1 /*
2 * Copyright (c) 1999, 2018, 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 "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/callGenerator.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/intrinsicnode.hpp"
45 #include "opto/idealKit.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/movenode.hpp"
48 #include "opto/mulnode.hpp"
49 #include "opto/narrowptrnode.hpp"
50 #include "opto/opaquenode.hpp"
51 #include "opto/parse.hpp"
52 #include "opto/runtime.hpp"
53 #include "opto/rootnode.hpp"
54 #include "opto/subnode.hpp"
55 #include "prims/nativeLookup.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "utilities/macros.hpp"
60
61
62 class LibraryIntrinsic : public InlineCallGenerator {
63 // Extend the set of intrinsics known to the runtime:
64 public:
65 private:
66 bool _is_virtual;
67 bool _does_virtual_dispatch;
68 int8_t _predicates_count; // Intrinsic is predicated by several conditions
69 int8_t _last_predicate; // Last generated predicate
70 vmIntrinsics::ID _intrinsic_id;
71
72 public:
73 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
74 : InlineCallGenerator(m),
75 _is_virtual(is_virtual),
76 _does_virtual_dispatch(does_virtual_dispatch),
77 _predicates_count((int8_t)predicates_count),
78 _last_predicate((int8_t)-1),
79 _intrinsic_id(id)
80 {
81 }
82 virtual bool is_intrinsic() const { return true; }
83 virtual bool is_virtual() const { return _is_virtual; }
84 virtual bool is_predicated() const { return _predicates_count > 0; }
85 virtual int predicates_count() const { return _predicates_count; }
86 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
87 virtual JVMState* generate(JVMState* jvms);
88 virtual Node* generate_predicate(JVMState* jvms, int predicate);
89 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
90 };
91
92
93 // Local helper class for LibraryIntrinsic:
94 class LibraryCallKit : public GraphKit {
95 private:
96 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
97 Node* _result; // the result node, if any
98 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
99
100 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
101
102 public:
103 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
104 : GraphKit(jvms),
105 _intrinsic(intrinsic),
106 _result(NULL)
107 {
108 // Check if this is a root compile. In that case we don't have a caller.
109 if (!jvms->has_method()) {
110 _reexecute_sp = sp();
111 } else {
112 // Find out how many arguments the interpreter needs when deoptimizing
113 // and save the stack pointer value so it can used by uncommon_trap.
114 // We find the argument count by looking at the declared signature.
115 bool ignored_will_link;
116 ciSignature* declared_signature = NULL;
117 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
118 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
119 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
120 }
121 }
122
123 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
124
125 ciMethod* caller() const { return jvms()->method(); }
126 int bci() const { return jvms()->bci(); }
127 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
128 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
129 ciMethod* callee() const { return _intrinsic->method(); }
130
131 bool try_to_inline(int predicate);
132 Node* try_to_predicate(int predicate);
133
134 void push_result() {
135 // Push the result onto the stack.
136 if (!stopped() && result() != NULL) {
137 BasicType bt = result()->bottom_type()->basic_type();
138 push_node(bt, result());
139 }
140 }
141
142 private:
143 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
144 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
145 }
146
147 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
148 void set_result(RegionNode* region, PhiNode* value);
149 Node* result() { return _result; }
150
151 virtual int reexecute_sp() { return _reexecute_sp; }
152
153 // Helper functions to inline natives
154 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
155 Node* generate_slow_guard(Node* test, RegionNode* region);
156 Node* generate_fair_guard(Node* test, RegionNode* region);
157 Node* generate_negative_guard(Node* index, RegionNode* region,
158 // resulting CastII of index:
159 Node* *pos_index = NULL);
160 Node* generate_limit_guard(Node* offset, Node* subseq_length,
161 Node* array_length,
162 RegionNode* region);
163 void generate_string_range_check(Node* array, Node* offset,
164 Node* length, bool char_count);
165 Node* generate_current_thread(Node* &tls_output);
166 Node* load_mirror_from_klass(Node* klass);
167 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path,
169 int offset);
170 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
171 RegionNode* region, int null_path) {
172 int offset = java_lang_Class::klass_offset_in_bytes();
173 return load_klass_from_mirror_common(mirror, never_see_null,
174 region, null_path,
175 offset);
176 }
177 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
178 RegionNode* region, int null_path) {
179 int offset = java_lang_Class::array_klass_offset_in_bytes();
180 return load_klass_from_mirror_common(mirror, never_see_null,
181 region, null_path,
182 offset);
183 }
184 Node* generate_access_flags_guard(Node* kls,
185 int modifier_mask, int modifier_bits,
186 RegionNode* region);
187 Node* generate_interface_guard(Node* kls, RegionNode* region);
188 Node* generate_array_guard(Node* kls, RegionNode* region) {
189 return generate_array_guard_common(kls, region, false, false);
190 }
191 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
192 return generate_array_guard_common(kls, region, false, true);
193 }
194 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
195 return generate_array_guard_common(kls, region, true, false);
196 }
197 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
198 return generate_array_guard_common(kls, region, true, true);
199 }
200 Node* generate_array_guard_common(Node* kls, RegionNode* region,
201 bool obj_array, bool not_array);
202 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
203 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
204 bool is_virtual = false, bool is_static = false);
205 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
206 return generate_method_call(method_id, false, true);
207 }
208 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
209 return generate_method_call(method_id, true, false);
210 }
211 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
212 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
213
214 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
215 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
218 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
219 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
220 bool inline_string_indexOfChar();
221 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
222 bool inline_string_toBytesU();
223 bool inline_string_getCharsU();
224 bool inline_string_copy(bool compress);
225 bool inline_string_char_access(bool is_store);
226 Node* round_double_node(Node* n);
227 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
228 bool inline_math_native(vmIntrinsics::ID id);
229 bool inline_math(vmIntrinsics::ID id);
230 template <typename OverflowOp>
231 bool inline_math_overflow(Node* arg1, Node* arg2);
232 void inline_math_mathExact(Node* math, Node* test);
233 bool inline_math_addExactI(bool is_increment);
234 bool inline_math_addExactL(bool is_increment);
235 bool inline_math_multiplyExactI();
236 bool inline_math_multiplyExactL();
237 bool inline_math_multiplyHigh();
238 bool inline_math_negateExactI();
239 bool inline_math_negateExactL();
240 bool inline_math_subtractExactI(bool is_decrement);
241 bool inline_math_subtractExactL(bool is_decrement);
242 bool inline_min_max(vmIntrinsics::ID id);
243 bool inline_notify(vmIntrinsics::ID id);
244 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
245 // This returns Type::AnyPtr, RawPtr, or OopPtr.
246 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
247 Node* make_unsafe_address(Node*& base, Node* offset, BasicType type = T_ILLEGAL, bool can_cast = false);
248
249 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
250 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
251 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
252 static bool klass_needs_init_guard(Node* kls);
253 bool inline_unsafe_allocate();
254 bool inline_unsafe_newArray(bool uninitialized);
255 bool inline_unsafe_copyMemory();
256 bool inline_native_currentThread();
257
258 bool inline_native_time_funcs(address method, const char* funcName);
259 #ifdef JFR_HAVE_INTRINSICS
260 bool inline_native_classID();
261 bool inline_native_getEventWriter();
262 #endif
263 bool inline_native_isInterrupted();
264 bool inline_native_Class_query(vmIntrinsics::ID id);
265 bool inline_native_subtype_check();
266 bool inline_native_getLength();
267 bool inline_array_copyOf(bool is_copyOfRange);
268 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
269 bool inline_preconditions_checkIndex();
270 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
271 bool inline_native_clone(bool is_virtual);
272 bool inline_native_Reflection_getCallerClass();
273 // Helper function for inlining native object hash method
274 bool inline_native_hashcode(bool is_virtual, bool is_static);
275 bool inline_native_getClass();
276
277 // Helper functions for inlining arraycopy
278 bool inline_arraycopy();
279 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
280 RegionNode* slow_region);
281 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
282 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
283 uint new_idx);
284
285 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
286 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
287 bool inline_unsafe_fence(vmIntrinsics::ID id);
288 bool inline_onspinwait();
289 bool inline_fp_conversions(vmIntrinsics::ID id);
290 bool inline_number_methods(vmIntrinsics::ID id);
291 bool inline_reference_get();
292 bool inline_Class_cast();
293 bool inline_aescrypt_Block(vmIntrinsics::ID id);
294 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
295 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
296 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
297 Node* inline_counterMode_AESCrypt_predicate();
298 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
299 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
300 bool inline_ghash_processBlocks();
301 bool inline_base64_encodeBlock();
302 bool inline_sha_implCompress(vmIntrinsics::ID id);
303 bool inline_digestBase_implCompressMB(int predicate);
304 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
305 bool long_state, address stubAddr, const char *stubName,
306 Node* src_start, Node* ofs, Node* limit);
307 Node* get_state_from_sha_object(Node *sha_object);
308 Node* get_state_from_sha5_object(Node *sha_object);
309 Node* inline_digestBase_implCompressMB_predicate(int predicate);
310 bool inline_encodeISOArray();
311 bool inline_updateCRC32();
312 bool inline_updateBytesCRC32();
313 bool inline_updateByteBufferCRC32();
314 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
315 bool inline_updateBytesCRC32C();
316 bool inline_updateDirectByteBufferCRC32C();
317 bool inline_updateBytesAdler32();
318 bool inline_updateByteBufferAdler32();
319 bool inline_multiplyToLen();
320 bool inline_hasNegatives();
321 bool inline_squareToLen();
322 bool inline_mulAdd();
323 bool inline_montgomeryMultiply();
324 bool inline_montgomerySquare();
325 bool inline_vectorizedMismatch();
326 bool inline_fma(vmIntrinsics::ID id);
327
328 bool inline_profileBoolean();
329 bool inline_isCompileConstant();
330 void clear_upper_avx() {
331 #ifdef X86
332 if (UseAVX >= 2) {
333 C->set_clear_upper_avx(true);
334 }
335 #endif
336 }
337 };
338
339 //---------------------------make_vm_intrinsic----------------------------
340 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
341 vmIntrinsics::ID id = m->intrinsic_id();
342 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
343
344 if (!m->is_loaded()) {
345 // Do not attempt to inline unloaded methods.
346 return NULL;
347 }
348
349 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
350 bool is_available = false;
351
352 {
353 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
354 // the compiler must transition to '_thread_in_vm' state because both
355 // methods access VM-internal data.
356 VM_ENTRY_MARK;
357 methodHandle mh(THREAD, m->get_Method());
358 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
359 !C->directive()->is_intrinsic_disabled(mh) &&
360 !vmIntrinsics::is_disabled_by_flags(mh);
361
362 }
363
364 if (is_available) {
365 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
366 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
367 return new LibraryIntrinsic(m, is_virtual,
368 vmIntrinsics::predicates_needed(id),
369 vmIntrinsics::does_virtual_dispatch(id),
370 (vmIntrinsics::ID) id);
371 } else {
372 return NULL;
373 }
374 }
375
376 //----------------------register_library_intrinsics-----------------------
377 // Initialize this file's data structures, for each Compile instance.
378 void Compile::register_library_intrinsics() {
379 // Nothing to do here.
380 }
381
382 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
383 LibraryCallKit kit(jvms, this);
384 Compile* C = kit.C;
385 int nodes = C->unique();
386 #ifndef PRODUCT
387 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
388 char buf[1000];
389 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
390 tty->print_cr("Intrinsic %s", str);
391 }
392 #endif
393 ciMethod* callee = kit.callee();
394 const int bci = kit.bci();
395
396 // Try to inline the intrinsic.
397 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
398 kit.try_to_inline(_last_predicate)) {
399 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
400 : "(intrinsic)";
401 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
402 if (C->print_intrinsics() || C->print_inlining()) {
403 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
404 }
405 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
406 if (C->log()) {
407 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
408 vmIntrinsics::name_at(intrinsic_id()),
409 (is_virtual() ? " virtual='1'" : ""),
410 C->unique() - nodes);
411 }
412 // Push the result from the inlined method onto the stack.
413 kit.push_result();
414 C->print_inlining_update(this);
415 return kit.transfer_exceptions_into_jvms();
416 }
417
418 // The intrinsic bailed out
419 if (jvms->has_method()) {
420 // Not a root compile.
421 const char* msg;
422 if (callee->intrinsic_candidate()) {
423 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
424 } else {
425 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
426 : "failed to inline (intrinsic), method not annotated";
427 }
428 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
429 if (C->print_intrinsics() || C->print_inlining()) {
430 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
431 }
432 } else {
433 // Root compile
434 ResourceMark rm;
435 stringStream msg_stream;
436 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
437 vmIntrinsics::name_at(intrinsic_id()),
438 is_virtual() ? " (virtual)" : "", bci);
439 const char *msg = msg_stream.as_string();
440 log_debug(jit, inlining)("%s", msg);
441 if (C->print_intrinsics() || C->print_inlining()) {
442 tty->print("%s", msg);
443 }
444 }
445 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
446 C->print_inlining_update(this);
447 return NULL;
448 }
449
450 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
451 LibraryCallKit kit(jvms, this);
452 Compile* C = kit.C;
453 int nodes = C->unique();
454 _last_predicate = predicate;
455 #ifndef PRODUCT
456 assert(is_predicated() && predicate < predicates_count(), "sanity");
457 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
458 char buf[1000];
459 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
460 tty->print_cr("Predicate for intrinsic %s", str);
461 }
462 #endif
463 ciMethod* callee = kit.callee();
464 const int bci = kit.bci();
465
466 Node* slow_ctl = kit.try_to_predicate(predicate);
467 if (!kit.failing()) {
468 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
469 : "(intrinsic, predicate)";
470 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
471 if (C->print_intrinsics() || C->print_inlining()) {
472 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
473 }
474 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
475 if (C->log()) {
476 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
477 vmIntrinsics::name_at(intrinsic_id()),
478 (is_virtual() ? " virtual='1'" : ""),
479 C->unique() - nodes);
480 }
481 return slow_ctl; // Could be NULL if the check folds.
482 }
483
484 // The intrinsic bailed out
485 if (jvms->has_method()) {
486 // Not a root compile.
487 const char* msg = "failed to generate predicate for intrinsic";
488 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
489 if (C->print_intrinsics() || C->print_inlining()) {
490 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
491 }
492 } else {
493 // Root compile
494 ResourceMark rm;
495 stringStream msg_stream;
496 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
497 vmIntrinsics::name_at(intrinsic_id()),
498 is_virtual() ? " (virtual)" : "", bci);
499 const char *msg = msg_stream.as_string();
500 log_debug(jit, inlining)("%s", msg);
501 if (C->print_intrinsics() || C->print_inlining()) {
502 C->print_inlining_stream()->print("%s", msg);
503 }
504 }
505 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
506 return NULL;
507 }
508
509 bool LibraryCallKit::try_to_inline(int predicate) {
510 // Handle symbolic names for otherwise undistinguished boolean switches:
511 const bool is_store = true;
512 const bool is_compress = true;
513 const bool is_static = true;
514 const bool is_volatile = true;
515
516 if (!jvms()->has_method()) {
517 // Root JVMState has a null method.
518 assert(map()->memory()->Opcode() == Op_Parm, "");
519 // Insert the memory aliasing node
520 set_all_memory(reset_memory());
521 }
522 assert(merged_memory(), "");
523
524
525 switch (intrinsic_id()) {
526 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
527 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
528 case vmIntrinsics::_getClass: return inline_native_getClass();
529
530 case vmIntrinsics::_dsin:
531 case vmIntrinsics::_dcos:
532 case vmIntrinsics::_dtan:
533 case vmIntrinsics::_dabs:
534 case vmIntrinsics::_datan2:
535 case vmIntrinsics::_dsqrt:
536 case vmIntrinsics::_dexp:
537 case vmIntrinsics::_dlog:
538 case vmIntrinsics::_dlog10:
539 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
540
541 case vmIntrinsics::_min:
542 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
543
544 case vmIntrinsics::_notify:
545 case vmIntrinsics::_notifyAll:
546 if (ObjectMonitor::Knob_InlineNotify) {
547 return inline_notify(intrinsic_id());
548 }
549 return false;
550
551 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
552 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
553 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
554 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
555 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
556 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
557 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
558 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
559 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
560 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
561 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
562 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
563 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
564
565 case vmIntrinsics::_arraycopy: return inline_arraycopy();
566
567 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
568 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
569 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
570 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
571
572 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
573 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
574 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
575 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
576 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
577 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
578 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
579
580 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
581 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
582
583 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
584 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
585 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
586 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
587
588 case vmIntrinsics::_compressStringC:
589 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
590 case vmIntrinsics::_inflateStringC:
591 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
592
593 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
594 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
595 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
596 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
597 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
598 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
599 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
600 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
601 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
602
603 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
604 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
605 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
606 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
607 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
608 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
609 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
610 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
611 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
612
613 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
614 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
615 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
616 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
617 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
618 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
619 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
620 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
621 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
622
623 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
624 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
625 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
626 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
627 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
628 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
629 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
630 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
631 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
632
633 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
634 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
635 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
636 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
637
638 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
639 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
640 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
641 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
642
643 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
644 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
645 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
646 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
647 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
648 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
649 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
650 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
651 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
652
653 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
654 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
655 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
656 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
657 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
658 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
659 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
660 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
661 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
662
663 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
664 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
665 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
666 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
667 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
668 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
669 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
670 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
671 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
672
673 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
674 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
675 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
676 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
677 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
678 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
679 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
680 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
681 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
682
683 case vmIntrinsics::_compareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
684 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
685 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
686 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
687 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
688
689 case vmIntrinsics::_weakCompareAndSetObjectPlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
690 case vmIntrinsics::_weakCompareAndSetObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
691 case vmIntrinsics::_weakCompareAndSetObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
692 case vmIntrinsics::_weakCompareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
693 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
694 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
695 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
696 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
697 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
698 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
699 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
700 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
701 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
702 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
703 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
704 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
705 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
706 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
707 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
708 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
709
710 case vmIntrinsics::_compareAndExchangeObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
711 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
712 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
713 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
714 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
715 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
716 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
717 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
718 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
719 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
720 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
721 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
722 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
723 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
724 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
725
726 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
727 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
728 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
729 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
730
731 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
732 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
733 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
734 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
735 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
736
737 case vmIntrinsics::_loadFence:
738 case vmIntrinsics::_storeFence:
739 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
740
741 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
742
743 case vmIntrinsics::_currentThread: return inline_native_currentThread();
744 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
745
746 #ifdef JFR_HAVE_INTRINSICS
747 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
748 case vmIntrinsics::_getClassId: return inline_native_classID();
749 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
750 #endif
751 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
752 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
753 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
754 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
755 case vmIntrinsics::_getLength: return inline_native_getLength();
756 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
757 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
758 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
759 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
760 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
761 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
762
763 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
764 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
765
766 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
767
768 case vmIntrinsics::_isInstance:
769 case vmIntrinsics::_getModifiers:
770 case vmIntrinsics::_isInterface:
771 case vmIntrinsics::_isArray:
772 case vmIntrinsics::_isPrimitive:
773 case vmIntrinsics::_getSuperclass:
774 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
775
776 case vmIntrinsics::_floatToRawIntBits:
777 case vmIntrinsics::_floatToIntBits:
778 case vmIntrinsics::_intBitsToFloat:
779 case vmIntrinsics::_doubleToRawLongBits:
780 case vmIntrinsics::_doubleToLongBits:
781 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
782
783 case vmIntrinsics::_numberOfLeadingZeros_i:
784 case vmIntrinsics::_numberOfLeadingZeros_l:
785 case vmIntrinsics::_numberOfTrailingZeros_i:
786 case vmIntrinsics::_numberOfTrailingZeros_l:
787 case vmIntrinsics::_bitCount_i:
788 case vmIntrinsics::_bitCount_l:
789 case vmIntrinsics::_reverseBytes_i:
790 case vmIntrinsics::_reverseBytes_l:
791 case vmIntrinsics::_reverseBytes_s:
792 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
793
794 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
795
796 case vmIntrinsics::_Reference_get: return inline_reference_get();
797
798 case vmIntrinsics::_Class_cast: return inline_Class_cast();
799
800 case vmIntrinsics::_aescrypt_encryptBlock:
801 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
802
803 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
804 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
805 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
806
807 case vmIntrinsics::_counterMode_AESCrypt:
808 return inline_counterMode_AESCrypt(intrinsic_id());
809
810 case vmIntrinsics::_sha_implCompress:
811 case vmIntrinsics::_sha2_implCompress:
812 case vmIntrinsics::_sha5_implCompress:
813 return inline_sha_implCompress(intrinsic_id());
814
815 case vmIntrinsics::_digestBase_implCompressMB:
816 return inline_digestBase_implCompressMB(predicate);
817
818 case vmIntrinsics::_multiplyToLen:
819 return inline_multiplyToLen();
820
821 case vmIntrinsics::_squareToLen:
822 return inline_squareToLen();
823
824 case vmIntrinsics::_mulAdd:
825 return inline_mulAdd();
826
827 case vmIntrinsics::_montgomeryMultiply:
828 return inline_montgomeryMultiply();
829 case vmIntrinsics::_montgomerySquare:
830 return inline_montgomerySquare();
831
832 case vmIntrinsics::_vectorizedMismatch:
833 return inline_vectorizedMismatch();
834
835 case vmIntrinsics::_ghash_processBlocks:
836 return inline_ghash_processBlocks();
837 case vmIntrinsics::_base64_encodeBlock:
838 return inline_base64_encodeBlock();
839
840 case vmIntrinsics::_encodeISOArray:
841 case vmIntrinsics::_encodeByteISOArray:
842 return inline_encodeISOArray();
843
844 case vmIntrinsics::_updateCRC32:
845 return inline_updateCRC32();
846 case vmIntrinsics::_updateBytesCRC32:
847 return inline_updateBytesCRC32();
848 case vmIntrinsics::_updateByteBufferCRC32:
849 return inline_updateByteBufferCRC32();
850
851 case vmIntrinsics::_updateBytesCRC32C:
852 return inline_updateBytesCRC32C();
853 case vmIntrinsics::_updateDirectByteBufferCRC32C:
854 return inline_updateDirectByteBufferCRC32C();
855
856 case vmIntrinsics::_updateBytesAdler32:
857 return inline_updateBytesAdler32();
858 case vmIntrinsics::_updateByteBufferAdler32:
859 return inline_updateByteBufferAdler32();
860
861 case vmIntrinsics::_profileBoolean:
862 return inline_profileBoolean();
863 case vmIntrinsics::_isCompileConstant:
864 return inline_isCompileConstant();
865
866 case vmIntrinsics::_hasNegatives:
867 return inline_hasNegatives();
868
869 case vmIntrinsics::_fmaD:
870 case vmIntrinsics::_fmaF:
871 return inline_fma(intrinsic_id());
872
873 default:
874 // If you get here, it may be that someone has added a new intrinsic
875 // to the list in vmSymbols.hpp without implementing it here.
876 #ifndef PRODUCT
877 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
878 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
879 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
880 }
881 #endif
882 return false;
883 }
884 }
885
886 Node* LibraryCallKit::try_to_predicate(int predicate) {
887 if (!jvms()->has_method()) {
888 // Root JVMState has a null method.
889 assert(map()->memory()->Opcode() == Op_Parm, "");
890 // Insert the memory aliasing node
891 set_all_memory(reset_memory());
892 }
893 assert(merged_memory(), "");
894
895 switch (intrinsic_id()) {
896 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
897 return inline_cipherBlockChaining_AESCrypt_predicate(false);
898 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
899 return inline_cipherBlockChaining_AESCrypt_predicate(true);
900 case vmIntrinsics::_counterMode_AESCrypt:
901 return inline_counterMode_AESCrypt_predicate();
902 case vmIntrinsics::_digestBase_implCompressMB:
903 return inline_digestBase_implCompressMB_predicate(predicate);
904
905 default:
906 // If you get here, it may be that someone has added a new intrinsic
907 // to the list in vmSymbols.hpp without implementing it here.
908 #ifndef PRODUCT
909 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
910 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
911 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
912 }
913 #endif
914 Node* slow_ctl = control();
915 set_control(top()); // No fast path instrinsic
916 return slow_ctl;
917 }
918 }
919
920 //------------------------------set_result-------------------------------
921 // Helper function for finishing intrinsics.
922 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
923 record_for_igvn(region);
924 set_control(_gvn.transform(region));
925 set_result( _gvn.transform(value));
926 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
927 }
928
929 //------------------------------generate_guard---------------------------
930 // Helper function for generating guarded fast-slow graph structures.
931 // The given 'test', if true, guards a slow path. If the test fails
932 // then a fast path can be taken. (We generally hope it fails.)
933 // In all cases, GraphKit::control() is updated to the fast path.
