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_writeback0();
256 bool inline_unsafe_writebackSync0(bool isPre);
257 bool inline_unsafe_copyMemory();
258 bool inline_native_currentThread();
259
260 bool inline_native_time_funcs(address method, const char* funcName);
261 #ifdef JFR_HAVE_INTRINSICS
262 bool inline_native_classID();
263 bool inline_native_getEventWriter();
264 #endif
265 bool inline_native_isInterrupted();
266 bool inline_native_Class_query(vmIntrinsics::ID id);
267 bool inline_native_subtype_check();
268 bool inline_native_getLength();
269 bool inline_array_copyOf(bool is_copyOfRange);
270 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
271 bool inline_preconditions_checkIndex();
272 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
273 bool inline_native_clone(bool is_virtual);
274 bool inline_native_Reflection_getCallerClass();
275 // Helper function for inlining native object hash method
276 bool inline_native_hashcode(bool is_virtual, bool is_static);
277 bool inline_native_getClass();
278
279 // Helper functions for inlining arraycopy
280 bool inline_arraycopy();
281 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
282 RegionNode* slow_region);
283 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
284 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
285 uint new_idx);
286
287 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
288 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
289 bool inline_unsafe_fence(vmIntrinsics::ID id);
290 bool inline_onspinwait();
291 bool inline_fp_conversions(vmIntrinsics::ID id);
292 bool inline_number_methods(vmIntrinsics::ID id);
293 bool inline_reference_get();
294 bool inline_Class_cast();
295 bool inline_aescrypt_Block(vmIntrinsics::ID id);
296 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
297 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
298 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
299 Node* inline_counterMode_AESCrypt_predicate();
300 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
301 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
302 bool inline_ghash_processBlocks();
303 bool inline_base64_encodeBlock();
304 bool inline_sha_implCompress(vmIntrinsics::ID id);
305 bool inline_digestBase_implCompressMB(int predicate);
306 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
307 bool long_state, address stubAddr, const char *stubName,
308 Node* src_start, Node* ofs, Node* limit);
309 Node* get_state_from_sha_object(Node *sha_object);
310 Node* get_state_from_sha5_object(Node *sha_object);
311 Node* inline_digestBase_implCompressMB_predicate(int predicate);
312 bool inline_encodeISOArray();
313 bool inline_updateCRC32();
314 bool inline_updateBytesCRC32();
315 bool inline_updateByteBufferCRC32();
316 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
317 bool inline_updateBytesCRC32C();
318 bool inline_updateDirectByteBufferCRC32C();
319 bool inline_updateBytesAdler32();
320 bool inline_updateByteBufferAdler32();
321 bool inline_multiplyToLen();
322 bool inline_hasNegatives();
323 bool inline_squareToLen();
324 bool inline_mulAdd();
325 bool inline_montgomeryMultiply();
326 bool inline_montgomerySquare();
327 bool inline_vectorizedMismatch();
328 bool inline_fma(vmIntrinsics::ID id);
329
330 bool inline_profileBoolean();
331 bool inline_isCompileConstant();
332 void clear_upper_avx() {
333 #ifdef X86
334 if (UseAVX >= 2) {
335 C->set_clear_upper_avx(true);
336 }
337 #endif
338 }
339 };
340
341 //---------------------------make_vm_intrinsic----------------------------
342 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
343 vmIntrinsics::ID id = m->intrinsic_id();
344 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
345
346 if (!m->is_loaded()) {
347 // Do not attempt to inline unloaded methods.
348 return NULL;
349 }
350
351 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
352 bool is_available = false;
353
354 {
355 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
356 // the compiler must transition to '_thread_in_vm' state because both
357 // methods access VM-internal data.
358 VM_ENTRY_MARK;
359 methodHandle mh(THREAD, m->get_Method());
360 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
361 !C->directive()->is_intrinsic_disabled(mh) &&
362 !vmIntrinsics::is_disabled_by_flags(mh);
363
364 }
365
366 if (is_available) {
367 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
368 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
369 return new LibraryIntrinsic(m, is_virtual,
370 vmIntrinsics::predicates_needed(id),
371 vmIntrinsics::does_virtual_dispatch(id),
372 (vmIntrinsics::ID) id);
373 } else {
374 return NULL;
375 }
376 }
377
378 //----------------------register_library_intrinsics-----------------------
379 // Initialize this file's data structures, for each Compile instance.
380 void Compile::register_library_intrinsics() {
381 // Nothing to do here.
382 }
383
384 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
385 LibraryCallKit kit(jvms, this);
386 Compile* C = kit.C;
387 int nodes = C->unique();
388 #ifndef PRODUCT
389 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
390 char buf[1000];
391 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
392 tty->print_cr("Intrinsic %s", str);
393 }
394 #endif
395 ciMethod* callee = kit.callee();
396 const int bci = kit.bci();
397
398 // Try to inline the intrinsic.
399 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
400 kit.try_to_inline(_last_predicate)) {
401 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
402 : "(intrinsic)";
403 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
404 if (C->print_intrinsics() || C->print_inlining()) {
405 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
406 }
407 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
408 if (C->log()) {
409 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
410 vmIntrinsics::name_at(intrinsic_id()),
411 (is_virtual() ? " virtual='1'" : ""),
412 C->unique() - nodes);
413 }
414 // Push the result from the inlined method onto the stack.
415 kit.push_result();
416 C->print_inlining_update(this);
417 return kit.transfer_exceptions_into_jvms();
418 }
419
420 // The intrinsic bailed out
421 if (jvms->has_method()) {
422 // Not a root compile.
423 const char* msg;
424 if (callee->intrinsic_candidate()) {
425 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
426 } else {
427 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
428 : "failed to inline (intrinsic), method not annotated";
429 }
430 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
431 if (C->print_intrinsics() || C->print_inlining()) {
432 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
433 }
434 } else {
435 // Root compile
436 ResourceMark rm;
437 stringStream msg_stream;
438 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
439 vmIntrinsics::name_at(intrinsic_id()),
440 is_virtual() ? " (virtual)" : "", bci);
441 const char *msg = msg_stream.as_string();
442 log_debug(jit, inlining)("%s", msg);
443 if (C->print_intrinsics() || C->print_inlining()) {
444 tty->print("%s", msg);
445 }
446 }
447 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
448 C->print_inlining_update(this);
449 return NULL;
450 }
451
452 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
453 LibraryCallKit kit(jvms, this);
454 Compile* C = kit.C;
455 int nodes = C->unique();
456 _last_predicate = predicate;
457 #ifndef PRODUCT
458 assert(is_predicated() && predicate < predicates_count(), "sanity");
459 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
460 char buf[1000];
461 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
462 tty->print_cr("Predicate for intrinsic %s", str);
463 }
464 #endif
465 ciMethod* callee = kit.callee();
466 const int bci = kit.bci();
467
468 Node* slow_ctl = kit.try_to_predicate(predicate);
469 if (!kit.failing()) {
470 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
471 : "(intrinsic, predicate)";
472 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
473 if (C->print_intrinsics() || C->print_inlining()) {
474 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
475 }
476 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
477 if (C->log()) {
478 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
479 vmIntrinsics::name_at(intrinsic_id()),
480 (is_virtual() ? " virtual='1'" : ""),
481 C->unique() - nodes);
482 }
483 return slow_ctl; // Could be NULL if the check folds.
484 }
485
486 // The intrinsic bailed out
487 if (jvms->has_method()) {
488 // Not a root compile.
489 const char* msg = "failed to generate predicate for intrinsic";
490 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
491 if (C->print_intrinsics() || C->print_inlining()) {
492 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
493 }
494 } else {
495 // Root compile
496 ResourceMark rm;
497 stringStream msg_stream;
498 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
499 vmIntrinsics::name_at(intrinsic_id()),
500 is_virtual() ? " (virtual)" : "", bci);
501 const char *msg = msg_stream.as_string();
502 log_debug(jit, inlining)("%s", msg);
503 if (C->print_intrinsics() || C->print_inlining()) {
504 C->print_inlining_stream()->print("%s", msg);
505 }
506 }
507 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
508 return NULL;
509 }
510
511 bool LibraryCallKit::try_to_inline(int predicate) {
512 // Handle symbolic names for otherwise undistinguished boolean switches:
513 const bool is_store = true;
514 const bool is_compress = true;
515 const bool is_static = true;
516 const bool is_volatile = true;
517
518 if (!jvms()->has_method()) {
519 // Root JVMState has a null method.
