5299 }
5300 const char* stubName = "multiplyToLen";
5301
5302 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5303
5304 // no receiver because it is a static method
5305 Node* x = argument(0);
5306 Node* xlen = argument(1);
5307 Node* y = argument(2);
5308 Node* ylen = argument(3);
5309 Node* z = argument(4);
5310
5311 const Type* x_type = x->Value(&_gvn);
5312 const Type* y_type = y->Value(&_gvn);
5313 const TypeAryPtr* top_x = x_type->isa_aryptr();
5314 const TypeAryPtr* top_y = y_type->isa_aryptr();
5315 if (top_x == NULL || top_x->klass() == NULL ||
5316 top_y == NULL || top_y->klass() == NULL) {
5317 // failed array check
5318 return false;
5319 }
5320
5321 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5322 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5323 if (x_elem != T_INT || y_elem != T_INT) {
5324 return false;
5325 }
5326
5327 // Set the original stack and the reexecute bit for the interpreter to reexecute
5328 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5329 // on the return from z array allocation in runtime.
5330 { PreserveReexecuteState preexecs(this);
5331 jvms()->set_should_reexecute(true);
5332
5333 Node* x_start = array_element_address(x, intcon(0), x_elem);
5334 Node* y_start = array_element_address(y, intcon(0), y_elem);
5335 // 'x_start' points to x array + scaled xlen
5336 // 'y_start' points to y array + scaled ylen
5337
5338 // Allocate the result array
5339 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5340 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5341 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5342
5343 IdealKit ideal(this);
5344
5345 #define __ ideal.
5346 Node* one = __ ConI(1);
5347 Node* zero = __ ConI(0);
5348 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5349 __ set(need_alloc, zero);
5350 __ set(z_alloc, z);
5351 __ if_then(z, BoolTest::eq, null()); {
5352 __ increment (need_alloc, one);
5353 } __ else_(); {
5354 // Update graphKit memory and control from IdealKit.
5355 sync_kit(ideal);
5356 Node* zlen_arg = load_array_length(z);
5357 // Update IdealKit memory and control from graphKit.
5358 __ sync_kit(this);
5359 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5360 __ increment (need_alloc, one);
5361 } __ end_if();
5362 } __ end_if();
5363
5364 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5365 // Update graphKit memory and control from IdealKit.
5366 sync_kit(ideal);
5367 Node * narr = new_array(klass_node, zlen, 1);
5368 // Update IdealKit memory and control from graphKit.
5369 __ sync_kit(this);
5370 __ set(z_alloc, narr);
5371 } __ end_if();
5372
5373 sync_kit(ideal);
5374 z = __ value(z_alloc);
5375 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5376 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5377 // Final sync IdealKit and GraphKit.
5378 final_sync(ideal);
5379 #undef __
5380
5381 Node* z_start = array_element_address(z, intcon(0), T_INT);
5382
5383 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5384 OptoRuntime::multiplyToLen_Type(),
5385 stubAddr, stubName, TypePtr::BOTTOM,
5386 x_start, xlen, y_start, ylen, z_start, zlen);
5387 } // original reexecute is set back here
5388
5389 C->set_has_split_ifs(true); // Has chance for split-if optimization
5390 set_result(z);
5391 return true;
5392 }
5393
5394 //-------------inline_squareToLen------------------------------------
5395 bool LibraryCallKit::inline_squareToLen() {
5396 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5397
5398 address stubAddr = StubRoutines::squareToLen();
5399 if (stubAddr == NULL) {
5400 return false; // Intrinsic's stub is not implemented on this platform
5401 }
5402 const char* stubName = "squareToLen";
5403
5404 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5405
5406 Node* x = argument(0);
5407 Node* len = argument(1);
5408 Node* z = argument(2);
5409 Node* zlen = argument(3);
5410
5411 const Type* x_type = x->Value(&_gvn);
5412 const Type* z_type = z->Value(&_gvn);
5413 const TypeAryPtr* top_x = x_type->isa_aryptr();
5414 const TypeAryPtr* top_z = z_type->isa_aryptr();
5415 if (top_x == NULL || top_x->klass() == NULL ||
5416 top_z == NULL || top_z->klass() == NULL) {
5417 // failed array check
5418 return false;
5419 }
5420
5421 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5422 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5423 if (x_elem != T_INT || z_elem != T_INT) {
5424 return false;
5425 }
5426
5427
5428 Node* x_start = array_element_address(x, intcon(0), x_elem);
5429 Node* z_start = array_element_address(z, intcon(0), z_elem);
5430
5431 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5432 OptoRuntime::squareToLen_Type(),
5433 stubAddr, stubName, TypePtr::BOTTOM,
5434 x_start, len, z_start, zlen);
5435
5436 set_result(z);
5437 return true;
5438 }
5439
5440 //-------------inline_mulAdd------------------------------------------
5441 bool LibraryCallKit::inline_mulAdd() {
5442 assert(UseMulAddIntrinsic, "not implemented on this platform");
5443
5444 address stubAddr = StubRoutines::mulAdd();
5445 if (stubAddr == NULL) {
5446 return false; // Intrinsic's stub is not implemented on this platform
5447 }
5448 const char* stubName = "mulAdd";
5449
5450 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5451
5452 Node* out = argument(0);
5453 Node* in = argument(1);
5454 Node* offset = argument(2);
5455 Node* len = argument(3);
5456 Node* k = argument(4);
5457
5458 const Type* out_type = out->Value(&_gvn);
5459 const Type* in_type = in->Value(&_gvn);
5460 const TypeAryPtr* top_out = out_type->isa_aryptr();
5461 const TypeAryPtr* top_in = in_type->isa_aryptr();
5462 if (top_out == NULL || top_out->klass() == NULL ||
5463 top_in == NULL || top_in->klass() == NULL) {
5464 // failed array check
5465 return false;
5466 }
5467
5468 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5469 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5470 if (out_elem != T_INT || in_elem != T_INT) {
5471 return false;
5472 }
5473
5474 Node* outlen = load_array_length(out);