934 // The returned value represents the control for the slow path.
935 // The return value is never 'top'; it is either a valid control
936 // or NULL if it is obvious that the slow path can never be taken.
937 // Also, if region and the slow control are not NULL, the slow edge
938 // is appended to the region.
939 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
940 if (stopped()) {
941 // Already short circuited.
942 return NULL;
943 }
944
945 // Build an if node and its projections.
946 // If test is true we take the slow path, which we assume is uncommon.
947 if (_gvn.type(test) == TypeInt::ZERO) {
948 // The slow branch is never taken. No need to build this guard.
949 return NULL;
950 }
951
952 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
953
954 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
955 if (if_slow == top()) {
956 // The slow branch is never taken. No need to build this guard.
957 return NULL;
958 }
959
960 if (region != NULL)
961 region->add_req(if_slow);
962
963 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
964 set_control(if_fast);
965
966 return if_slow;
967 }
968
969 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
970 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
971 }
972 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
973 return generate_guard(test, region, PROB_FAIR);
974 }
975
976 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
977 Node* *pos_index) {
978 if (stopped())
979 return NULL; // already stopped
980 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
981 return NULL; // index is already adequately typed
982 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
983 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
984 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
985 if (is_neg != NULL && pos_index != NULL) {
986 // Emulate effect of Parse::adjust_map_after_if.
987 Node* ccast = new CastIINode(index, TypeInt::POS);
988 ccast->set_req(0, control());
989 (*pos_index) = _gvn.transform(ccast);
990 }
991 return is_neg;
992 }
993
994 // Make sure that 'position' is a valid limit index, in [0..length].
995 // There are two equivalent plans for checking this:
996 // A. (offset + copyLength) unsigned<= arrayLength
997 // B. offset <= (arrayLength - copyLength)
998 // We require that all of the values above, except for the sum and
999 // difference, are already known to be non-negative.
1000 // Plan A is robust in the face of overflow, if offset and copyLength
1001 // are both hugely positive.
1002 //
1003 // Plan B is less direct and intuitive, but it does not overflow at
1004 // all, since the difference of two non-negatives is always
1005 // representable. Whenever Java methods must perform the equivalent
1006 // check they generally use Plan B instead of Plan A.
1007 // For the moment we use Plan A.
1008 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1009 Node* subseq_length,
1010 Node* array_length,
1011 RegionNode* region) {
1012 if (stopped())
1013 return NULL; // already stopped
1014 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1015 if (zero_offset && subseq_length->eqv_uncast(array_length))
1016 return NULL; // common case of whole-array copy
1017 Node* last = subseq_length;
1018 if (!zero_offset) // last += offset
1019 last = _gvn.transform(new AddINode(last, offset));
1020 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1021 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1022 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1023 return is_over;
1024 }
1025
1026 // Emit range checks for the given String.value byte array
1027 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1028 if (stopped()) {
1029 return; // already stopped
1030 }
1031 RegionNode* bailout = new RegionNode(1);
1032 record_for_igvn(bailout);
1033 if (char_count) {
1034 // Convert char count to byte count
1035 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1036 }
1037
1038 // Offset and count must not be negative
1039 generate_negative_guard(offset, bailout);
1040 generate_negative_guard(count, bailout);
1041 // Offset + count must not exceed length of array
1042 generate_limit_guard(offset, count, load_array_length(array), bailout);
1043
1044 if (bailout->req() > 1) {
1045 PreserveJVMState pjvms(this);
1046 set_control(_gvn.transform(bailout));
1047 uncommon_trap(Deoptimization::Reason_intrinsic,
1048 Deoptimization::Action_maybe_recompile);
1049 }
1050 }
1051
1052 //--------------------------generate_current_thread--------------------
1053 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1054 ciKlass* thread_klass = env()->Thread_klass();
1055 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1056 Node* thread = _gvn.transform(new ThreadLocalNode());
1057 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1058 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1059 tls_output = thread;
1060 return threadObj;
1061 }
1062
1063
1064 //------------------------------make_string_method_node------------------------
1065 // Helper method for String intrinsic functions. This version is called with
1066 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1067 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1068 // containing the lengths of str1 and str2.
1069 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1070 Node* result = NULL;
1071 switch (opcode) {
1072 case Op_StrIndexOf:
1073 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1074 str1_start, cnt1, str2_start, cnt2, ae);
1075 break;
1076 case Op_StrComp:
1077 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1078 str1_start, cnt1, str2_start, cnt2, ae);
1079 break;
1080 case Op_StrEquals:
1081 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1082 // Use the constant length if there is one because optimized match rule may exist.
1083 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1084 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1085 break;
1086 default:
1087 ShouldNotReachHere();
1088 return NULL;
1089 }
1090
1091 // All these intrinsics have checks.
1092 C->set_has_split_ifs(true); // Has chance for split-if optimization
1093 clear_upper_avx();
1094
1095 return _gvn.transform(result);
1096 }
1097
1098 //------------------------------inline_string_compareTo------------------------
1099 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1100 Node* arg1 = argument(0);
1101 Node* arg2 = argument(1);
1102
1103 // Get start addr and length of first argument
1104 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1105 Node* arg1_cnt = load_array_length(arg1);
1106
1107 // Get start addr and length of second argument
1108 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1109 Node* arg2_cnt = load_array_length(arg2);
1110
1111 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1112 set_result(result);
1113 return true;
1114 }
1115
1116 //------------------------------inline_string_equals------------------------
1117 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1118 Node* arg1 = argument(0);
1119 Node* arg2 = argument(1);
1120
1121 // paths (plus control) merge
1122 RegionNode* region = new RegionNode(3);
1123 Node* phi = new PhiNode(region, TypeInt::BOOL);
1124
1125 if (!stopped()) {
1126 // Get start addr and length of first argument
1127 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1128 Node* arg1_cnt = load_array_length(arg1);
1129
1130 // Get start addr and length of second argument
1131 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1132 Node* arg2_cnt = load_array_length(arg2);
1133
1134 // Check for arg1_cnt != arg2_cnt
1135 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1136 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1137 Node* if_ne = generate_slow_guard(bol, NULL);
1138 if (if_ne != NULL) {
1139 phi->init_req(2, intcon(0));
1140 region->init_req(2, if_ne);
1141 }
1142
1143 // Check for count == 0 is done by assembler code for StrEquals.
1144
1145 if (!stopped()) {
1146 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1147 phi->init_req(1, equals);
1148 region->init_req(1, control());
1149 }
1150 }
1151
1152 // post merge
1153 set_control(_gvn.transform(region));
1154 record_for_igvn(region);
1155
1156 set_result(_gvn.transform(phi));
1157 return true;
1158 }
1159
1160 //------------------------------inline_array_equals----------------------------
1161 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1162 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1163 Node* arg1 = argument(0);
1164 Node* arg2 = argument(1);
1165
1166 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1167 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1168 clear_upper_avx();
1169
1170 return true;
1171 }
1172
1173 //------------------------------inline_hasNegatives------------------------------
1174 bool LibraryCallKit::inline_hasNegatives() {
1175 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1176 return false;
1177 }
1178
1179 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1180 // no receiver since it is static method
1181 Node* ba = argument(0);
1182 Node* offset = argument(1);
1183 Node* len = argument(2);
1184
1185 // Range checks
1186 generate_string_range_check(ba, offset, len, false);
1187 if (stopped()) {
1188 return true;
1189 }
1190 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1191 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1192 set_result(_gvn.transform(result));
1193 return true;
1194 }
1195
1196 bool LibraryCallKit::inline_preconditions_checkIndex() {
1197 Node* index = argument(0);
1198 Node* length = argument(1);
1199 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1200 return false;
1201 }
1202
1203 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1204 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1205
1206 {
1207 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1208 uncommon_trap(Deoptimization::Reason_intrinsic,
1209 Deoptimization::Action_make_not_entrant);
1210 }
1211
1212 if (stopped()) {
1213 return false;
1214 }
1215
1216 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1217 BoolTest::mask btest = BoolTest::lt;
1218 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1219 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1220 _gvn.set_type(rc, rc->Value(&_gvn));
1221 if (!rc_bool->is_Con()) {
1222 record_for_igvn(rc);
1223 }
1224 set_control(_gvn.transform(new IfTrueNode(rc)));
1225 {
1226 PreserveJVMState pjvms(this);
1227 set_control(_gvn.transform(new IfFalseNode(rc)));
1228 uncommon_trap(Deoptimization::Reason_range_check,
1229 Deoptimization::Action_make_not_entrant);
1230 }
1231
1232 if (stopped()) {
1233 return false;
1234 }
1235
1236 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1237 result->set_req(0, control());
1238 result = _gvn.transform(result);
1239 set_result(result);
1240 replace_in_map(index, result);
1241 clear_upper_avx();
1242 return true;
1243 }
1244
1245 //------------------------------inline_string_indexOf------------------------
1246 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1247 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1248 return false;
1249 }
1250 Node* src = argument(0);
1251 Node* tgt = argument(1);
1252
1253 // Make the merge point
1254 RegionNode* result_rgn = new RegionNode(4);
1255 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1256
1257 // Get start addr and length of source string
1258 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1259 Node* src_count = load_array_length(src);
1260
1261 // Get start addr and length of substring
1262 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1263 Node* tgt_count = load_array_length(tgt);
1264
1265 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1266 // Divide src size by 2 if String is UTF16 encoded
1267 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1268 }
1269 if (ae == StrIntrinsicNode::UU) {
1270 // Divide substring size by 2 if String is UTF16 encoded
1271 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1272 }
1273
1274 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1275 if (result != NULL) {
1276 result_phi->init_req(3, result);
1277 result_rgn->init_req(3, control());
1278 }
1279 set_control(_gvn.transform(result_rgn));
1280 record_for_igvn(result_rgn);
1281 set_result(_gvn.transform(result_phi));
1282
1283 return true;
1284 }
1285
1286 //-----------------------------inline_string_indexOf-----------------------
1287 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1288 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1289 return false;
1290 }
1291 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1292 return false;
1293 }
1294 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1295 Node* src = argument(0); // byte[]
1296 Node* src_count = argument(1); // char count
1297 Node* tgt = argument(2); // byte[]
1298 Node* tgt_count = argument(3); // char count
1299 Node* from_index = argument(4); // char index
1300
1301 // Multiply byte array index by 2 if String is UTF16 encoded
1302 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1303 src_count = _gvn.transform(new SubINode(src_count, from_index));
1304 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1305 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1306
1307 // Range checks
1308 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1309 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1310 if (stopped()) {
1311 return true;
1312 }
1313
1314 RegionNode* region = new RegionNode(5);
1315 Node* phi = new PhiNode(region, TypeInt::INT);
1316
1317 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1318 if (result != NULL) {
1319 // The result is index relative to from_index if substring was found, -1 otherwise.
1320 // Generate code which will fold into cmove.
1321 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1322 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1323
1324 Node* if_lt = generate_slow_guard(bol, NULL);
1325 if (if_lt != NULL) {
1326 // result == -1
1327 phi->init_req(3, result);
1328 region->init_req(3, if_lt);
1329 }
1330 if (!stopped()) {
1331 result = _gvn.transform(new AddINode(result, from_index));
1332 phi->init_req(4, result);
1333 region->init_req(4, control());
1334 }
1335 }
1336
1337 set_control(_gvn.transform(region));
1338 record_for_igvn(region);
1339 set_result(_gvn.transform(phi));
1340 clear_upper_avx();
1341
1342 return true;
1343 }
1344
1345 // Create StrIndexOfNode with fast path checks
1346 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1347 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1348 // Check for substr count > string count
1349 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1350 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1351 Node* if_gt = generate_slow_guard(bol, NULL);
1352 if (if_gt != NULL) {
1353 phi->init_req(1, intcon(-1));
1354 region->init_req(1, if_gt);
1355 }
1356 if (!stopped()) {
1357 // Check for substr count == 0
1358 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1359 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1360 Node* if_zero = generate_slow_guard(bol, NULL);
1361 if (if_zero != NULL) {
1362 phi->init_req(2, intcon(0));
1363 region->init_req(2, if_zero);
1364 }
1365 }
1366 if (!stopped()) {
1367 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1368 }
1369 return NULL;
1370 }
1371
1372 //-----------------------------inline_string_indexOfChar-----------------------
1373 bool LibraryCallKit::inline_string_indexOfChar() {
1374 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1375 return false;
1376 }
1377 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1378 return false;
1379 }
1380 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1381 Node* src = argument(0); // byte[]
1382 Node* tgt = argument(1); // tgt is int ch
1383 Node* from_index = argument(2);
1384 Node* max = argument(3);
1385
1386 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1387 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1388 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1389
1390 // Range checks
1391 generate_string_range_check(src, src_offset, src_count, true);
1392 if (stopped()) {
1393 return true;
1394 }
1395
1396 RegionNode* region = new RegionNode(3);
1397 Node* phi = new PhiNode(region, TypeInt::INT);
1398
1399 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1400 C->set_has_split_ifs(true); // Has chance for split-if optimization
1401 _gvn.transform(result);
1402
1403 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1404 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1405
1406 Node* if_lt = generate_slow_guard(bol, NULL);
1407 if (if_lt != NULL) {
1408 // result == -1
1409 phi->init_req(2, result);
1410 region->init_req(2, if_lt);
1411 }
1412 if (!stopped()) {
1413 result = _gvn.transform(new AddINode(result, from_index));
1414 phi->init_req(1, result);
1415 region->init_req(1, control());
1416 }
1417 set_control(_gvn.transform(region));
1418 record_for_igvn(region);
1419 set_result(_gvn.transform(phi));
1420
1421 return true;
1422 }
1423 //---------------------------inline_string_copy---------------------
1424 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1425 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1426 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1427 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1428 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1429 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1430 bool LibraryCallKit::inline_string_copy(bool compress) {
1431 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1432 return false;
1433 }
1434 int nargs = 5; // 2 oops, 3 ints
1435 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1436
1437 Node* src = argument(0);
1438 Node* src_offset = argument(1);
1439 Node* dst = argument(2);
1440 Node* dst_offset = argument(3);
1441 Node* length = argument(4);
1442
1443 // Check for allocation before we add nodes that would confuse
1444 // tightly_coupled_allocation()
1445 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1446
1447 // Figure out the size and type of the elements we will be copying.
1448 const Type* src_type = src->Value(&_gvn);
1449 const Type* dst_type = dst->Value(&_gvn);
1450 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1451 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1452 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1453 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1454 "Unsupported array types for inline_string_copy");
1455
1456 // Convert char[] offsets to byte[] offsets
1457 bool convert_src = (compress && src_elem == T_BYTE);
1458 bool convert_dst = (!compress && dst_elem == T_BYTE);
1459 if (convert_src) {
1460 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1461 } else if (convert_dst) {
1462 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1463 }
1464
1465 // Range checks
1466 generate_string_range_check(src, src_offset, length, convert_src);
1467 generate_string_range_check(dst, dst_offset, length, convert_dst);
1468 if (stopped()) {
1469 return true;
1470 }
1471
1472 Node* src_start = array_element_address(src, src_offset, src_elem);
1473 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1474 // 'src_start' points to src array + scaled offset
1475 // 'dst_start' points to dst array + scaled offset
1476 Node* count = NULL;
1477 if (compress) {
1478 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1479 } else {
1480 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1481 }
1482
1483 if (alloc != NULL) {
1484 if (alloc->maybe_set_complete(&_gvn)) {
1485 // "You break it, you buy it."
1486 InitializeNode* init = alloc->initialization();
1487 assert(init->is_complete(), "we just did this");
1488 init->set_complete_with_arraycopy();
1489 assert(dst->is_CheckCastPP(), "sanity");
1490 assert(dst->in(0)->in(0) == init, "dest pinned");
1491 }
1492 // Do not let stores that initialize this object be reordered with
1493 // a subsequent store that would make this object accessible by
1494 // other threads.
1495 // Record what AllocateNode this StoreStore protects so that
1496 // escape analysis can go from the MemBarStoreStoreNode to the
1497 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1498 // based on the escape status of the AllocateNode.
1499 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1500 }
1501 if (compress) {
1502 set_result(_gvn.transform(count));
1503 }
1504 clear_upper_avx();
1505
1506 return true;
1507 }
1508
1509 #ifdef _LP64
1510 #define XTOP ,top() /*additional argument*/
1511 #else //_LP64
1512 #define XTOP /*no additional argument*/
1513 #endif //_LP64
1514
1515 //------------------------inline_string_toBytesU--------------------------
1516 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1517 bool LibraryCallKit::inline_string_toBytesU() {
1518 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1519 return false;
1520 }
1521 // Get the arguments.
1522 Node* value = argument(0);
1523 Node* offset = argument(1);
1524 Node* length = argument(2);
1525
1526 Node* newcopy = NULL;
1527
1528 // Set the original stack and the reexecute bit for the interpreter to reexecute
1529 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1530 { PreserveReexecuteState preexecs(this);
1531 jvms()->set_should_reexecute(true);
1532
1533 // Check if a null path was taken unconditionally.
1534 value = null_check(value);
1535
1536 RegionNode* bailout = new RegionNode(1);
1537 record_for_igvn(bailout);
1538
1539 // Range checks
1540 generate_negative_guard(offset, bailout);
1541 generate_negative_guard(length, bailout);
1542 generate_limit_guard(offset, length, load_array_length(value), bailout);
1543 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1544 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1545
1546 if (bailout->req() > 1) {
1547 PreserveJVMState pjvms(this);
1548 set_control(_gvn.transform(bailout));
1549 uncommon_trap(Deoptimization::Reason_intrinsic,
1550 Deoptimization::Action_maybe_recompile);
1551 }
1552 if (stopped()) {
1553 return true;
1554 }
1555
1556 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1557 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1558 newcopy = new_array(klass_node, size, 0); // no arguments to push
1559 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1560
1561 // Calculate starting addresses.
1562 Node* src_start = array_element_address(value, offset, T_CHAR);
1563 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1564
1565 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1566 const TypeInt* toffset = gvn().type(offset)->is_int();
1567 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1568
1569 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1570 const char* copyfunc_name = "arraycopy";
1571 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1572 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1573 OptoRuntime::fast_arraycopy_Type(),
1574 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1575 src_start, dst_start, ConvI2X(length) XTOP);
1576 // Do not let reads from the cloned object float above the arraycopy.
1577 if (alloc != NULL) {
1578 if (alloc->maybe_set_complete(&_gvn)) {
1579 // "You break it, you buy it."
1580 InitializeNode* init = alloc->initialization();
1581 assert(init->is_complete(), "we just did this");
1582 init->set_complete_with_arraycopy();
1583 assert(newcopy->is_CheckCastPP(), "sanity");
1584 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1585 }
1586 // Do not let stores that initialize this object be reordered with
1587 // a subsequent store that would make this object accessible by
1588 // other threads.
1589 // Record what AllocateNode this StoreStore protects so that
1590 // escape analysis can go from the MemBarStoreStoreNode to the
1591 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1592 // based on the escape status of the AllocateNode.
1593 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1594 } else {
1595 insert_mem_bar(Op_MemBarCPUOrder);
1596 }
1597 } // original reexecute is set back here
1598
1599 C->set_has_split_ifs(true); // Has chance for split-if optimization
1600 if (!stopped()) {
1601 set_result(newcopy);
1602 }
1603 clear_upper_avx();
1604
1605 return true;
1606 }
1607
1608 //------------------------inline_string_getCharsU--------------------------
1609 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1610 bool LibraryCallKit::inline_string_getCharsU() {
1611 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1612 return false;
1613 }
1614
1615 // Get the arguments.
1616 Node* src = argument(0);
1617 Node* src_begin = argument(1);
1618 Node* src_end = argument(2); // exclusive offset (i < src_end)
1619 Node* dst = argument(3);
1620 Node* dst_begin = argument(4);
1621
1622 // Check for allocation before we add nodes that would confuse
1623 // tightly_coupled_allocation()
1624 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1625
1626 // Check if a null path was taken unconditionally.
1627 src = null_check(src);
1628 dst = null_check(dst);
1629 if (stopped()) {
1630 return true;
1631 }
1632
1633 // Get length and convert char[] offset to byte[] offset
1634 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1635 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1636
1637 // Range checks
1638 generate_string_range_check(src, src_begin, length, true);
1639 generate_string_range_check(dst, dst_begin, length, false);
1640 if (stopped()) {
1641 return true;
1642 }
1643
1644 if (!stopped()) {
1645 // Calculate starting addresses.
1646 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1647 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1648
1649 // Check if array addresses are aligned to HeapWordSize
1650 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1651 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1652 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1653 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1654
1655 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1656 const char* copyfunc_name = "arraycopy";
1657 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1658 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1659 OptoRuntime::fast_arraycopy_Type(),
1660 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1661 src_start, dst_start, ConvI2X(length) XTOP);
1662 // Do not let reads from the cloned object float above the arraycopy.
1663 if (alloc != NULL) {
1664 if (alloc->maybe_set_complete(&_gvn)) {
1665 // "You break it, you buy it."
1666 InitializeNode* init = alloc->initialization();
1667 assert(init->is_complete(), "we just did this");
1668 init->set_complete_with_arraycopy();
1669 assert(dst->is_CheckCastPP(), "sanity");
1670 assert(dst->in(0)->in(0) == init, "dest pinned");
1671 }
1672 // Do not let stores that initialize this object be reordered with
1673 // a subsequent store that would make this object accessible by
1674 // other threads.
1675 // Record what AllocateNode this StoreStore protects so that
1676 // escape analysis can go from the MemBarStoreStoreNode to the
1677 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1678 // based on the escape status of the AllocateNode.
1679 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1680 } else {
1681 insert_mem_bar(Op_MemBarCPUOrder);
1682 }
1683 }
1684
1685 C->set_has_split_ifs(true); // Has chance for split-if optimization
1686 return true;
1687 }
1688
1689 //----------------------inline_string_char_access----------------------------
1690 // Store/Load char to/from byte[] array.
1691 // static void StringUTF16.putChar(byte[] val, int index, int c)
1692 // static char StringUTF16.getChar(byte[] val, int index)
1693 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1694 Node* value = argument(0);
1695 Node* index = argument(1);
1696 Node* ch = is_store ? argument(2) : NULL;
1697
1698 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1699 // correctly requires matched array shapes.
1700 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1701 "sanity: byte[] and char[] bases agree");
1702 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1703 "sanity: byte[] and char[] scales agree");
1704
1705 // Bail when getChar over constants is requested: constant folding would
1706 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1707 // Java method would constant fold nicely instead.
1708 if (!is_store && value->is_Con() && index->is_Con()) {
1709 return false;
1710 }
1711
1712 Node* adr = array_element_address(value, index, T_CHAR);
1713 if (adr->is_top()) {
1714 return false;
1715 }
1716 if (is_store) {
1717 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1718 false, false, true /* mismatched */);
1719 } else {
1720 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1721 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1722 set_result(ch);
1723 }
1724 return true;
1725 }
1726
1727 //--------------------------round_double_node--------------------------------
1728 // Round a double node if necessary.
1729 Node* LibraryCallKit::round_double_node(Node* n) {
1730 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1731 n = _gvn.transform(new RoundDoubleNode(0, n));
1732 return n;
1733 }
1734
1735 //------------------------------inline_math-----------------------------------
1736 // public static double Math.abs(double)
1737 // public static double Math.sqrt(double)
1738 // public static double Math.log(double)
1739 // public static double Math.log10(double)
1740 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1741 Node* arg = round_double_node(argument(0));
1742 Node* n = NULL;
1743 switch (id) {
1744 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1745 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1746 default: fatal_unexpected_iid(id); break;
1747 }
1748 set_result(_gvn.transform(n));
1749 return true;
1750 }
1751
1752 //------------------------------runtime_math-----------------------------
1753 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1754 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1755 "must be (DD)D or (D)D type");
1756
1757 // Inputs
1758 Node* a = round_double_node(argument(0));
1759 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1760
1761 const TypePtr* no_memory_effects = NULL;
1762 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1763 no_memory_effects,
1764 a, top(), b, b ? top() : NULL);
1765 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1766 #ifdef ASSERT
1767 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1768 assert(value_top == top(), "second value must be top");
1769 #endif
1770
1771 set_result(value);
1772 return true;
1773 }
1774
1775 //------------------------------inline_math_native-----------------------------
1776 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1777 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1778 switch (id) {
1779 // These intrinsics are not properly supported on all hardware
1780 case vmIntrinsics::_dsin:
1781 return StubRoutines::dsin() != NULL ?
1782 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1783 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1784 case vmIntrinsics::_dcos:
1785 return StubRoutines::dcos() != NULL ?
1786 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1787 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1788 case vmIntrinsics::_dtan:
1789 return StubRoutines::dtan() != NULL ?