520 assert(map()->memory()->Opcode() == Op_Parm, "");
521 // Insert the memory aliasing node
522 set_all_memory(reset_memory());
523 }
524 assert(merged_memory(), "");
525
526
527 switch (intrinsic_id()) {
528 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
529 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
530 case vmIntrinsics::_getClass: return inline_native_getClass();
531
532 case vmIntrinsics::_dsin:
533 case vmIntrinsics::_dcos:
534 case vmIntrinsics::_dtan:
535 case vmIntrinsics::_dabs:
536 case vmIntrinsics::_datan2:
537 case vmIntrinsics::_dsqrt:
538 case vmIntrinsics::_dexp:
539 case vmIntrinsics::_dlog:
540 case vmIntrinsics::_dlog10:
541 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
542
543 case vmIntrinsics::_min:
544 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
545
546 case vmIntrinsics::_notify:
547 case vmIntrinsics::_notifyAll:
548 if (ObjectMonitor::Knob_InlineNotify) {
549 return inline_notify(intrinsic_id());
550 }
551 return false;
552
553 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
554 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
555 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
556 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
557 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
558 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
559 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
560 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
561 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
562 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
563 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
564 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
565 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
566
567 case vmIntrinsics::_arraycopy: return inline_arraycopy();
568
569 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
570 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
571 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
572 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
573
574 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
575 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
576 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
577 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
578 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
579 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
580 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
581
582 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
583 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
584
585 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
586 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
587 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
588 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
589
590 case vmIntrinsics::_compressStringC:
591 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
592 case vmIntrinsics::_inflateStringC:
593 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
594
595 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
596 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
597 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
598 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
599 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
600 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
601 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
602 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
603 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
604
605 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
606 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
607 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
608 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
609 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
610 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
611 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
612 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
613 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
614
615 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
616 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
617 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
618 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
619 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
620 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
621 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
622 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
623 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
624
625 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
626 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
627 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
628 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
629 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
630 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
631 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
632 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
633 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
634
635 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
636 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
637 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
638 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
639
640 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
641 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
642 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
643 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
644
645 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
646 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
647 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
648 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
649 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
650 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
651 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
652 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
653 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
654
655 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
656 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
657 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
658 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
659 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
660 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
661 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
662 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
663 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
664
665 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
666 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
667 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
668 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
669 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
670 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
671 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
672 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
673 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
674
675 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
676 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
677 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
678 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
679 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
680 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
681 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
682 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
683 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
684
685 case vmIntrinsics::_compareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
686 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
687 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
688 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
689 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
690
691 case vmIntrinsics::_weakCompareAndSetObjectPlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
692 case vmIntrinsics::_weakCompareAndSetObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
693 case vmIntrinsics::_weakCompareAndSetObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
694 case vmIntrinsics::_weakCompareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
695 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
696 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
697 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
698 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
699 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
700 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
701 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
702 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
703 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
704 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
705 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
706 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
707 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
708 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
709 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
710 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
711
712 case vmIntrinsics::_compareAndExchangeObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
713 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
714 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
715 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
716 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
717 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
718 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
719 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
720 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
721 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
722 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
723 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
724 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
725 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
726 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
727
728 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
729 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
730 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
731 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
732
733 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
734 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
735 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
736 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
737 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
738
739 case vmIntrinsics::_loadFence:
740 case vmIntrinsics::_storeFence:
741 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
742
743 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
744
745 case vmIntrinsics::_currentThread: return inline_native_currentThread();
746 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
747
748 #ifdef JFR_HAVE_INTRINSICS
749 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
750 case vmIntrinsics::_getClassId: return inline_native_classID();
751 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
752 #endif
753 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
754 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
755 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
756 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
757 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
758 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
759 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
760 case vmIntrinsics::_getLength: return inline_native_getLength();
761 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
762 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
763 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
764 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
765 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
766 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
767
768 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
769 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
770
771 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
772
773 case vmIntrinsics::_isInstance:
774 case vmIntrinsics::_getModifiers:
775 case vmIntrinsics::_isInterface:
776 case vmIntrinsics::_isArray:
777 case vmIntrinsics::_isPrimitive:
778 case vmIntrinsics::_getSuperclass:
779 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
780
781 case vmIntrinsics::_floatToRawIntBits:
782 case vmIntrinsics::_floatToIntBits:
783 case vmIntrinsics::_intBitsToFloat:
784 case vmIntrinsics::_doubleToRawLongBits:
785 case vmIntrinsics::_doubleToLongBits:
786 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
787
788 case vmIntrinsics::_numberOfLeadingZeros_i:
789 case vmIntrinsics::_numberOfLeadingZeros_l:
790 case vmIntrinsics::_numberOfTrailingZeros_i:
791 case vmIntrinsics::_numberOfTrailingZeros_l:
792 case vmIntrinsics::_bitCount_i:
793 case vmIntrinsics::_bitCount_l:
794 case vmIntrinsics::_reverseBytes_i:
795 case vmIntrinsics::_reverseBytes_l:
796 case vmIntrinsics::_reverseBytes_s:
797 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
798
799 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
800
801 case vmIntrinsics::_Reference_get: return inline_reference_get();
802
803 case vmIntrinsics::_Class_cast: return inline_Class_cast();
804
805 case vmIntrinsics::_aescrypt_encryptBlock:
806 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
807
808 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
809 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
810 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
811
812 case vmIntrinsics::_counterMode_AESCrypt:
813 return inline_counterMode_AESCrypt(intrinsic_id());
814
815 case vmIntrinsics::_sha_implCompress:
816 case vmIntrinsics::_sha2_implCompress:
817 case vmIntrinsics::_sha5_implCompress:
818 return inline_sha_implCompress(intrinsic_id());
819
820 case vmIntrinsics::_digestBase_implCompressMB:
821 return inline_digestBase_implCompressMB(predicate);
822
823 case vmIntrinsics::_multiplyToLen:
824 return inline_multiplyToLen();
825
826 case vmIntrinsics::_squareToLen:
827 return inline_squareToLen();
828
829 case vmIntrinsics::_mulAdd:
830 return inline_mulAdd();
831
832 case vmIntrinsics::_montgomeryMultiply:
833 return inline_montgomeryMultiply();
834 case vmIntrinsics::_montgomerySquare:
835 return inline_montgomerySquare();
836
837 case vmIntrinsics::_vectorizedMismatch:
838 return inline_vectorizedMismatch();
839
840 case vmIntrinsics::_ghash_processBlocks:
841 return inline_ghash_processBlocks();
842 case vmIntrinsics::_base64_encodeBlock:
843 return inline_base64_encodeBlock();
844
845 case vmIntrinsics::_encodeISOArray:
846 case vmIntrinsics::_encodeByteISOArray:
847 return inline_encodeISOArray();
848
849 case vmIntrinsics::_updateCRC32:
850 return inline_updateCRC32();
851 case vmIntrinsics::_updateBytesCRC32:
852 return inline_updateBytesCRC32();
853 case vmIntrinsics::_updateByteBufferCRC32:
854 return inline_updateByteBufferCRC32();
855
856 case vmIntrinsics::_updateBytesCRC32C:
857 return inline_updateBytesCRC32C();
858 case vmIntrinsics::_updateDirectByteBufferCRC32C:
859 return inline_updateDirectByteBufferCRC32C();
860
861 case vmIntrinsics::_updateBytesAdler32:
862 return inline_updateBytesAdler32();
863 case vmIntrinsics::_updateByteBufferAdler32:
864 return inline_updateByteBufferAdler32();
865
866 case vmIntrinsics::_profileBoolean:
867 return inline_profileBoolean();
868 case vmIntrinsics::_isCompileConstant:
869 return inline_isCompileConstant();
870
871 case vmIntrinsics::_hasNegatives:
872 return inline_hasNegatives();
873
874 case vmIntrinsics::_fmaD:
875 case vmIntrinsics::_fmaF:
876 return inline_fma(intrinsic_id());
877
878 default:
879 // If you get here, it may be that someone has added a new intrinsic
880 // to the list in vmSymbols.hpp without implementing it here.
881 #ifndef PRODUCT
882 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
883 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
884 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
885 }
886 #endif
887 return false;
888 }
889 }
890
891 Node* LibraryCallKit::try_to_predicate(int predicate) {
892 if (!jvms()->has_method()) {
893 // Root JVMState has a null method.
894 assert(map()->memory()->Opcode() == Op_Parm, "");
895 // Insert the memory aliasing node
896 set_all_memory(reset_memory());
897 }
898 assert(merged_memory(), "");
899
900 switch (intrinsic_id()) {
901 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
902 return inline_cipherBlockChaining_AESCrypt_predicate(false);
903 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
904 return inline_cipherBlockChaining_AESCrypt_predicate(true);
905 case vmIntrinsics::_counterMode_AESCrypt:
906 return inline_counterMode_AESCrypt_predicate();
907 case vmIntrinsics::_digestBase_implCompressMB:
908 return inline_digestBase_implCompressMB_predicate(predicate);
909
910 default:
911 // If you get here, it may be that someone has added a new intrinsic
912 // to the list in vmSymbols.hpp without implementing it here.
913 #ifndef PRODUCT
914 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
915 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
916 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
917 }
918 #endif
919 Node* slow_ctl = control();
920 set_control(top()); // No fast path instrinsic
921 return slow_ctl;
922 }
923 }
924
925 //------------------------------set_result-------------------------------
926 // Helper function for finishing intrinsics.
927 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
928 record_for_igvn(region);
929 set_control(_gvn.transform(region));
930 set_result( _gvn.transform(value));
931 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
932 }
933
934 //------------------------------generate_guard---------------------------
935 // Helper function for generating guarded fast-slow graph structures.
936 // The given 'test', if true, guards a slow path. If the test fails
937 // then a fast path can be taken. (We generally hope it fails.)
938 // In all cases, GraphKit::control() is updated to the fast path.
939 // The returned value represents the control for the slow path.
940 // The return value is never 'top'; it is either a valid control
941 // or NULL if it is obvious that the slow path can never be taken.
942 // Also, if region and the slow control are not NULL, the slow edge
943 // is appended to the region.
944 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
945 if (stopped()) {
946 // Already short circuited.
947 return NULL;
948 }
949
950 // Build an if node and its projections.
951 // If test is true we take the slow path, which we assume is uncommon.
952 if (_gvn.type(test) == TypeInt::ZERO) {
953 // The slow branch is never taken. No need to build this guard.
954 return NULL;
955 }
956
957 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
958
959 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
960 if (if_slow == top()) {
961 // The slow branch is never taken. No need to build this guard.
962 return NULL;
963 }
964
965 if (region != NULL)
966 region->add_req(if_slow);
967
968 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
969 set_control(if_fast);
970
971 return if_slow;
972 }
973
974 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
975 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
976 }
977 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
978 return generate_guard(test, region, PROB_FAIR);
979 }
980
981 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
982 Node* *pos_index) {
983 if (stopped())
984 return NULL; // already stopped
985 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
986 return NULL; // index is already adequately typed
987 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
988 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
989 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
990 if (is_neg != NULL && pos_index != NULL) {
991 // Emulate effect of Parse::adjust_map_after_if.
992 Node* ccast = new CastIINode(index, TypeInt::POS);
993 ccast->set_req(0, control());
994 (*pos_index) = _gvn.transform(ccast);
995 }
996 return is_neg;
997 }
998
999 // Make sure that 'position' is a valid limit index, in [0..length].
1000 // There are two equivalent plans for checking this:
1001 // A. (offset + copyLength) unsigned<= arrayLength
1002 // B. offset <= (arrayLength - copyLength)
1003 // We require that all of the values above, except for the sum and
1004 // difference, are already known to be non-negative.
1005 // Plan A is robust in the face of overflow, if offset and copyLength
1006 // are both hugely positive.
1007 //
1008 // Plan B is less direct and intuitive, but it does not overflow at
1009 // all, since the difference of two non-negatives is always
1010 // representable. Whenever Java methods must perform the equivalent
1011 // check they generally use Plan B instead of Plan A.
1012 // For the moment we use Plan A.
1013 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1014 Node* subseq_length,
1015 Node* array_length,
1016 RegionNode* region) {
1017 if (stopped())
1018 return NULL; // already stopped
1019 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1020 if (zero_offset && subseq_length->eqv_uncast(array_length))
1021 return NULL; // common case of whole-array copy
1022 Node* last = subseq_length;
1023 if (!zero_offset) // last += offset
1024 last = _gvn.transform(new AddINode(last, offset));
1025 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1026 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1027 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1028 return is_over;
1029 }
1030
1031 // Emit range checks for the given String.value byte array
1032 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1033 if (stopped()) {
1034 return; // already stopped
1035 }
1036 RegionNode* bailout = new RegionNode(1);
1037 record_for_igvn(bailout);
1038 if (char_count) {
1039 // Convert char count to byte count
1040 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1041 }
1042
1043 // Offset and count must not be negative
1044 generate_negative_guard(offset, bailout);
1045 generate_negative_guard(count, bailout);
1046 // Offset + count must not exceed length of array
1047 generate_limit_guard(offset, count, load_array_length(array), bailout);
1048
1049 if (bailout->req() > 1) {
1050 PreserveJVMState pjvms(this);
1051 set_control(_gvn.transform(bailout));
1052 uncommon_trap(Deoptimization::Reason_intrinsic,
1053 Deoptimization::Action_maybe_recompile);
1054 }
1055 }
1056
1057 //--------------------------generate_current_thread--------------------
1058 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1059 ciKlass* thread_klass = env()->Thread_klass();
1060 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1061 Node* thread = _gvn.transform(new ThreadLocalNode());
1062 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1063 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1064 tls_output = thread;
1065 return threadObj;
1066 }
1067
1068
1069 //------------------------------make_string_method_node------------------------
1070 // Helper method for String intrinsic functions. This version is called with
1071 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1072 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1073 // containing the lengths of str1 and str2.
1074 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1075 Node* result = NULL;
1076 switch (opcode) {
1077 case Op_StrIndexOf:
1078 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1079 str1_start, cnt1, str2_start, cnt2, ae);
1080 break;
1081 case Op_StrComp:
1082 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1083 str1_start, cnt1, str2_start, cnt2, ae);
1084 break;
1085 case Op_StrEquals:
1086 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1087 // Use the constant length if there is one because optimized match rule may exist.
1088 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1089 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1090 break;
1091 default:
1092 ShouldNotReachHere();
1093 return NULL;
1094 }
1095
1096 // All these intrinsics have checks.
1097 C->set_has_split_ifs(true); // Has chance for split-if optimization
1098 clear_upper_avx();
1099
1100 return _gvn.transform(result);
1101 }
1102
1103 //------------------------------inline_string_compareTo------------------------
1104 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1105 Node* arg1 = argument(0);
1106 Node* arg2 = argument(1);
1107
1108 // Get start addr and length of first argument
1109 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1110 Node* arg1_cnt = load_array_length(arg1);
1111
1112 // Get start addr and length of second argument
1113 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1114 Node* arg2_cnt = load_array_length(arg2);
1115
1116 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1117 set_result(result);
1118 return true;
1119 }
1120
1121 //------------------------------inline_string_equals------------------------
1122 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1123 Node* arg1 = argument(0);
1124 Node* arg2 = argument(1);
1125
1126 // paths (plus control) merge
1127 RegionNode* region = new RegionNode(3);
1128 Node* phi = new PhiNode(region, TypeInt::BOOL);
1129
1130 if (!stopped()) {
1131 // Get start addr and length of first argument
1132 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1133 Node* arg1_cnt = load_array_length(arg1);
1134
1135 // Get start addr and length of second argument
1136 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1137 Node* arg2_cnt = load_array_length(arg2);
1138
1139 // Check for arg1_cnt != arg2_cnt
1140 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1141 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1142 Node* if_ne = generate_slow_guard(bol, NULL);
1143 if (if_ne != NULL) {
1144 phi->init_req(2, intcon(0));
1145 region->init_req(2, if_ne);
1146 }
1147
1148 // Check for count == 0 is done by assembler code for StrEquals.