5475 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5476 Node* out_start = array_element_address(out, intcon(0), out_elem);
5477 Node* in_start = array_element_address(in, intcon(0), in_elem);
5478
5479 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5480 OptoRuntime::mulAdd_Type(),
5481 stubAddr, stubName, TypePtr::BOTTOM,
5482 out_start,in_start, new_offset, len, k);
5483 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5484 set_result(result);
5485 return true;
5486 }
5487
5488 //-------------inline_montgomeryMultiply-----------------------------------
5489 bool LibraryCallKit::inline_montgomeryMultiply() {
5490 address stubAddr = StubRoutines::montgomeryMultiply();
5491 if (stubAddr == NULL) {
5492 return false; // Intrinsic's stub is not implemented on this platform
5493 }
5494
5495 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5496 const char* stubName = "montgomery_square";
5497
5498 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5499
5500 Node* a = argument(0);
5501 Node* b = argument(1);
5502 Node* n = argument(2);
5503 Node* len = argument(3);
5504 Node* inv = argument(4);
5505 Node* m = argument(6);
5506
5507 const Type* a_type = a->Value(&_gvn);
5508 const TypeAryPtr* top_a = a_type->isa_aryptr();
5509 const Type* b_type = b->Value(&_gvn);
5510 const TypeAryPtr* top_b = b_type->isa_aryptr();
5511 const Type* n_type = a->Value(&_gvn);
5512 const TypeAryPtr* top_n = n_type->isa_aryptr();
5513 const Type* m_type = a->Value(&_gvn);
5514 const TypeAryPtr* top_m = m_type->isa_aryptr();
5515 if (top_a == NULL || top_a->klass() == NULL ||
5516 top_b == NULL || top_b->klass() == NULL ||
5517 top_n == NULL || top_n->klass() == NULL ||
5518 top_m == NULL || top_m->klass() == NULL) {
5519 // failed array check
5520 return false;
5521 }
5522
5523 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5524 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5525 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5526 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5527 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5528 return false;
5529 }
5530
5531 // Make the call
5532 {
5533 Node* a_start = array_element_address(a, intcon(0), a_elem);
5534 Node* b_start = array_element_address(b, intcon(0), b_elem);
5535 Node* n_start = array_element_address(n, intcon(0), n_elem);
5536 Node* m_start = array_element_address(m, intcon(0), m_elem);
5537
5538 Node* call = make_runtime_call(RC_LEAF,
5539 OptoRuntime::montgomeryMultiply_Type(),
5540 stubAddr, stubName, TypePtr::BOTTOM,
5541 a_start, b_start, n_start, len, inv, top(),
5542 m_start);
5543 set_result(m);
5544 }
5545
5546 return true;
5547 }
5548
5549 bool LibraryCallKit::inline_montgomerySquare() {
5550 address stubAddr = StubRoutines::montgomerySquare();
5551 if (stubAddr == NULL) {
5552 return false; // Intrinsic's stub is not implemented on this platform
5553 }
5554
5555 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5556 const char* stubName = "montgomery_square";
5557
5558 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5559
5560 Node* a = argument(0);
5561 Node* n = argument(1);
5562 Node* len = argument(2);
5563 Node* inv = argument(3);
5564 Node* m = argument(5);
5565
5566 const Type* a_type = a->Value(&_gvn);
5567 const TypeAryPtr* top_a = a_type->isa_aryptr();
5568 const Type* n_type = a->Value(&_gvn);
5569 const TypeAryPtr* top_n = n_type->isa_aryptr();
5570 const Type* m_type = a->Value(&_gvn);
5571 const TypeAryPtr* top_m = m_type->isa_aryptr();
5572 if (top_a == NULL || top_a->klass() == NULL ||
5573 top_n == NULL || top_n->klass() == NULL ||
5574 top_m == NULL || top_m->klass() == NULL) {
5575 // failed array check
5576 return false;
5577 }
5578
5579 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5580 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5581 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5582 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5583 return false;
5584 }
5585
5586 // Make the call
5587 {
5588 Node* a_start = array_element_address(a, intcon(0), a_elem);
5589 Node* n_start = array_element_address(n, intcon(0), n_elem);
5590 Node* m_start = array_element_address(m, intcon(0), m_elem);
5591
5592 Node* call = make_runtime_call(RC_LEAF,
5593 OptoRuntime::montgomerySquare_Type(),
5594 stubAddr, stubName, TypePtr::BOTTOM,
5595 a_start, n_start, len, inv, top(),
5596 m_start);
5597 set_result(m);
5598 }
5599
5600 return true;
5601 }
5602
5603 //-------------inline_vectorizedMismatch------------------------------
5604 bool LibraryCallKit::inline_vectorizedMismatch() {
5605 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5606
5607 address stubAddr = StubRoutines::vectorizedMismatch();
5608 if (stubAddr == NULL) {
5609 return false; // Intrinsic's stub is not implemented on this platform
5610 }
5611 const char* stubName = "vectorizedMismatch";
5612 int size_l = callee()->signature()->size();
5613 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5614
5615 Node* obja = argument(0);
5616 Node* aoffset = argument(1);
5617 Node* objb = argument(3);
5618 Node* boffset = argument(4);
5619 Node* length = argument(6);
5620 Node* scale = argument(7);
5621
5622 const Type* a_type = obja->Value(&_gvn);
5623 const Type* b_type = objb->Value(&_gvn);
5624 const TypeAryPtr* top_a = a_type->isa_aryptr();
5625 const TypeAryPtr* top_b = b_type->isa_aryptr();
5626 if (top_a == NULL || top_a->klass() == NULL ||
5627 top_b == NULL || top_b->klass() == NULL) {
5628 // failed array check
5629 return false;
5630 }
5631
5632 Node* call;
5633 jvms()->set_should_reexecute(true);
5634
5635 Node* obja_adr = make_unsafe_address(obja, aoffset);
5636 Node* objb_adr = make_unsafe_address(objb, boffset);
5637
5638 call = make_runtime_call(RC_LEAF,
5639 OptoRuntime::vectorizedMismatch_Type(),
5640 stubAddr, stubName, TypePtr::BOTTOM,
5641 obja_adr, objb_adr, length, scale);
5642
5643 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5644 set_result(result);
5645 return true;
5646 }
5647
5648 /**
5649 * Calculate CRC32 for byte.