1790 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1791 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1792 case vmIntrinsics::_dlog:
1793 return StubRoutines::dlog() != NULL ?
1794 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1795 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1796 case vmIntrinsics::_dlog10:
1797 return StubRoutines::dlog10() != NULL ?
1798 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1799 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1800
1801 // These intrinsics are supported on all hardware
1802 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1803 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1804
1805 case vmIntrinsics::_dexp:
1806 return StubRoutines::dexp() != NULL ?
1807 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1808 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1809 case vmIntrinsics::_dpow: {
1810 Node* exp = round_double_node(argument(2));
1811 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1812 if (d != NULL && d->getd() == 2.0) {
1813 // Special case: pow(x, 2.0) => x * x
1814 Node* base = round_double_node(argument(0));
1815 set_result(_gvn.transform(new MulDNode(base, base)));
1816 return true;
1817 }
1818 return StubRoutines::dexp() != NULL ?
1819 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1820 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1821 }
1822 #undef FN_PTR
1823
1824 // These intrinsics are not yet correctly implemented
1825 case vmIntrinsics::_datan2:
1826 return false;
1827
1828 default:
1829 fatal_unexpected_iid(id);
1830 return false;
1831 }
1832 }
1833
1834 static bool is_simple_name(Node* n) {
1835 return (n->req() == 1 // constant
1836 || (n->is_Type() && n->as_Type()->type()->singleton())
1837 || n->is_Proj() // parameter or return value
1838 || n->is_Phi() // local of some sort
1839 );
1840 }
1841
1842 //----------------------------inline_notify-----------------------------------*
1843 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1844 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1845 address func;
1846 if (id == vmIntrinsics::_notify) {
1847 func = OptoRuntime::monitor_notify_Java();
1848 } else {
1849 func = OptoRuntime::monitor_notifyAll_Java();
1850 }
1851 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1852 make_slow_call_ex(call, env()->Throwable_klass(), false);
1853 return true;
1854 }
1855
1856
1857 //----------------------------inline_min_max-----------------------------------
1858 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1859 set_result(generate_min_max(id, argument(0), argument(1)));
1860 return true;
1861 }
1862
1863 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1864 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1865 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1866 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1867 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1868
1869 {
1870 PreserveJVMState pjvms(this);
1871 PreserveReexecuteState preexecs(this);
1872 jvms()->set_should_reexecute(true);
1873
1874 set_control(slow_path);
1875 set_i_o(i_o());
1876
1877 uncommon_trap(Deoptimization::Reason_intrinsic,
1878 Deoptimization::Action_none);
1879 }
1880
1881 set_control(fast_path);
1882 set_result(math);
1883 }
1884
1885 template <typename OverflowOp>
1886 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1887 typedef typename OverflowOp::MathOp MathOp;
1888
1889 MathOp* mathOp = new MathOp(arg1, arg2);
1890 Node* operation = _gvn.transform( mathOp );
1891 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1892 inline_math_mathExact(operation, ofcheck);
1893 return true;
1894 }
1895
1896 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1897 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1898 }
1899
1900 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1901 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1902 }
1903
1904 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1905 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1906 }
1907
1908 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1909 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1910 }
1911
1912 bool LibraryCallKit::inline_math_negateExactI() {
1913 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1914 }
1915
1916 bool LibraryCallKit::inline_math_negateExactL() {
1917 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1918 }
1919
1920 bool LibraryCallKit::inline_math_multiplyExactI() {
1921 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1922 }
1923
1924 bool LibraryCallKit::inline_math_multiplyExactL() {
1925 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1926 }
1927
1928 bool LibraryCallKit::inline_math_multiplyHigh() {
1929 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1930 return true;
1931 }
1932
1933 Node*
1934 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1935 // These are the candidate return value:
1936 Node* xvalue = x0;
1937 Node* yvalue = y0;
1938
1939 if (xvalue == yvalue) {
1940 return xvalue;
1941 }
1942
1943 bool want_max = (id == vmIntrinsics::_max);
1944
1945 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1946 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1947 if (txvalue == NULL || tyvalue == NULL) return top();
1948 // This is not really necessary, but it is consistent with a
1949 // hypothetical MaxINode::Value method:
1950 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1951
1952 // %%% This folding logic should (ideally) be in a different place.
1953 // Some should be inside IfNode, and there to be a more reliable
1954 // transformation of ?: style patterns into cmoves. We also want
1955 // more powerful optimizations around cmove and min/max.
1956
1957 // Try to find a dominating comparison of these guys.
1958 // It can simplify the index computation for Arrays.copyOf
1959 // and similar uses of System.arraycopy.
1960 // First, compute the normalized version of CmpI(x, y).
1961 int cmp_op = Op_CmpI;
1962 Node* xkey = xvalue;
1963 Node* ykey = yvalue;
1964 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1965 if (ideal_cmpxy->is_Cmp()) {
1966 // E.g., if we have CmpI(length - offset, count),
1967 // it might idealize to CmpI(length, count + offset)
1968 cmp_op = ideal_cmpxy->Opcode();
1969 xkey = ideal_cmpxy->in(1);
1970 ykey = ideal_cmpxy->in(2);
1971 }
1972
1973 // Start by locating any relevant comparisons.
1974 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1975 Node* cmpxy = NULL;
1976 Node* cmpyx = NULL;
1977 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1978 Node* cmp = start_from->fast_out(k);
1979 if (cmp->outcnt() > 0 && // must have prior uses
1980 cmp->in(0) == NULL && // must be context-independent
1981 cmp->Opcode() == cmp_op) { // right kind of compare
1982 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1983 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1984 }
1985 }
1986
1987 const int NCMPS = 2;
1988 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1989 int cmpn;
1990 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1991 if (cmps[cmpn] != NULL) break; // find a result
1992 }
1993 if (cmpn < NCMPS) {
1994 // Look for a dominating test that tells us the min and max.
1995 int depth = 0; // Limit search depth for speed
1996 Node* dom = control();
1997 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1998 if (++depth >= 100) break;
1999 Node* ifproj = dom;
2000 if (!ifproj->is_Proj()) continue;
2001 Node* iff = ifproj->in(0);
2002 if (!iff->is_If()) continue;
2003 Node* bol = iff->in(1);
2004 if (!bol->is_Bool()) continue;
2005 Node* cmp = bol->in(1);
2006 if (cmp == NULL) continue;
2007 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2008 if (cmps[cmpn] == cmp) break;
2009 if (cmpn == NCMPS) continue;
2010 BoolTest::mask btest = bol->as_Bool()->_test._test;
2011 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2012 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2013 // At this point, we know that 'x btest y' is true.
2014 switch (btest) {
2015 case BoolTest::eq:
2016 // They are proven equal, so we can collapse the min/max.
2017 // Either value is the answer. Choose the simpler.
2018 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2019 return yvalue;
2020 return xvalue;
2021 case BoolTest::lt: // x < y
2022 case BoolTest::le: // x <= y
2023 return (want_max ? yvalue : xvalue);
2024 case BoolTest::gt: // x > y
2025 case BoolTest::ge: // x >= y
2026 return (want_max ? xvalue : yvalue);
2027 default:
2028 break;
2029 }
2030 }
2031 }
2032
2033 // We failed to find a dominating test.
2034 // Let's pick a test that might GVN with prior tests.
2035 Node* best_bol = NULL;
2036 BoolTest::mask best_btest = BoolTest::illegal;
2037 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2038 Node* cmp = cmps[cmpn];
2039 if (cmp == NULL) continue;
2040 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2041 Node* bol = cmp->fast_out(j);
2042 if (!bol->is_Bool()) continue;
2043 BoolTest::mask btest = bol->as_Bool()->_test._test;
2044 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2045 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2046 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2047 best_bol = bol->as_Bool();
2048 best_btest = btest;
2049 }
2050 }
2051 }
2052
2053 Node* answer_if_true = NULL;
2054 Node* answer_if_false = NULL;
2055 switch (best_btest) {
2056 default:
2057 if (cmpxy == NULL)
2058 cmpxy = ideal_cmpxy;
2059 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2060 // and fall through:
2061 case BoolTest::lt: // x < y
2062 case BoolTest::le: // x <= y
2063 answer_if_true = (want_max ? yvalue : xvalue);
2064 answer_if_false = (want_max ? xvalue : yvalue);
2065 break;
2066 case BoolTest::gt: // x > y
2067 case BoolTest::ge: // x >= y
2068 answer_if_true = (want_max ? xvalue : yvalue);
2069 answer_if_false = (want_max ? yvalue : xvalue);
2070 break;
2071 }
2072
2073 jint hi, lo;
2074 if (want_max) {
2075 // We can sharpen the minimum.
2076 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2077 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2078 } else {
2079 // We can sharpen the maximum.
2080 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2081 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2082 }
2083
2084 // Use a flow-free graph structure, to avoid creating excess control edges
2085 // which could hinder other optimizations.
2086 // Since Math.min/max is often used with arraycopy, we want
2087 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2088 Node* cmov = CMoveNode::make(NULL, best_bol,
2089 answer_if_false, answer_if_true,
2090 TypeInt::make(lo, hi, widen));
2091
2092 return _gvn.transform(cmov);
2093
2094 /*
2095 // This is not as desirable as it may seem, since Min and Max
2096 // nodes do not have a full set of optimizations.
2097 // And they would interfere, anyway, with 'if' optimizations
2098 // and with CMoveI canonical forms.
2099 switch (id) {
2100 case vmIntrinsics::_min:
2101 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2102 case vmIntrinsics::_max:
2103 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2104 default:
2105 ShouldNotReachHere();
2106 }
2107 */
2108 }
2109
2110 inline int
2111 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2112 const TypePtr* base_type = TypePtr::NULL_PTR;
2113 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2114 if (base_type == NULL) {
2115 // Unknown type.
2116 return Type::AnyPtr;
2117 } else if (base_type == TypePtr::NULL_PTR) {
2118 // Since this is a NULL+long form, we have to switch to a rawptr.
2119 base = _gvn.transform(new CastX2PNode(offset));
2120 offset = MakeConX(0);
2121 return Type::RawPtr;
2122 } else if (base_type->base() == Type::RawPtr) {
2123 return Type::RawPtr;
2124 } else if (base_type->isa_oopptr()) {
2125 // Base is never null => always a heap address.
2126 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2127 return Type::OopPtr;
2128 }
2129 // Offset is small => always a heap address.
2130 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2131 if (offset_type != NULL &&
2132 base_type->offset() == 0 && // (should always be?)
2133 offset_type->_lo >= 0 &&
2134 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2135 return Type::OopPtr;
2136 } else if (type == T_OBJECT) {
2137 // off heap access to an oop doesn't make any sense. Has to be on
2138 // heap.
2139 return Type::OopPtr;
2140 }
2141 // Otherwise, it might either be oop+off or NULL+addr.
2142 return Type::AnyPtr;
2143 } else {
2144 // No information:
2145 return Type::AnyPtr;
2146 }
2147 }
2148
2149 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2150 Node* uncasted_base = base;
2151 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2152 if (kind == Type::RawPtr) {
2153 return basic_plus_adr(top(), uncasted_base, offset);
2154 } else if (kind == Type::AnyPtr) {
2155 assert(base == uncasted_base, "unexpected base change");
2156 if (can_cast) {
2157 if (!_gvn.type(base)->speculative_maybe_null() &&
2158 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2159 // According to profiling, this access is always on
2160 // heap. Casting the base to not null and thus avoiding membars
2161 // around the access should allow better optimizations
2162 Node* null_ctl = top();
2163 base = null_check_oop(base, &null_ctl, true, true, true);
2164 assert(null_ctl->is_top(), "no null control here");
2165 return basic_plus_adr(base, offset);
2166 } else if (_gvn.type(base)->speculative_always_null() &&
2167 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2168 // According to profiling, this access is always off
2169 // heap.
2170 base = null_assert(base);
2171 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2172 offset = MakeConX(0);
2173 return basic_plus_adr(top(), raw_base, offset);
2174 }
2175 }
2176 // We don't know if it's an on heap or off heap access. Fall back
2177 // to raw memory access.
2178 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2179 return basic_plus_adr(top(), raw, offset);
2180 } else {
2181 assert(base == uncasted_base, "unexpected base change");
2182 // We know it's an on heap access so base can't be null
2183 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2184 base = must_be_not_null(base, true);
2185 }
2186 return basic_plus_adr(base, offset);
2187 }
2188 }
2189
2190 //--------------------------inline_number_methods-----------------------------
2191 // inline int Integer.numberOfLeadingZeros(int)
2192 // inline int Long.numberOfLeadingZeros(long)
2193 //
2194 // inline int Integer.numberOfTrailingZeros(int)
2195 // inline int Long.numberOfTrailingZeros(long)
2196 //
2197 // inline int Integer.bitCount(int)
2198 // inline int Long.bitCount(long)
2199 //
2200 // inline char Character.reverseBytes(char)
2201 // inline short Short.reverseBytes(short)
2202 // inline int Integer.reverseBytes(int)
2203 // inline long Long.reverseBytes(long)
2204 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2205 Node* arg = argument(0);
2206 Node* n = NULL;
2207 switch (id) {
2208 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2209 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2210 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2211 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2212 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2213 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2214 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2215 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2216 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2217 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2218 default: fatal_unexpected_iid(id); break;
2219 }
2220 set_result(_gvn.transform(n));
2221 return true;
2222 }
2223
2224 //----------------------------inline_unsafe_access----------------------------
2225
2226 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2227 // Attempt to infer a sharper value type from the offset and base type.
2228 ciKlass* sharpened_klass = NULL;
2229
2230 // See if it is an instance field, with an object type.
2231 if (alias_type->field() != NULL) {
2232 if (alias_type->field()->type()->is_klass()) {
2233 sharpened_klass = alias_type->field()->type()->as_klass();
2234 }
2235 }
2236
2237 // See if it is a narrow oop array.
2238 if (adr_type->isa_aryptr()) {
2239 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2240 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2241 if (elem_type != NULL) {
2242 sharpened_klass = elem_type->klass();
2243 }
2244 }
2245 }
2246
2247 // The sharpened class might be unloaded if there is no class loader
2248 // contraint in place.
2249 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2250 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2251
2252 #ifndef PRODUCT
2253 if (C->print_intrinsics() || C->print_inlining()) {
2254 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2255 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2256 }
2257 #endif
2258 // Sharpen the value type.
2259 return tjp;
2260 }
2261 return NULL;
2262 }
2263
2264 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2265 switch (kind) {
2266 case Relaxed:
2267 return MO_UNORDERED;
2268 case Opaque:
2269 return MO_RELAXED;
2270 case Acquire:
2271 return MO_ACQUIRE;
2272 case Release:
2273 return MO_RELEASE;
2274 case Volatile:
2275 return MO_SEQ_CST;
2276 default:
2277 ShouldNotReachHere();
2278 return 0;
2279 }
2280 }
2281
2282 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2283 if (callee()->is_static()) return false; // caller must have the capability!
2284 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2285 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2286 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2287 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2288
2289 if (type == T_OBJECT || type == T_ARRAY) {
2290 decorators |= ON_UNKNOWN_OOP_REF;
2291 }
2292
2293 if (unaligned) {
2294 decorators |= C2_UNALIGNED;
2295 }
2296
2297 #ifndef PRODUCT
2298 {
2299 ResourceMark rm;
2300 // Check the signatures.
2301 ciSignature* sig = callee()->signature();
2302 #ifdef ASSERT
2303 if (!is_store) {
2304 // Object getObject(Object base, int/long offset), etc.
2305 BasicType rtype = sig->return_type()->basic_type();
2306 assert(rtype == type, "getter must return the expected value");
2307 assert(sig->count() == 2, "oop getter has 2 arguments");
2308 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2309 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2310 } else {
2311 // void putObject(Object base, int/long offset, Object x), etc.
2312 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2313 assert(sig->count() == 3, "oop putter has 3 arguments");
2314 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2315 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2316 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2317 assert(vtype == type, "putter must accept the expected value");
2318 }
2319 #endif // ASSERT
2320 }
2321 #endif //PRODUCT
2322
2323 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2324
2325 Node* receiver = argument(0); // type: oop
2326
2327 // Build address expression.
2328 Node* adr;
2329 Node* heap_base_oop = top();
2330 Node* offset = top();
2331 Node* val;
2332
2333 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2334 Node* base = argument(1); // type: oop
2335 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2336 offset = argument(2); // type: long
2337 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2338 // to be plain byte offsets, which are also the same as those accepted
2339 // by oopDesc::field_addr.
2340 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2341 "fieldOffset must be byte-scaled");
2342 // 32-bit machines ignore the high half!
2343 offset = ConvL2X(offset);
2344 adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2345
2346 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2347 heap_base_oop = base;
2348 } else if (type == T_OBJECT) {
2349 return false; // off-heap oop accesses are not supported
2350 }
2351
2352 // Can base be NULL? Otherwise, always on-heap access.
2353 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2354
2355 if (!can_access_non_heap) {
2356 decorators |= IN_HEAP;
2357 }
2358
2359 val = is_store ? argument(4) : NULL;
2360
2361 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2362
2363 // Try to categorize the address.
2364 Compile::AliasType* alias_type = C->alias_type(adr_type);
2365 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2366
2367 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2368 alias_type->adr_type() == TypeAryPtr::RANGE) {
2369 return false; // not supported
2370 }
2371
2372 bool mismatched = false;
2373 BasicType bt = alias_type->basic_type();
2374 if (bt != T_ILLEGAL) {
2375 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2376 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2377 // Alias type doesn't differentiate between byte[] and boolean[]).
2378 // Use address type to get the element type.
2379 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2380 }
2381 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2382 // accessing an array field with getObject is not a mismatch
2383 bt = T_OBJECT;
2384 }
2385 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2386 // Don't intrinsify mismatched object accesses
2387 return false;
2388 }
2389 mismatched = (bt != type);
2390 } else if (alias_type->adr_type()->isa_oopptr()) {
2391 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2392 }
2393
2394 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2395
2396 if (mismatched) {
2397 decorators |= C2_MISMATCHED;
2398 }
2399
2400 // First guess at the value type.
2401 const Type *value_type = Type::get_const_basic_type(type);
2402
2403 // Figure out the memory ordering.
2404 decorators |= mo_decorator_for_access_kind(kind);
2405
2406 if (!is_store && type == T_OBJECT) {
2407 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2408 if (tjp != NULL) {
2409 value_type = tjp;
2410 }
2411 }
2412
2413 receiver = null_check(receiver);
2414 if (stopped()) {
2415 return true;
2416 }
2417 // Heap pointers get a null-check from the interpreter,
2418 // as a courtesy. However, this is not guaranteed by Unsafe,
2419 // and it is not possible to fully distinguish unintended nulls
2420 // from intended ones in this API.
2421
2422 if (!is_store) {
2423 Node* p = NULL;
2424 // Try to constant fold a load from a constant field
2425 ciField* field = alias_type->field();
2426 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2427 // final or stable field
2428 p = make_constant_from_field(field, heap_base_oop);
2429 }
2430
2431 if (p == NULL) { // Could not constant fold the load
2432 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2433 // Normalize the value returned by getBoolean in the following cases
2434 if (type == T_BOOLEAN &&
2435 (mismatched ||
2436 heap_base_oop == top() || // - heap_base_oop is NULL or
2437 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2438 // and the unsafe access is made to large offset
2439 // (i.e., larger than the maximum offset necessary for any
2440 // field access)
2441 ) {
2442 IdealKit ideal = IdealKit(this);
2443 #define __ ideal.
2444 IdealVariable normalized_result(ideal);
2445 __ declarations_done();
2446 __ set(normalized_result, p);
2447 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2448 __ set(normalized_result, ideal.ConI(1));
2449 ideal.end_if();
2450 final_sync(ideal);
2451 p = __ value(normalized_result);
2452 #undef __
2453 }
2454 }
2455 if (type == T_ADDRESS) {
2456 p = gvn().transform(new CastP2XNode(NULL, p));
2457 p = ConvX2UL(p);
2458 }
2459 // The load node has the control of the preceding MemBarCPUOrder. All
2460 // following nodes will have the control of the MemBarCPUOrder inserted at
2461 // the end of this method. So, pushing the load onto the stack at a later
2462 // point is fine.
2463 set_result(p);
2464 } else {
2465 if (bt == T_ADDRESS) {
2466 // Repackage the long as a pointer.
2467 val = ConvL2X(val);
2468 val = gvn().transform(new CastX2PNode(val));
2469 }
2470 access_store_at(control(), heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2471 }
2472
2473 return true;
2474 }
2475
2476 //----------------------------inline_unsafe_load_store----------------------------
2477 // This method serves a couple of different customers (depending on LoadStoreKind):
2478 //
2479 // LS_cmp_swap:
2480 //
2481 // boolean compareAndSetObject(Object o, long offset, Object expected, Object x);
2482 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2483 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2484 //
2485 // LS_cmp_swap_weak:
2486 //
2487 // boolean weakCompareAndSetObject( Object o, long offset, Object expected, Object x);
2488 // boolean weakCompareAndSetObjectPlain( Object o, long offset, Object expected, Object x);
2489 // boolean weakCompareAndSetObjectAcquire(Object o, long offset, Object expected, Object x);
2490 // boolean weakCompareAndSetObjectRelease(Object o, long offset, Object expected, Object x);
2491 //
2492 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2493 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2494 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2495 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2496 //
2497 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2498 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2499 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2500 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2501 //
2502 // LS_cmp_exchange:
2503 //
2504 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2505 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2506 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2507 //
2508 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2509 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2510 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2511 //
2512 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2513 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2514 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2515 //
2516 // LS_get_add:
2517 //
2518 // int getAndAddInt( Object o, long offset, int delta)
2519 // long getAndAddLong(Object o, long offset, long delta)
2520 //
2521 // LS_get_set:
2522 //
2523 // int getAndSet(Object o, long offset, int newValue)
2524 // long getAndSet(Object o, long offset, long newValue)
2525 // Object getAndSet(Object o, long offset, Object newValue)
2526 //
2527 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2528 // This basic scheme here is the same as inline_unsafe_access, but
2529 // differs in enough details that combining them would make the code
2530 // overly confusing. (This is a true fact! I originally combined
2531 // them, but even I was confused by it!) As much code/comments as
2532 // possible are retained from inline_unsafe_access though to make
2533 // the correspondences clearer. - dl
2534
2535 if (callee()->is_static()) return false; // caller must have the capability!
2536
2537 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2538 decorators |= mo_decorator_for_access_kind(access_kind);
2539
2540 #ifndef PRODUCT
2541 BasicType rtype;
2542 {
2543 ResourceMark rm;
2544 // Check the signatures.
2545 ciSignature* sig = callee()->signature();
2546 rtype = sig->return_type()->basic_type();
2547 switch(kind) {
2548 case LS_get_add:
2549 case LS_get_set: {
2550 // Check the signatures.
2551 #ifdef ASSERT
2552 assert(rtype == type, "get and set must return the expected type");
2553 assert(sig->count() == 3, "get and set has 3 arguments");
2554 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2555 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2556 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2557 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2558 #endif // ASSERT
2559 break;
2560 }
2561 case LS_cmp_swap:
2562 case LS_cmp_swap_weak: {
2563 // Check the signatures.
2564 #ifdef ASSERT
2565 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2566 assert(sig->count() == 4, "CAS has 4 arguments");
2567 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2568 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2569 #endif // ASSERT
2570 break;
2571 }
2572 case LS_cmp_exchange: {
2573 // Check the signatures.
2574 #ifdef ASSERT
2575 assert(rtype == type, "CAS must return the expected type");
2576 assert(sig->count() == 4, "CAS has 4 arguments");
2577 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2578 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2579 #endif // ASSERT
2580 break;
2581 }
2582 default:
2583 ShouldNotReachHere();
2584 }
2585 }
2586 #endif //PRODUCT
2587
2588 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2589
2590 // Get arguments:
2591 Node* receiver = NULL;
2592 Node* base = NULL;
2593 Node* offset = NULL;
2594 Node* oldval = NULL;
2595 Node* newval = NULL;
2596 switch(kind) {
2597 case LS_cmp_swap:
2598 case LS_cmp_swap_weak:
2599 case LS_cmp_exchange: {
2600 const bool two_slot_type = type2size[type] == 2;
2601 receiver = argument(0); // type: oop
2602 base = argument(1); // type: oop
2603 offset = argument(2); // type: long
2604 oldval = argument(4); // type: oop, int, or long
2605 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2606 break;
2607 }
2608 case LS_get_add:
2609 case LS_get_set: {
2610 receiver = argument(0); // type: oop
2611 base = argument(1); // type: oop
2612 offset = argument(2); // type: long
2613 oldval = NULL;
2614 newval = argument(4); // type: oop, int, or long
2615 break;
2616 }
2617 default:
2618 ShouldNotReachHere();
2619 }
2620
2621 // Build field offset expression.
