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