5650 * int java.util.zip.CRC32.update(int crc, int b)
5651 */
5652 bool LibraryCallKit::inline_updateCRC32() {
5653 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5654 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5655 // no receiver since it is static method
5656 Node* crc = argument(0); // type: int
5657 Node* b = argument(1); // type: int
5658
5659 /*
5660 * int c = ~ crc;
5661 * b = timesXtoThe32[(b ^ c) & 0xFF];
5662 * b = b ^ (c >>> 8);
5663 * crc = ~b;
5664 */
5665
5666 Node* M1 = intcon(-1);
5667 crc = _gvn.transform(new XorINode(crc, M1));
5668 Node* result = _gvn.transform(new XorINode(crc, b));
5669 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5670
5671 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5672 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5673 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5674 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5675
5676 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5677 result = _gvn.transform(new XorINode(crc, result));
5678 result = _gvn.transform(new XorINode(result, M1));
5679 set_result(result);
5680 return true;
5681 }
5682
5683 /**
5684 * Calculate CRC32 for byte[] array.
5685 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5686 */
5687 bool LibraryCallKit::inline_updateBytesCRC32() {
5688 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5689 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5690 // no receiver since it is static method
5691 Node* crc = argument(0); // type: int
5692 Node* src = argument(1); // type: oop
5693 Node* offset = argument(2); // type: int
5694 Node* length = argument(3); // type: int
5695
5696 const Type* src_type = src->Value(&_gvn);
5697 const TypeAryPtr* top_src = src_type->isa_aryptr();
5698 if (top_src == NULL || top_src->klass() == NULL) {
5699 // failed array check
5700 return false;
5701 }
5702
5703 // Figure out the size and type of the elements we will be copying.
5704 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5705 if (src_elem != T_BYTE) {
5706 return false;
5707 }
5708
5709 // 'src_start' points to src array + scaled offset
5710 Node* src_start = array_element_address(src, offset, src_elem);
5711
5712 // We assume that range check is done by caller.
5713 // TODO: generate range check (offset+length < src.length) in debug VM.
5714
5715 // Call the stub.
5716 address stubAddr = StubRoutines::updateBytesCRC32();
5717 const char *stubName = "updateBytesCRC32";
5718
5719 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5720 stubAddr, stubName, TypePtr::BOTTOM,
5721 crc, src_start, length);
5722 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5723 set_result(result);
5724 return true;
5725 }
5726
5727 /**
5728 * Calculate CRC32 for ByteBuffer.
5729 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5730 */
5731 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5732 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5733 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5734 // no receiver since it is static method
5735 Node* crc = argument(0); // type: int
5736 Node* src = argument(1); // type: long
5737 Node* offset = argument(3); // type: int
5738 Node* length = argument(4); // type: int
5739
5740 src = ConvL2X(src); // adjust Java long to machine word
5741 Node* base = _gvn.transform(new CastX2PNode(src));
5742 offset = ConvI2X(offset);
5743
5744 // 'src_start' points to src array + scaled offset
5745 Node* src_start = basic_plus_adr(top(), base, offset);
5746
5747 // Call the stub.
5748 address stubAddr = StubRoutines::updateBytesCRC32();
5749 const char *stubName = "updateBytesCRC32";
5750
5751 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5752 stubAddr, stubName, TypePtr::BOTTOM,
5753 crc, src_start, length);
5754 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5755 set_result(result);
5756 return true;
5757 }
5758
5759 //------------------------------get_table_from_crc32c_class-----------------------
5760 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5761 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5762 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5763
5764 return table;
5765 }
5766
5767 //------------------------------inline_updateBytesCRC32C-----------------------
5768 //
5769 // Calculate CRC32C for byte[] array.
5770 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5771 //
5772 bool LibraryCallKit::inline_updateBytesCRC32C() {
5773 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5774 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5775 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5776 // no receiver since it is a static method
5777 Node* crc = argument(0); // type: int
5778 Node* src = argument(1); // type: oop
5779 Node* offset = argument(2); // type: int
5780 Node* end = argument(3); // type: int
5781
5782 Node* length = _gvn.transform(new SubINode(end, offset));
5783
5784 const Type* src_type = src->Value(&_gvn);
5785 const TypeAryPtr* top_src = src_type->isa_aryptr();
5786 if (top_src == NULL || top_src->klass() == NULL) {
5787 // failed array check
5788 return false;
5789 }
5790
5791 // Figure out the size and type of the elements we will be copying.
5792 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5793 if (src_elem != T_BYTE) {
5794 return false;
5795 }
5796
5797 // 'src_start' points to src array + scaled offset
5798 Node* src_start = array_element_address(src, offset, src_elem);
5799
5800 // static final int[] byteTable in class CRC32C
5801 Node* table = get_table_from_crc32c_class(callee()->holder());
5802 Node* table_start = array_element_address(table, intcon(0), T_INT);
5803
5804 // We assume that range check is done by caller.
5805 // TODO: generate range check (offset+length < src.length) in debug VM.
5806
5807 // Call the stub.
5808 address stubAddr = StubRoutines::updateBytesCRC32C();
5809 const char *stubName = "updateBytesCRC32C";
5810
5811 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5812 stubAddr, stubName, TypePtr::BOTTOM,
5813 crc, src_start, length, table_start);
5814 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5815 set_result(result);
5816 return true;
5817 }
5818
5819 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5820 //
5821 // Calculate CRC32C for DirectByteBuffer.