2622 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2623 // to be plain byte offsets, which are also the same as those accepted
2624 // by oopDesc::field_addr.
2625 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2626 // 32-bit machines ignore the high half of long offsets
2627 offset = ConvL2X(offset);
2628 Node* adr = make_unsafe_address(base, offset, type, false);
2629 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2630
2631 Compile::AliasType* alias_type = C->alias_type(adr_type);
2632 BasicType bt = alias_type->basic_type();
2633 if (bt != T_ILLEGAL &&
2634 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2635 // Don't intrinsify mismatched object accesses.
2636 return false;
2637 }
2638
2639 // For CAS, unlike inline_unsafe_access, there seems no point in
2640 // trying to refine types. Just use the coarse types here.
2641 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2642 const Type *value_type = Type::get_const_basic_type(type);
2643
2644 switch (kind) {
2645 case LS_get_set:
2646 case LS_cmp_exchange: {
2647 if (type == T_OBJECT) {
2648 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2649 if (tjp != NULL) {
2650 value_type = tjp;
2651 }
2652 }
2653 break;
2654 }
2655 case LS_cmp_swap:
2656 case LS_cmp_swap_weak:
2657 case LS_get_add:
2658 break;
2659 default:
2660 ShouldNotReachHere();
2661 }
2662
2663 // Null check receiver.
2664 receiver = null_check(receiver);
2665 if (stopped()) {
2666 return true;
2667 }
2668
2669 int alias_idx = C->get_alias_index(adr_type);
2670
2671 if (type == T_OBJECT || type == T_ARRAY) {
2672 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2673
2674 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2675 // could be delayed during Parse (for example, in adjust_map_after_if()).
2676 // Execute transformation here to avoid barrier generation in such case.
2677 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2678 newval = _gvn.makecon(TypePtr::NULL_PTR);
2679
2680 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2681 // Refine the value to a null constant, when it is known to be null
2682 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2683 }
2684 }
2685
2686 Node* result = NULL;
2687 switch (kind) {
2688 case LS_cmp_exchange: {
2689 result = access_atomic_cmpxchg_val_at(control(), base, adr, adr_type, alias_idx,
2690 oldval, newval, value_type, type, decorators);
2691 break;
2692 }
2693 case LS_cmp_swap_weak:
2694 decorators |= C2_WEAK_CMPXCHG;
2695 case LS_cmp_swap: {
2696 result = access_atomic_cmpxchg_bool_at(control(), base, adr, adr_type, alias_idx,
2697 oldval, newval, value_type, type, decorators);
2698 break;
2699 }
2700 case LS_get_set: {
2701 result = access_atomic_xchg_at(control(), base, adr, adr_type, alias_idx,
2702 newval, value_type, type, decorators);
2703 break;
2704 }
2705 case LS_get_add: {
2706 result = access_atomic_add_at(control(), base, adr, adr_type, alias_idx,
2707 newval, value_type, type, decorators);
2708 break;
2709 }
2710 default:
2711 ShouldNotReachHere();
2712 }
2713
2714 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2715 set_result(result);
2716 return true;
2717 }
2718
2719 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2720 // Regardless of form, don't allow previous ld/st to move down,
2721 // then issue acquire, release, or volatile mem_bar.
2722 insert_mem_bar(Op_MemBarCPUOrder);
2723 switch(id) {
2724 case vmIntrinsics::_loadFence:
2725 insert_mem_bar(Op_LoadFence);
2726 return true;
2727 case vmIntrinsics::_storeFence:
2728 insert_mem_bar(Op_StoreFence);
2729 return true;
2730 case vmIntrinsics::_fullFence:
2731 insert_mem_bar(Op_MemBarVolatile);
2732 return true;
2733 default:
2734 fatal_unexpected_iid(id);
2735 return false;
2736 }
2737 }
2738
2739 bool LibraryCallKit::inline_onspinwait() {
2740 insert_mem_bar(Op_OnSpinWait);
2741 return true;
2742 }
2743
2744 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2745 if (!kls->is_Con()) {
2746 return true;
2747 }
2748 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2749 if (klsptr == NULL) {
2750 return true;
2751 }
2752 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2753 // don't need a guard for a klass that is already initialized
2754 return !ik->is_initialized();
2755 }
2756
2757 //----------------------------inline_unsafe_allocate---------------------------
2758 // public native Object Unsafe.allocateInstance(Class<?> cls);
2759 bool LibraryCallKit::inline_unsafe_allocate() {
2760 if (callee()->is_static()) return false; // caller must have the capability!
2761
2762 null_check_receiver(); // null-check, then ignore
2763 Node* cls = null_check(argument(1));
2764 if (stopped()) return true;
2765
2766 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2767 kls = null_check(kls);
2768 if (stopped()) return true; // argument was like int.class
2769
2770 Node* test = NULL;
2771 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2772 // Note: The argument might still be an illegal value like
2773 // Serializable.class or Object[].class. The runtime will handle it.
2774 // But we must make an explicit check for initialization.
2775 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2776 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2777 // can generate code to load it as unsigned byte.
2778 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2779 Node* bits = intcon(InstanceKlass::fully_initialized);
2780 test = _gvn.transform(new SubINode(inst, bits));
2781 // The 'test' is non-zero if we need to take a slow path.
2782 }
2783
2784 Node* obj = new_instance(kls, test);
2785 set_result(obj);
2786 return true;
2787 }
2788
2789 //------------------------inline_native_time_funcs--------------
2790 // inline code for System.currentTimeMillis() and System.nanoTime()
2791 // these have the same type and signature
2792 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2793 const TypeFunc* tf = OptoRuntime::void_long_Type();
2794 const TypePtr* no_memory_effects = NULL;
2795 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2796 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2797 #ifdef ASSERT
2798 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2799 assert(value_top == top(), "second value must be top");
2800 #endif
2801 set_result(value);
2802 return true;
2803 }
2804
2805 #ifdef JFR_HAVE_INTRINSICS
2806
2807 /*
2808 * oop -> myklass
2809 * myklass->trace_id |= USED
2810 * return myklass->trace_id & ~0x3
2811 */
2812 bool LibraryCallKit::inline_native_classID() {
2813 Node* cls = null_check(argument(0), T_OBJECT);
2814 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2815 kls = null_check(kls, T_OBJECT);
2816
2817 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2818 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2819 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2820
2821 Node* clsused = longcon(0x01l); // set the class bit
2822 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2823 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2824 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2825
2826 #ifdef TRACE_ID_META_BITS
2827 Node* mbits = longcon(~TRACE_ID_META_BITS);
2828 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2829 #endif
2830 #ifdef TRACE_ID_SHIFT
2831 Node* cbits = intcon(TRACE_ID_SHIFT);
2832 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2833 #endif
2834
2835 set_result(tvalue);
2836 return true;
2837
2838 }
2839
2840 bool LibraryCallKit::inline_native_getEventWriter() {
2841 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2842
2843 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2844 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)
2845 );
2846
2847 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2848
2849 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2850 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2851
2852 IfNode* iff_jobj_null =
2853 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2854
2855 enum { _normal_path = 1,
2856 _null_path = 2,
2857 PATH_LIMIT };
2858
2859 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2860 PhiNode* result_val = new PhiNode(result_rgn, TypePtr::BOTTOM);
2861
2862 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2863 result_rgn->init_req(_null_path, jobj_is_null);
2864 result_val->init_req(_null_path, null());
2865
2866 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2867 result_rgn->init_req(_normal_path, jobj_is_not_null);
2868
2869 Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
2870 result_val->init_req(_normal_path, res);
2871
2872 set_result(result_rgn, result_val);
2873
2874 return true;
2875 }
2876
2877 #endif // JFR_HAVE_INTRINSICS
2878
2879 //------------------------inline_native_currentThread------------------
2880 bool LibraryCallKit::inline_native_currentThread() {
2881 Node* junk = NULL;
2882 set_result(generate_current_thread(junk));
2883 return true;
2884 }
2885
2886 //------------------------inline_native_isInterrupted------------------
2887 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
2888 bool LibraryCallKit::inline_native_isInterrupted() {
2889 // Add a fast path to t.isInterrupted(clear_int):
2890 // (t == Thread.current() &&
2891 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
2892 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
2893 // So, in the common case that the interrupt bit is false,
2894 // we avoid making a call into the VM. Even if the interrupt bit
2895 // is true, if the clear_int argument is false, we avoid the VM call.
2896 // However, if the receiver is not currentThread, we must call the VM,
2897 // because there must be some locking done around the operation.
2898
2899 // We only go to the fast case code if we pass two guards.
2900 // Paths which do not pass are accumulated in the slow_region.
2901
2902 enum {
2903 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
2904 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
2905 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
2906 PATH_LIMIT
2907 };
2908
2909 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
2910 // out of the function.
2911 insert_mem_bar(Op_MemBarCPUOrder);
2912
2913 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2914 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
2915
2916 RegionNode* slow_region = new RegionNode(1);
2917 record_for_igvn(slow_region);
2918
2919 // (a) Receiving thread must be the current thread.
2920 Node* rec_thr = argument(0);
2921 Node* tls_ptr = NULL;
2922 Node* cur_thr = generate_current_thread(tls_ptr);
2923 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
2924 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
2925
2926 generate_slow_guard(bol_thr, slow_region);
2927
2928 // (b) Interrupt bit on TLS must be false.
2929 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
2930 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
2931 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
2932
2933 // Set the control input on the field _interrupted read to prevent it floating up.
2934 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
2935 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
2936 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
2937
2938 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2939
2940 // First fast path: if (!TLS._interrupted) return false;
2941 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
2942 result_rgn->init_req(no_int_result_path, false_bit);
2943 result_val->init_req(no_int_result_path, intcon(0));
2944
2945 // drop through to next case
2946 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
2947
2948 #ifndef _WINDOWS
2949 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
2950 Node* clr_arg = argument(1);
2951 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
2952 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
2953 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
2954
2955 // Second fast path: ... else if (!clear_int) return true;
2956 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
2957 result_rgn->init_req(no_clear_result_path, false_arg);
2958 result_val->init_req(no_clear_result_path, intcon(1));
2959
2960 // drop through to next case
2961 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
2962 #else
2963 // To return true on Windows you must read the _interrupted field
2964 // and check the event state i.e. take the slow path.
2965 #endif // _WINDOWS
2966
2967 // (d) Otherwise, go to the slow path.
2968 slow_region->add_req(control());
2969 set_control( _gvn.transform(slow_region));
2970
2971 if (stopped()) {
2972 // There is no slow path.
2973 result_rgn->init_req(slow_result_path, top());
2974 result_val->init_req(slow_result_path, top());
2975 } else {
2976 // non-virtual because it is a private non-static
2977 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
2978
2979 Node* slow_val = set_results_for_java_call(slow_call);
2980 // this->control() comes from set_results_for_java_call
2981
2982 Node* fast_io = slow_call->in(TypeFunc::I_O);
2983 Node* fast_mem = slow_call->in(TypeFunc::Memory);
2984
2985 // These two phis are pre-filled with copies of of the fast IO and Memory
2986 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
2987 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
2988
2989 result_rgn->init_req(slow_result_path, control());
2990 result_io ->init_req(slow_result_path, i_o());
2991 result_mem->init_req(slow_result_path, reset_memory());
2992 result_val->init_req(slow_result_path, slow_val);
2993
2994 set_all_memory(_gvn.transform(result_mem));
2995 set_i_o( _gvn.transform(result_io));
2996 }
2997
2998 C->set_has_split_ifs(true); // Has chance for split-if optimization
2999 set_result(result_rgn, result_val);
3000 return true;
3001 }
3002
3003 //---------------------------load_mirror_from_klass----------------------------
3004 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3005 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3006 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3007 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3008 // mirror = ((OopHandle)mirror)->resolve();
3009 return make_load(NULL, load, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3010 }
3011
3012 //-----------------------load_klass_from_mirror_common-------------------------
3013 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3014 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3015 // and branch to the given path on the region.
3016 // If never_see_null, take an uncommon trap on null, so we can optimistically
3017 // compile for the non-null case.
3018 // If the region is NULL, force never_see_null = true.
3019 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3020 bool never_see_null,
3021 RegionNode* region,
3022 int null_path,
3023 int offset) {
3024 if (region == NULL) never_see_null = true;
3025 Node* p = basic_plus_adr(mirror, offset);
3026 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3027 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3028 Node* null_ctl = top();
3029 kls = null_check_oop(kls, &null_ctl, never_see_null);
3030 if (region != NULL) {
3031 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3032 region->init_req(null_path, null_ctl);
3033 } else {
3034 assert(null_ctl == top(), "no loose ends");
3035 }
3036 return kls;
3037 }
3038
3039 //--------------------(inline_native_Class_query helpers)---------------------
3040 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3041 // Fall through if (mods & mask) == bits, take the guard otherwise.
3042 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3043 // Branch around if the given klass has the given modifier bit set.
3044 // Like generate_guard, adds a new path onto the region.
3045 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3046 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3047 Node* mask = intcon(modifier_mask);
3048 Node* bits = intcon(modifier_bits);
3049 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3050 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3051 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3052 return generate_fair_guard(bol, region);
3053 }
3054 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3055 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3056 }
3057
3058 //-------------------------inline_native_Class_query-------------------
3059 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3060 const Type* return_type = TypeInt::BOOL;
3061 Node* prim_return_value = top(); // what happens if it's a primitive class?
3062 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3063 bool expect_prim = false; // most of these guys expect to work on refs
3064
3065 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3066
3067 Node* mirror = argument(0);
3068 Node* obj = top();
3069
3070 switch (id) {
3071 case vmIntrinsics::_isInstance:
3072 // nothing is an instance of a primitive type
3073 prim_return_value = intcon(0);
3074 obj = argument(1);
3075 break;
3076 case vmIntrinsics::_getModifiers:
3077 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3078 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3079 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3080 break;
3081 case vmIntrinsics::_isInterface:
3082 prim_return_value = intcon(0);
3083 break;
3084 case vmIntrinsics::_isArray:
3085 prim_return_value = intcon(0);
3086 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3087 break;
3088 case vmIntrinsics::_isPrimitive:
3089 prim_return_value = intcon(1);
3090 expect_prim = true; // obviously
3091 break;
3092 case vmIntrinsics::_getSuperclass:
3093 prim_return_value = null();
3094 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3095 break;
3096 case vmIntrinsics::_getClassAccessFlags:
3097 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3098 return_type = TypeInt::INT; // not bool! 6297094
3099 break;
3100 default:
3101 fatal_unexpected_iid(id);
3102 break;
3103 }
3104
3105 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3106 if (mirror_con == NULL) return false; // cannot happen?
3107
3108 #ifndef PRODUCT
3109 if (C->print_intrinsics() || C->print_inlining()) {
3110 ciType* k = mirror_con->java_mirror_type();
3111 if (k) {
3112 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3113 k->print_name();
3114 tty->cr();
3115 }
3116 }
3117 #endif
3118
3119 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3120 RegionNode* region = new RegionNode(PATH_LIMIT);
3121 record_for_igvn(region);
3122 PhiNode* phi = new PhiNode(region, return_type);
3123
3124 // The mirror will never be null of Reflection.getClassAccessFlags, however
3125 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3126 // if it is. See bug 4774291.
3127
3128 // For Reflection.getClassAccessFlags(), the null check occurs in
3129 // the wrong place; see inline_unsafe_access(), above, for a similar
3130 // situation.
3131 mirror = null_check(mirror);
3132 // If mirror or obj is dead, only null-path is taken.
3133 if (stopped()) return true;
3134
3135 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3136
3137 // Now load the mirror's klass metaobject, and null-check it.
3138 // Side-effects region with the control path if the klass is null.
3139 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3140 // If kls is null, we have a primitive mirror.
3141 phi->init_req(_prim_path, prim_return_value);
3142 if (stopped()) { set_result(region, phi); return true; }
3143 bool safe_for_replace = (region->in(_prim_path) == top());
3144
3145 Node* p; // handy temp
3146 Node* null_ctl;
3147
3148 // Now that we have the non-null klass, we can perform the real query.
3149 // For constant classes, the query will constant-fold in LoadNode::Value.
3150 Node* query_value = top();
3151 switch (id) {
3152 case vmIntrinsics::_isInstance:
3153 // nothing is an instance of a primitive type
3154 query_value = gen_instanceof(obj, kls, safe_for_replace);
3155 break;
3156
3157 case vmIntrinsics::_getModifiers:
3158 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3159 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3160 break;
3161
3162 case vmIntrinsics::_isInterface:
3163 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3164 if (generate_interface_guard(kls, region) != NULL)
3165 // A guard was added. If the guard is taken, it was an interface.
3166 phi->add_req(intcon(1));
3167 // If we fall through, it's a plain class.
3168 query_value = intcon(0);
3169 break;
3170
3171 case vmIntrinsics::_isArray:
3172 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3173 if (generate_array_guard(kls, region) != NULL)
3174 // A guard was added. If the guard is taken, it was an array.
3175 phi->add_req(intcon(1));
3176 // If we fall through, it's a plain class.
3177 query_value = intcon(0);
3178 break;
3179
3180 case vmIntrinsics::_isPrimitive:
3181 query_value = intcon(0); // "normal" path produces false
3182 break;
3183
3184 case vmIntrinsics::_getSuperclass:
3185 // The rules here are somewhat unfortunate, but we can still do better
3186 // with random logic than with a JNI call.
3187 // Interfaces store null or Object as _super, but must report null.
3188 // Arrays store an intermediate super as _super, but must report Object.
3189 // Other types can report the actual _super.
3190 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3191 if (generate_interface_guard(kls, region) != NULL)
3192 // A guard was added. If the guard is taken, it was an interface.
3193 phi->add_req(null());
3194 if (generate_array_guard(kls, region) != NULL)
3195 // A guard was added. If the guard is taken, it was an array.
3196 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3197 // If we fall through, it's a plain class. Get its _super.
3198 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3199 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3200 null_ctl = top();
3201 kls = null_check_oop(kls, &null_ctl);
3202 if (null_ctl != top()) {
3203 // If the guard is taken, Object.superClass is null (both klass and mirror).
3204 region->add_req(null_ctl);
3205 phi ->add_req(null());
3206 }
3207 if (!stopped()) {
3208 query_value = load_mirror_from_klass(kls);
3209 }
3210 break;
3211
3212 case vmIntrinsics::_getClassAccessFlags:
3213 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3214 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3215 break;
3216
3217 default:
3218 fatal_unexpected_iid(id);
3219 break;
3220 }
3221
3222 // Fall-through is the normal case of a query to a real class.
3223 phi->init_req(1, query_value);
3224 region->init_req(1, control());
3225
3226 C->set_has_split_ifs(true); // Has chance for split-if optimization
3227 set_result(region, phi);
3228 return true;
3229 }
3230
3231 //-------------------------inline_Class_cast-------------------
3232 bool LibraryCallKit::inline_Class_cast() {
3233 Node* mirror = argument(0); // Class
3234 Node* obj = argument(1);
3235 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3236 if (mirror_con == NULL) {
3237 return false; // dead path (mirror->is_top()).
3238 }
3239 if (obj == NULL || obj->is_top()) {
3240 return false; // dead path
3241 }
3242 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3243
3244 // First, see if Class.cast() can be folded statically.
3245 // java_mirror_type() returns non-null for compile-time Class constants.
3246 ciType* tm = mirror_con->java_mirror_type();
3247 if (tm != NULL && tm->is_klass() &&
3248 tp != NULL && tp->klass() != NULL) {
3249 if (!tp->klass()->is_loaded()) {
3250 // Don't use intrinsic when class is not loaded.
3251 return false;
3252 } else {
3253 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3254 if (static_res == Compile::SSC_always_true) {
3255 // isInstance() is true - fold the code.
3256 set_result(obj);
3257 return true;
3258 } else if (static_res == Compile::SSC_always_false) {
3259 // Don't use intrinsic, have to throw ClassCastException.
3260 // If the reference is null, the non-intrinsic bytecode will
3261 // be optimized appropriately.
3262 return false;
3263 }
3264 }
3265 }
3266
3267 // Bailout intrinsic and do normal inlining if exception path is frequent.
3268 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3269 return false;
3270 }
3271
3272 // Generate dynamic checks.
3273 // Class.cast() is java implementation of _checkcast bytecode.
3274 // Do checkcast (Parse::do_checkcast()) optimizations here.
3275
3276 mirror = null_check(mirror);
3277 // If mirror is dead, only null-path is taken.
3278 if (stopped()) {
3279 return true;
3280 }
3281
3282 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3283 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3284 RegionNode* region = new RegionNode(PATH_LIMIT);
3285 record_for_igvn(region);
3286
3287 // Now load the mirror's klass metaobject, and null-check it.
3288 // If kls is null, we have a primitive mirror and
3289 // nothing is an instance of a primitive type.
3290 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3291
3292 Node* res = top();
3293 if (!stopped()) {
3294 Node* bad_type_ctrl = top();
3295 // Do checkcast optimizations.
3296 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3297 region->init_req(_bad_type_path, bad_type_ctrl);
3298 }
3299 if (region->in(_prim_path) != top() ||
3300 region->in(_bad_type_path) != top()) {
3301 // Let Interpreter throw ClassCastException.
3302 PreserveJVMState pjvms(this);
3303 set_control(_gvn.transform(region));
3304 uncommon_trap(Deoptimization::Reason_intrinsic,
3305 Deoptimization::Action_maybe_recompile);
3306 }
3307 if (!stopped()) {
3308 set_result(res);
3309 }
3310 return true;
3311 }
3312
3313
3314 //--------------------------inline_native_subtype_check------------------------
3315 // This intrinsic takes the JNI calls out of the heart of
3316 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3317 bool LibraryCallKit::inline_native_subtype_check() {
3318 // Pull both arguments off the stack.
3319 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3320 args[0] = argument(0);
3321 args[1] = argument(1);
3322 Node* klasses[2]; // corresponding Klasses: superk, subk
3323 klasses[0] = klasses[1] = top();
3324
3325 enum {
3326 // A full decision tree on {superc is prim, subc is prim}:
3327 _prim_0_path = 1, // {P,N} => false
3328 // {P,P} & superc!=subc => false
3329 _prim_same_path, // {P,P} & superc==subc => true
3330 _prim_1_path, // {N,P} => false
3331 _ref_subtype_path, // {N,N} & subtype check wins => true
3332 _both_ref_path, // {N,N} & subtype check loses => false
3333 PATH_LIMIT
3334 };
3335
3336 RegionNode* region = new RegionNode(PATH_LIMIT);
3337 Node* phi = new PhiNode(region, TypeInt::BOOL);
3338 record_for_igvn(region);
3339
3340 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3341 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3342 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3343
3344 // First null-check both mirrors and load each mirror's klass metaobject.
3345 int which_arg;
3346 for (which_arg = 0; which_arg <= 1; which_arg++) {
3347 Node* arg = args[which_arg];
3348 arg = null_check(arg);
3349 if (stopped()) break;
3350 args[which_arg] = arg;
3351
3352 Node* p = basic_plus_adr(arg, class_klass_offset);
3353 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3354 klasses[which_arg] = _gvn.transform(kls);
3355 }
3356
3357 // Having loaded both klasses, test each for null.
3358 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3359 for (which_arg = 0; which_arg <= 1; which_arg++) {
3360 Node* kls = klasses[which_arg];
3361 Node* null_ctl = top();
3362 kls = null_check_oop(kls, &null_ctl, never_see_null);
3363 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3364 region->init_req(prim_path, null_ctl);
3365 if (stopped()) break;
3366 klasses[which_arg] = kls;
3367 }
3368
3369 if (!stopped()) {
3370 // now we have two reference types, in klasses[0..1]
3371 Node* subk = klasses[1]; // the argument to isAssignableFrom
3372 Node* superk = klasses[0]; // the receiver
3373 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3374 // now we have a successful reference subtype check
3375 region->set_req(_ref_subtype_path, control());
3376 }
3377
3378 // If both operands are primitive (both klasses null), then
3379 // we must return true when they are identical primitives.
3380 // It is convenient to test this after the first null klass check.
3381 set_control(region->in(_prim_0_path)); // go back to first null check
3382 if (!stopped()) {
3383 // Since superc is primitive, make a guard for the superc==subc case.
3384 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3385 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3386 generate_guard(bol_eq, region, PROB_FAIR);
3387 if (region->req() == PATH_LIMIT+1) {
3388 // A guard was added. If the added guard is taken, superc==subc.
3389 region->swap_edges(PATH_LIMIT, _prim_same_path);
3390 region->del_req(PATH_LIMIT);
3391 }
3392 region->set_req(_prim_0_path, control()); // Not equal after all.