1149
1150 if (!stopped()) {
1151 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1152 phi->init_req(1, equals);
1153 region->init_req(1, control());
1154 }
1155 }
1156
1157 // post merge
1158 set_control(_gvn.transform(region));
1159 record_for_igvn(region);
1160
1161 set_result(_gvn.transform(phi));
1162 return true;
1163 }
1164
1165 //------------------------------inline_array_equals----------------------------
1166 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1167 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1168 Node* arg1 = argument(0);
1169 Node* arg2 = argument(1);
1170
1171 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1172 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1173 clear_upper_avx();
1174
1175 return true;
1176 }
1177
1178 //------------------------------inline_hasNegatives------------------------------
1179 bool LibraryCallKit::inline_hasNegatives() {
1180 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1181 return false;
1182 }
1183
1184 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1185 // no receiver since it is static method
1186 Node* ba = argument(0);
1187 Node* offset = argument(1);
1188 Node* len = argument(2);
1189
1190 // Range checks
1191 generate_string_range_check(ba, offset, len, false);
1192 if (stopped()) {
1193 return true;
1194 }
1195 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1196 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1197 set_result(_gvn.transform(result));
1198 return true;
1199 }
1200
1201 bool LibraryCallKit::inline_preconditions_checkIndex() {
1202 Node* index = argument(0);
1203 Node* length = argument(1);
1204 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1205 return false;
1206 }
1207
1208 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1209 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1210
1211 {
1212 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1213 uncommon_trap(Deoptimization::Reason_intrinsic,
1214 Deoptimization::Action_make_not_entrant);
1215 }
1216
1217 if (stopped()) {
1218 return false;
1219 }
1220
1221 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1222 BoolTest::mask btest = BoolTest::lt;
1223 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1224 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1225 _gvn.set_type(rc, rc->Value(&_gvn));
1226 if (!rc_bool->is_Con()) {
1227 record_for_igvn(rc);
1228 }
1229 set_control(_gvn.transform(new IfTrueNode(rc)));
1230 {
1231 PreserveJVMState pjvms(this);
1232 set_control(_gvn.transform(new IfFalseNode(rc)));
1233 uncommon_trap(Deoptimization::Reason_range_check,
1234 Deoptimization::Action_make_not_entrant);
1235 }
1236
1237 if (stopped()) {
1238 return false;
1239 }
1240
1241 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1242 result->set_req(0, control());
1243 result = _gvn.transform(result);
1244 set_result(result);
1245 replace_in_map(index, result);
1246 clear_upper_avx();
1247 return true;
1248 }
1249
1250 //------------------------------inline_string_indexOf------------------------
1251 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1252 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1253 return false;
1254 }
1255 Node* src = argument(0);
1256 Node* tgt = argument(1);
1257
1258 // Make the merge point
1259 RegionNode* result_rgn = new RegionNode(4);
1260 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1261
1262 // Get start addr and length of source string
1263 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1264 Node* src_count = load_array_length(src);
1265
1266 // Get start addr and length of substring
1267 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1268 Node* tgt_count = load_array_length(tgt);
1269
1270 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1271 // Divide src size by 2 if String is UTF16 encoded
1272 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1273 }
1274 if (ae == StrIntrinsicNode::UU) {
1275 // Divide substring size by 2 if String is UTF16 encoded
1276 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1277 }
1278
1279 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1280 if (result != NULL) {
1281 result_phi->init_req(3, result);
1282 result_rgn->init_req(3, control());
1283 }
1284 set_control(_gvn.transform(result_rgn));
1285 record_for_igvn(result_rgn);
1286 set_result(_gvn.transform(result_phi));
1287
1288 return true;
1289 }
1290
1291 //-----------------------------inline_string_indexOf-----------------------
1292 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1293 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1294 return false;
1295 }
1296 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1297 return false;
1298 }
1299 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1300 Node* src = argument(0); // byte[]
1301 Node* src_count = argument(1); // char count
1302 Node* tgt = argument(2); // byte[]
1303 Node* tgt_count = argument(3); // char count
1304 Node* from_index = argument(4); // char index
1305
1306 // Multiply byte array index by 2 if String is UTF16 encoded
1307 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1308 src_count = _gvn.transform(new SubINode(src_count, from_index));
1309 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1310 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1311
1312 // Range checks
1313 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1314 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1315 if (stopped()) {
1316 return true;
1317 }
1318
1319 RegionNode* region = new RegionNode(5);
1320 Node* phi = new PhiNode(region, TypeInt::INT);
1321
1322 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1323 if (result != NULL) {
1324 // The result is index relative to from_index if substring was found, -1 otherwise.
1325 // Generate code which will fold into cmove.
1326 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1327 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1328
1329 Node* if_lt = generate_slow_guard(bol, NULL);
1330 if (if_lt != NULL) {
1331 // result == -1
1332 phi->init_req(3, result);
1333 region->init_req(3, if_lt);
1334 }
1335 if (!stopped()) {
1336 result = _gvn.transform(new AddINode(result, from_index));
1337 phi->init_req(4, result);
1338 region->init_req(4, control());
1339 }
1340 }
1341
1342 set_control(_gvn.transform(region));
1343 record_for_igvn(region);
1344 set_result(_gvn.transform(phi));
1345 clear_upper_avx();
1346
1347 return true;
1348 }
1349
1350 // Create StrIndexOfNode with fast path checks
1351 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1352 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1353 // Check for substr count > string count
1354 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1355 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1356 Node* if_gt = generate_slow_guard(bol, NULL);
1357 if (if_gt != NULL) {
1358 phi->init_req(1, intcon(-1));
1359 region->init_req(1, if_gt);
1360 }
1361 if (!stopped()) {
1362 // Check for substr count == 0
1363 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1364 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1365 Node* if_zero = generate_slow_guard(bol, NULL);
1366 if (if_zero != NULL) {
1367 phi->init_req(2, intcon(0));
1368 region->init_req(2, if_zero);
1369 }
1370 }
1371 if (!stopped()) {
1372 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1373 }
1374 return NULL;
1375 }
1376
1377 //-----------------------------inline_string_indexOfChar-----------------------
1378 bool LibraryCallKit::inline_string_indexOfChar() {
1379 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1380 return false;
1381 }
1382 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1383 return false;
1384 }
1385 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1386 Node* src = argument(0); // byte[]
1387 Node* tgt = argument(1); // tgt is int ch
1388 Node* from_index = argument(2);
1389 Node* max = argument(3);
1390
1391 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1392 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1393 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1394
1395 // Range checks
1396 generate_string_range_check(src, src_offset, src_count, true);
1397 if (stopped()) {
1398 return true;
1399 }
1400
1401 RegionNode* region = new RegionNode(3);
1402 Node* phi = new PhiNode(region, TypeInt::INT);
1403
1404 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1405 C->set_has_split_ifs(true); // Has chance for split-if optimization
1406 _gvn.transform(result);
1407
1408 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1409 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1410
1411 Node* if_lt = generate_slow_guard(bol, NULL);
1412 if (if_lt != NULL) {
1413 // result == -1
1414 phi->init_req(2, result);
1415 region->init_req(2, if_lt);
1416 }
1417 if (!stopped()) {
1418 result = _gvn.transform(new AddINode(result, from_index));
1419 phi->init_req(1, result);
1420 region->init_req(1, control());
1421 }
1422 set_control(_gvn.transform(region));
1423 record_for_igvn(region);
1424 set_result(_gvn.transform(phi));
1425
1426 return true;
1427 }
1428 //---------------------------inline_string_copy---------------------
1429 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1430 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1431 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1432 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1433 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1434 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1435 bool LibraryCallKit::inline_string_copy(bool compress) {
1436 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1437 return false;
1438 }
1439 int nargs = 5; // 2 oops, 3 ints
1440 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1441
1442 Node* src = argument(0);
1443 Node* src_offset = argument(1);
1444 Node* dst = argument(2);
1445 Node* dst_offset = argument(3);
1446 Node* length = argument(4);
1447
1448 // Check for allocation before we add nodes that would confuse
1449 // tightly_coupled_allocation()
1450 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1451
1452 // Figure out the size and type of the elements we will be copying.
1453 const Type* src_type = src->Value(&_gvn);
1454 const Type* dst_type = dst->Value(&_gvn);
1455 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1456 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1457 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1458 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1459 "Unsupported array types for inline_string_copy");
1460
1461 // Convert char[] offsets to byte[] offsets
1462 bool convert_src = (compress && src_elem == T_BYTE);
1463 bool convert_dst = (!compress && dst_elem == T_BYTE);
1464 if (convert_src) {
1465 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1466 } else if (convert_dst) {
1467 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1468 }
1469
1470 // Range checks
1471 generate_string_range_check(src, src_offset, length, convert_src);
1472 generate_string_range_check(dst, dst_offset, length, convert_dst);
1473 if (stopped()) {
1474 return true;
1475 }
1476
1477 Node* src_start = array_element_address(src, src_offset, src_elem);
1478 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1479 // 'src_start' points to src array + scaled offset
1480 // 'dst_start' points to dst array + scaled offset
1481 Node* count = NULL;
1482 if (compress) {
1483 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1484 } else {
1485 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1486 }
1487
1488 if (alloc != NULL) {
1489 if (alloc->maybe_set_complete(&_gvn)) {
1490 // "You break it, you buy it."
1491 InitializeNode* init = alloc->initialization();
1492 assert(init->is_complete(), "we just did this");
1493 init->set_complete_with_arraycopy();
1494 assert(dst->is_CheckCastPP(), "sanity");
1495 assert(dst->in(0)->in(0) == init, "dest pinned");
1496 }
1497 // Do not let stores that initialize this object be reordered with
1498 // a subsequent store that would make this object accessible by
1499 // other threads.
1500 // Record what AllocateNode this StoreStore protects so that
1501 // escape analysis can go from the MemBarStoreStoreNode to the
1502 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1503 // based on the escape status of the AllocateNode.
1504 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1505 }
1506 if (compress) {
1507 set_result(_gvn.transform(count));
1508 }
1509 clear_upper_avx();
1510
1511 return true;
1512 }
1513
1514 #ifdef _LP64
1515 #define XTOP ,top() /*additional argument*/
1516 #else //_LP64
1517 #define XTOP /*no additional argument*/
1518 #endif //_LP64
1519
1520 //------------------------inline_string_toBytesU--------------------------
1521 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1522 bool LibraryCallKit::inline_string_toBytesU() {
1523 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1524 return false;
1525 }
1526 // Get the arguments.
1527 Node* value = argument(0);
1528 Node* offset = argument(1);
1529 Node* length = argument(2);
1530
1531 Node* newcopy = NULL;
1532
1533 // Set the original stack and the reexecute bit for the interpreter to reexecute
1534 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1535 { PreserveReexecuteState preexecs(this);
1536 jvms()->set_should_reexecute(true);
1537
1538 // Check if a null path was taken unconditionally.
1539 value = null_check(value);
1540
1541 RegionNode* bailout = new RegionNode(1);
1542 record_for_igvn(bailout);
1543
1544 // Range checks
1545 generate_negative_guard(offset, bailout);
1546 generate_negative_guard(length, bailout);
1547 generate_limit_guard(offset, length, load_array_length(value), bailout);
1548 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1549 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1550
1551 if (bailout->req() > 1) {
1552 PreserveJVMState pjvms(this);
1553 set_control(_gvn.transform(bailout));
1554 uncommon_trap(Deoptimization::Reason_intrinsic,
1555 Deoptimization::Action_maybe_recompile);
1556 }
1557 if (stopped()) {
1558 return true;
1559 }
1560
1561 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1562 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1563 newcopy = new_array(klass_node, size, 0); // no arguments to push
1564 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1565
1566 // Calculate starting addresses.