5822 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5823 //
5824 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5825 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5826 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5827 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5828 // no receiver since it is a static method
5829 Node* crc = argument(0); // type: int
5830 Node* src = argument(1); // type: long
5831 Node* offset = argument(3); // type: int
5832 Node* end = argument(4); // type: int
5833
5834 Node* length = _gvn.transform(new SubINode(end, offset));
5835
5836 src = ConvL2X(src); // adjust Java long to machine word
5837 Node* base = _gvn.transform(new CastX2PNode(src));
5838 offset = ConvI2X(offset);
5839
5840 // 'src_start' points to src array + scaled offset
5841 Node* src_start = basic_plus_adr(top(), base, offset);
5842
5843 // static final int[] byteTable in class CRC32C
5844 Node* table = get_table_from_crc32c_class(callee()->holder());
5845 Node* table_start = array_element_address(table, intcon(0), T_INT);
5846
5847 // Call the stub.
5848 address stubAddr = StubRoutines::updateBytesCRC32C();
5849 const char *stubName = "updateBytesCRC32C";
5850
5851 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5852 stubAddr, stubName, TypePtr::BOTTOM,
5853 crc, src_start, length, table_start);
5854 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5855 set_result(result);
5856 return true;
5857 }
5858
5859 //------------------------------inline_updateBytesAdler32----------------------
5860 //
5861 // Calculate Adler32 checksum for byte[] array.
5862 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5863 //
5864 bool LibraryCallKit::inline_updateBytesAdler32() {
5865 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5866 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5867 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5868 // no receiver since it is static method
5869 Node* crc = argument(0); // type: int
5870 Node* src = argument(1); // type: oop
5871 Node* offset = argument(2); // type: int
5872 Node* length = argument(3); // type: int
5873
5874 const Type* src_type = src->Value(&_gvn);
5875 const TypeAryPtr* top_src = src_type->isa_aryptr();
5876 if (top_src == NULL || top_src->klass() == NULL) {
5877 // failed array check
5878 return false;
5879 }
5880
5881 // Figure out the size and type of the elements we will be copying.
5882 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5883 if (src_elem != T_BYTE) {
5884 return false;
5885 }
5886
5887 // 'src_start' points to src array + scaled offset
5888 Node* src_start = array_element_address(src, offset, src_elem);
5889
5890 // We assume that range check is done by caller.
5891 // TODO: generate range check (offset+length < src.length) in debug VM.
5892
5893 // Call the stub.
5894 address stubAddr = StubRoutines::updateBytesAdler32();
5895 const char *stubName = "updateBytesAdler32";
5896
5897 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5898 stubAddr, stubName, TypePtr::BOTTOM,
5899 crc, src_start, length);
5900 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5901 set_result(result);
5902 return true;
5903 }
5904
5905 //------------------------------inline_updateByteBufferAdler32---------------
5906 //
5907 // Calculate Adler32 checksum for DirectByteBuffer.
5908 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5909 //
5910 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5911 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5912 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5913 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5914 // no receiver since it is static method
5915 Node* crc = argument(0); // type: int
5916 Node* src = argument(1); // type: long
5917 Node* offset = argument(3); // type: int
5918 Node* length = argument(4); // type: int
5919
5920 src = ConvL2X(src); // adjust Java long to machine word
5921 Node* base = _gvn.transform(new CastX2PNode(src));
5922 offset = ConvI2X(offset);
5923
5924 // 'src_start' points to src array + scaled offset
5925 Node* src_start = basic_plus_adr(top(), base, offset);
5926
5927 // Call the stub.
5928 address stubAddr = StubRoutines::updateBytesAdler32();
5929 const char *stubName = "updateBytesAdler32";
5930
5931 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5932 stubAddr, stubName, TypePtr::BOTTOM,
5933 crc, src_start, length);
5934
5935 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5936 set_result(result);
5937 return true;
5938 }
5939
5940 //----------------------------inline_reference_get----------------------------
5941 // public T java.lang.ref.Reference.get();
5942 bool LibraryCallKit::inline_reference_get() {
5943 const int referent_offset = java_lang_ref_Reference::referent_offset;
5944 guarantee(referent_offset > 0, "should have already been set");
5945
5946 // Get the argument:
5947 Node* reference_obj = null_check_receiver();
5948 if (stopped()) return true;
5949
5950 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5951
5952 ciInstanceKlass* klass = env()->Object_klass();
5953 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5954
5955 Node* no_ctrl = NULL;
5956 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5957
5958 // Use the pre-barrier to record the value in the referent field
5959 pre_barrier(false /* do_load */,
5960 control(),
5961 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5962 result /* pre_val */,
5963 T_OBJECT);
5964
5965 // Add memory barrier to prevent commoning reads from this field
5966 // across safepoint since GC can change its value.
5967 insert_mem_bar(Op_MemBarCPUOrder);
5968
5969 set_result(result);
5970 return true;
5971 }
5972
5973
5974 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5975 bool is_exact=true, bool is_static=false,
5976 ciInstanceKlass * fromKls=NULL) {
5977 if (fromKls == NULL) {
5978 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5979 assert(tinst != NULL, "obj is null");
5980 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5981 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5982 fromKls = tinst->klass()->as_instance_klass();
5983 } else {
5984 assert(is_static, "only for static field access");
5985 }
5986 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5987 ciSymbol::make(fieldTypeString),
5988 is_static);
5989
5990 assert (field != NULL, "undefined field");
5991 if (field == NULL) return (Node *) NULL;
5992
5993 if (is_static) {
5994 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5995 fromObj = makecon(tip);
5996 }
5997
5998 // Next code copied from Parse::do_get_xxx():
5999
6000 // Compute address and memory type.
6001 int offset = field->offset_in_bytes();
6002 bool is_vol = field->is_volatile();
6003 ciType* field_klass = field->type();
6004 assert(field_klass->is_loaded(), "should be loaded");
6005 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6006 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6007 BasicType bt = field->layout_type();
6008
6009 // Build the resultant type of the load
6010 const Type *type;
6011 if (bt == T_OBJECT) {
6012 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6013 } else {
6014 type = Type::get_const_basic_type(bt);
6015 }
6016
6017 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
6018 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
6019 }
6020 // Build the load.