3393 }
3394
3395 // these are the only paths that produce 'true':
3396 phi->set_req(_prim_same_path, intcon(1));
3397 phi->set_req(_ref_subtype_path, intcon(1));
3398
3399 // pull together the cases:
3400 assert(region->req() == PATH_LIMIT, "sane region");
3401 for (uint i = 1; i < region->req(); i++) {
3402 Node* ctl = region->in(i);
3403 if (ctl == NULL || ctl == top()) {
3404 region->set_req(i, top());
3405 phi ->set_req(i, top());
3406 } else if (phi->in(i) == NULL) {
3407 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3408 }
3409 }
3410
3411 set_control(_gvn.transform(region));
3412 set_result(_gvn.transform(phi));
3413 return true;
3414 }
3415
3416 //---------------------generate_array_guard_common------------------------
3417 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3418 bool obj_array, bool not_array) {
3419
3420 if (stopped()) {
3421 return NULL;
3422 }
3423
3424 // If obj_array/non_array==false/false:
3425 // Branch around if the given klass is in fact an array (either obj or prim).
3426 // If obj_array/non_array==false/true:
3427 // Branch around if the given klass is not an array klass of any kind.
3428 // If obj_array/non_array==true/true:
3429 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3430 // If obj_array/non_array==true/false:
3431 // Branch around if the kls is an oop array (Object[] or subtype)
3432 //
3433 // Like generate_guard, adds a new path onto the region.
3434 jint layout_con = 0;
3435 Node* layout_val = get_layout_helper(kls, layout_con);
3436 if (layout_val == NULL) {
3437 bool query = (obj_array
3438 ? Klass::layout_helper_is_objArray(layout_con)
3439 : Klass::layout_helper_is_array(layout_con));
3440 if (query == not_array) {
3441 return NULL; // never a branch
3442 } else { // always a branch
3443 Node* always_branch = control();
3444 if (region != NULL)
3445 region->add_req(always_branch);
3446 set_control(top());
3447 return always_branch;
3448 }
3449 }
3450 // Now test the correct condition.
3451 jint nval = (obj_array
3452 ? (jint)(Klass::_lh_array_tag_type_value
3453 << Klass::_lh_array_tag_shift)
3454 : Klass::_lh_neutral_value);
3455 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3456 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3457 // invert the test if we are looking for a non-array
3458 if (not_array) btest = BoolTest(btest).negate();
3459 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3460 return generate_fair_guard(bol, region);
3461 }
3462
3463
3464 //-----------------------inline_native_newArray--------------------------
3465 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3466 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3467 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3468 Node* mirror;
3469 Node* count_val;
3470 if (uninitialized) {
3471 mirror = argument(1);
3472 count_val = argument(2);
3473 } else {
3474 mirror = argument(0);
3475 count_val = argument(1);
3476 }
3477
3478 mirror = null_check(mirror);
3479 // If mirror or obj is dead, only null-path is taken.
3480 if (stopped()) return true;
3481
3482 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3483 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3484 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3485 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3486 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3487
3488 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3489 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3490 result_reg, _slow_path);
3491 Node* normal_ctl = control();
3492 Node* no_array_ctl = result_reg->in(_slow_path);
3493
3494 // Generate code for the slow case. We make a call to newArray().
3495 set_control(no_array_ctl);
3496 if (!stopped()) {
3497 // Either the input type is void.class, or else the
3498 // array klass has not yet been cached. Either the
3499 // ensuing call will throw an exception, or else it
3500 // will cache the array klass for next time.
3501 PreserveJVMState pjvms(this);
3502 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3503 Node* slow_result = set_results_for_java_call(slow_call);
3504 // this->control() comes from set_results_for_java_call
3505 result_reg->set_req(_slow_path, control());
3506 result_val->set_req(_slow_path, slow_result);
3507 result_io ->set_req(_slow_path, i_o());
3508 result_mem->set_req(_slow_path, reset_memory());
3509 }
3510
3511 set_control(normal_ctl);
3512 if (!stopped()) {
3513 // Normal case: The array type has been cached in the java.lang.Class.
3514 // The following call works fine even if the array type is polymorphic.
3515 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3516 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3517 result_reg->init_req(_normal_path, control());
3518 result_val->init_req(_normal_path, obj);
3519 result_io ->init_req(_normal_path, i_o());
3520 result_mem->init_req(_normal_path, reset_memory());
3521
3522 if (uninitialized) {
3523 // Mark the allocation so that zeroing is skipped
3524 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3525 alloc->maybe_set_complete(&_gvn);
3526 }
3527 }
3528
3529 // Return the combined state.
3530 set_i_o( _gvn.transform(result_io) );
3531 set_all_memory( _gvn.transform(result_mem));
3532
3533 C->set_has_split_ifs(true); // Has chance for split-if optimization
3534 set_result(result_reg, result_val);
3535 return true;
3536 }
3537
3538 //----------------------inline_native_getLength--------------------------
3539 // public static native int java.lang.reflect.Array.getLength(Object array);
3540 bool LibraryCallKit::inline_native_getLength() {
3541 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3542
3543 Node* array = null_check(argument(0));
3544 // If array is dead, only null-path is taken.
3545 if (stopped()) return true;
3546
3547 // Deoptimize if it is a non-array.
3548 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3549
3550 if (non_array != NULL) {
3551 PreserveJVMState pjvms(this);
3552 set_control(non_array);
3553 uncommon_trap(Deoptimization::Reason_intrinsic,
3554 Deoptimization::Action_maybe_recompile);
3555 }
3556
3557 // If control is dead, only non-array-path is taken.
3558 if (stopped()) return true;
3559
3560 // The works fine even if the array type is polymorphic.
3561 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3562 Node* result = load_array_length(array);
3563
3564 C->set_has_split_ifs(true); // Has chance for split-if optimization
3565 set_result(result);
3566 return true;
3567 }
3568
3569 //------------------------inline_array_copyOf----------------------------
3570 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3571 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3572 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3573 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3574
3575 // Get the arguments.
3576 Node* original = argument(0);
3577 Node* start = is_copyOfRange? argument(1): intcon(0);
3578 Node* end = is_copyOfRange? argument(2): argument(1);
3579 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3580
3581 Node* newcopy = NULL;
3582
3583 // Set the original stack and the reexecute bit for the interpreter to reexecute
3584 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3585 { PreserveReexecuteState preexecs(this);
3586 jvms()->set_should_reexecute(true);
3587
3588 array_type_mirror = null_check(array_type_mirror);
3589 original = null_check(original);
3590
3591 // Check if a null path was taken unconditionally.
3592 if (stopped()) return true;
3593
3594 Node* orig_length = load_array_length(original);
3595
3596 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3597 klass_node = null_check(klass_node);
3598
3599 RegionNode* bailout = new RegionNode(1);
3600 record_for_igvn(bailout);
3601
3602 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3603 // Bail out if that is so.
3604 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3605 if (not_objArray != NULL) {
3606 // Improve the klass node's type from the new optimistic assumption:
3607 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3608 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3609 Node* cast = new CastPPNode(klass_node, akls);
3610 cast->init_req(0, control());
3611 klass_node = _gvn.transform(cast);
3612 }
3613
3614 // Bail out if either start or end is negative.
3615 generate_negative_guard(start, bailout, &start);
3616 generate_negative_guard(end, bailout, &end);
3617
3618 Node* length = end;
3619 if (_gvn.type(start) != TypeInt::ZERO) {
3620 length = _gvn.transform(new SubINode(end, start));
3621 }
3622
3623 // Bail out if length is negative.
3624 // Without this the new_array would throw
3625 // NegativeArraySizeException but IllegalArgumentException is what
3626 // should be thrown
3627 generate_negative_guard(length, bailout, &length);
3628
3629 if (bailout->req() > 1) {
3630 PreserveJVMState pjvms(this);
3631 set_control(_gvn.transform(bailout));
3632 uncommon_trap(Deoptimization::Reason_intrinsic,
3633 Deoptimization::Action_maybe_recompile);
3634 }
3635
3636 if (!stopped()) {
3637 // How many elements will we copy from the original?
3638 // The answer is MinI(orig_length - start, length).
3639 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3640 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3641
3642 // Generate a direct call to the right arraycopy function(s).
3643 // We know the copy is disjoint but we might not know if the
3644 // oop stores need checking.
3645 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3646 // This will fail a store-check if x contains any non-nulls.
3647
3648 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3649 // loads/stores but it is legal only if we're sure the
3650 // Arrays.copyOf would succeed. So we need all input arguments
3651 // to the copyOf to be validated, including that the copy to the
3652 // new array won't trigger an ArrayStoreException. That subtype
3653 // check can be optimized if we know something on the type of
3654 // the input array from type speculation.
3655 if (_gvn.type(klass_node)->singleton()) {
3656 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3657 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3658
3659 int test = C->static_subtype_check(superk, subk);
3660 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3661 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3662 if (t_original->speculative_type() != NULL) {
3663 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3664 }
3665 }
3666 }
3667
3668 bool validated = false;
3669 // Reason_class_check rather than Reason_intrinsic because we
3670 // want to intrinsify even if this traps.
3671 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3672 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3673 klass_node);
3674
3675 if (not_subtype_ctrl != top()) {
3676 PreserveJVMState pjvms(this);
3677 set_control(not_subtype_ctrl);
3678 uncommon_trap(Deoptimization::Reason_class_check,
3679 Deoptimization::Action_make_not_entrant);
3680 assert(stopped(), "Should be stopped");
3681 }
3682 validated = true;
3683 }
3684
3685 if (!stopped()) {
3686 newcopy = new_array(klass_node, length, 0); // no arguments to push
3687
3688 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3689 load_object_klass(original), klass_node);
3690 if (!is_copyOfRange) {
3691 ac->set_copyof(validated);
3692 } else {
3693 ac->set_copyofrange(validated);
3694 }
3695 Node* n = _gvn.transform(ac);
3696 if (n == ac) {
3697 ac->connect_outputs(this);
3698 } else {
3699 assert(validated, "shouldn't transform if all arguments not validated");
3700 set_all_memory(n);
3701 }
3702 }
3703 }
3704 } // original reexecute is set back here
3705
3706 C->set_has_split_ifs(true); // Has chance for split-if optimization
3707 if (!stopped()) {
3708 set_result(newcopy);
3709 }
3710 return true;
3711 }
3712
3713
3714 //----------------------generate_virtual_guard---------------------------
3715 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3716 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3717 RegionNode* slow_region) {
3718 ciMethod* method = callee();
3719 int vtable_index = method->vtable_index();
3720 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3721 "bad index %d", vtable_index);
3722 // Get the Method* out of the appropriate vtable entry.
3723 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3724 vtable_index*vtableEntry::size_in_bytes() +
3725 vtableEntry::method_offset_in_bytes();
3726 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3727 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3728
3729 // Compare the target method with the expected method (e.g., Object.hashCode).
3730 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3731
3732 Node* native_call = makecon(native_call_addr);
3733 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3734 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3735
3736 return generate_slow_guard(test_native, slow_region);
3737 }
3738
3739 //-----------------------generate_method_call----------------------------
3740 // Use generate_method_call to make a slow-call to the real
3741 // method if the fast path fails. An alternative would be to
3742 // use a stub like OptoRuntime::slow_arraycopy_Java.
3743 // This only works for expanding the current library call,
3744 // not another intrinsic. (E.g., don't use this for making an
3745 // arraycopy call inside of the copyOf intrinsic.)
3746 CallJavaNode*
3747 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3748 // When compiling the intrinsic method itself, do not use this technique.
3749 guarantee(callee() != C->method(), "cannot make slow-call to self");
3750
3751 ciMethod* method = callee();
3752 // ensure the JVMS we have will be correct for this call
3753 guarantee(method_id == method->intrinsic_id(), "must match");
3754
3755 const TypeFunc* tf = TypeFunc::make(method);
3756 CallJavaNode* slow_call;
3757 if (is_static) {
3758 assert(!is_virtual, "");
3759 slow_call = new CallStaticJavaNode(C, tf,
3760 SharedRuntime::get_resolve_static_call_stub(),
3761 method, bci());
3762 } else if (is_virtual) {
3763 null_check_receiver();
3764 int vtable_index = Method::invalid_vtable_index;
3765 if (UseInlineCaches) {
3766 // Suppress the vtable call
3767 } else {
3768 // hashCode and clone are not a miranda methods,
3769 // so the vtable index is fixed.
3770 // No need to use the linkResolver to get it.
3771 vtable_index = method->vtable_index();
3772 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3773 "bad index %d", vtable_index);
3774 }
3775 slow_call = new CallDynamicJavaNode(tf,
3776 SharedRuntime::get_resolve_virtual_call_stub(),
3777 method, vtable_index, bci());
3778 } else { // neither virtual nor static: opt_virtual
3779 null_check_receiver();
3780 slow_call = new CallStaticJavaNode(C, tf,
3781 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3782 method, bci());
3783 slow_call->set_optimized_virtual(true);
3784 }
3785 set_arguments_for_java_call(slow_call);
3786 set_edges_for_java_call(slow_call);
3787 return slow_call;
3788 }
3789
3790
3791 /**
3792 * Build special case code for calls to hashCode on an object. This call may
3793 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3794 * slightly different code.
3795 */
3796 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3797 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3798 assert(!(is_virtual && is_static), "either virtual, special, or static");
3799
3800 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3801
3802 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3803 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3804 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3805 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3806 Node* obj = NULL;
3807 if (!is_static) {
3808 // Check for hashing null object
3809 obj = null_check_receiver();
3810 if (stopped()) return true; // unconditionally null
3811 result_reg->init_req(_null_path, top());
3812 result_val->init_req(_null_path, top());
3813 } else {
3814 // Do a null check, and return zero if null.
3815 // System.identityHashCode(null) == 0
3816 obj = argument(0);
3817 Node* null_ctl = top();
3818 obj = null_check_oop(obj, &null_ctl);
3819 result_reg->init_req(_null_path, null_ctl);
3820 result_val->init_req(_null_path, _gvn.intcon(0));
3821 }
3822
3823 // Unconditionally null? Then return right away.
3824 if (stopped()) {
3825 set_control( result_reg->in(_null_path));
3826 if (!stopped())
3827 set_result(result_val->in(_null_path));
3828 return true;
3829 }
3830
3831 // We only go to the fast case code if we pass a number of guards. The
3832 // paths which do not pass are accumulated in the slow_region.
3833 RegionNode* slow_region = new RegionNode(1);
3834 record_for_igvn(slow_region);
3835
3836 // If this is a virtual call, we generate a funny guard. We pull out
3837 // the vtable entry corresponding to hashCode() from the target object.
3838 // If the target method which we are calling happens to be the native
3839 // Object hashCode() method, we pass the guard. We do not need this
3840 // guard for non-virtual calls -- the caller is known to be the native
3841 // Object hashCode().
3842 if (is_virtual) {
3843 // After null check, get the object's klass.
3844 Node* obj_klass = load_object_klass(obj);
3845 generate_virtual_guard(obj_klass, slow_region);
3846 }
3847
3848 // Get the header out of the object, use LoadMarkNode when available
3849 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3850 // The control of the load must be NULL. Otherwise, the load can move before
3851 // the null check after castPP removal.
3852 Node* no_ctrl = NULL;
3853 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3854
3855 // Test the header to see if it is unlocked.
3856 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
3857 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3858 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
3859 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3860 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3861
3862 generate_slow_guard(test_unlocked, slow_region);
3863
3864 // Get the hash value and check to see that it has been properly assigned.
3865 // We depend on hash_mask being at most 32 bits and avoid the use of
3866 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3867 // vm: see markOop.hpp.
3868 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
3869 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
3870 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3871 // This hack lets the hash bits live anywhere in the mark object now, as long
3872 // as the shift drops the relevant bits into the low 32 bits. Note that
3873 // Java spec says that HashCode is an int so there's no point in capturing
3874 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3875 hshifted_header = ConvX2I(hshifted_header);
3876 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3877
3878 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
3879 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3880 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3881
3882 generate_slow_guard(test_assigned, slow_region);
3883
3884 Node* init_mem = reset_memory();
3885 // fill in the rest of the null path:
3886 result_io ->init_req(_null_path, i_o());
3887 result_mem->init_req(_null_path, init_mem);
3888
3889 result_val->init_req(_fast_path, hash_val);
3890 result_reg->init_req(_fast_path, control());
3891 result_io ->init_req(_fast_path, i_o());
3892 result_mem->init_req(_fast_path, init_mem);
3893
3894 // Generate code for the slow case. We make a call to hashCode().
3895 set_control(_gvn.transform(slow_region));
3896 if (!stopped()) {
3897 // No need for PreserveJVMState, because we're using up the present state.
3898 set_all_memory(init_mem);
3899 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3900 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3901 Node* slow_result = set_results_for_java_call(slow_call);
3902 // this->control() comes from set_results_for_java_call
3903 result_reg->init_req(_slow_path, control());
3904 result_val->init_req(_slow_path, slow_result);
3905 result_io ->set_req(_slow_path, i_o());
3906 result_mem ->set_req(_slow_path, reset_memory());
3907 }
3908
3909 // Return the combined state.
3910 set_i_o( _gvn.transform(result_io) );
3911 set_all_memory( _gvn.transform(result_mem));
3912
3913 set_result(result_reg, result_val);
3914 return true;
3915 }
3916
3917 //---------------------------inline_native_getClass----------------------------
3918 // public final native Class<?> java.lang.Object.getClass();
3919 //
3920 // Build special case code for calls to getClass on an object.
3921 bool LibraryCallKit::inline_native_getClass() {
3922 Node* obj = null_check_receiver();
3923 if (stopped()) return true;
3924 set_result(load_mirror_from_klass(load_object_klass(obj)));
3925 return true;
3926 }
3927
3928 //-----------------inline_native_Reflection_getCallerClass---------------------
3929 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3930 //
3931 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3932 //
3933 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3934 // in that it must skip particular security frames and checks for
3935 // caller sensitive methods.
3936 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3937 #ifndef PRODUCT
3938 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3939 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3940 }
3941 #endif
3942
3943 if (!jvms()->has_method()) {
3944 #ifndef PRODUCT
3945 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3946 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
3947 }
3948 #endif
3949 return false;
3950 }
3951
3952 // Walk back up the JVM state to find the caller at the required
3953 // depth.
3954 JVMState* caller_jvms = jvms();
3955
3956 // Cf. JVM_GetCallerClass
3957 // NOTE: Start the loop at depth 1 because the current JVM state does
3958 // not include the Reflection.getCallerClass() frame.
3959 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
3960 ciMethod* m = caller_jvms->method();
3961 switch (n) {
3962 case 0:
3963 fatal("current JVM state does not include the Reflection.getCallerClass frame");
3964 break;
3965 case 1:
3966 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
3967 if (!m->caller_sensitive()) {
3968 #ifndef PRODUCT
3969 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3970 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
3971 }
3972 #endif
3973 return false; // bail-out; let JVM_GetCallerClass do the work
3974 }
3975 break;
3976 default:
3977 if (!m->is_ignored_by_security_stack_walk()) {
3978 // We have reached the desired frame; return the holder class.
3979 // Acquire method holder as java.lang.Class and push as constant.
3980 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
3981 ciInstance* caller_mirror = caller_klass->java_mirror();
3982 set_result(makecon(TypeInstPtr::make(caller_mirror)));
3983
3984 #ifndef PRODUCT
3985 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3986 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());
3987 tty->print_cr(" JVM state at this point:");
3988 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
3989 ciMethod* m = jvms()->of_depth(i)->method();
3990 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
3991 }
3992 }
3993 #endif
3994 return true;
3995 }
3996 break;
3997 }
3998 }
3999
4000 #ifndef PRODUCT
4001 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4002 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4003 tty->print_cr(" JVM state at this point:");
4004 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4005 ciMethod* m = jvms()->of_depth(i)->method();
4006 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4007 }
4008 }
4009 #endif
4010
4011 return false; // bail-out; let JVM_GetCallerClass do the work
4012 }
4013
4014 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4015 Node* arg = argument(0);
4016 Node* result = NULL;
4017
4018 switch (id) {
4019 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4020 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4021 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4022 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4023
4024 case vmIntrinsics::_doubleToLongBits: {
4025 // two paths (plus control) merge in a wood
4026 RegionNode *r = new RegionNode(3);
4027 Node *phi = new PhiNode(r, TypeLong::LONG);
4028
4029 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4030 // Build the boolean node
4031 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4032
4033 // Branch either way.
4034 // NaN case is less traveled, which makes all the difference.
4035 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4036 Node *opt_isnan = _gvn.transform(ifisnan);
4037 assert( opt_isnan->is_If(), "Expect an IfNode");
4038 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4039 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4040
4041 set_control(iftrue);
4042
4043 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4044 Node *slow_result = longcon(nan_bits); // return NaN
4045 phi->init_req(1, _gvn.transform( slow_result ));
4046 r->init_req(1, iftrue);
4047
4048 // Else fall through
4049 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4050 set_control(iffalse);
4051
4052 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4053 r->init_req(2, iffalse);
4054
4055 // Post merge
4056 set_control(_gvn.transform(r));
4057 record_for_igvn(r);
4058
4059 C->set_has_split_ifs(true); // Has chance for split-if optimization
4060 result = phi;
4061 assert(result->bottom_type()->isa_long(), "must be");
4062 break;
4063 }
4064
4065 case vmIntrinsics::_floatToIntBits: {
4066 // two paths (plus control) merge in a wood
4067 RegionNode *r = new RegionNode(3);
4068 Node *phi = new PhiNode(r, TypeInt::INT);
4069
4070 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4071 // Build the boolean node
4072 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4073
4074 // Branch either way.
4075 // NaN case is less traveled, which makes all the difference.
4076 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4077 Node *opt_isnan = _gvn.transform(ifisnan);
4078 assert( opt_isnan->is_If(), "Expect an IfNode");
4079 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4080 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4081
4082 set_control(iftrue);
4083
4084 static const jint nan_bits = 0x7fc00000;
4085 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4086 phi->init_req(1, _gvn.transform( slow_result ));
4087 r->init_req(1, iftrue);
4088
4089 // Else fall through
4090 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4091 set_control(iffalse);
4092
4093 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4094 r->init_req(2, iffalse);
4095
4096 // Post merge
4097 set_control(_gvn.transform(r));
4098 record_for_igvn(r);
4099
4100 C->set_has_split_ifs(true); // Has chance for split-if optimization
4101 result = phi;
4102 assert(result->bottom_type()->isa_int(), "must be");
4103 break;
4104 }
4105
4106 default:
4107 fatal_unexpected_iid(id);
4108 break;
4109 }
4110 set_result(_gvn.transform(result));
4111 return true;
4112 }
4113
4114 //----------------------inline_unsafe_copyMemory-------------------------
4115 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4116 bool LibraryCallKit::inline_unsafe_copyMemory() {
4117 if (callee()->is_static()) return false; // caller must have the capability!
4118 null_check_receiver(); // null-check receiver
4119 if (stopped()) return true;
4120
4121 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4122
4123 Node* src_ptr = argument(1); // type: oop
4124 Node* src_off = ConvL2X(argument(2)); // type: long
4125 Node* dst_ptr = argument(4); // type: oop
4126 Node* dst_off = ConvL2X(argument(5)); // type: long
4127 Node* size = ConvL2X(argument(7)); // type: long
4128
4129 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4130 "fieldOffset must be byte-scaled");
4131
4132 Node* src = make_unsafe_address(src_ptr, src_off);
4133 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4134
4135 // Conservatively insert a memory barrier on all memory slices.
4136 // Do not let writes of the copy source or destination float below the copy.
4137 insert_mem_bar(Op_MemBarCPUOrder);
4138
4139 // Call it. Note that the length argument is not scaled.
4140 make_runtime_call(RC_LEAF|RC_NO_FP,
4141 OptoRuntime::fast_arraycopy_Type(),
4142 StubRoutines::unsafe_arraycopy(),
4143 "unsafe_arraycopy",
4144 TypeRawPtr::BOTTOM,
4145 src, dst, size XTOP);
4146
4147 // Do not let reads of the copy destination float above the copy.
4148 insert_mem_bar(Op_MemBarCPUOrder);
4149
4150 return true;
4151 }
4152
4153 //------------------------clone_coping-----------------------------------
4154 // Helper function for inline_native_clone.
4155 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4156 assert(obj_size != NULL, "");
4157 Node* raw_obj = alloc_obj->in(1);
4158 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4159
4160 AllocateNode* alloc = NULL;
4161 if (ReduceBulkZeroing) {
4162 // We will be completely responsible for initializing this object -
4163 // mark Initialize node as complete.