1567 Node* src_start = array_element_address(value, offset, T_CHAR);
1568 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1569
1570 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1571 const TypeInt* toffset = gvn().type(offset)->is_int();
1572 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1573
1574 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1575 const char* copyfunc_name = "arraycopy";
1576 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1577 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1578 OptoRuntime::fast_arraycopy_Type(),
1579 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1580 src_start, dst_start, ConvI2X(length) XTOP);
1581 // Do not let reads from the cloned object float above the arraycopy.
1582 if (alloc != NULL) {
1583 if (alloc->maybe_set_complete(&_gvn)) {
1584 // "You break it, you buy it."
1585 InitializeNode* init = alloc->initialization();
1586 assert(init->is_complete(), "we just did this");
1587 init->set_complete_with_arraycopy();
1588 assert(newcopy->is_CheckCastPP(), "sanity");
1589 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1590 }
1591 // Do not let stores that initialize this object be reordered with
1592 // a subsequent store that would make this object accessible by
1593 // other threads.
1594 // Record what AllocateNode this StoreStore protects so that
1595 // escape analysis can go from the MemBarStoreStoreNode to the
1596 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1597 // based on the escape status of the AllocateNode.
1598 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1599 } else {
1600 insert_mem_bar(Op_MemBarCPUOrder);
1601 }
1602 } // original reexecute is set back here
1603
1604 C->set_has_split_ifs(true); // Has chance for split-if optimization
1605 if (!stopped()) {
1606 set_result(newcopy);
1607 }
1608 clear_upper_avx();
1609
1610 return true;
1611 }
1612
1613 //------------------------inline_string_getCharsU--------------------------
1614 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1615 bool LibraryCallKit::inline_string_getCharsU() {
1616 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1617 return false;
1618 }
1619
1620 // Get the arguments.
1621 Node* src = argument(0);
1622 Node* src_begin = argument(1);
1623 Node* src_end = argument(2); // exclusive offset (i < src_end)
1624 Node* dst = argument(3);
1625 Node* dst_begin = argument(4);
1626
1627 // Check for allocation before we add nodes that would confuse
1628 // tightly_coupled_allocation()
1629 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1630
1631 // Check if a null path was taken unconditionally.
1632 src = null_check(src);
1633 dst = null_check(dst);
1634 if (stopped()) {
1635 return true;
1636 }
1637
1638 // Get length and convert char[] offset to byte[] offset
1639 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1640 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1641
1642 // Range checks
1643 generate_string_range_check(src, src_begin, length, true);
1644 generate_string_range_check(dst, dst_begin, length, false);
1645 if (stopped()) {
1646 return true;
1647 }
1648
1649 if (!stopped()) {
1650 // Calculate starting addresses.
1651 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1652 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1653
1654 // Check if array addresses are aligned to HeapWordSize
1655 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1656 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1657 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1658 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1659
1660 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1661 const char* copyfunc_name = "arraycopy";
1662 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1663 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1664 OptoRuntime::fast_arraycopy_Type(),
1665 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1666 src_start, dst_start, ConvI2X(length) XTOP);
1667 // Do not let reads from the cloned object float above the arraycopy.
1668 if (alloc != NULL) {
1669 if (alloc->maybe_set_complete(&_gvn)) {
1670 // "You break it, you buy it."
1671 InitializeNode* init = alloc->initialization();
1672 assert(init->is_complete(), "we just did this");
1673 init->set_complete_with_arraycopy();
1674 assert(dst->is_CheckCastPP(), "sanity");
1675 assert(dst->in(0)->in(0) == init, "dest pinned");
1676 }
1677 // Do not let stores that initialize this object be reordered with
1678 // a subsequent store that would make this object accessible by
1679 // other threads.
1680 // Record what AllocateNode this StoreStore protects so that
1681 // escape analysis can go from the MemBarStoreStoreNode to the
1682 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1683 // based on the escape status of the AllocateNode.
1684 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1685 } else {
1686 insert_mem_bar(Op_MemBarCPUOrder);
1687 }
1688 }
1689
1690 C->set_has_split_ifs(true); // Has chance for split-if optimization
1691 return true;
1692 }
1693
1694 //----------------------inline_string_char_access----------------------------
1695 // Store/Load char to/from byte[] array.
1696 // static void StringUTF16.putChar(byte[] val, int index, int c)
1697 // static char StringUTF16.getChar(byte[] val, int index)
1698 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1699 Node* value = argument(0);
1700 Node* index = argument(1);
1701 Node* ch = is_store ? argument(2) : NULL;
1702
1703 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1704 // correctly requires matched array shapes.
1705 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1706 "sanity: byte[] and char[] bases agree");
1707 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1708 "sanity: byte[] and char[] scales agree");
1709
1710 // Bail when getChar over constants is requested: constant folding would
1711 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1712 // Java method would constant fold nicely instead.
1713 if (!is_store && value->is_Con() && index->is_Con()) {
1714 return false;
1715 }
1716
1717 Node* adr = array_element_address(value, index, T_CHAR);
1718 if (adr->is_top()) {
1719 return false;
1720 }
1721 if (is_store) {
1722 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1723 false, false, true /* mismatched */);
1724 } else {
1725 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1726 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1727 set_result(ch);
1728 }
1729 return true;
1730 }
1731
1732 //--------------------------round_double_node--------------------------------
1733 // Round a double node if necessary.
1734 Node* LibraryCallKit::round_double_node(Node* n) {
1735 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1736 n = _gvn.transform(new RoundDoubleNode(0, n));
1737 return n;
1738 }
1739
1740 //------------------------------inline_math-----------------------------------
1741 // public static double Math.abs(double)
1742 // public static double Math.sqrt(double)
1743 // public static double Math.log(double)
1744 // public static double Math.log10(double)
1745 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1746 Node* arg = round_double_node(argument(0));
1747 Node* n = NULL;
1748 switch (id) {
1749 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1750 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1751 default: fatal_unexpected_iid(id); break;
1752 }
1753 set_result(_gvn.transform(n));
1754 return true;
1755 }
1756
1757 //------------------------------runtime_math-----------------------------
1758 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1759 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1760 "must be (DD)D or (D)D type");
1761
1762 // Inputs
1763 Node* a = round_double_node(argument(0));
1764 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1765
1766 const TypePtr* no_memory_effects = NULL;
1767 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1768 no_memory_effects,
1769 a, top(), b, b ? top() : NULL);
1770 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1771 #ifdef ASSERT
1772 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1773 assert(value_top == top(), "second value must be top");
1774 #endif
1775
1776 set_result(value);
1777 return true;
1778 }
1779
1780 //------------------------------inline_math_native-----------------------------
1781 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1782 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1783 switch (id) {
1784 // These intrinsics are not properly supported on all hardware
1785 case vmIntrinsics::_dsin:
1786 return StubRoutines::dsin() != NULL ?
1787 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1788 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1789 case vmIntrinsics::_dcos:
1790 return StubRoutines::dcos() != NULL ?
1791 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1792 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1793 case vmIntrinsics::_dtan:
1794 return StubRoutines::dtan() != NULL ?
1795 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1796 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1797 case vmIntrinsics::_dlog:
1798 return StubRoutines::dlog() != NULL ?
1799 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1800 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1801 case vmIntrinsics::_dlog10:
1802 return StubRoutines::dlog10() != NULL ?
1803 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1804 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1805
1806 // These intrinsics are supported on all hardware
1807 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1808 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1809
1810 case vmIntrinsics::_dexp:
1811 return StubRoutines::dexp() != NULL ?
1812 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1813 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1814 case vmIntrinsics::_dpow: {
1815 Node* exp = round_double_node(argument(2));
1816 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1817 if (d != NULL && d->getd() == 2.0) {
1818 // Special case: pow(x, 2.0) => x * x
1819 Node* base = round_double_node(argument(0));
1820 set_result(_gvn.transform(new MulDNode(base, base)));
1821 return true;
1822 }
1823 return StubRoutines::dexp() != NULL ?
1824 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1825 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1826 }
1827 #undef FN_PTR
1828
1829 // These intrinsics are not yet correctly implemented
1830 case vmIntrinsics::_datan2:
1831 return false;
1832
1833 default:
1834 fatal_unexpected_iid(id);
1835 return false;
1836 }
1837 }
1838
1839 static bool is_simple_name(Node* n) {
1840 return (n->req() == 1 // constant
1841 || (n->is_Type() && n->as_Type()->type()->singleton())
1842 || n->is_Proj() // parameter or return value
1843 || n->is_Phi() // local of some sort
1844 );
1845 }
1846
1847 //----------------------------inline_notify-----------------------------------*
1848 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1849 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1850 address func;
1851 if (id == vmIntrinsics::_notify) {
1852 func = OptoRuntime::monitor_notify_Java();
1853 } else {
1854 func = OptoRuntime::monitor_notifyAll_Java();
1855 }
1856 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1857 make_slow_call_ex(call, env()->Throwable_klass(), false);
1858 return true;
1859 }
1860
1861
1862 //----------------------------inline_min_max-----------------------------------
1863 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1864 set_result(generate_min_max(id, argument(0), argument(1)));
1865 return true;
1866 }
1867
1868 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1869 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1870 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1871 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1872 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1873
1874 {
1875 PreserveJVMState pjvms(this);
1876 PreserveReexecuteState preexecs(this);
1877 jvms()->set_should_reexecute(true);
1878
1879 set_control(slow_path);
1880 set_i_o(i_o());
1881
1882 uncommon_trap(Deoptimization::Reason_intrinsic,
1883 Deoptimization::Action_none);
1884 }
1885
1886 set_control(fast_path);
1887 set_result(math);
1888 }
1889
1890 template <typename OverflowOp>
1891 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1892 typedef typename OverflowOp::MathOp MathOp;
1893
1894 MathOp* mathOp = new MathOp(arg1, arg2);
1895 Node* operation = _gvn.transform( mathOp );
1896 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1897 inline_math_mathExact(operation, ofcheck);
1898 return true;
1899 }
1900
1901 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1902 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1903 }
1904
1905 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1906 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1907 }
1908
1909 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1910 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1911 }
1912
1913 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1914 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1915 }
1916
1917 bool LibraryCallKit::inline_math_negateExactI() {
1918 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1919 }
1920
1921 bool LibraryCallKit::inline_math_negateExactL() {
1922 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1923 }
1924
1925 bool LibraryCallKit::inline_math_multiplyExactI() {
1926 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1927 }
1928
1929 bool LibraryCallKit::inline_math_multiplyExactL() {
1930 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1931 }
1932
1933 bool LibraryCallKit::inline_math_multiplyHigh() {
1934 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1935 return true;
1936 }
1937
1938 Node*
1939 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1940 // These are the candidate return value:
1941 Node* xvalue = x0;
1942 Node* yvalue = y0;
1943
1944 if (xvalue == yvalue) {
1945 return xvalue;
1946 }
1947
1948 bool want_max = (id == vmIntrinsics::_max);
1949
1950 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1951 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1952 if (txvalue == NULL || tyvalue == NULL) return top();
1953 // This is not really necessary, but it is consistent with a
1954 // hypothetical MaxINode::Value method:
1955 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1956
1957 // %%% This folding logic should (ideally) be in a different place.
1958 // Some should be inside IfNode, and there to be a more reliable
1959 // transformation of ?: style patterns into cmoves. We also want
1960 // more powerful optimizations around cmove and min/max.
1961
1962 // Try to find a dominating comparison of these guys.
1963 // It can simplify the index computation for Arrays.copyOf
1964 // and similar uses of System.arraycopy.
1965 // First, compute the normalized version of CmpI(x, y).