6021 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
6022 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
6023 // If reference is volatile, prevent following memory ops from
6024 // floating up past the volatile read. Also prevents commoning
6025 // another volatile read.
6026 if (is_vol) {
6027 // Memory barrier includes bogus read of value to force load BEFORE membar
6028 insert_mem_bar(Op_MemBarAcquire, loadedField);
6029 }
6030 return loadedField;
6031 }
6032
6033 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6034 bool is_exact = true, bool is_static = false,
6035 ciInstanceKlass * fromKls = NULL) {
6036 if (fromKls == NULL) {
6037 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6038 assert(tinst != NULL, "obj is null");
6039 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6040 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6041 fromKls = tinst->klass()->as_instance_klass();
6042 }
6043 else {
6044 assert(is_static, "only for static field access");
6045 }
6046 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6047 ciSymbol::make(fieldTypeString),
6048 is_static);
6049
6050 assert(field != NULL, "undefined field");
6051 assert(!field->is_volatile(), "not defined for volatile fields");
6052
6053 if (is_static) {
6054 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6055 fromObj = makecon(tip);
6056 }
6057
6058 // Next code copied from Parse::do_get_xxx():
6059
6060 // Compute address and memory type.
6061 int offset = field->offset_in_bytes();
6062 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6063
6064 return adr;
6065 }
6066
6067 //------------------------------inline_aescrypt_Block-----------------------
6068 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6069 address stubAddr = NULL;
6070 const char *stubName;
6071 assert(UseAES, "need AES instruction support");
6072
6073 switch(id) {
6074 case vmIntrinsics::_aescrypt_encryptBlock:
6075 stubAddr = StubRoutines::aescrypt_encryptBlock();
6076 stubName = "aescrypt_encryptBlock";
6077 break;
6078 case vmIntrinsics::_aescrypt_decryptBlock:
6079 stubAddr = StubRoutines::aescrypt_decryptBlock();
6080 stubName = "aescrypt_decryptBlock";
6081 break;
6082 }
6083 if (stubAddr == NULL) return false;
6084
6085 Node* aescrypt_object = argument(0);
6086 Node* src = argument(1);
6087 Node* src_offset = argument(2);
6088 Node* dest = argument(3);
6089 Node* dest_offset = argument(4);
6090
6091 // (1) src and dest are arrays.
6092 const Type* src_type = src->Value(&_gvn);
6093 const Type* dest_type = dest->Value(&_gvn);
6094 const TypeAryPtr* top_src = src_type->isa_aryptr();
6095 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6096 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6097
6098 // for the quick and dirty code we will skip all the checks.
6099 // we are just trying to get the call to be generated.
6100 Node* src_start = src;
6101 Node* dest_start = dest;
6102 if (src_offset != NULL || dest_offset != NULL) {
6103 assert(src_offset != NULL && dest_offset != NULL, "");
6104 src_start = array_element_address(src, src_offset, T_BYTE);
6105 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6106 }
6107
6108 // now need to get the start of its expanded key array
6109 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6110 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6111 if (k_start == NULL) return false;
6112
6113 if (Matcher::pass_original_key_for_aes()) {
6114 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6115 // compatibility issues between Java key expansion and SPARC crypto instructions
6116 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6117 if (original_k_start == NULL) return false;
6118
6119 // Call the stub.
6120 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6121 stubAddr, stubName, TypePtr::BOTTOM,
6122 src_start, dest_start, k_start, original_k_start);
6123 } else {
6124 // Call the stub.
6125 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6126 stubAddr, stubName, TypePtr::BOTTOM,
6127 src_start, dest_start, k_start);
6128 }
6129
6130 return true;
6131 }
6132
6133 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6134 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6135 address stubAddr = NULL;
6136 const char *stubName = NULL;
6137
6138 assert(UseAES, "need AES instruction support");
6139
6140 switch(id) {
6141 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6142 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6143 stubName = "cipherBlockChaining_encryptAESCrypt";
6144 break;
6145 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6146 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6147 stubName = "cipherBlockChaining_decryptAESCrypt";
6148 break;
6149 }
6150 if (stubAddr == NULL) return false;
6151
6152 Node* cipherBlockChaining_object = argument(0);
6153 Node* src = argument(1);
6154 Node* src_offset = argument(2);
6155 Node* len = argument(3);
6156 Node* dest = argument(4);
6157 Node* dest_offset = argument(5);
6158
6159 // (1) src and dest are arrays.
6160 const Type* src_type = src->Value(&_gvn);
6161 const Type* dest_type = dest->Value(&_gvn);
6162 const TypeAryPtr* top_src = src_type->isa_aryptr();
6163 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6164 assert (top_src != NULL && top_src->klass() != NULL
6165 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6166
6167 // checks are the responsibility of the caller
6168 Node* src_start = src;
6169 Node* dest_start = dest;
6170 if (src_offset != NULL || dest_offset != NULL) {
6171 assert(src_offset != NULL && dest_offset != NULL, "");
6172 src_start = array_element_address(src, src_offset, T_BYTE);
6173 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6174 }
6175
6176 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6177 // (because of the predicated logic executed earlier).
6178 // so we cast it here safely.