4164 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4165 // The object was just allocated - there should be no any stores!
4166 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4167 // Mark as complete_with_arraycopy so that on AllocateNode
4168 // expansion, we know this AllocateNode is initialized by an array
4169 // copy and a StoreStore barrier exists after the array copy.
4170 alloc->initialization()->set_complete_with_arraycopy();
4171 }
4172
4173 // Copy the fastest available way.
4174 // TODO: generate fields copies for small objects instead.
4175 Node* size = _gvn.transform(obj_size);
4176
4177 access_clone(control(), obj, alloc_obj, size, is_array);
4178
4179 // Do not let reads from the cloned object float above the arraycopy.
4180 if (alloc != NULL) {
4181 // Do not let stores that initialize this object be reordered with
4182 // a subsequent store that would make this object accessible by
4183 // other threads.
4184 // Record what AllocateNode this StoreStore protects so that
4185 // escape analysis can go from the MemBarStoreStoreNode to the
4186 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4187 // based on the escape status of the AllocateNode.
4188 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4189 } else {
4190 insert_mem_bar(Op_MemBarCPUOrder);
4191 }
4192 }
4193
4194 //------------------------inline_native_clone----------------------------
4195 // protected native Object java.lang.Object.clone();
4196 //
4197 // Here are the simple edge cases:
4198 // null receiver => normal trap
4199 // virtual and clone was overridden => slow path to out-of-line clone
4200 // not cloneable or finalizer => slow path to out-of-line Object.clone
4201 //
4202 // The general case has two steps, allocation and copying.
4203 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4204 //
4205 // Copying also has two cases, oop arrays and everything else.
4206 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4207 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4208 //
4209 // These steps fold up nicely if and when the cloned object's klass
4210 // can be sharply typed as an object array, a type array, or an instance.
4211 //
4212 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4213 PhiNode* result_val;
4214
4215 // Set the reexecute bit for the interpreter to reexecute
4216 // the bytecode that invokes Object.clone if deoptimization happens.
4217 { PreserveReexecuteState preexecs(this);
4218 jvms()->set_should_reexecute(true);
4219
4220 Node* obj = null_check_receiver();
4221 if (stopped()) return true;
4222
4223 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4224
4225 // If we are going to clone an instance, we need its exact type to
4226 // know the number and types of fields to convert the clone to
4227 // loads/stores. Maybe a speculative type can help us.
4228 if (!obj_type->klass_is_exact() &&
4229 obj_type->speculative_type() != NULL &&
4230 obj_type->speculative_type()->is_instance_klass()) {
4231 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4232 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4233 !spec_ik->has_injected_fields()) {
4234 ciKlass* k = obj_type->klass();
4235 if (!k->is_instance_klass() ||
4236 k->as_instance_klass()->is_interface() ||
4237 k->as_instance_klass()->has_subklass()) {
4238 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4239 }
4240 }
4241 }
4242
4243 Node* obj_klass = load_object_klass(obj);
4244 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4245 const TypeOopPtr* toop = ((tklass != NULL)
4246 ? tklass->as_instance_type()
4247 : TypeInstPtr::NOTNULL);
4248
4249 // Conservatively insert a memory barrier on all memory slices.
4250 // Do not let writes into the original float below the clone.
4251 insert_mem_bar(Op_MemBarCPUOrder);
4252
4253 // paths into result_reg:
4254 enum {
4255 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4256 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4257 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4258 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4259 PATH_LIMIT
4260 };
4261 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4262 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4263 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4264 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4265 record_for_igvn(result_reg);
4266
4267 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4268 if (array_ctl != NULL) {
4269 // It's an array.
4270 PreserveJVMState pjvms(this);
4271 set_control(array_ctl);
4272 Node* obj_length = load_array_length(obj);
4273 Node* obj_size = NULL;
4274 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4275
4276 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4277 if (bs->array_copy_requires_gc_barriers(T_OBJECT)) {
4278 // If it is an oop array, it requires very special treatment,
4279 // because gc barriers are required when accessing the array.
4280 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4281 if (is_obja != NULL) {
4282 PreserveJVMState pjvms2(this);
4283 set_control(is_obja);
4284 // Generate a direct call to the right arraycopy function(s).
4285 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4286 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4287 ac->set_cloneoop();
4288 Node* n = _gvn.transform(ac);
4289 assert(n == ac, "cannot disappear");
4290 ac->connect_outputs(this);
4291
4292 result_reg->init_req(_objArray_path, control());
4293 result_val->init_req(_objArray_path, alloc_obj);
4294 result_i_o ->set_req(_objArray_path, i_o());
4295 result_mem ->set_req(_objArray_path, reset_memory());
4296 }
4297 }
4298 // Otherwise, there are no barriers to worry about.
4299 // (We can dispense with card marks if we know the allocation
4300 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4301 // causes the non-eden paths to take compensating steps to
4302 // simulate a fresh allocation, so that no further
4303 // card marks are required in compiled code to initialize
4304 // the object.)
4305
4306 if (!stopped()) {
4307 copy_to_clone(obj, alloc_obj, obj_size, true);
4308
4309 // Present the results of the copy.
4310 result_reg->init_req(_array_path, control());
4311 result_val->init_req(_array_path, alloc_obj);
4312 result_i_o ->set_req(_array_path, i_o());
4313 result_mem ->set_req(_array_path, reset_memory());
4314 }
4315 }
4316
4317 // We only go to the instance fast case code if we pass a number of guards.
4318 // The paths which do not pass are accumulated in the slow_region.
4319 RegionNode* slow_region = new RegionNode(1);
4320 record_for_igvn(slow_region);
4321 if (!stopped()) {
4322 // It's an instance (we did array above). Make the slow-path tests.
4323 // If this is a virtual call, we generate a funny guard. We grab
4324 // the vtable entry corresponding to clone() from the target object.
4325 // If the target method which we are calling happens to be the
4326 // Object clone() method, we pass the guard. We do not need this
4327 // guard for non-virtual calls; the caller is known to be the native
4328 // Object clone().
4329 if (is_virtual) {
4330 generate_virtual_guard(obj_klass, slow_region);
4331 }
4332
4333 // The object must be easily cloneable and must not have a finalizer.
4334 // Both of these conditions may be checked in a single test.
4335 // We could optimize the test further, but we don't care.
4336 generate_access_flags_guard(obj_klass,
4337 // Test both conditions:
4338 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4339 // Must be cloneable but not finalizer:
4340 JVM_ACC_IS_CLONEABLE_FAST,
4341 slow_region);
4342 }
4343
4344 if (!stopped()) {
4345 // It's an instance, and it passed the slow-path tests.
4346 PreserveJVMState pjvms(this);
4347 Node* obj_size = NULL;
4348 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4349 // is reexecuted if deoptimization occurs and there could be problems when merging
4350 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4351 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4352
4353 copy_to_clone(obj, alloc_obj, obj_size, false);
4354
4355 // Present the results of the slow call.
4356 result_reg->init_req(_instance_path, control());
4357 result_val->init_req(_instance_path, alloc_obj);
4358 result_i_o ->set_req(_instance_path, i_o());
4359 result_mem ->set_req(_instance_path, reset_memory());
4360 }
4361
4362 // Generate code for the slow case. We make a call to clone().
4363 set_control(_gvn.transform(slow_region));
4364 if (!stopped()) {
4365 PreserveJVMState pjvms(this);
4366 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4367 Node* slow_result = set_results_for_java_call(slow_call);
4368 // this->control() comes from set_results_for_java_call
4369 result_reg->init_req(_slow_path, control());
4370 result_val->init_req(_slow_path, slow_result);
4371 result_i_o ->set_req(_slow_path, i_o());
4372 result_mem ->set_req(_slow_path, reset_memory());
4373 }
4374
4375 // Return the combined state.
4376 set_control( _gvn.transform(result_reg));
4377 set_i_o( _gvn.transform(result_i_o));
4378 set_all_memory( _gvn.transform(result_mem));
4379 } // original reexecute is set back here
4380
4381 set_result(_gvn.transform(result_val));
4382 return true;
4383 }
4384
4385 // If we have a tighly coupled allocation, the arraycopy may take care
4386 // of the array initialization. If one of the guards we insert between
4387 // the allocation and the arraycopy causes a deoptimization, an
4388 // unitialized array will escape the compiled method. To prevent that
4389 // we set the JVM state for uncommon traps between the allocation and
4390 // the arraycopy to the state before the allocation so, in case of
4391 // deoptimization, we'll reexecute the allocation and the
4392 // initialization.
4393 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4394 if (alloc != NULL) {
4395 ciMethod* trap_method = alloc->jvms()->method();
4396 int trap_bci = alloc->jvms()->bci();
4397
4398 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4399 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4400 // Make sure there's no store between the allocation and the
4401 // arraycopy otherwise visible side effects could be rexecuted
4402 // in case of deoptimization and cause incorrect execution.
4403 bool no_interfering_store = true;
4404 Node* mem = alloc->in(TypeFunc::Memory);
4405 if (mem->is_MergeMem()) {
4406 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4407 Node* n = mms.memory();
4408 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4409 assert(n->is_Store(), "what else?");
4410 no_interfering_store = false;
4411 break;
4412 }
4413 }
4414 } else {
4415 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4416 Node* n = mms.memory();
4417 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4418 assert(n->is_Store(), "what else?");
4419 no_interfering_store = false;
4420 break;
4421 }
4422 }
4423 }
4424
4425 if (no_interfering_store) {
4426 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4427 uint size = alloc->req();
4428 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4429 old_jvms->set_map(sfpt);
4430 for (uint i = 0; i < size; i++) {
4431 sfpt->init_req(i, alloc->in(i));
4432 }
4433 // re-push array length for deoptimization
4434 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4435 old_jvms->set_sp(old_jvms->sp()+1);
4436 old_jvms->set_monoff(old_jvms->monoff()+1);
4437 old_jvms->set_scloff(old_jvms->scloff()+1);
4438 old_jvms->set_endoff(old_jvms->endoff()+1);
4439 old_jvms->set_should_reexecute(true);
4440
4441 sfpt->set_i_o(map()->i_o());
4442 sfpt->set_memory(map()->memory());
4443 sfpt->set_control(map()->control());
4444
4445 JVMState* saved_jvms = jvms();
4446 saved_reexecute_sp = _reexecute_sp;
4447
4448 set_jvms(sfpt->jvms());
4449 _reexecute_sp = jvms()->sp();
4450
4451 return saved_jvms;
4452 }
4453 }
4454 }
4455 return NULL;
4456 }
4457
4458 // In case of a deoptimization, we restart execution at the
4459 // allocation, allocating a new array. We would leave an uninitialized
4460 // array in the heap that GCs wouldn't expect. Move the allocation
4461 // after the traps so we don't allocate the array if we
4462 // deoptimize. This is possible because tightly_coupled_allocation()
4463 // guarantees there's no observer of the allocated array at this point
4464 // and the control flow is simple enough.
4465 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4466 int saved_reexecute_sp, uint new_idx) {
4467 if (saved_jvms != NULL && !stopped()) {
4468 assert(alloc != NULL, "only with a tightly coupled allocation");
4469 // restore JVM state to the state at the arraycopy
4470 saved_jvms->map()->set_control(map()->control());
4471 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4472 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4473 // If we've improved the types of some nodes (null check) while
4474 // emitting the guards, propagate them to the current state
4475 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4476 set_jvms(saved_jvms);
4477 _reexecute_sp = saved_reexecute_sp;
4478
4479 // Remove the allocation from above the guards
4480 CallProjections callprojs;
4481 alloc->extract_projections(&callprojs, true);
4482 InitializeNode* init = alloc->initialization();
4483 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4484 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4485 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4486 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4487
4488 // move the allocation here (after the guards)
4489 _gvn.hash_delete(alloc);
4490 alloc->set_req(TypeFunc::Control, control());
4491 alloc->set_req(TypeFunc::I_O, i_o());
4492 Node *mem = reset_memory();
4493 set_all_memory(mem);
4494 alloc->set_req(TypeFunc::Memory, mem);
4495 set_control(init->proj_out_or_null(TypeFunc::Control));
4496 set_i_o(callprojs.fallthrough_ioproj);
4497
4498 // Update memory as done in GraphKit::set_output_for_allocation()
4499 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4500 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4501 if (ary_type->isa_aryptr() && length_type != NULL) {
4502 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4503 }
4504 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4505 int elemidx = C->get_alias_index(telemref);
4506 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4507 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4508
4509 Node* allocx = _gvn.transform(alloc);
4510 assert(allocx == alloc, "where has the allocation gone?");
4511 assert(dest->is_CheckCastPP(), "not an allocation result?");
4512
4513 _gvn.hash_delete(dest);
4514 dest->set_req(0, control());
4515 Node* destx = _gvn.transform(dest);
4516 assert(destx == dest, "where has the allocation result gone?");
4517 }
4518 }
4519
4520
4521 //------------------------------inline_arraycopy-----------------------
4522 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4523 // Object dest, int destPos,
4524 // int length);
4525 bool LibraryCallKit::inline_arraycopy() {
4526 // Get the arguments.
4527 Node* src = argument(0); // type: oop
4528 Node* src_offset = argument(1); // type: int
4529 Node* dest = argument(2); // type: oop
4530 Node* dest_offset = argument(3); // type: int
4531 Node* length = argument(4); // type: int
4532
4533 uint new_idx = C->unique();
4534
4535 // Check for allocation before we add nodes that would confuse
4536 // tightly_coupled_allocation()
4537 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4538
4539 int saved_reexecute_sp = -1;
4540 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4541 // See arraycopy_restore_alloc_state() comment
4542 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4543 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4544 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4545 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4546
4547 // The following tests must be performed
4548 // (1) src and dest are arrays.
4549 // (2) src and dest arrays must have elements of the same BasicType
4550 // (3) src and dest must not be null.
4551 // (4) src_offset must not be negative.
4552 // (5) dest_offset must not be negative.
4553 // (6) length must not be negative.
4554 // (7) src_offset + length must not exceed length of src.
4555 // (8) dest_offset + length must not exceed length of dest.
4556 // (9) each element of an oop array must be assignable
4557
4558 // (3) src and dest must not be null.
4559 // always do this here because we need the JVM state for uncommon traps
4560 Node* null_ctl = top();
4561 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4562 assert(null_ctl->is_top(), "no null control here");
4563 dest = null_check(dest, T_ARRAY);
4564
4565 if (!can_emit_guards) {
4566 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4567 // guards but the arraycopy node could still take advantage of a
4568 // tightly allocated allocation. tightly_coupled_allocation() is
4569 // called again to make sure it takes the null check above into
4570 // account: the null check is mandatory and if it caused an
4571 // uncommon trap to be emitted then the allocation can't be
4572 // considered tightly coupled in this context.
4573 alloc = tightly_coupled_allocation(dest, NULL);
4574 }
4575
4576 bool validated = false;
4577
4578 const Type* src_type = _gvn.type(src);
4579 const Type* dest_type = _gvn.type(dest);
4580 const TypeAryPtr* top_src = src_type->isa_aryptr();
4581 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4582
4583 // Do we have the type of src?
4584 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4585 // Do we have the type of dest?
4586 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4587 // Is the type for src from speculation?
4588 bool src_spec = false;
4589 // Is the type for dest from speculation?
4590 bool dest_spec = false;
4591
4592 if ((!has_src || !has_dest) && can_emit_guards) {
4593 // We don't have sufficient type information, let's see if
4594 // speculative types can help. We need to have types for both src
4595 // and dest so that it pays off.
4596
4597 // Do we already have or could we have type information for src
4598 bool could_have_src = has_src;
4599 // Do we already have or could we have type information for dest
4600 bool could_have_dest = has_dest;
4601
4602 ciKlass* src_k = NULL;
4603 if (!has_src) {
4604 src_k = src_type->speculative_type_not_null();
4605 if (src_k != NULL && src_k->is_array_klass()) {
4606 could_have_src = true;
4607 }
4608 }
4609
4610 ciKlass* dest_k = NULL;
4611 if (!has_dest) {
4612 dest_k = dest_type->speculative_type_not_null();
4613 if (dest_k != NULL && dest_k->is_array_klass()) {
4614 could_have_dest = true;
4615 }
4616 }
4617
4618 if (could_have_src && could_have_dest) {
4619 // This is going to pay off so emit the required guards
4620 if (!has_src) {
4621 src = maybe_cast_profiled_obj(src, src_k, true);
4622 src_type = _gvn.type(src);
4623 top_src = src_type->isa_aryptr();
4624 has_src = (top_src != NULL && top_src->klass() != NULL);
4625 src_spec = true;
4626 }
4627 if (!has_dest) {
4628 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4629 dest_type = _gvn.type(dest);
4630 top_dest = dest_type->isa_aryptr();
4631 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4632 dest_spec = true;
4633 }
4634 }
4635 }
4636
4637 if (has_src && has_dest && can_emit_guards) {
4638 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4639 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4640 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
4641 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
4642
4643 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4644 // If both arrays are object arrays then having the exact types
4645 // for both will remove the need for a subtype check at runtime
4646 // before the call and may make it possible to pick a faster copy
4647 // routine (without a subtype check on every element)
4648 // Do we have the exact type of src?
4649 bool could_have_src = src_spec;
4650 // Do we have the exact type of dest?
4651 bool could_have_dest = dest_spec;
4652 ciKlass* src_k = top_src->klass();
4653 ciKlass* dest_k = top_dest->klass();
4654 if (!src_spec) {
4655 src_k = src_type->speculative_type_not_null();
4656 if (src_k != NULL && src_k->is_array_klass()) {
4657 could_have_src = true;
4658 }
4659 }
4660 if (!dest_spec) {
4661 dest_k = dest_type->speculative_type_not_null();
4662 if (dest_k != NULL && dest_k->is_array_klass()) {
4663 could_have_dest = true;
4664 }
4665 }
4666 if (could_have_src && could_have_dest) {
4667 // If we can have both exact types, emit the missing guards
4668 if (could_have_src && !src_spec) {
4669 src = maybe_cast_profiled_obj(src, src_k, true);
4670 }
4671 if (could_have_dest && !dest_spec) {
4672 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4673 }
4674 }
4675 }
4676 }
4677
4678 ciMethod* trap_method = method();
4679 int trap_bci = bci();
4680 if (saved_jvms != NULL) {
4681 trap_method = alloc->jvms()->method();
4682 trap_bci = alloc->jvms()->bci();
4683 }
4684
4685 bool negative_length_guard_generated = false;
4686
4687 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4688 can_emit_guards &&
4689 !src->is_top() && !dest->is_top()) {
4690 // validate arguments: enables transformation the ArrayCopyNode
4691 validated = true;
4692
4693 RegionNode* slow_region = new RegionNode(1);
4694 record_for_igvn(slow_region);
4695
4696 // (1) src and dest are arrays.
4697 generate_non_array_guard(load_object_klass(src), slow_region);
4698 generate_non_array_guard(load_object_klass(dest), slow_region);
4699
4700 // (2) src and dest arrays must have elements of the same BasicType
4701 // done at macro expansion or at Ideal transformation time
4702
4703 // (4) src_offset must not be negative.
4704 generate_negative_guard(src_offset, slow_region);
4705
4706 // (5) dest_offset must not be negative.
4707 generate_negative_guard(dest_offset, slow_region);
4708
4709 // (7) src_offset + length must not exceed length of src.
4710 generate_limit_guard(src_offset, length,
4711 load_array_length(src),
4712 slow_region);
4713
4714 // (8) dest_offset + length must not exceed length of dest.
4715 generate_limit_guard(dest_offset, length,
4716 load_array_length(dest),
4717 slow_region);
4718
4719 // (6) length must not be negative.
4720 // This is also checked in generate_arraycopy() during macro expansion, but
4721 // we also have to check it here for the case where the ArrayCopyNode will
4722 // be eliminated by Escape Analysis.
4723 if (EliminateAllocations) {
4724 generate_negative_guard(length, slow_region);
4725 negative_length_guard_generated = true;
4726 }
4727
4728 // (9) each element of an oop array must be assignable
4729 Node* src_klass = load_object_klass(src);
4730 Node* dest_klass = load_object_klass(dest);
4731 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4732
4733 if (not_subtype_ctrl != top()) {
4734 PreserveJVMState pjvms(this);
4735 set_control(not_subtype_ctrl);
4736 uncommon_trap(Deoptimization::Reason_intrinsic,
4737 Deoptimization::Action_make_not_entrant);
4738 assert(stopped(), "Should be stopped");
4739 }
4740 {
4741 PreserveJVMState pjvms(this);
4742 set_control(_gvn.transform(slow_region));
4743 uncommon_trap(Deoptimization::Reason_intrinsic,
4744 Deoptimization::Action_make_not_entrant);
4745 assert(stopped(), "Should be stopped");
4746 }
4747
4748 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4749 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4750 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4751 }
4752
4753 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4754
4755 if (stopped()) {
4756 return true;
4757 }
4758
4759 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4760 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4761 // so the compiler has a chance to eliminate them: during macro expansion,
4762 // we have to set their control (CastPP nodes are eliminated).
4763 load_object_klass(src), load_object_klass(dest),
4764 load_array_length(src), load_array_length(dest));
4765
4766 ac->set_arraycopy(validated);
4767
4768 Node* n = _gvn.transform(ac);
4769 if (n == ac) {
4770 ac->connect_outputs(this);
4771 } else {
4772 assert(validated, "shouldn't transform if all arguments not validated");
4773 set_all_memory(n);
4774 }
4775 clear_upper_avx();
4776
4777
4778 return true;
4779 }
4780
4781
4782 // Helper function which determines if an arraycopy immediately follows
4783 // an allocation, with no intervening tests or other escapes for the object.
4784 AllocateArrayNode*
4785 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4786 RegionNode* slow_region) {
4787 if (stopped()) return NULL; // no fast path
4788 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4789
4790 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4791 if (alloc == NULL) return NULL;
4792
4793 Node* rawmem = memory(Compile::AliasIdxRaw);
4794 // Is the allocation's memory state untouched?
4795 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4796 // Bail out if there have been raw-memory effects since the allocation.
4797 // (Example: There might have been a call or safepoint.)
4798 return NULL;
4799 }
4800 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4801 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4802 return NULL;
4803 }
4804
4805 // There must be no unexpected observers of this allocation.
4806 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4807 Node* obs = ptr->fast_out(i);
4808 if (obs != this->map()) {
4809 return NULL;
4810 }
4811 }
4812
4813 // This arraycopy must unconditionally follow the allocation of the ptr.
4814 Node* alloc_ctl = ptr->in(0);
4815 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4816
4817 Node* ctl = control();
4818 while (ctl != alloc_ctl) {
4819 // There may be guards which feed into the slow_region.
4820 // Any other control flow means that we might not get a chance
4821 // to finish initializing the allocated object.
4822 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4823 IfNode* iff = ctl->in(0)->as_If();
4824 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4825 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4826 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4827 ctl = iff->in(0); // This test feeds the known slow_region.
4828 continue;
4829 }
4830 // One more try: Various low-level checks bottom out in
4831 // uncommon traps. If the debug-info of the trap omits
4832 // any reference to the allocation, as we've already
4833 // observed, then there can be no objection to the trap.
4834 bool found_trap = false;
4835 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4836 Node* obs = not_ctl->fast_out(j);
4837 if (obs->in(0) == not_ctl && obs->is_Call() &&
4838 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4839 found_trap = true; break;
4840 }
4841 }
4842 if (found_trap) {
4843 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4844 continue;
4845 }
4846 }
4847 return NULL;
4848 }
4849
4850 // If we get this far, we have an allocation which immediately
4851 // precedes the arraycopy, and we can take over zeroing the new object.
4852 // The arraycopy will finish the initialization, and provide
4853 // a new control state to which we will anchor the destination pointer.
4854
4855 return alloc;
4856 }
4857
4858 //-------------inline_encodeISOArray-----------------------------------
4859 // encode char[] to byte[] in ISO_8859_1
4860 bool LibraryCallKit::inline_encodeISOArray() {
4861 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4862 // no receiver since it is static method
4863 Node *src = argument(0);
4864 Node *src_offset = argument(1);
4865 Node *dst = argument(2);
4866 Node *dst_offset = argument(3);
4867 Node *length = argument(4);
4868
4869 const Type* src_type = src->Value(&_gvn);
4870 const Type* dst_type = dst->Value(&_gvn);
4871 const TypeAryPtr* top_src = src_type->isa_aryptr();
4872 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4873 if (top_src == NULL || top_src->klass() == NULL ||
4874 top_dest == NULL || top_dest->klass() == NULL) {
4875 // failed array check
4876 return false;
4877 }
4878
4879 // Figure out the size and type of the elements we will be copying.