1966 int cmp_op = Op_CmpI;
1967 Node* xkey = xvalue;
1968 Node* ykey = yvalue;
1969 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1970 if (ideal_cmpxy->is_Cmp()) {
1971 // E.g., if we have CmpI(length - offset, count),
1972 // it might idealize to CmpI(length, count + offset)
1973 cmp_op = ideal_cmpxy->Opcode();
1974 xkey = ideal_cmpxy->in(1);
1975 ykey = ideal_cmpxy->in(2);
1976 }
1977
1978 // Start by locating any relevant comparisons.
1979 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1980 Node* cmpxy = NULL;
1981 Node* cmpyx = NULL;
1982 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1983 Node* cmp = start_from->fast_out(k);
1984 if (cmp->outcnt() > 0 && // must have prior uses
1985 cmp->in(0) == NULL && // must be context-independent
1986 cmp->Opcode() == cmp_op) { // right kind of compare
1987 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1988 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1989 }
1990 }
1991
1992 const int NCMPS = 2;
1993 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1994 int cmpn;
1995 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1996 if (cmps[cmpn] != NULL) break; // find a result
1997 }
1998 if (cmpn < NCMPS) {
1999 // Look for a dominating test that tells us the min and max.
2000 int depth = 0; // Limit search depth for speed
2001 Node* dom = control();
2002 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2003 if (++depth >= 100) break;
2004 Node* ifproj = dom;
2005 if (!ifproj->is_Proj()) continue;
2006 Node* iff = ifproj->in(0);
2007 if (!iff->is_If()) continue;
2008 Node* bol = iff->in(1);
2009 if (!bol->is_Bool()) continue;
2010 Node* cmp = bol->in(1);
2011 if (cmp == NULL) continue;
2012 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2013 if (cmps[cmpn] == cmp) break;
2014 if (cmpn == NCMPS) continue;
2015 BoolTest::mask btest = bol->as_Bool()->_test._test;
2016 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2017 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2018 // At this point, we know that 'x btest y' is true.
2019 switch (btest) {
2020 case BoolTest::eq:
2021 // They are proven equal, so we can collapse the min/max.
2022 // Either value is the answer. Choose the simpler.
2023 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2024 return yvalue;
2025 return xvalue;
2026 case BoolTest::lt: // x < y
2027 case BoolTest::le: // x <= y
2028 return (want_max ? yvalue : xvalue);
2029 case BoolTest::gt: // x > y
2030 case BoolTest::ge: // x >= y
2031 return (want_max ? xvalue : yvalue);
2032 default:
2033 break;
2034 }
2035 }
2036 }
2037
2038 // We failed to find a dominating test.
2039 // Let's pick a test that might GVN with prior tests.
2040 Node* best_bol = NULL;
2041 BoolTest::mask best_btest = BoolTest::illegal;
2042 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2043 Node* cmp = cmps[cmpn];
2044 if (cmp == NULL) continue;
2045 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2046 Node* bol = cmp->fast_out(j);
2047 if (!bol->is_Bool()) continue;
2048 BoolTest::mask btest = bol->as_Bool()->_test._test;
2049 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2050 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2051 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2052 best_bol = bol->as_Bool();
2053 best_btest = btest;
2054 }
2055 }
2056 }
2057
2058 Node* answer_if_true = NULL;
2059 Node* answer_if_false = NULL;
2060 switch (best_btest) {
2061 default:
2062 if (cmpxy == NULL)
2063 cmpxy = ideal_cmpxy;
2064 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2065 // and fall through:
2066 case BoolTest::lt: // x < y
2067 case BoolTest::le: // x <= y
2068 answer_if_true = (want_max ? yvalue : xvalue);
2069 answer_if_false = (want_max ? xvalue : yvalue);
2070 break;
2071 case BoolTest::gt: // x > y
2072 case BoolTest::ge: // x >= y
2073 answer_if_true = (want_max ? xvalue : yvalue);
2074 answer_if_false = (want_max ? yvalue : xvalue);
2075 break;
2076 }
2077
2078 jint hi, lo;
2079 if (want_max) {
2080 // We can sharpen the minimum.
2081 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2082 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2083 } else {
2084 // We can sharpen the maximum.
2085 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2086 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2087 }
2088
2089 // Use a flow-free graph structure, to avoid creating excess control edges
2090 // which could hinder other optimizations.
2091 // Since Math.min/max is often used with arraycopy, we want
2092 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2093 Node* cmov = CMoveNode::make(NULL, best_bol,
2094 answer_if_false, answer_if_true,
2095 TypeInt::make(lo, hi, widen));
2096
2097 return _gvn.transform(cmov);
2098
2099 /*
2100 // This is not as desirable as it may seem, since Min and Max
2101 // nodes do not have a full set of optimizations.
2102 // And they would interfere, anyway, with 'if' optimizations
2103 // and with CMoveI canonical forms.
2104 switch (id) {
2105 case vmIntrinsics::_min:
2106 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2107 case vmIntrinsics::_max:
2108 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2109 default:
2110 ShouldNotReachHere();
2111 }
2112 */
2113 }
2114
2115 inline int
2116 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2117 const TypePtr* base_type = TypePtr::NULL_PTR;
2118 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2119 if (base_type == NULL) {
2120 // Unknown type.
2121 return Type::AnyPtr;
2122 } else if (base_type == TypePtr::NULL_PTR) {
2123 // Since this is a NULL+long form, we have to switch to a rawptr.
2124 base = _gvn.transform(new CastX2PNode(offset));
2125 offset = MakeConX(0);
2126 return Type::RawPtr;
2127 } else if (base_type->base() == Type::RawPtr) {
2128 return Type::RawPtr;
2129 } else if (base_type->isa_oopptr()) {
2130 // Base is never null => always a heap address.
2131 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2132 return Type::OopPtr;
2133 }
2134 // Offset is small => always a heap address.
2135 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2136 if (offset_type != NULL &&
2137 base_type->offset() == 0 && // (should always be?)
2138 offset_type->_lo >= 0 &&
2139 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2140 return Type::OopPtr;
2141 } else if (type == T_OBJECT) {
2142 // off heap access to an oop doesn't make any sense. Has to be on
2143 // heap.
2144 return Type::OopPtr;
2145 }
2146 // Otherwise, it might either be oop+off or NULL+addr.
2147 return Type::AnyPtr;
2148 } else {
2149 // No information:
2150 return Type::AnyPtr;
2151 }
2152 }
2153
2154 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2155 Node* uncasted_base = base;
2156 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2157 if (kind == Type::RawPtr) {
2158 return basic_plus_adr(top(), uncasted_base, offset);
2159 } else if (kind == Type::AnyPtr) {
2160 assert(base == uncasted_base, "unexpected base change");
2161 if (can_cast) {
2162 if (!_gvn.type(base)->speculative_maybe_null() &&
2163 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2164 // According to profiling, this access is always on
2165 // heap. Casting the base to not null and thus avoiding membars
2166 // around the access should allow better optimizations
2167 Node* null_ctl = top();
2168 base = null_check_oop(base, &null_ctl, true, true, true);
2169 assert(null_ctl->is_top(), "no null control here");
2170 return basic_plus_adr(base, offset);
2171 } else if (_gvn.type(base)->speculative_always_null() &&
2172 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2173 // According to profiling, this access is always off
2174 // heap.
2175 base = null_assert(base);
2176 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2177 offset = MakeConX(0);
2178 return basic_plus_adr(top(), raw_base, offset);
2179 }
2180 }
2181 // We don't know if it's an on heap or off heap access. Fall back
2182 // to raw memory access.
2183 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2184 return basic_plus_adr(top(), raw, offset);
2185 } else {
2186 assert(base == uncasted_base, "unexpected base change");
2187 // We know it's an on heap access so base can't be null
2188 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2189 base = must_be_not_null(base, true);
2190 }
2191 return basic_plus_adr(base, offset);
2192 }
2193 }
2194
2195 //--------------------------inline_number_methods-----------------------------
2196 // inline int Integer.numberOfLeadingZeros(int)
2197 // inline int Long.numberOfLeadingZeros(long)
2198 //
2199 // inline int Integer.numberOfTrailingZeros(int)
2200 // inline int Long.numberOfTrailingZeros(long)
2201 //
2202 // inline int Integer.bitCount(int)
2203 // inline int Long.bitCount(long)
2204 //
2205 // inline char Character.reverseBytes(char)
2206 // inline short Short.reverseBytes(short)
2207 // inline int Integer.reverseBytes(int)
2208 // inline long Long.reverseBytes(long)
2209 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2210 Node* arg = argument(0);
2211 Node* n = NULL;
2212 switch (id) {
2213 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2214 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2215 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2216 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2217 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2218 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2219 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2220 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2221 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2222 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2223 default: fatal_unexpected_iid(id); break;
2224 }
2225 set_result(_gvn.transform(n));
2226 return true;
2227 }
2228
2229 //----------------------------inline_unsafe_access----------------------------
2230
2231 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2232 // Attempt to infer a sharper value type from the offset and base type.
2233 ciKlass* sharpened_klass = NULL;
2234
2235 // See if it is an instance field, with an object type.
2236 if (alias_type->field() != NULL) {
2237 if (alias_type->field()->type()->is_klass()) {
2238 sharpened_klass = alias_type->field()->type()->as_klass();
2239 }
2240 }
2241
2242 // See if it is a narrow oop array.
2243 if (adr_type->isa_aryptr()) {
2244 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2245 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2246 if (elem_type != NULL) {
2247 sharpened_klass = elem_type->klass();
2248 }
2249 }
2250 }
2251
2252 // The sharpened class might be unloaded if there is no class loader
2253 // contraint in place.
2254 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2255 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2256
2257 #ifndef PRODUCT
2258 if (C->print_intrinsics() || C->print_inlining()) {
2259 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2260 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2261 }
2262 #endif
2263 // Sharpen the value type.
2264 return tjp;
2265 }
2266 return NULL;
2267 }
2268
2269 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2270 switch (kind) {
2271 case Relaxed:
2272 return MO_UNORDERED;
2273 case Opaque:
2274 return MO_RELAXED;
2275 case Acquire:
2276 return MO_ACQUIRE;
2277 case Release:
2278 return MO_RELEASE;
2279 case Volatile:
2280 return MO_SEQ_CST;
2281 default:
2282 ShouldNotReachHere();
2283 return 0;
2284 }
2285 }
2286
2287 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2288 if (callee()->is_static()) return false; // caller must have the capability!
2289 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2290 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2291 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2292 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2293
2294 if (type == T_OBJECT || type == T_ARRAY) {
2295 decorators |= ON_UNKNOWN_OOP_REF;
2296 }
2297
2298 if (unaligned) {
2299 decorators |= C2_UNALIGNED;
2300 }
2301
2302 #ifndef PRODUCT
2303 {
2304 ResourceMark rm;
2305 // Check the signatures.
2306 ciSignature* sig = callee()->signature();
2307 #ifdef ASSERT
2308 if (!is_store) {
2309 // Object getObject(Object base, int/long offset), etc.
2310 BasicType rtype = sig->return_type()->basic_type();
2311 assert(rtype == type, "getter must return the expected value");
2312 assert(sig->count() == 2, "oop getter has 2 arguments");
2313 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2314 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2315 } else {
2316 // void putObject(Object base, int/long offset, Object x), etc.
2317 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2318 assert(sig->count() == 3, "oop putter has 3 arguments");
2319 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2320 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2321 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2322 assert(vtype == type, "putter must accept the expected value");
2323 }
2324 #endif // ASSERT
2325 }
2326 #endif //PRODUCT
2327
2328 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2329
2330 Node* receiver = argument(0); // type: oop
2331
2332 // Build address expression.
2333 Node* adr;
2334 Node* heap_base_oop = top();
2335 Node* offset = top();
2336 Node* val;
2337
2338 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2339 Node* base = argument(1); // type: oop
2340 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2341 offset = argument(2); // type: long
2342 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2343 // to be plain byte offsets, which are also the same as those accepted
2344 // by oopDesc::field_addr.