6179 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6180
6181 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6182 if (embeddedCipherObj == NULL) return false;
6183
6184 // cast it to what we know it will be at runtime
6185 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6186 assert(tinst != NULL, "CBC obj is null");
6187 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6188 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6189 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6190
6191 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6192 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6193 const TypeOopPtr* xtype = aklass->as_instance_type();
6194 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6195 aescrypt_object = _gvn.transform(aescrypt_object);
6196
6197 // we need to get the start of the aescrypt_object's expanded key array
6198 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6199 if (k_start == NULL) return false;
6200
6201 // similarly, get the start address of the r vector
6202 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6203 if (objRvec == NULL) return false;
6204 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6205
6206 Node* cbcCrypt;
6207 if (Matcher::pass_original_key_for_aes()) {
6208 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6209 // compatibility issues between Java key expansion and SPARC crypto instructions
6210 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6211 if (original_k_start == NULL) return false;
6212
6213 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6214 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6215 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6216 stubAddr, stubName, TypePtr::BOTTOM,
6217 src_start, dest_start, k_start, r_start, len, original_k_start);
6218 } else {
6219 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6220 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6221 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6222 stubAddr, stubName, TypePtr::BOTTOM,
6223 src_start, dest_start, k_start, r_start, len);
6224 }
6225
6226 // return cipher length (int)
6227 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6228 set_result(retvalue);
6229 return true;
6230 }
6231
6232 //------------------------------inline_counterMode_AESCrypt-----------------------
6233 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6234 assert(UseAES, "need AES instruction support");
6235 if (!UseAESCTRIntrinsics) return false;
6236
6237 address stubAddr = NULL;
6238 const char *stubName = NULL;
6239 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6240 stubAddr = StubRoutines::counterMode_AESCrypt();
6241 stubName = "counterMode_AESCrypt";
6242 }
6243 if (stubAddr == NULL) return false;
6244
6245 Node* counterMode_object = argument(0);
6246 Node* src = argument(1);
6247 Node* src_offset = argument(2);
6248 Node* len = argument(3);
6249 Node* dest = argument(4);
6250 Node* dest_offset = argument(5);
6251
6252 // (1) src and dest are arrays.
6253 const Type* src_type = src->Value(&_gvn);
6254 const Type* dest_type = dest->Value(&_gvn);
6255 const TypeAryPtr* top_src = src_type->isa_aryptr();
6256 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6257 assert(top_src != NULL && top_src->klass() != NULL &&
6258 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6259
6260 // checks are the responsibility of the caller
6261 Node* src_start = src;
6262 Node* dest_start = dest;
6263 if (src_offset != NULL || dest_offset != NULL) {
6264 assert(src_offset != NULL && dest_offset != NULL, "");
6265 src_start = array_element_address(src, src_offset, T_BYTE);
6266 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6267 }
6268
6269 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6270 // (because of the predicated logic executed earlier).
6271 // so we cast it here safely.
6272 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6273 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6274 if (embeddedCipherObj == NULL) return false;
6275 // cast it to what we know it will be at runtime
6276 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6277 assert(tinst != NULL, "CTR obj is null");
6278 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6279 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6280 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6281 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6282 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6283 const TypeOopPtr* xtype = aklass->as_instance_type();
6284 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6285 aescrypt_object = _gvn.transform(aescrypt_object);
6286 // we need to get the start of the aescrypt_object's expanded key array
6287 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6288 if (k_start == NULL) return false;
6289 // similarly, get the start address of the r vector
6290 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6291 if (obj_counter == NULL) return false;
6292 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6293
6294 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6295 if (saved_encCounter == NULL) return false;
6296 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6297 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6298
6299 Node* ctrCrypt;
6300 if (Matcher::pass_original_key_for_aes()) {
6301 // no SPARC version for AES/CTR intrinsics now.
6302 return false;
6303 }
6304 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6305 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6306 OptoRuntime::counterMode_aescrypt_Type(),
6307 stubAddr, stubName, TypePtr::BOTTOM,
6308 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6309
6310 // return cipher length (int)
6311 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6312 set_result(retvalue);
6313 return true;
6314 }
6315
6316 //------------------------------get_key_start_from_aescrypt_object-----------------------
6317 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6318 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6319 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6320 if (objAESCryptKey == NULL) return (Node *) NULL;
6321
6322 // now have the array, need to get the start address of the K array
6323 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6324 return k_start;
6325 }
6326
6327 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6328 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6329 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6330 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6331 if (objAESCryptKey == NULL) return (Node *) NULL;
6332
6333 // now have the array, need to get the start address of the lastKey array
6334 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6335 return original_k_start;
6336 }
6337
6338 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6339 // Return node representing slow path of predicate check.
6340 // the pseudo code we want to emulate with this predicate is:
6341 // for encryption:
6342 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6343 // for decryption:
6344 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6345 // note cipher==plain is more conservative than the original java code but that's OK
6346 //
6347 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6348 // The receiver was checked for NULL already.
6349 Node* objCBC = argument(0);
6350
6351 // Load embeddedCipher field of CipherBlockChaining object.
6352 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6353
6354 // get AESCrypt klass for instanceOf check
6355 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6356 // will have same classloader as CipherBlockChaining object
6357 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6358 assert(tinst != NULL, "CBCobj is null");
6359 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6360
6361 // we want to do an instanceof comparison against the AESCrypt class
6362 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6363 if (!klass_AESCrypt->is_loaded()) {
6364 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6365 Node* ctrl = control();
6366 set_control(top()); // no regular fast path
6367 return ctrl;
6368 }
6369 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6370
6371 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6372 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6373 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6374
6375 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6376
6377 // for encryption, we are done
6378 if (!decrypting)
6379 return instof_false; // even if it is NULL
6380
6381 // for decryption, we need to add a further check to avoid
6382 // taking the intrinsic path when cipher and plain are the same
6383 // see the original java code for why.
6384 RegionNode* region = new RegionNode(3);
6385 region->init_req(1, instof_false);
6386 Node* src = argument(1);
6387 Node* dest = argument(4);
6388 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6389 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6390 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6391 region->init_req(2, src_dest_conjoint);
6392
6393 record_for_igvn(region);
6394 return _gvn.transform(region);
6395 }
6396
6397 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6398 // Return node representing slow path of predicate check.
6399 // the pseudo code we want to emulate with this predicate is:
6400 // for encryption:
6401 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6402 // for decryption:
6403 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6404 // note cipher==plain is more conservative than the original java code but that's OK
6405 //
6406
6407 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6408 // The receiver was checked for NULL already.
6409 Node* objCTR = argument(0);
6410
6411 // Load embeddedCipher field of CipherBlockChaining object.