4880 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4881 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4882 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4883 return false;
4884 }
4885
4886 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4887 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4888 // 'src_start' points to src array + scaled offset
4889 // 'dst_start' points to dst array + scaled offset
4890
4891 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4892 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4893 enc = _gvn.transform(enc);
4894 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4895 set_memory(res_mem, mtype);
4896 set_result(enc);
4897 clear_upper_avx();
4898
4899 return true;
4900 }
4901
4902 //-------------inline_multiplyToLen-----------------------------------
4903 bool LibraryCallKit::inline_multiplyToLen() {
4904 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4905
4906 address stubAddr = StubRoutines::multiplyToLen();
4907 if (stubAddr == NULL) {
4908 return false; // Intrinsic's stub is not implemented on this platform
4909 }
4910 const char* stubName = "multiplyToLen";
4911
4912 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4913
4914 // no receiver because it is a static method
4915 Node* x = argument(0);
4916 Node* xlen = argument(1);
4917 Node* y = argument(2);
4918 Node* ylen = argument(3);
4919 Node* z = argument(4);
4920
4921 const Type* x_type = x->Value(&_gvn);
4922 const Type* y_type = y->Value(&_gvn);
4923 const TypeAryPtr* top_x = x_type->isa_aryptr();
4924 const TypeAryPtr* top_y = y_type->isa_aryptr();
4925 if (top_x == NULL || top_x->klass() == NULL ||
4926 top_y == NULL || top_y->klass() == NULL) {
4927 // failed array check
4928 return false;
4929 }
4930
4931 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4932 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4933 if (x_elem != T_INT || y_elem != T_INT) {
4934 return false;
4935 }
4936
4937 // Set the original stack and the reexecute bit for the interpreter to reexecute
4938 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
4939 // on the return from z array allocation in runtime.
4940 { PreserveReexecuteState preexecs(this);
4941 jvms()->set_should_reexecute(true);
4942
4943 Node* x_start = array_element_address(x, intcon(0), x_elem);
4944 Node* y_start = array_element_address(y, intcon(0), y_elem);
4945 // 'x_start' points to x array + scaled xlen
4946 // 'y_start' points to y array + scaled ylen
4947
4948 // Allocate the result array
4949 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
4950 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
4951 Node* klass_node = makecon(TypeKlassPtr::make(klass));
4952
4953 IdealKit ideal(this);
4954
4955 #define __ ideal.
4956 Node* one = __ ConI(1);
4957 Node* zero = __ ConI(0);
4958 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
4959 __ set(need_alloc, zero);
4960 __ set(z_alloc, z);
4961 __ if_then(z, BoolTest::eq, null()); {
4962 __ increment (need_alloc, one);
4963 } __ else_(); {
4964 // Update graphKit memory and control from IdealKit.
4965 sync_kit(ideal);
4966 Node* zlen_arg = load_array_length(z);
4967 // Update IdealKit memory and control from graphKit.
4968 __ sync_kit(this);
4969 __ if_then(zlen_arg, BoolTest::lt, zlen); {
4970 __ increment (need_alloc, one);
4971 } __ end_if();
4972 } __ end_if();
4973
4974 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
4975 // Update graphKit memory and control from IdealKit.
4976 sync_kit(ideal);
4977 Node * narr = new_array(klass_node, zlen, 1);
4978 // Update IdealKit memory and control from graphKit.
4979 __ sync_kit(this);
4980 __ set(z_alloc, narr);
4981 } __ end_if();
4982
4983 sync_kit(ideal);
4984 z = __ value(z_alloc);
4985 // Can't use TypeAryPtr::INTS which uses Bottom offset.
4986 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
4987 // Final sync IdealKit and GraphKit.
4988 final_sync(ideal);
4989 #undef __
4990
4991 Node* z_start = array_element_address(z, intcon(0), T_INT);
4992
4993 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
4994 OptoRuntime::multiplyToLen_Type(),
4995 stubAddr, stubName, TypePtr::BOTTOM,
4996 x_start, xlen, y_start, ylen, z_start, zlen);
4997 } // original reexecute is set back here
4998
4999 C->set_has_split_ifs(true); // Has chance for split-if optimization
5000 set_result(z);
5001 return true;
5002 }
5003
5004 //-------------inline_squareToLen------------------------------------
5005 bool LibraryCallKit::inline_squareToLen() {
5006 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5007
5008 address stubAddr = StubRoutines::squareToLen();
5009 if (stubAddr == NULL) {
5010 return false; // Intrinsic's stub is not implemented on this platform
5011 }
5012 const char* stubName = "squareToLen";
5013
5014 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5015
5016 Node* x = argument(0);
5017 Node* len = argument(1);
5018 Node* z = argument(2);
5019 Node* zlen = argument(3);
5020
5021 const Type* x_type = x->Value(&_gvn);
5022 const Type* z_type = z->Value(&_gvn);
5023 const TypeAryPtr* top_x = x_type->isa_aryptr();
5024 const TypeAryPtr* top_z = z_type->isa_aryptr();
5025 if (top_x == NULL || top_x->klass() == NULL ||
5026 top_z == NULL || top_z->klass() == NULL) {
5027 // failed array check
5028 return false;
5029 }
5030
5031 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5032 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5033 if (x_elem != T_INT || z_elem != T_INT) {
5034 return false;
5035 }
5036
5037
5038 Node* x_start = array_element_address(x, intcon(0), x_elem);
5039 Node* z_start = array_element_address(z, intcon(0), z_elem);
5040
5041 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5042 OptoRuntime::squareToLen_Type(),
5043 stubAddr, stubName, TypePtr::BOTTOM,
5044 x_start, len, z_start, zlen);
5045
5046 set_result(z);
5047 return true;
5048 }
5049
5050 //-------------inline_mulAdd------------------------------------------
5051 bool LibraryCallKit::inline_mulAdd() {
5052 assert(UseMulAddIntrinsic, "not implemented on this platform");
5053
5054 address stubAddr = StubRoutines::mulAdd();
5055 if (stubAddr == NULL) {
5056 return false; // Intrinsic's stub is not implemented on this platform
5057 }
5058 const char* stubName = "mulAdd";
5059
5060 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5061
5062 Node* out = argument(0);
5063 Node* in = argument(1);
5064 Node* offset = argument(2);
5065 Node* len = argument(3);
5066 Node* k = argument(4);
5067
5068 const Type* out_type = out->Value(&_gvn);
5069 const Type* in_type = in->Value(&_gvn);
5070 const TypeAryPtr* top_out = out_type->isa_aryptr();
5071 const TypeAryPtr* top_in = in_type->isa_aryptr();
5072 if (top_out == NULL || top_out->klass() == NULL ||
5073 top_in == NULL || top_in->klass() == NULL) {
5074 // failed array check
5075 return false;
5076 }
5077
5078 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5079 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5080 if (out_elem != T_INT || in_elem != T_INT) {
5081 return false;
5082 }
5083
5084 Node* outlen = load_array_length(out);
5085 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5086 Node* out_start = array_element_address(out, intcon(0), out_elem);
5087 Node* in_start = array_element_address(in, intcon(0), in_elem);
5088
5089 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5090 OptoRuntime::mulAdd_Type(),
5091 stubAddr, stubName, TypePtr::BOTTOM,
5092 out_start,in_start, new_offset, len, k);
5093 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5094 set_result(result);
5095 return true;
5096 }
5097
5098 //-------------inline_montgomeryMultiply-----------------------------------
5099 bool LibraryCallKit::inline_montgomeryMultiply() {
5100 address stubAddr = StubRoutines::montgomeryMultiply();
5101 if (stubAddr == NULL) {
5102 return false; // Intrinsic's stub is not implemented on this platform
5103 }
5104
5105 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5106 const char* stubName = "montgomery_multiply";
5107
5108 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5109
5110 Node* a = argument(0);
5111 Node* b = argument(1);
5112 Node* n = argument(2);
5113 Node* len = argument(3);
5114 Node* inv = argument(4);
5115 Node* m = argument(6);
5116
5117 const Type* a_type = a->Value(&_gvn);
5118 const TypeAryPtr* top_a = a_type->isa_aryptr();
5119 const Type* b_type = b->Value(&_gvn);
5120 const TypeAryPtr* top_b = b_type->isa_aryptr();
5121 const Type* n_type = a->Value(&_gvn);
5122 const TypeAryPtr* top_n = n_type->isa_aryptr();
5123 const Type* m_type = a->Value(&_gvn);
5124 const TypeAryPtr* top_m = m_type->isa_aryptr();
5125 if (top_a == NULL || top_a->klass() == NULL ||
5126 top_b == NULL || top_b->klass() == NULL ||
5127 top_n == NULL || top_n->klass() == NULL ||
5128 top_m == NULL || top_m->klass() == NULL) {
5129 // failed array check
5130 return false;
5131 }
5132
5133 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5134 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5135 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5136 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5137 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5138 return false;
5139 }
5140
5141 // Make the call
5142 {
5143 Node* a_start = array_element_address(a, intcon(0), a_elem);
5144 Node* b_start = array_element_address(b, intcon(0), b_elem);
5145 Node* n_start = array_element_address(n, intcon(0), n_elem);
5146 Node* m_start = array_element_address(m, intcon(0), m_elem);
5147
5148 Node* call = make_runtime_call(RC_LEAF,
5149 OptoRuntime::montgomeryMultiply_Type(),
5150 stubAddr, stubName, TypePtr::BOTTOM,
5151 a_start, b_start, n_start, len, inv, top(),
5152 m_start);
5153 set_result(m);
5154 }
5155
5156 return true;
5157 }
5158
5159 bool LibraryCallKit::inline_montgomerySquare() {
5160 address stubAddr = StubRoutines::montgomerySquare();
5161 if (stubAddr == NULL) {
5162 return false; // Intrinsic's stub is not implemented on this platform
5163 }
5164
5165 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5166 const char* stubName = "montgomery_square";
5167
5168 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5169
5170 Node* a = argument(0);
5171 Node* n = argument(1);
5172 Node* len = argument(2);
5173 Node* inv = argument(3);
5174 Node* m = argument(5);
5175
5176 const Type* a_type = a->Value(&_gvn);
5177 const TypeAryPtr* top_a = a_type->isa_aryptr();
5178 const Type* n_type = a->Value(&_gvn);
5179 const TypeAryPtr* top_n = n_type->isa_aryptr();
5180 const Type* m_type = a->Value(&_gvn);
5181 const TypeAryPtr* top_m = m_type->isa_aryptr();
5182 if (top_a == NULL || top_a->klass() == NULL ||
5183 top_n == NULL || top_n->klass() == NULL ||
5184 top_m == NULL || top_m->klass() == NULL) {
5185 // failed array check
5186 return false;
5187 }
5188
5189 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5190 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5191 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5192 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5193 return false;
5194 }
5195
5196 // Make the call
5197 {
5198 Node* a_start = array_element_address(a, intcon(0), a_elem);
5199 Node* n_start = array_element_address(n, intcon(0), n_elem);
5200 Node* m_start = array_element_address(m, intcon(0), m_elem);
5201
5202 Node* call = make_runtime_call(RC_LEAF,
5203 OptoRuntime::montgomerySquare_Type(),
5204 stubAddr, stubName, TypePtr::BOTTOM,
5205 a_start, n_start, len, inv, top(),
5206 m_start);
5207 set_result(m);
5208 }
5209
5210 return true;
5211 }
5212
5213 //-------------inline_vectorizedMismatch------------------------------
5214 bool LibraryCallKit::inline_vectorizedMismatch() {
5215 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5216
5217 address stubAddr = StubRoutines::vectorizedMismatch();
5218 if (stubAddr == NULL) {
5219 return false; // Intrinsic's stub is not implemented on this platform
5220 }
5221 const char* stubName = "vectorizedMismatch";
5222 int size_l = callee()->signature()->size();
5223 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5224
5225 Node* obja = argument(0);
5226 Node* aoffset = argument(1);
5227 Node* objb = argument(3);
5228 Node* boffset = argument(4);
5229 Node* length = argument(6);
5230 Node* scale = argument(7);
5231
5232 const Type* a_type = obja->Value(&_gvn);
5233 const Type* b_type = objb->Value(&_gvn);
5234 const TypeAryPtr* top_a = a_type->isa_aryptr();
5235 const TypeAryPtr* top_b = b_type->isa_aryptr();
5236 if (top_a == NULL || top_a->klass() == NULL ||
5237 top_b == NULL || top_b->klass() == NULL) {
5238 // failed array check
5239 return false;
5240 }
5241
5242 Node* call;
5243 jvms()->set_should_reexecute(true);
5244
5245 Node* obja_adr = make_unsafe_address(obja, aoffset);
5246 Node* objb_adr = make_unsafe_address(objb, boffset);
5247
5248 call = make_runtime_call(RC_LEAF,
5249 OptoRuntime::vectorizedMismatch_Type(),
5250 stubAddr, stubName, TypePtr::BOTTOM,
5251 obja_adr, objb_adr, length, scale);
5252
5253 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5254 set_result(result);
5255 return true;
5256 }
5257
5258 /**
5259 * Calculate CRC32 for byte.
5260 * int java.util.zip.CRC32.update(int crc, int b)
5261 */
5262 bool LibraryCallKit::inline_updateCRC32() {
5263 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5264 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5265 // no receiver since it is static method
5266 Node* crc = argument(0); // type: int
5267 Node* b = argument(1); // type: int
5268
5269 /*
5270 * int c = ~ crc;
5271 * b = timesXtoThe32[(b ^ c) & 0xFF];
5272 * b = b ^ (c >>> 8);
5273 * crc = ~b;
5274 */
5275
5276 Node* M1 = intcon(-1);
5277 crc = _gvn.transform(new XorINode(crc, M1));
5278 Node* result = _gvn.transform(new XorINode(crc, b));
5279 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5280
5281 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5282 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5283 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5284 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5285
5286 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5287 result = _gvn.transform(new XorINode(crc, result));
5288 result = _gvn.transform(new XorINode(result, M1));
5289 set_result(result);
5290 return true;
5291 }
5292
5293 /**
5294 * Calculate CRC32 for byte[] array.
5295 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5296 */
5297 bool LibraryCallKit::inline_updateBytesCRC32() {
5298 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5299 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5300 // no receiver since it is static method
5301 Node* crc = argument(0); // type: int
5302 Node* src = argument(1); // type: oop
5303 Node* offset = argument(2); // type: int
5304 Node* length = argument(3); // type: int
5305
5306 const Type* src_type = src->Value(&_gvn);
5307 const TypeAryPtr* top_src = src_type->isa_aryptr();
5308 if (top_src == NULL || top_src->klass() == NULL) {
5309 // failed array check
5310 return false;
5311 }
5312
5313 // Figure out the size and type of the elements we will be copying.
5314 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5315 if (src_elem != T_BYTE) {
5316 return false;
5317 }
5318
5319 // 'src_start' points to src array + scaled offset
5320 Node* src_start = array_element_address(src, offset, src_elem);
5321
5322 // We assume that range check is done by caller.
5323 // TODO: generate range check (offset+length < src.length) in debug VM.
5324
5325 // Call the stub.
5326 address stubAddr = StubRoutines::updateBytesCRC32();
5327 const char *stubName = "updateBytesCRC32";
5328
5329 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5330 stubAddr, stubName, TypePtr::BOTTOM,
5331 crc, src_start, length);
5332 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5333 set_result(result);
5334 return true;
5335 }
5336
5337 /**
5338 * Calculate CRC32 for ByteBuffer.
5339 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5340 */
5341 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5342 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5343 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5344 // no receiver since it is static method
5345 Node* crc = argument(0); // type: int
5346 Node* src = argument(1); // type: long
5347 Node* offset = argument(3); // type: int
5348 Node* length = argument(4); // type: int
5349
5350 src = ConvL2X(src); // adjust Java long to machine word
5351 Node* base = _gvn.transform(new CastX2PNode(src));
5352 offset = ConvI2X(offset);
5353
5354 // 'src_start' points to src array + scaled offset
5355 Node* src_start = basic_plus_adr(top(), base, offset);
5356
5357 // Call the stub.
5358 address stubAddr = StubRoutines::updateBytesCRC32();
5359 const char *stubName = "updateBytesCRC32";
5360
5361 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5362 stubAddr, stubName, TypePtr::BOTTOM,
5363 crc, src_start, length);
5364 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5365 set_result(result);
5366 return true;
5367 }
5368
5369 //------------------------------get_table_from_crc32c_class-----------------------
5370 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5371 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5372 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5373
5374 return table;
5375 }
5376
5377 //------------------------------inline_updateBytesCRC32C-----------------------
5378 //
5379 // Calculate CRC32C for byte[] array.
5380 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5381 //
5382 bool LibraryCallKit::inline_updateBytesCRC32C() {
5383 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5384 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5385 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5386 // no receiver since it is a static method
5387 Node* crc = argument(0); // type: int
5388 Node* src = argument(1); // type: oop
5389 Node* offset = argument(2); // type: int
5390 Node* end = argument(3); // type: int
5391
5392 Node* length = _gvn.transform(new SubINode(end, offset));
5393
5394 const Type* src_type = src->Value(&_gvn);
5395 const TypeAryPtr* top_src = src_type->isa_aryptr();
5396 if (top_src == NULL || top_src->klass() == NULL) {
5397 // failed array check
5398 return false;
5399 }
5400
5401 // Figure out the size and type of the elements we will be copying.
5402 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5403 if (src_elem != T_BYTE) {
5404 return false;
5405 }
5406
5407 // 'src_start' points to src array + scaled offset
5408 Node* src_start = array_element_address(src, offset, src_elem);
5409
5410 // static final int[] byteTable in class CRC32C
5411 Node* table = get_table_from_crc32c_class(callee()->holder());
5412 Node* table_start = array_element_address(table, intcon(0), T_INT);
5413
5414 // We assume that range check is done by caller.
5415 // TODO: generate range check (offset+length < src.length) in debug VM.
5416
5417 // Call the stub.
5418 address stubAddr = StubRoutines::updateBytesCRC32C();
5419 const char *stubName = "updateBytesCRC32C";
5420
5421 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5422 stubAddr, stubName, TypePtr::BOTTOM,
5423 crc, src_start, length, table_start);
5424 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5425 set_result(result);
5426 return true;
5427 }
5428
5429 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5430 //
5431 // Calculate CRC32C for DirectByteBuffer.
5432 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5433 //
5434 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5435 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5436 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5437 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5438 // no receiver since it is a static method
5439 Node* crc = argument(0); // type: int
5440 Node* src = argument(1); // type: long
5441 Node* offset = argument(3); // type: int
5442 Node* end = argument(4); // type: int
5443
5444 Node* length = _gvn.transform(new SubINode(end, offset));
5445
5446 src = ConvL2X(src); // adjust Java long to machine word
5447 Node* base = _gvn.transform(new CastX2PNode(src));
5448 offset = ConvI2X(offset);
5449
5450 // 'src_start' points to src array + scaled offset
5451 Node* src_start = basic_plus_adr(top(), base, offset);
5452
5453 // static final int[] byteTable in class CRC32C
5454 Node* table = get_table_from_crc32c_class(callee()->holder());
5455 Node* table_start = array_element_address(table, intcon(0), T_INT);
5456
5457 // Call the stub.
5458 address stubAddr = StubRoutines::updateBytesCRC32C();
5459 const char *stubName = "updateBytesCRC32C";
5460
5461 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5462 stubAddr, stubName, TypePtr::BOTTOM,
5463 crc, src_start, length, table_start);
5464 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5465 set_result(result);
5466 return true;
5467 }
5468
5469 //------------------------------inline_updateBytesAdler32----------------------
5470 //
5471 // Calculate Adler32 checksum for byte[] array.
5472 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5473 //
5474 bool LibraryCallKit::inline_updateBytesAdler32() {
5475 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5476 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5477 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5478 // no receiver since it is static method
5479 Node* crc = argument(0); // type: int
5480 Node* src = argument(1); // type: oop
5481 Node* offset = argument(2); // type: int
5482 Node* length = argument(3); // type: int
5483
5484 const Type* src_type = src->Value(&_gvn);
5485 const TypeAryPtr* top_src = src_type->isa_aryptr();
5486 if (top_src == NULL || top_src->klass() == NULL) {
5487 // failed array check
5488 return false;
5489 }
5490
5491 // Figure out the size and type of the elements we will be copying.
5492 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5493 if (src_elem != T_BYTE) {
5494 return false;
5495 }
5496
5497 // 'src_start' points to src array + scaled offset
5498 Node* src_start = array_element_address(src, offset, src_elem);
5499
5500 // We assume that range check is done by caller.
5501 // TODO: generate range check (offset+length < src.length) in debug VM.
5502
5503 // Call the stub.
5504 address stubAddr = StubRoutines::updateBytesAdler32();
5505 const char *stubName = "updateBytesAdler32";
5506
5507 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5508 stubAddr, stubName, TypePtr::BOTTOM,
5509 crc, src_start, length);
5510 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5511 set_result(result);
5512 return true;
5513 }
5514
5515 //------------------------------inline_updateByteBufferAdler32---------------
5516 //
5517 // Calculate Adler32 checksum for DirectByteBuffer.
5518 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5519 //
5520 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5521 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5522 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5523 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5524 // no receiver since it is static method
5525 Node* crc = argument(0); // type: int
5526 Node* src = argument(1); // type: long
5527 Node* offset = argument(3); // type: int
5528 Node* length = argument(4); // type: int
5529
5530 src = ConvL2X(src); // adjust Java long to machine word
5531 Node* base = _gvn.transform(new CastX2PNode(src));
5532 offset = ConvI2X(offset);
5533
5534 // 'src_start' points to src array + scaled offset
5535 Node* src_start = basic_plus_adr(top(), base, offset);
5536
5537 // Call the stub.
5538 address stubAddr = StubRoutines::updateBytesAdler32();
5539 const char *stubName = "updateBytesAdler32";
5540
5541 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5542 stubAddr, stubName, TypePtr::BOTTOM,
5543 crc, src_start, length);
5544
5545 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5546 set_result(result);
5547 return true;
5548 }
5549
5550 //----------------------------inline_reference_get----------------------------
5551 // public T java.lang.ref.Reference.get();
5552 bool LibraryCallKit::inline_reference_get() {
5553 const int referent_offset = java_lang_ref_Reference::referent_offset;
5554 guarantee(referent_offset > 0, "should have already been set");
5555
5556 // Get the argument:
5557 Node* reference_obj = null_check_receiver();
5558 if (stopped()) return true;
5559
5560 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5561 assert(tinst != NULL, "obj is null");
5562 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5563 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5564 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5565 ciSymbol::make("Ljava/lang/Object;"),
5566 false);
5567 assert (field != NULL, "undefined field");
5568
5569 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5570 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5571
5572 ciInstanceKlass* klass = env()->Object_klass();
5573 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5574
5575 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5576 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5577 // Add memory barrier to prevent commoning reads from this field
5578 // across safepoint since GC can change its value.
5579 insert_mem_bar(Op_MemBarCPUOrder);
5580
5581 set_result(result);
5582 return true;
5583 }
5584
5585
5586 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5587 bool is_exact=true, bool is_static=false,
5588 ciInstanceKlass * fromKls=NULL) {
5589 if (fromKls == NULL) {
5590 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5591 assert(tinst != NULL, "obj is null");
5592 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5593 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5594 fromKls = tinst->klass()->as_instance_klass();
5595 } else {
5596 assert(is_static, "only for static field access");
5597 }
5598 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5599 ciSymbol::make(fieldTypeString),
5600 is_static);
5601
5602 assert (field != NULL, "undefined field");
5603 if (field == NULL) return (Node *) NULL;
5604
5605 if (is_static) {
5606 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5607 fromObj = makecon(tip);
5608 }
5609
5610 // Next code copied from Parse::do_get_xxx():
5611
5612 // Compute address and memory type.
5613 int offset = field->offset_in_bytes();
5614 bool is_vol = field->is_volatile();
5615 ciType* field_klass = field->type();
5616 assert(field_klass->is_loaded(), "should be loaded");
5617 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5618 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5619 BasicType bt = field->layout_type();
5620
5621 // Build the resultant type of the load
5622 const Type *type;
5623 if (bt == T_OBJECT) {
5624 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5625 } else {
5626 type = Type::get_const_basic_type(bt);
5627 }
5628
5629 DecoratorSet decorators = IN_HEAP;
5630
5631 if (is_vol) {
5632 decorators |= MO_SEQ_CST;
5633 }
5634
5635 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5636 }
5637
5638 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5639 bool is_exact = true, bool is_static = false,
5640 ciInstanceKlass * fromKls = NULL) {
5641 if (fromKls == NULL) {
5642 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5643 assert(tinst != NULL, "obj is null");
5644 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5645 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5646 fromKls = tinst->klass()->as_instance_klass();
5647 }
5648 else {
5649 assert(is_static, "only for static field access");
5650 }
5651 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5652 ciSymbol::make(fieldTypeString),
5653 is_static);
5654
5655 assert(field != NULL, "undefined field");
5656 assert(!field->is_volatile(), "not defined for volatile fields");
5657
5658 if (is_static) {
5659 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5660 fromObj = makecon(tip);
5661 }
5662
5663 // Next code copied from Parse::do_get_xxx():
5664
5665 // Compute address and memory type.