2345 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2346 "fieldOffset must be byte-scaled");
2347 // 32-bit machines ignore the high half!
2348 offset = ConvL2X(offset);
2349 adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2350
2351 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2352 heap_base_oop = base;
2353 } else if (type == T_OBJECT) {
2354 return false; // off-heap oop accesses are not supported
2355 }
2356
2357 // Can base be NULL? Otherwise, always on-heap access.
2358 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2359
2360 if (!can_access_non_heap) {
2361 decorators |= IN_HEAP;
2362 }
2363
2364 val = is_store ? argument(4) : NULL;
2365
2366 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2367
2368 // Try to categorize the address.
2369 Compile::AliasType* alias_type = C->alias_type(adr_type);
2370 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2371
2372 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2373 alias_type->adr_type() == TypeAryPtr::RANGE) {
2374 return false; // not supported
2375 }
2376
2377 bool mismatched = false;
2378 BasicType bt = alias_type->basic_type();
2379 if (bt != T_ILLEGAL) {
2380 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2381 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2382 // Alias type doesn't differentiate between byte[] and boolean[]).
2383 // Use address type to get the element type.
2384 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2385 }
2386 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2387 // accessing an array field with getObject is not a mismatch
2388 bt = T_OBJECT;
2389 }
2390 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2391 // Don't intrinsify mismatched object accesses
2392 return false;
2393 }
2394 mismatched = (bt != type);
2395 } else if (alias_type->adr_type()->isa_oopptr()) {
2396 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2397 }
2398
2399 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2400
2401 if (mismatched) {
2402 decorators |= C2_MISMATCHED;
2403 }
2404
2405 // First guess at the value type.
2406 const Type *value_type = Type::get_const_basic_type(type);
2407
2408 // Figure out the memory ordering.
2409 decorators |= mo_decorator_for_access_kind(kind);
2410
2411 if (!is_store && type == T_OBJECT) {
2412 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2413 if (tjp != NULL) {
2414 value_type = tjp;
2415 }
2416 }
2417
2418 receiver = null_check(receiver);
2419 if (stopped()) {
2420 return true;
2421 }
2422 // Heap pointers get a null-check from the interpreter,
2423 // as a courtesy. However, this is not guaranteed by Unsafe,
2424 // and it is not possible to fully distinguish unintended nulls
2425 // from intended ones in this API.
2426
2427 if (!is_store) {
2428 Node* p = NULL;
2429 // Try to constant fold a load from a constant field
2430 ciField* field = alias_type->field();
2431 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2432 // final or stable field
2433 p = make_constant_from_field(field, heap_base_oop);
2434 }
2435
2436 if (p == NULL) { // Could not constant fold the load
2437 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2438 // Normalize the value returned by getBoolean in the following cases
2439 if (type == T_BOOLEAN &&
2440 (mismatched ||
2441 heap_base_oop == top() || // - heap_base_oop is NULL or
2442 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2443 // and the unsafe access is made to large offset
2444 // (i.e., larger than the maximum offset necessary for any
2445 // field access)
2446 ) {
2447 IdealKit ideal = IdealKit(this);
2448 #define __ ideal.
2449 IdealVariable normalized_result(ideal);
2450 __ declarations_done();
2451 __ set(normalized_result, p);
2452 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2453 __ set(normalized_result, ideal.ConI(1));
2454 ideal.end_if();
2455 final_sync(ideal);
2456 p = __ value(normalized_result);
2457 #undef __
2458 }
2459 }
2460 if (type == T_ADDRESS) {
2461 p = gvn().transform(new CastP2XNode(NULL, p));
2462 p = ConvX2UL(p);
2463 }
2464 // The load node has the control of the preceding MemBarCPUOrder. All
2465 // following nodes will have the control of the MemBarCPUOrder inserted at
2466 // the end of this method. So, pushing the load onto the stack at a later
2467 // point is fine.
2468 set_result(p);
2469 } else {
2470 if (bt == T_ADDRESS) {
2471 // Repackage the long as a pointer.
2472 val = ConvL2X(val);
2473 val = gvn().transform(new CastX2PNode(val));
2474 }
2475 access_store_at(control(), heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2476 }
2477
2478 return true;
2479 }
2480
2481 //----------------------------inline_unsafe_load_store----------------------------
2482 // This method serves a couple of different customers (depending on LoadStoreKind):
2483 //
2484 // LS_cmp_swap:
2485 //
2486 // boolean compareAndSetObject(Object o, long offset, Object expected, Object x);
2487 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2488 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2489 //
2490 // LS_cmp_swap_weak:
2491 //
2492 // boolean weakCompareAndSetObject( Object o, long offset, Object expected, Object x);
2493 // boolean weakCompareAndSetObjectPlain( Object o, long offset, Object expected, Object x);
2494 // boolean weakCompareAndSetObjectAcquire(Object o, long offset, Object expected, Object x);
2495 // boolean weakCompareAndSetObjectRelease(Object o, long offset, Object expected, Object x);
2496 //
2497 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2498 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2499 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2500 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2501 //
2502 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2503 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2504 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2505 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2506 //
2507 // LS_cmp_exchange:
2508 //
2509 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2510 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2511 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2512 //
2513 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2514 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2515 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2516 //
2517 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2518 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2519 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2520 //
2521 // LS_get_add:
2522 //
2523 // int getAndAddInt( Object o, long offset, int delta)
2524 // long getAndAddLong(Object o, long offset, long delta)
2525 //
2526 // LS_get_set:
2527 //
2528 // int getAndSet(Object o, long offset, int newValue)
2529 // long getAndSet(Object o, long offset, long newValue)
2530 // Object getAndSet(Object o, long offset, Object newValue)
2531 //
2532 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2533 // This basic scheme here is the same as inline_unsafe_access, but
2534 // differs in enough details that combining them would make the code
2535 // overly confusing. (This is a true fact! I originally combined
2536 // them, but even I was confused by it!) As much code/comments as
2537 // possible are retained from inline_unsafe_access though to make
2538 // the correspondences clearer. - dl
2539
2540 if (callee()->is_static()) return false; // caller must have the capability!
2541
2542 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2543 decorators |= mo_decorator_for_access_kind(access_kind);
2544
2545 #ifndef PRODUCT
2546 BasicType rtype;
2547 {
2548 ResourceMark rm;
2549 // Check the signatures.
2550 ciSignature* sig = callee()->signature();
2551 rtype = sig->return_type()->basic_type();
2552 switch(kind) {
2553 case LS_get_add:
2554 case LS_get_set: {
2555 // Check the signatures.
2556 #ifdef ASSERT
2557 assert(rtype == type, "get and set must return the expected type");
2558 assert(sig->count() == 3, "get and set has 3 arguments");
2559 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2560 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2561 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2562 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2563 #endif // ASSERT
2564 break;
2565 }
2566 case LS_cmp_swap:
2567 case LS_cmp_swap_weak: {
2568 // Check the signatures.
2569 #ifdef ASSERT
2570 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2571 assert(sig->count() == 4, "CAS has 4 arguments");
2572 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2573 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2574 #endif // ASSERT
2575 break;
2576 }
2577 case LS_cmp_exchange: {
2578 // Check the signatures.
2579 #ifdef ASSERT
2580 assert(rtype == type, "CAS must return the expected type");
2581 assert(sig->count() == 4, "CAS has 4 arguments");
2582 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2583 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2584 #endif // ASSERT
2585 break;
2586 }
2587 default:
2588 ShouldNotReachHere();
2589 }
2590 }
2591 #endif //PRODUCT
2592
2593 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2594
2595 // Get arguments:
2596 Node* receiver = NULL;
2597 Node* base = NULL;
2598 Node* offset = NULL;
2599 Node* oldval = NULL;
2600 Node* newval = NULL;
2601 switch(kind) {
2602 case LS_cmp_swap:
2603 case LS_cmp_swap_weak:
2604 case LS_cmp_exchange: {
2605 const bool two_slot_type = type2size[type] == 2;
2606 receiver = argument(0); // type: oop
2607 base = argument(1); // type: oop
2608 offset = argument(2); // type: long
2609 oldval = argument(4); // type: oop, int, or long
2610 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2611 break;
2612 }
2613 case LS_get_add:
2614 case LS_get_set: {
2615 receiver = argument(0); // type: oop
2616 base = argument(1); // type: oop
2617 offset = argument(2); // type: long
2618 oldval = NULL;
2619 newval = argument(4); // type: oop, int, or long
2620 break;
2621 }
2622 default:
2623 ShouldNotReachHere();
2624 }
2625
2626 // Build field offset expression.
2627 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2628 // to be plain byte offsets, which are also the same as those accepted
2629 // by oopDesc::field_addr.
2630 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2631 // 32-bit machines ignore the high half of long offsets
2632 offset = ConvL2X(offset);
2633 Node* adr = make_unsafe_address(base, offset, type, false);
2634 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2635
2636 Compile::AliasType* alias_type = C->alias_type(adr_type);
2637 BasicType bt = alias_type->basic_type();
2638 if (bt != T_ILLEGAL &&
2639 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2640 // Don't intrinsify mismatched object accesses.
2641 return false;
2642 }
2643
2644 // For CAS, unlike inline_unsafe_access, there seems no point in
2645 // trying to refine types. Just use the coarse types here.
2646 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2647 const Type *value_type = Type::get_const_basic_type(type);
2648
2649 switch (kind) {
2650 case LS_get_set:
2651 case LS_cmp_exchange: {
2652 if (type == T_OBJECT) {
2653 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2654 if (tjp != NULL) {
2655 value_type = tjp;
2656 }
2657 }
2658 break;
2659 }
2660 case LS_cmp_swap:
2661 case LS_cmp_swap_weak:
2662 case LS_get_add:
2663 break;
2664 default:
2665 ShouldNotReachHere();
2666 }
2667
2668 // Null check receiver.
2669 receiver = null_check(receiver);
2670 if (stopped()) {
2671 return true;
2672 }
2673
2674 int alias_idx = C->get_alias_index(adr_type);
2675
2676 if (type == T_OBJECT || type == T_ARRAY) {
2677 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2678
2679 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2680 // could be delayed during Parse (for example, in adjust_map_after_if()).
2681 // Execute transformation here to avoid barrier generation in such case.
2682 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2683 newval = _gvn.makecon(TypePtr::NULL_PTR);
2684
2685 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2686 // Refine the value to a null constant, when it is known to be null
2687 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2688 }
2689 }
2690
2691 Node* result = NULL;
2692 switch (kind) {
2693 case LS_cmp_exchange: {
2694 result = access_atomic_cmpxchg_val_at(control(), base, adr, adr_type, alias_idx,
2695 oldval, newval, value_type, type, decorators);
2696 break;
2697 }
2698 case LS_cmp_swap_weak:
2699 decorators |= C2_WEAK_CMPXCHG;
2700 case LS_cmp_swap: {
2701 result = access_atomic_cmpxchg_bool_at(control(), base, adr, adr_type, alias_idx,
2702 oldval, newval, value_type, type, decorators);
2703 break;
2704 }
2705 case LS_get_set: {
2706 result = access_atomic_xchg_at(control(), base, adr, adr_type, alias_idx,
2707 newval, value_type, type, decorators);
2708 break;
2709 }
2710 case LS_get_add: {
2711 result = access_atomic_add_at(control(), base, adr, adr_type, alias_idx,
2712 newval, value_type, type, decorators);
2713 break;
2714 }
2715 default:
2716 ShouldNotReachHere();
2717 }
2718
2719 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2720 set_result(result);
2721 return true;
2722 }
2723
2724 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2725 // Regardless of form, don't allow previous ld/st to move down,
2726 // then issue acquire, release, or volatile mem_bar.