6412 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6413
6414 // get AESCrypt klass for instanceOf check
6415 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6416 // will have same classloader as CipherBlockChaining object
6417 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6418 assert(tinst != NULL, "CTRobj is null");
6419 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6420
6421 // we want to do an instanceof comparison against the AESCrypt class
6422 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6423 if (!klass_AESCrypt->is_loaded()) {
6424 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6425 Node* ctrl = control();
6426 set_control(top()); // no regular fast path
6427 return ctrl;
6428 }
6429
6430 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6431 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6432 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6433 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6434 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6435
6436 return instof_false; // even if it is NULL
6437 }
6438
6439 //------------------------------inline_ghash_processBlocks
6440 bool LibraryCallKit::inline_ghash_processBlocks() {
6441 address stubAddr;
6442 const char *stubName;
6443 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6444
6445 stubAddr = StubRoutines::ghash_processBlocks();
6446 stubName = "ghash_processBlocks";
6447
6448 Node* data = argument(0);
6449 Node* offset = argument(1);
6450 Node* len = argument(2);
6451 Node* state = argument(3);
6452 Node* subkeyH = argument(4);
6453
6454 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6455 assert(state_start, "state is NULL");
6456 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6457 assert(subkeyH_start, "subkeyH is NULL");
6458 Node* data_start = array_element_address(data, offset, T_BYTE);
6459 assert(data_start, "data is NULL");
6460
6461 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6462 OptoRuntime::ghash_processBlocks_Type(),
6463 stubAddr, stubName, TypePtr::BOTTOM,
6464 state_start, subkeyH_start, data_start, len);
6465 return true;
6466 }
6467
6468 //------------------------------inline_sha_implCompress-----------------------
6469 //
6470 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6471 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6472 //
6473 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6474 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6475 //
6476 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6477 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6478 //
6479 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6480 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6481
6482 Node* sha_obj = argument(0);
6483 Node* src = argument(1); // type oop
6484 Node* ofs = argument(2); // type int
6485
6486 const Type* src_type = src->Value(&_gvn);
6487 const TypeAryPtr* top_src = src_type->isa_aryptr();
6488 if (top_src == NULL || top_src->klass() == NULL) {
6489 // failed array check
6490 return false;
6491 }
6492 // Figure out the size and type of the elements we will be copying.
6493 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6494 if (src_elem != T_BYTE) {
6495 return false;
6496 }
6497 // 'src_start' points to src array + offset
6498 Node* src_start = array_element_address(src, ofs, src_elem);
6499 Node* state = NULL;
6500 address stubAddr;
6501 const char *stubName;
6502
6503 switch(id) {
6504 case vmIntrinsics::_sha_implCompress:
6505 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6506 state = get_state_from_sha_object(sha_obj);
6507 stubAddr = StubRoutines::sha1_implCompress();
6508 stubName = "sha1_implCompress";
6509 break;
6510 case vmIntrinsics::_sha2_implCompress:
6511 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6512 state = get_state_from_sha_object(sha_obj);
6513 stubAddr = StubRoutines::sha256_implCompress();
6514 stubName = "sha256_implCompress";
6515 break;
6516 case vmIntrinsics::_sha5_implCompress:
6517 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6518 state = get_state_from_sha5_object(sha_obj);
6519 stubAddr = StubRoutines::sha512_implCompress();
6520 stubName = "sha512_implCompress";
6521 break;
6522 default:
6523 fatal_unexpected_iid(id);
6524 return false;
6525 }
6526 if (state == NULL) return false;
6527
6528 // Call the stub.
6529 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6530 stubAddr, stubName, TypePtr::BOTTOM,
6531 src_start, state);
6532
6533 return true;
6534 }
6535
6536 //------------------------------inline_digestBase_implCompressMB-----------------------
6537 //
6538 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6539 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6540 //
6541 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6542 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6543 "need SHA1/SHA256/SHA512 instruction support");
6544 assert((uint)predicate < 3, "sanity");
6545 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6546
6547 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6548 Node* src = argument(1); // byte[] array
6549 Node* ofs = argument(2); // type int
6550 Node* limit = argument(3); // type int
6551
6552 const Type* src_type = src->Value(&_gvn);
6553 const TypeAryPtr* top_src = src_type->isa_aryptr();
6554 if (top_src == NULL || top_src->klass() == NULL) {
6555 // failed array check
6556 return false;
6557 }
6558 // Figure out the size and type of the elements we will be copying.
6559 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6560 if (src_elem != T_BYTE) {
6561 return false;
6562 }
6563 // 'src_start' points to src array + offset
6564 Node* src_start = array_element_address(src, ofs, src_elem);
6565
6566 const char* klass_SHA_name = NULL;
6567 const char* stub_name = NULL;
6568 address stub_addr = NULL;
6569 bool long_state = false;
6570
6571 switch (predicate) {
6572 case 0:
6573 if (UseSHA1Intrinsics) {
6574 klass_SHA_name = "sun/security/provider/SHA";
6575 stub_name = "sha1_implCompressMB";
6576 stub_addr = StubRoutines::sha1_implCompressMB();
6577 }
6578 break;
6579 case 1:
6580 if (UseSHA256Intrinsics) {
6581 klass_SHA_name = "sun/security/provider/SHA2";
6582 stub_name = "sha256_implCompressMB";
6583 stub_addr = StubRoutines::sha256_implCompressMB();
6584 }
6585 break;
6586 case 2:
6587 if (UseSHA512Intrinsics) {
6588 klass_SHA_name = "sun/security/provider/SHA5";
6589 stub_name = "sha512_implCompressMB";
6590 stub_addr = StubRoutines::sha512_implCompressMB();
6591 long_state = true;
6592 }
6593 break;
6594 default:
6595 fatal("unknown SHA intrinsic predicate: %d", predicate);
6596 }
6597 if (klass_SHA_name != NULL) {
6598 // get DigestBase klass to lookup for SHA klass
6599 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6600 assert(tinst != NULL, "digestBase_obj is not instance???");
6601 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6602
6603 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6604 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6605 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6606 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6607 }
6608 return false;
6609 }
6610 //------------------------------inline_sha_implCompressMB-----------------------
6611 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6612 bool long_state, address stubAddr, const char *stubName,
6613 Node* src_start, Node* ofs, Node* limit) {
6614 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6615 const TypeOopPtr* xtype = aklass->as_instance_type();
6616 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6617 sha_obj = _gvn.transform(sha_obj);
6618
6619 Node* state;
6620 if (long_state) {
6621 state = get_state_from_sha5_object(sha_obj);
6622 } else {
6623 state = get_state_from_sha_object(sha_obj);
6624 }
6625 if (state == NULL) return false;
6626
6627 // Call the stub.