5666 int offset = field->offset_in_bytes();
5667 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5668
5669 return adr;
5670 }
5671
5672 //------------------------------inline_aescrypt_Block-----------------------
5673 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5674 address stubAddr = NULL;
5675 const char *stubName;
5676 assert(UseAES, "need AES instruction support");
5677
5678 switch(id) {
5679 case vmIntrinsics::_aescrypt_encryptBlock:
5680 stubAddr = StubRoutines::aescrypt_encryptBlock();
5681 stubName = "aescrypt_encryptBlock";
5682 break;
5683 case vmIntrinsics::_aescrypt_decryptBlock:
5684 stubAddr = StubRoutines::aescrypt_decryptBlock();
5685 stubName = "aescrypt_decryptBlock";
5686 break;
5687 default:
5688 break;
5689 }
5690 if (stubAddr == NULL) return false;
5691
5692 Node* aescrypt_object = argument(0);
5693 Node* src = argument(1);
5694 Node* src_offset = argument(2);
5695 Node* dest = argument(3);
5696 Node* dest_offset = argument(4);
5697
5698 // (1) src and dest are arrays.
5699 const Type* src_type = src->Value(&_gvn);
5700 const Type* dest_type = dest->Value(&_gvn);
5701 const TypeAryPtr* top_src = src_type->isa_aryptr();
5702 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5703 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5704
5705 // for the quick and dirty code we will skip all the checks.
5706 // we are just trying to get the call to be generated.
5707 Node* src_start = src;
5708 Node* dest_start = dest;
5709 if (src_offset != NULL || dest_offset != NULL) {
5710 assert(src_offset != NULL && dest_offset != NULL, "");
5711 src_start = array_element_address(src, src_offset, T_BYTE);
5712 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5713 }
5714
5715 // now need to get the start of its expanded key array
5716 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5717 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5718 if (k_start == NULL) return false;
5719
5720 if (Matcher::pass_original_key_for_aes()) {
5721 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5722 // compatibility issues between Java key expansion and SPARC crypto instructions
5723 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5724 if (original_k_start == NULL) return false;
5725
5726 // Call the stub.
5727 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5728 stubAddr, stubName, TypePtr::BOTTOM,
5729 src_start, dest_start, k_start, original_k_start);
5730 } else {
5731 // Call the stub.
5732 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5733 stubAddr, stubName, TypePtr::BOTTOM,
5734 src_start, dest_start, k_start);
5735 }
5736
5737 return true;
5738 }
5739
5740 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5741 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5742 address stubAddr = NULL;
5743 const char *stubName = NULL;
5744
5745 assert(UseAES, "need AES instruction support");
5746
5747 switch(id) {
5748 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5749 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5750 stubName = "cipherBlockChaining_encryptAESCrypt";
5751 break;
5752 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5753 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5754 stubName = "cipherBlockChaining_decryptAESCrypt";
5755 break;
5756 default:
5757 break;
5758 }
5759 if (stubAddr == NULL) return false;
5760
5761 Node* cipherBlockChaining_object = argument(0);
5762 Node* src = argument(1);
5763 Node* src_offset = argument(2);
5764 Node* len = argument(3);
5765 Node* dest = argument(4);
5766 Node* dest_offset = argument(5);
5767
5768 // (1) src and dest are arrays.
5769 const Type* src_type = src->Value(&_gvn);
5770 const Type* dest_type = dest->Value(&_gvn);
5771 const TypeAryPtr* top_src = src_type->isa_aryptr();
5772 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5773 assert (top_src != NULL && top_src->klass() != NULL
5774 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5775
5776 // checks are the responsibility of the caller
5777 Node* src_start = src;
5778 Node* dest_start = dest;
5779 if (src_offset != NULL || dest_offset != NULL) {
5780 assert(src_offset != NULL && dest_offset != NULL, "");
5781 src_start = array_element_address(src, src_offset, T_BYTE);
5782 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5783 }
5784
5785 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5786 // (because of the predicated logic executed earlier).
5787 // so we cast it here safely.
5788 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5789
5790 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5791 if (embeddedCipherObj == NULL) return false;
5792
5793 // cast it to what we know it will be at runtime
5794 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5795 assert(tinst != NULL, "CBC obj is null");
5796 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5797 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5798 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5799
5800 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5801 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5802 const TypeOopPtr* xtype = aklass->as_instance_type();
5803 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5804 aescrypt_object = _gvn.transform(aescrypt_object);
5805
5806 // we need to get the start of the aescrypt_object's expanded key array
5807 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5808 if (k_start == NULL) return false;
5809
5810 // similarly, get the start address of the r vector
5811 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5812 if (objRvec == NULL) return false;
5813 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5814
5815 Node* cbcCrypt;
5816 if (Matcher::pass_original_key_for_aes()) {
5817 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5818 // compatibility issues between Java key expansion and SPARC crypto instructions
5819 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5820 if (original_k_start == NULL) return false;
5821
5822 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5823 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5824 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5825 stubAddr, stubName, TypePtr::BOTTOM,
5826 src_start, dest_start, k_start, r_start, len, original_k_start);
5827 } else {
5828 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5829 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5830 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5831 stubAddr, stubName, TypePtr::BOTTOM,
5832 src_start, dest_start, k_start, r_start, len);
5833 }
5834
5835 // return cipher length (int)
5836 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5837 set_result(retvalue);
5838 return true;
5839 }
5840
5841 //------------------------------inline_counterMode_AESCrypt-----------------------
5842 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
5843 assert(UseAES, "need AES instruction support");
5844 if (!UseAESCTRIntrinsics) return false;
5845
5846 address stubAddr = NULL;
5847 const char *stubName = NULL;
5848 if (id == vmIntrinsics::_counterMode_AESCrypt) {
5849 stubAddr = StubRoutines::counterMode_AESCrypt();
5850 stubName = "counterMode_AESCrypt";
5851 }
5852 if (stubAddr == NULL) return false;
5853
5854 Node* counterMode_object = argument(0);
5855 Node* src = argument(1);
5856 Node* src_offset = argument(2);
5857 Node* len = argument(3);
5858 Node* dest = argument(4);
5859 Node* dest_offset = argument(5);
5860
5861 // (1) src and dest are arrays.
5862 const Type* src_type = src->Value(&_gvn);
5863 const Type* dest_type = dest->Value(&_gvn);
5864 const TypeAryPtr* top_src = src_type->isa_aryptr();
5865 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5866 assert(top_src != NULL && top_src->klass() != NULL &&
5867 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5868
5869 // checks are the responsibility of the caller
5870 Node* src_start = src;
5871 Node* dest_start = dest;
5872 if (src_offset != NULL || dest_offset != NULL) {
5873 assert(src_offset != NULL && dest_offset != NULL, "");
5874 src_start = array_element_address(src, src_offset, T_BYTE);
5875 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5876 }
5877
5878 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5879 // (because of the predicated logic executed earlier).
5880 // so we cast it here safely.
5881 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5882 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5883 if (embeddedCipherObj == NULL) return false;
5884 // cast it to what we know it will be at runtime
5885 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
5886 assert(tinst != NULL, "CTR obj is null");
5887 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
5888 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5889 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5890 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5891 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5892 const TypeOopPtr* xtype = aklass->as_instance_type();
5893 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5894 aescrypt_object = _gvn.transform(aescrypt_object);
5895 // we need to get the start of the aescrypt_object's expanded key array
5896 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5897 if (k_start == NULL) return false;
5898 // similarly, get the start address of the r vector
5899 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
5900 if (obj_counter == NULL) return false;
5901 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
5902
5903 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
5904 if (saved_encCounter == NULL) return false;
5905 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
5906 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
5907
5908 Node* ctrCrypt;
5909 if (Matcher::pass_original_key_for_aes()) {
5910 // no SPARC version for AES/CTR intrinsics now.
5911 return false;
5912 }
5913 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5914 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5915 OptoRuntime::counterMode_aescrypt_Type(),
5916 stubAddr, stubName, TypePtr::BOTTOM,
5917 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
5918
5919 // return cipher length (int)
5920 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
5921 set_result(retvalue);
5922 return true;
5923 }
5924
5925 //------------------------------get_key_start_from_aescrypt_object-----------------------
5926 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
5927 #if defined(PPC64) || defined(S390)
5928 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
5929 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
5930 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
5931 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
5932 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
5933 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5934 if (objSessionK == NULL) {
5935 return (Node *) NULL;
5936 }
5937 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
5938 #else
5939 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
5940 #endif // PPC64
5941 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5942 if (objAESCryptKey == NULL) return (Node *) NULL;
5943
5944 // now have the array, need to get the start address of the K array
5945 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
5946 return k_start;
5947 }
5948
5949 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
5950 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
5951 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
5952 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5953 if (objAESCryptKey == NULL) return (Node *) NULL;
5954
5955 // now have the array, need to get the start address of the lastKey array
5956 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
5957 return original_k_start;
5958 }
5959
5960 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
5961 // Return node representing slow path of predicate check.
5962 // the pseudo code we want to emulate with this predicate is:
5963 // for encryption:
5964 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
5965 // for decryption:
5966 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
5967 // note cipher==plain is more conservative than the original java code but that's OK
5968 //
5969 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
5970 // The receiver was checked for NULL already.
5971 Node* objCBC = argument(0);
5972
5973 // Load embeddedCipher field of CipherBlockChaining object.
5974 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5975
5976 // get AESCrypt klass for instanceOf check
5977 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
5978 // will have same classloader as CipherBlockChaining object
5979 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
5980 assert(tinst != NULL, "CBCobj is null");
5981 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
5982
5983 // we want to do an instanceof comparison against the AESCrypt class
5984 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5985 if (!klass_AESCrypt->is_loaded()) {
5986 // if AESCrypt is not even loaded, we never take the intrinsic fast path
5987 Node* ctrl = control();
5988 set_control(top()); // no regular fast path
5989 return ctrl;
5990 }
5991 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5992
5993 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
5994 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
5995 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
5996
5997 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
5998
5999 // for encryption, we are done
6000 if (!decrypting)
6001 return instof_false; // even if it is NULL
6002
6003 // for decryption, we need to add a further check to avoid
6004 // taking the intrinsic path when cipher and plain are the same
6005 // see the original java code for why.
6006 RegionNode* region = new RegionNode(3);
6007 region->init_req(1, instof_false);
6008 Node* src = argument(1);
6009 Node* dest = argument(4);
6010 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6011 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6012 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6013 region->init_req(2, src_dest_conjoint);
6014
6015 record_for_igvn(region);
6016 return _gvn.transform(region);
6017 }
6018
6019 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6020 // Return node representing slow path of predicate check.
6021 // the pseudo code we want to emulate with this predicate is:
6022 // for encryption:
6023 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6024 // for decryption:
6025 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6026 // note cipher==plain is more conservative than the original java code but that's OK
6027 //
6028
6029 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6030 // The receiver was checked for NULL already.
6031 Node* objCTR = argument(0);
6032
6033 // Load embeddedCipher field of CipherBlockChaining object.
6034 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6035
6036 // get AESCrypt klass for instanceOf check
6037 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6038 // will have same classloader as CipherBlockChaining object
6039 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6040 assert(tinst != NULL, "CTRobj is null");
6041 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6042
6043 // we want to do an instanceof comparison against the AESCrypt class
6044 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6045 if (!klass_AESCrypt->is_loaded()) {
6046 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6047 Node* ctrl = control();
6048 set_control(top()); // no regular fast path
6049 return ctrl;
6050 }
6051
6052 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6053 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6054 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6055 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6056 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6057
6058 return instof_false; // even if it is NULL
6059 }
6060
6061 //------------------------------inline_ghash_processBlocks
6062 bool LibraryCallKit::inline_ghash_processBlocks() {
6063 address stubAddr;
6064 const char *stubName;
6065 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6066
6067 stubAddr = StubRoutines::ghash_processBlocks();
6068 stubName = "ghash_processBlocks";
6069
6070 Node* data = argument(0);
6071 Node* offset = argument(1);
6072 Node* len = argument(2);
6073 Node* state = argument(3);
6074 Node* subkeyH = argument(4);
6075
6076 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6077 assert(state_start, "state is NULL");
6078 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6079 assert(subkeyH_start, "subkeyH is NULL");
6080 Node* data_start = array_element_address(data, offset, T_BYTE);
6081 assert(data_start, "data is NULL");
6082
6083 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6084 OptoRuntime::ghash_processBlocks_Type(),
6085 stubAddr, stubName, TypePtr::BOTTOM,
6086 state_start, subkeyH_start, data_start, len);
6087 return true;
6088 }
6089
6090 bool LibraryCallKit::inline_base64_encodeBlock() {
6091 address stubAddr;
6092 const char *stubName;
6093 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6094 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6095 stubAddr = StubRoutines::base64_encodeBlock();
6096 stubName = "encodeBlock";
6097
6098 if (!stubAddr) return false;
6099 Node* base64obj = argument(0);
6100 Node* src = argument(1);
6101 Node* offset = argument(2);
6102 Node* len = argument(3);
6103 Node* dest = argument(4);
6104 Node* dp = argument(5);
6105 Node* isURL = argument(6);
6106
6107 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6108 assert(src_start, "source array is NULL");
6109 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6110 assert(dest_start, "destination array is NULL");
6111
6112 Node* base64 = make_runtime_call(RC_LEAF,
6113 OptoRuntime::base64_encodeBlock_Type(),
6114 stubAddr, stubName, TypePtr::BOTTOM,
6115 src_start, offset, len, dest_start, dp, isURL);
6116 return true;
6117 }
6118
6119 //------------------------------inline_sha_implCompress-----------------------
6120 //
6121 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6122 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6123 //
6124 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6125 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6126 //
6127 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6128 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6129 //
6130 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6131 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6132
6133 Node* sha_obj = argument(0);
6134 Node* src = argument(1); // type oop
6135 Node* ofs = argument(2); // type int
6136
6137 const Type* src_type = src->Value(&_gvn);
6138 const TypeAryPtr* top_src = src_type->isa_aryptr();
6139 if (top_src == NULL || top_src->klass() == NULL) {
6140 // failed array check
6141 return false;
6142 }
6143 // Figure out the size and type of the elements we will be copying.
6144 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6145 if (src_elem != T_BYTE) {
6146 return false;
6147 }
6148 // 'src_start' points to src array + offset
6149 Node* src_start = array_element_address(src, ofs, src_elem);
6150 Node* state = NULL;
6151 address stubAddr;
6152 const char *stubName;
6153
6154 switch(id) {
6155 case vmIntrinsics::_sha_implCompress:
6156 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6157 state = get_state_from_sha_object(sha_obj);
6158 stubAddr = StubRoutines::sha1_implCompress();
6159 stubName = "sha1_implCompress";
6160 break;
6161 case vmIntrinsics::_sha2_implCompress:
6162 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6163 state = get_state_from_sha_object(sha_obj);
6164 stubAddr = StubRoutines::sha256_implCompress();
6165 stubName = "sha256_implCompress";
6166 break;
6167 case vmIntrinsics::_sha5_implCompress:
6168 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6169 state = get_state_from_sha5_object(sha_obj);
6170 stubAddr = StubRoutines::sha512_implCompress();
6171 stubName = "sha512_implCompress";
6172 break;
6173 default:
6174 fatal_unexpected_iid(id);
6175 return false;
6176 }
6177 if (state == NULL) return false;
6178
6179 // Call the stub.
6180 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6181 stubAddr, stubName, TypePtr::BOTTOM,
6182 src_start, state);
6183
6184 return true;
6185 }
6186
6187 //------------------------------inline_digestBase_implCompressMB-----------------------
6188 //
6189 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6190 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6191 //
6192 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6193 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6194 "need SHA1/SHA256/SHA512 instruction support");
6195 assert((uint)predicate < 3, "sanity");
6196 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6197
6198 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6199 Node* src = argument(1); // byte[] array
6200 Node* ofs = argument(2); // type int
6201 Node* limit = argument(3); // type int
6202
6203 const Type* src_type = src->Value(&_gvn);
6204 const TypeAryPtr* top_src = src_type->isa_aryptr();
6205 if (top_src == NULL || top_src->klass() == NULL) {
6206 // failed array check
6207 return false;
6208 }
6209 // Figure out the size and type of the elements we will be copying.
6210 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6211 if (src_elem != T_BYTE) {
6212 return false;
6213 }
6214 // 'src_start' points to src array + offset
6215 Node* src_start = array_element_address(src, ofs, src_elem);
6216
6217 const char* klass_SHA_name = NULL;
6218 const char* stub_name = NULL;
6219 address stub_addr = NULL;
6220 bool long_state = false;
6221
6222 switch (predicate) {
6223 case 0:
6224 if (UseSHA1Intrinsics) {
6225 klass_SHA_name = "sun/security/provider/SHA";
6226 stub_name = "sha1_implCompressMB";
6227 stub_addr = StubRoutines::sha1_implCompressMB();
6228 }
6229 break;
6230 case 1:
6231 if (UseSHA256Intrinsics) {
6232 klass_SHA_name = "sun/security/provider/SHA2";
6233 stub_name = "sha256_implCompressMB";
6234 stub_addr = StubRoutines::sha256_implCompressMB();
6235 }
6236 break;
6237 case 2:
6238 if (UseSHA512Intrinsics) {
6239 klass_SHA_name = "sun/security/provider/SHA5";
6240 stub_name = "sha512_implCompressMB";
6241 stub_addr = StubRoutines::sha512_implCompressMB();
6242 long_state = true;
6243 }
6244 break;
6245 default:
6246 fatal("unknown SHA intrinsic predicate: %d", predicate);
6247 }
6248 if (klass_SHA_name != NULL) {
6249 // get DigestBase klass to lookup for SHA klass
6250 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6251 assert(tinst != NULL, "digestBase_obj is not instance???");
6252 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6253
6254 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6255 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6256 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6257 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6258 }
6259 return false;
6260 }
6261 //------------------------------inline_sha_implCompressMB-----------------------
6262 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6263 bool long_state, address stubAddr, const char *stubName,
6264 Node* src_start, Node* ofs, Node* limit) {
6265 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6266 const TypeOopPtr* xtype = aklass->as_instance_type();
6267 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6268 sha_obj = _gvn.transform(sha_obj);
6269
6270 Node* state;
6271 if (long_state) {
6272 state = get_state_from_sha5_object(sha_obj);
6273 } else {
6274 state = get_state_from_sha_object(sha_obj);
6275 }
6276 if (state == NULL) return false;
6277
6278 // Call the stub.
6279 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6280 OptoRuntime::digestBase_implCompressMB_Type(),
6281 stubAddr, stubName, TypePtr::BOTTOM,
6282 src_start, state, ofs, limit);
6283 // return ofs (int)
6284 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6285 set_result(result);
6286
6287 return true;
6288 }
6289
6290 //------------------------------get_state_from_sha_object-----------------------
6291 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6292 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6293 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6294 if (sha_state == NULL) return (Node *) NULL;
6295
6296 // now have the array, need to get the start address of the state array
6297 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6298 return state;
6299 }
6300
6301 //------------------------------get_state_from_sha5_object-----------------------
6302 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6303 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6304 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6305 if (sha_state == NULL) return (Node *) NULL;
6306
6307 // now have the array, need to get the start address of the state array
6308 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6309 return state;
6310 }
6311
6312 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6313 // Return node representing slow path of predicate check.
6314 // the pseudo code we want to emulate with this predicate is:
6315 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6316 //
6317 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6318 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6319 "need SHA1/SHA256/SHA512 instruction support");
6320 assert((uint)predicate < 3, "sanity");
6321
6322 // The receiver was checked for NULL already.
6323 Node* digestBaseObj = argument(0);
6324
6325 // get DigestBase klass for instanceOf check
6326 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6327 assert(tinst != NULL, "digestBaseObj is null");
6328 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6329
6330 const char* klass_SHA_name = NULL;
6331 switch (predicate) {
6332 case 0:
6333 if (UseSHA1Intrinsics) {
6334 // we want to do an instanceof comparison against the SHA class
6335 klass_SHA_name = "sun/security/provider/SHA";
6336 }
6337 break;
6338 case 1:
6339 if (UseSHA256Intrinsics) {
6340 // we want to do an instanceof comparison against the SHA2 class
6341 klass_SHA_name = "sun/security/provider/SHA2";
6342 }
6343 break;
6344 case 2:
6345 if (UseSHA512Intrinsics) {
6346 // we want to do an instanceof comparison against the SHA5 class
6347 klass_SHA_name = "sun/security/provider/SHA5";
6348 }
6349 break;
6350 default:
6351 fatal("unknown SHA intrinsic predicate: %d", predicate);
6352 }
6353
6354 ciKlass* klass_SHA = NULL;
6355 if (klass_SHA_name != NULL) {
6356 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6357 }
6358 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6359 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6360 Node* ctrl = control();
6361 set_control(top()); // no intrinsic path
6362 return ctrl;
6363 }
6364 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6365
6366 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6367 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6368 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6369 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6370
6371 return instof_false; // even if it is NULL
6372 }
6373
6374 //-------------inline_fma-----------------------------------
6375 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6376 Node *a = NULL;
6377 Node *b = NULL;
6378 Node *c = NULL;
6379 Node* result = NULL;
6380 switch (id) {
6381 case vmIntrinsics::_fmaD:
6382 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6383 // no receiver since it is static method
6384 a = round_double_node(argument(0));
6385 b = round_double_node(argument(2));
6386 c = round_double_node(argument(4));
6387 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6388 break;
6389 case vmIntrinsics::_fmaF:
6390 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6391 a = argument(0);
6392 b = argument(1);
6393 c = argument(2);
6394 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6395 break;
6396 default:
6397 fatal_unexpected_iid(id); break;
6398 }
6399 set_result(result);
6400 return true;
6401 }
6402
6403 bool LibraryCallKit::inline_profileBoolean() {
6404 Node* counts = argument(1);
6405 const TypeAryPtr* ary = NULL;
6406 ciArray* aobj = NULL;
6407 if (counts->is_Con()
6408 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6409 && (aobj = ary->const_oop()->as_array()) != NULL
6410 && (aobj->length() == 2)) {
6411 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6412 jint false_cnt = aobj->element_value(0).as_int();
6413 jint true_cnt = aobj->element_value(1).as_int();
6414
6415 if (C->log() != NULL) {
6416 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6417 false_cnt, true_cnt);
6418 }
6419
6420 if (false_cnt + true_cnt == 0) {
6421 // According to profile, never executed.
6422 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6423 Deoptimization::Action_reinterpret);
6424 return true;
6425 }
6426
6427 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6428 // is a number of each value occurrences.
6429 Node* result = argument(0);
6430 if (false_cnt == 0 || true_cnt == 0) {
6431 // According to profile, one value has been never seen.
6432 int expected_val = (false_cnt == 0) ? 1 : 0;
6433
6434 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6435 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6436
6437 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6438 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6439 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6440
6441 { // Slow path: uncommon trap for never seen value and then reexecute
6442 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6443 // the value has been seen at least once.
6444 PreserveJVMState pjvms(this);
6445 PreserveReexecuteState preexecs(this);
6446 jvms()->set_should_reexecute(true);
6447
6448 set_control(slow_path);
6449 set_i_o(i_o());
6450
6451 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6452 Deoptimization::Action_reinterpret);
6453 }
6454 // The guard for never seen value enables sharpening of the result and
6455 // returning a constant. It allows to eliminate branches on the same value
6456 // later on.
6457 set_control(fast_path);
6458 result = intcon(expected_val);
6459 }
6460 // Stop profiling.
6461 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6462 // By replacing method body with profile data (represented as ProfileBooleanNode
6463 // on IR level) we effectively disable profiling.
6464 // It enables full speed execution once optimized code is generated.
6465 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6466 C->record_for_igvn(profile);
6467 set_result(profile);
6468 return true;
6469 } else {
6470 // Continue profiling.
6471 // Profile data isn't available at the moment. So, execute method's bytecode version.
6472 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6473 // is compiled and counters aren't available since corresponding MethodHandle
6474 // isn't a compile-time constant.
6475 return false;
6476 }
6477 }
6478
6479 bool LibraryCallKit::inline_isCompileConstant() {
6480 Node* n = argument(0);
6481 set_result(n->is_Con() ? intcon(1) : intcon(0));
6482 return true;
6483 }