2727 insert_mem_bar(Op_MemBarCPUOrder);
2728 switch(id) {
2729 case vmIntrinsics::_loadFence:
2730 insert_mem_bar(Op_LoadFence);
2731 return true;
2732 case vmIntrinsics::_storeFence:
2733 insert_mem_bar(Op_StoreFence);
2734 return true;
2735 case vmIntrinsics::_fullFence:
2736 insert_mem_bar(Op_MemBarVolatile);
2737 return true;
2738 default:
2739 fatal_unexpected_iid(id);
2740 return false;
2741 }
2742 }
2743
2744 bool LibraryCallKit::inline_onspinwait() {
2745 insert_mem_bar(Op_OnSpinWait);
2746 return true;
2747 }
2748
2749 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2750 if (!kls->is_Con()) {
2751 return true;
2752 }
2753 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2754 if (klsptr == NULL) {
2755 return true;
2756 }
2757 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2758 // don't need a guard for a klass that is already initialized
2759 return !ik->is_initialized();
2760 }
2761
2762 //----------------------------inline_unsafe_writeback0-------------------------
2763 // public native void Unsafe.writeback0(long address)
2764 bool LibraryCallKit::inline_unsafe_writeback0() {
2765 if (!Matcher::has_match_rule(Op_CacheWB)) {
2766 return false;
2767 }
2768 #ifndef PRODUCT
2769 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2770 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2771 ciSignature* sig = callee()->signature();
2772 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2773 #endif
2774 null_check_receiver(); // null-check, then ignore
2775 Node *addr = argument(1);
2776 addr = new CastX2PNode(addr);
2777 addr = _gvn.transform(addr);
2778 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2779 flush = _gvn.transform(flush);
2780 set_memory(flush, TypeRawPtr::BOTTOM);
2781 return true;
2782 }
2783
2784 //----------------------------inline_unsafe_writeback0-------------------------
2785 // public native void Unsafe.writeback0(long address)
2786 bool LibraryCallKit::inline_unsafe_writebackSync0(bool isPre) {
2787 if (isPre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2788 return false;
2789 }
2790 if (!isPre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2791 return false;
2792 }
2793 #ifndef PRODUCT
2794 assert(Matcher::has_match_rule(Op_CacheWB),
2795 (isPre ? "found match rule for CacheWBPreSync but not CacheWB"
2796 : "found match rule for CacheWBPostSync but not CacheWB"));
2797
2798 #endif
2799 null_check_receiver(); // null-check, then ignore
2800 Node *sync;
2801 if (isPre) {
2802 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2803 } else {
2804 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2805 }
2806 sync = _gvn.transform(sync);
2807 set_memory(sync, TypeRawPtr::BOTTOM);
2808 return true;
2809 }
2810
2811 //----------------------------inline_unsafe_allocate---------------------------
2812 // public native Object Unsafe.allocateInstance(Class<?> cls);
2813 bool LibraryCallKit::inline_unsafe_allocate() {
2814 if (callee()->is_static()) return false; // caller must have the capability!
2815
2816 null_check_receiver(); // null-check, then ignore
2817 Node* cls = null_check(argument(1));
2818 if (stopped()) return true;
2819
2820 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2821 kls = null_check(kls);
2822 if (stopped()) return true; // argument was like int.class
2823
2824 Node* test = NULL;
2825 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2826 // Note: The argument might still be an illegal value like
2827 // Serializable.class or Object[].class. The runtime will handle it.
2828 // But we must make an explicit check for initialization.
2829 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2830 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2831 // can generate code to load it as unsigned byte.
2832 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2833 Node* bits = intcon(InstanceKlass::fully_initialized);
2834 test = _gvn.transform(new SubINode(inst, bits));
2835 // The 'test' is non-zero if we need to take a slow path.
2836 }
2837
2838 Node* obj = new_instance(kls, test);
2839 set_result(obj);
2840 return true;
2841 }
2842
2843 //------------------------inline_native_time_funcs--------------
2844 // inline code for System.currentTimeMillis() and System.nanoTime()
2845 // these have the same type and signature
2846 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2847 const TypeFunc* tf = OptoRuntime::void_long_Type();
2848 const TypePtr* no_memory_effects = NULL;
2849 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2850 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2851 #ifdef ASSERT
2852 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2853 assert(value_top == top(), "second value must be top");
2854 #endif
2855 set_result(value);
2856 return true;
2857 }
2858
2859 #ifdef JFR_HAVE_INTRINSICS
2860
2861 /*
2862 * oop -> myklass
2863 * myklass->trace_id |= USED
2864 * return myklass->trace_id & ~0x3
2865 */
2866 bool LibraryCallKit::inline_native_classID() {
2867 Node* cls = null_check(argument(0), T_OBJECT);
2868 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2869 kls = null_check(kls, T_OBJECT);
2870
2871 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2872 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2873 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2874
2875 Node* clsused = longcon(0x01l); // set the class bit
2876 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2877 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2878 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2879
2880 #ifdef TRACE_ID_META_BITS
2881 Node* mbits = longcon(~TRACE_ID_META_BITS);
2882 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2883 #endif
2884 #ifdef TRACE_ID_SHIFT
2885 Node* cbits = intcon(TRACE_ID_SHIFT);
2886 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2887 #endif
2888
2889 set_result(tvalue);
2890 return true;
2891
2892 }
2893
2894 bool LibraryCallKit::inline_native_getEventWriter() {
2895 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2896
2897 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2898 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)
2899 );
2900
2901 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2902
2903 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2904 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2905
2906 IfNode* iff_jobj_null =
2907 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2908
2909 enum { _normal_path = 1,
2910 _null_path = 2,
2911 PATH_LIMIT };
2912
2913 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2914 PhiNode* result_val = new PhiNode(result_rgn, TypePtr::BOTTOM);
2915
2916 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2917 result_rgn->init_req(_null_path, jobj_is_null);
2918 result_val->init_req(_null_path, null());
2919
2920 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2921 result_rgn->init_req(_normal_path, jobj_is_not_null);
2922
2923 Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
2924 result_val->init_req(_normal_path, res);
2925
2926 set_result(result_rgn, result_val);
2927
2928 return true;
2929 }
2930
2931 #endif // JFR_HAVE_INTRINSICS
2932
2933 //------------------------inline_native_currentThread------------------
2934 bool LibraryCallKit::inline_native_currentThread() {
2935 Node* junk = NULL;
2936 set_result(generate_current_thread(junk));
2937 return true;
2938 }
2939
2940 //------------------------inline_native_isInterrupted------------------
2941 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
2942 bool LibraryCallKit::inline_native_isInterrupted() {
2943 // Add a fast path to t.isInterrupted(clear_int):
2944 // (t == Thread.current() &&
2945 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
2946 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
2947 // So, in the common case that the interrupt bit is false,
2948 // we avoid making a call into the VM. Even if the interrupt bit
2949 // is true, if the clear_int argument is false, we avoid the VM call.
2950 // However, if the receiver is not currentThread, we must call the VM,
2951 // because there must be some locking done around the operation.
2952
2953 // We only go to the fast case code if we pass two guards.
2954 // Paths which do not pass are accumulated in the slow_region.
2955
2956 enum {
2957 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
2958 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
2959 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
2960 PATH_LIMIT
2961 };
2962
2963 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
2964 // out of the function.
2965 insert_mem_bar(Op_MemBarCPUOrder);
2966
2967 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2968 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
2969
2970 RegionNode* slow_region = new RegionNode(1);
2971 record_for_igvn(slow_region);
2972
2973 // (a) Receiving thread must be the current thread.
2974 Node* rec_thr = argument(0);
2975 Node* tls_ptr = NULL;
2976 Node* cur_thr = generate_current_thread(tls_ptr);
2977 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
2978 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
2979
2980 generate_slow_guard(bol_thr, slow_region);
2981
2982 // (b) Interrupt bit on TLS must be false.
2983 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
2984 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
2985 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
2986
2987 // Set the control input on the field _interrupted read to prevent it floating up.
2988 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
2989 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
2990 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
2991
2992 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2993
2994 // First fast path: if (!TLS._interrupted) return false;
2995 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
2996 result_rgn->init_req(no_int_result_path, false_bit);
2997 result_val->init_req(no_int_result_path, intcon(0));
2998
2999 // drop through to next case
3000 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3001
3002 #ifndef _WINDOWS
3003 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3004 Node* clr_arg = argument(1);
3005 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3006 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3007 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3008
3009 // Second fast path: ... else if (!clear_int) return true;
3010 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3011 result_rgn->init_req(no_clear_result_path, false_arg);
3012 result_val->init_req(no_clear_result_path, intcon(1));
3013
3014 // drop through to next case
3015 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3016 #else
3017 // To return true on Windows you must read the _interrupted field
3018 // and check the event state i.e. take the slow path.
3019 #endif // _WINDOWS
3020
3021 // (d) Otherwise, go to the slow path.
3022 slow_region->add_req(control());
3023 set_control( _gvn.transform(slow_region));
3024
3025 if (stopped()) {
3026 // There is no slow path.
3027 result_rgn->init_req(slow_result_path, top());
3028 result_val->init_req(slow_result_path, top());
3029 } else {
3030 // non-virtual because it is a private non-static
3031 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3032
3033 Node* slow_val = set_results_for_java_call(slow_call);
3034 // this->control() comes from set_results_for_java_call
3035
3036 Node* fast_io = slow_call->in(TypeFunc::I_O);
3037 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3038
3039 // These two phis are pre-filled with copies of of the fast IO and Memory
3040 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3041 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3042
3043 result_rgn->init_req(slow_result_path, control());
3044 result_io ->init_req(slow_result_path, i_o());
3045 result_mem->init_req(slow_result_path, reset_memory());
3046 result_val->init_req(slow_result_path, slow_val);
3047
3048 set_all_memory(_gvn.transform(result_mem));
3049 set_i_o( _gvn.transform(result_io));
3050 }
3051
3052 C->set_has_split_ifs(true); // Has chance for split-if optimization
3053 set_result(result_rgn, result_val);
3054 return true;
3055 }
3056
3057 //---------------------------load_mirror_from_klass----------------------------
3058 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3059 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3060 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3061 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3062 // mirror = ((OopHandle)mirror)->resolve();
3063 return make_load(NULL, load, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3064 }
3065
3066 //-----------------------load_klass_from_mirror_common-------------------------
3067 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3068 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3069 // and branch to the given path on the region.
3070 // If never_see_null, take an uncommon trap on null, so we can optimistically
3071 // compile for the non-null case.
3072 // If the region is NULL, force never_see_null = true.
3073 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3074 bool never_see_null,
3075 RegionNode* region,
3076 int null_path,
3077 int offset) {
3078 if (region == NULL) never_see_null = true;
3079 Node* p = basic_plus_adr(mirror, offset);
3080 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3081 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3082 Node* null_ctl = top();
3083 kls = null_check_oop(kls, &null_ctl, never_see_null);
3084 if (region != NULL) {
3085 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3086 region->init_req(null_path, null_ctl);
3087 } else {
3088 assert(null_ctl == top(), "no loose ends");
3089 }
3090 return kls;
3091 }
3092
3093 //--------------------(inline_native_Class_query helpers)---------------------
3094 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3095 // Fall through if (mods & mask) == bits, take the guard otherwise.
3096 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3097 // Branch around if the given klass has the given modifier bit set.
3098 // Like generate_guard, adds a new path onto the region.