6628 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6629 OptoRuntime::digestBase_implCompressMB_Type(),
6630 stubAddr, stubName, TypePtr::BOTTOM,
6631 src_start, state, ofs, limit);
6632 // return ofs (int)
6633 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6634 set_result(result);
6635
6636 return true;
6637 }
6638
6639 //------------------------------get_state_from_sha_object-----------------------
6640 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6641 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6642 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6643 if (sha_state == NULL) return (Node *) NULL;
6644
6645 // now have the array, need to get the start address of the state array
6646 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6647 return state;
6648 }
6649
6650 //------------------------------get_state_from_sha5_object-----------------------
6651 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6652 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6653 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6654 if (sha_state == NULL) return (Node *) NULL;
6655
6656 // now have the array, need to get the start address of the state array
6657 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6658 return state;
6659 }
6660
6661 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6662 // Return node representing slow path of predicate check.
6663 // the pseudo code we want to emulate with this predicate is:
6664 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6665 //
6666 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6667 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6668 "need SHA1/SHA256/SHA512 instruction support");
6669 assert((uint)predicate < 3, "sanity");
6670
6671 // The receiver was checked for NULL already.
6672 Node* digestBaseObj = argument(0);
6673
6674 // get DigestBase klass for instanceOf check
6675 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6676 assert(tinst != NULL, "digestBaseObj is null");
6677 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6678
6679 const char* klass_SHA_name = NULL;
6680 switch (predicate) {
6681 case 0:
6682 if (UseSHA1Intrinsics) {
6683 // we want to do an instanceof comparison against the SHA class
6684 klass_SHA_name = "sun/security/provider/SHA";
6685 }
6686 break;
6687 case 1:
6688 if (UseSHA256Intrinsics) {
6689 // we want to do an instanceof comparison against the SHA2 class
6690 klass_SHA_name = "sun/security/provider/SHA2";
6691 }
6692 break;
6693 case 2:
6694 if (UseSHA512Intrinsics) {
6695 // we want to do an instanceof comparison against the SHA5 class
6696 klass_SHA_name = "sun/security/provider/SHA5";
6697 }
6698 break;
6699 default:
6700 fatal("unknown SHA intrinsic predicate: %d", predicate);
6701 }
6702
6703 ciKlass* klass_SHA = NULL;
6704 if (klass_SHA_name != NULL) {
6705 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6706 }
6707 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6708 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6709 Node* ctrl = control();
6710 set_control(top()); // no intrinsic path
6711 return ctrl;
6712 }
6713 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6714
6715 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6716 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6717 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6718 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6719
6720 return instof_false; // even if it is NULL
6721 }
6722
6723 bool LibraryCallKit::inline_profileBoolean() {
6724 Node* counts = argument(1);
6725 const TypeAryPtr* ary = NULL;
6726 ciArray* aobj = NULL;
6727 if (counts->is_Con()
6728 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6729 && (aobj = ary->const_oop()->as_array()) != NULL
6730 && (aobj->length() == 2)) {
6731 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6732 jint false_cnt = aobj->element_value(0).as_int();
6733 jint true_cnt = aobj->element_value(1).as_int();
6734
6735 if (C->log() != NULL) {
6736 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6737 false_cnt, true_cnt);
6738 }
6739
6740 if (false_cnt + true_cnt == 0) {
6741 // According to profile, never executed.
6742 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6743 Deoptimization::Action_reinterpret);
6744 return true;
6745 }
6746
6747 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6748 // is a number of each value occurrences.
6749 Node* result = argument(0);
6750 if (false_cnt == 0 || true_cnt == 0) {
6751 // According to profile, one value has been never seen.
6752 int expected_val = (false_cnt == 0) ? 1 : 0;
6753
6754 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6755 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6756
6757 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6758 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6759 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6760
6761 { // Slow path: uncommon trap for never seen value and then reexecute
6762 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6763 // the value has been seen at least once.
6764 PreserveJVMState pjvms(this);
6765 PreserveReexecuteState preexecs(this);
6766 jvms()->set_should_reexecute(true);
6767
6768 set_control(slow_path);
6769 set_i_o(i_o());
6770
6771 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6772 Deoptimization::Action_reinterpret);
6773 }
6774 // The guard for never seen value enables sharpening of the result and
6775 // returning a constant. It allows to eliminate branches on the same value
6776 // later on.
6777 set_control(fast_path);
6778 result = intcon(expected_val);
6779 }
6780 // Stop profiling.
6781 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6782 // By replacing method body with profile data (represented as ProfileBooleanNode
6783 // on IR level) we effectively disable profiling.
6784 // It enables full speed execution once optimized code is generated.
6785 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6786 C->record_for_igvn(profile);
6787 set_result(profile);
6788 return true;
6789 } else {
6790 // Continue profiling.
6791 // Profile data isn't available at the moment. So, execute method's bytecode version.
6792 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6793 // is compiled and counters aren't available since corresponding MethodHandle
6794 // isn't a compile-time constant.
6795 return false;
6796 }
6797 }
6798
6799 bool LibraryCallKit::inline_isCompileConstant() {
6800 Node* n = argument(0);
6801 set_result(n->is_Con() ? intcon(1) : intcon(0));
6802 return true;
6803 }
6804
6805 bool LibraryCallKit::inline_deoptimize() {
6806 assert(WhiteBoxAPI, "");
6807 PreserveReexecuteState preexecs(this);
6808 jvms()->set_should_reexecute(false);
6809 uncommon_trap(Deoptimization::Reason_intrinsic,
6810 Deoptimization::Action_none);
6811 return true;
6812 }