3099 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3100 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3101 Node* mask = intcon(modifier_mask);
3102 Node* bits = intcon(modifier_bits);
3103 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3104 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3105 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3106 return generate_fair_guard(bol, region);
3107 }
3108 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3109 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3110 }
3111
3112 //-------------------------inline_native_Class_query-------------------
3113 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3114 const Type* return_type = TypeInt::BOOL;
3115 Node* prim_return_value = top(); // what happens if it's a primitive class?
3116 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3117 bool expect_prim = false; // most of these guys expect to work on refs
3118
3119 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3120
3121 Node* mirror = argument(0);
3122 Node* obj = top();
3123
3124 switch (id) {
3125 case vmIntrinsics::_isInstance:
3126 // nothing is an instance of a primitive type
3127 prim_return_value = intcon(0);
3128 obj = argument(1);
3129 break;
3130 case vmIntrinsics::_getModifiers:
3131 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3132 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3133 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3134 break;
3135 case vmIntrinsics::_isInterface:
3136 prim_return_value = intcon(0);
3137 break;
3138 case vmIntrinsics::_isArray:
3139 prim_return_value = intcon(0);
3140 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3141 break;
3142 case vmIntrinsics::_isPrimitive:
3143 prim_return_value = intcon(1);
3144 expect_prim = true; // obviously
3145 break;
3146 case vmIntrinsics::_getSuperclass:
3147 prim_return_value = null();
3148 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3149 break;
3150 case vmIntrinsics::_getClassAccessFlags:
3151 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3152 return_type = TypeInt::INT; // not bool! 6297094
3153 break;
3154 default:
3155 fatal_unexpected_iid(id);
3156 break;
3157 }
3158
3159 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3160 if (mirror_con == NULL) return false; // cannot happen?
3161
3162 #ifndef PRODUCT
3163 if (C->print_intrinsics() || C->print_inlining()) {
3164 ciType* k = mirror_con->java_mirror_type();
3165 if (k) {
3166 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3167 k->print_name();
3168 tty->cr();
3169 }
3170 }
3171 #endif
3172
3173 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3174 RegionNode* region = new RegionNode(PATH_LIMIT);
3175 record_for_igvn(region);
3176 PhiNode* phi = new PhiNode(region, return_type);
3177
3178 // The mirror will never be null of Reflection.getClassAccessFlags, however
3179 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3180 // if it is. See bug 4774291.
3181
3182 // For Reflection.getClassAccessFlags(), the null check occurs in
3183 // the wrong place; see inline_unsafe_access(), above, for a similar
3184 // situation.
3185 mirror = null_check(mirror);
3186 // If mirror or obj is dead, only null-path is taken.
3187 if (stopped()) return true;
3188
3189 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3190
3191 // Now load the mirror's klass metaobject, and null-check it.
3192 // Side-effects region with the control path if the klass is null.
3193 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3194 // If kls is null, we have a primitive mirror.
3195 phi->init_req(_prim_path, prim_return_value);
3196 if (stopped()) { set_result(region, phi); return true; }
3197 bool safe_for_replace = (region->in(_prim_path) == top());
3198
3199 Node* p; // handy temp
3200 Node* null_ctl;
3201
3202 // Now that we have the non-null klass, we can perform the real query.
3203 // For constant classes, the query will constant-fold in LoadNode::Value.
3204 Node* query_value = top();
3205 switch (id) {
3206 case vmIntrinsics::_isInstance:
3207 // nothing is an instance of a primitive type
3208 query_value = gen_instanceof(obj, kls, safe_for_replace);
3209 break;
3210
3211 case vmIntrinsics::_getModifiers:
3212 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3213 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3214 break;
3215
3216 case vmIntrinsics::_isInterface:
3217 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3218 if (generate_interface_guard(kls, region) != NULL)
3219 // A guard was added. If the guard is taken, it was an interface.
3220 phi->add_req(intcon(1));
3221 // If we fall through, it's a plain class.
3222 query_value = intcon(0);
3223 break;
3224
3225 case vmIntrinsics::_isArray:
3226 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3227 if (generate_array_guard(kls, region) != NULL)
3228 // A guard was added. If the guard is taken, it was an array.
3229 phi->add_req(intcon(1));
3230 // If we fall through, it's a plain class.
3231 query_value = intcon(0);
3232 break;
3233
3234 case vmIntrinsics::_isPrimitive:
3235 query_value = intcon(0); // "normal" path produces false
3236 break;
3237
3238 case vmIntrinsics::_getSuperclass:
3239 // The rules here are somewhat unfortunate, but we can still do better
3240 // with random logic than with a JNI call.
3241 // Interfaces store null or Object as _super, but must report null.
3242 // Arrays store an intermediate super as _super, but must report Object.
3243 // Other types can report the actual _super.
3244 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3245 if (generate_interface_guard(kls, region) != NULL)
3246 // A guard was added. If the guard is taken, it was an interface.
3247 phi->add_req(null());
3248 if (generate_array_guard(kls, region) != NULL)
3249 // A guard was added. If the guard is taken, it was an array.
3250 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3251 // If we fall through, it's a plain class. Get its _super.
3252 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3253 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3254 null_ctl = top();
3255 kls = null_check_oop(kls, &null_ctl);
3256 if (null_ctl != top()) {
3257 // If the guard is taken, Object.superClass is null (both klass and mirror).
3258 region->add_req(null_ctl);
3259 phi ->add_req(null());
3260 }
3261 if (!stopped()) {
3262 query_value = load_mirror_from_klass(kls);
3263 }
3264 break;
3265
3266 case vmIntrinsics::_getClassAccessFlags:
3267 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3268 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3269 break;
3270
3271 default:
3272 fatal_unexpected_iid(id);
3273 break;
3274 }
3275
3276 // Fall-through is the normal case of a query to a real class.
3277 phi->init_req(1, query_value);
3278 region->init_req(1, control());
3279
3280 C->set_has_split_ifs(true); // Has chance for split-if optimization
3281 set_result(region, phi);
3282 return true;
3283 }
3284
3285 //-------------------------inline_Class_cast-------------------
3286 bool LibraryCallKit::inline_Class_cast() {
3287 Node* mirror = argument(0); // Class
3288 Node* obj = argument(1);
3289 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3290 if (mirror_con == NULL) {
3291 return false; // dead path (mirror->is_top()).
3292 }
3293 if (obj == NULL || obj->is_top()) {
3294 return false; // dead path
3295 }
3296 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3297
3298 // First, see if Class.cast() can be folded statically.
3299 // java_mirror_type() returns non-null for compile-time Class constants.
3300 ciType* tm = mirror_con->java_mirror_type();
3301 if (tm != NULL && tm->is_klass() &&
3302 tp != NULL && tp->klass() != NULL) {
3303 if (!tp->klass()->is_loaded()) {
3304 // Don't use intrinsic when class is not loaded.
3305 return false;
3306 } else {
3307 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3308 if (static_res == Compile::SSC_always_true) {
3309 // isInstance() is true - fold the code.
3310 set_result(obj);
3311 return true;
3312 } else if (static_res == Compile::SSC_always_false) {
3313 // Don't use intrinsic, have to throw ClassCastException.
3314 // If the reference is null, the non-intrinsic bytecode will
3315 // be optimized appropriately.
3316 return false;
3317 }
3318 }
3319 }
3320
3321 // Bailout intrinsic and do normal inlining if exception path is frequent.
3322 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3323 return false;
3324 }
3325
3326 // Generate dynamic checks.
3327 // Class.cast() is java implementation of _checkcast bytecode.
3328 // Do checkcast (Parse::do_checkcast()) optimizations here.
3329
3330 mirror = null_check(mirror);
3331 // If mirror is dead, only null-path is taken.
3332 if (stopped()) {
3333 return true;
3334 }
3335
3336 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3337 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3338 RegionNode* region = new RegionNode(PATH_LIMIT);
3339 record_for_igvn(region);
3340
3341 // Now load the mirror's klass metaobject, and null-check it.
3342 // If kls is null, we have a primitive mirror and
3343 // nothing is an instance of a primitive type.
3344 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3345
3346 Node* res = top();
3347 if (!stopped()) {
3348 Node* bad_type_ctrl = top();
3349 // Do checkcast optimizations.
3350 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3351 region->init_req(_bad_type_path, bad_type_ctrl);
3352 }
3353 if (region->in(_prim_path) != top() ||
3354 region->in(_bad_type_path) != top()) {
3355 // Let Interpreter throw ClassCastException.
3356 PreserveJVMState pjvms(this);
3357 set_control(_gvn.transform(region));
3358 uncommon_trap(Deoptimization::Reason_intrinsic,
3359 Deoptimization::Action_maybe_recompile);
3360 }
3361 if (!stopped()) {
3362 set_result(res);
3363 }
3364 return true;
3365 }
3366
3367
3368 //--------------------------inline_native_subtype_check------------------------
3369 // This intrinsic takes the JNI calls out of the heart of
3370 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3371 bool LibraryCallKit::inline_native_subtype_check() {
3372 // Pull both arguments off the stack.
3373 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3374 args[0] = argument(0);
3375 args[1] = argument(1);
3376 Node* klasses[2]; // corresponding Klasses: superk, subk
3377 klasses[0] = klasses[1] = top();
3378
3379 enum {
3380 // A full decision tree on {superc is prim, subc is prim}:
3381 _prim_0_path = 1, // {P,N} => false
3382 // {P,P} & superc!=subc => false
3383 _prim_same_path, // {P,P} & superc==subc => true
3384 _prim_1_path, // {N,P} => false
3385 _ref_subtype_path, // {N,N} & subtype check wins => true
3386 _both_ref_path, // {N,N} & subtype check loses => false
3387 PATH_LIMIT
3388 };
3389
3390 RegionNode* region = new RegionNode(PATH_LIMIT);
3391 Node* phi = new PhiNode(region, TypeInt::BOOL);
3392 record_for_igvn(region);
3393
3394 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3395 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3396 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3397
3398 // First null-check both mirrors and load each mirror's klass metaobject.
3399 int which_arg;
3400 for (which_arg = 0; which_arg <= 1; which_arg++) {
3401 Node* arg = args[which_arg];
3402 arg = null_check(arg);
3403 if (stopped()) break;
3404 args[which_arg] = arg;
3405
3406 Node* p = basic_plus_adr(arg, class_klass_offset);
3407 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3408 klasses[which_arg] = _gvn.transform(kls);
3409 }
3410
3411 // Having loaded both klasses, test each for null.
3412 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3413 for (which_arg = 0; which_arg <= 1; which_arg++) {
3414 Node* kls = klasses[which_arg];
3415 Node* null_ctl = top();
3416 kls = null_check_oop(kls, &null_ctl, never_see_null);
3417 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3418 region->init_req(prim_path, null_ctl);
3419 if (stopped()) break;
3420 klasses[which_arg] = kls;
3421 }
3422
3423 if (!stopped()) {
3424 // now we have two reference types, in klasses[0..1]
3425 Node* subk = klasses[1]; // the argument to isAssignableFrom
3426 Node* superk = klasses[0]; // the receiver
3427 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3428 // now we have a successful reference subtype check
3429 region->set_req(_ref_subtype_path, control());
3430 }
3431
3432 // If both operands are primitive (both klasses null), then
3433 // we must return true when they are identical primitives.
3434 // It is convenient to test this after the first null klass check.
3435 set_control(region->in(_prim_0_path)); // go back to first null check
3436 if (!stopped()) {
3437 // Since superc is primitive, make a guard for the superc==subc case.
3438 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3439 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3440 generate_guard(bol_eq, region, PROB_FAIR);
3441 if (region->req() == PATH_LIMIT+1) {
3442 // A guard was added. If the added guard is taken, superc==subc.
3443 region->swap_edges(PATH_LIMIT, _prim_same_path);
3444 region->del_req(PATH_LIMIT);
3445 }
3446 region->set_req(_prim_0_path, control()); // Not equal after all.
3447 }
3448
3449 // these are the only paths that produce 'true':
3450 phi->set_req(_prim_same_path, intcon(1));
3451 phi->set_req(_ref_subtype_path, intcon(1));
3452
3453 // pull together the cases:
3454 assert(region->req() == PATH_LIMIT, "sane region");
3455 for (uint i = 1; i < region->req(); i++) {
3456 Node* ctl = region->in(i);
3457 if (ctl == NULL || ctl == top()) {
3458 region->set_req(i, top());
3459 phi ->set_req(i, top());
3460 } else if (phi->in(i) == NULL) {
3461 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3462 }
3463 }
3464
3465 set_control(_gvn.transform(region));
3466 set_result(_gvn.transform(phi));
3467 return true;
3468 }
3469
3470 //---------------------generate_array_guard_common------------------------
3471 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3472 bool obj_array, bool not_array) {
3473
3474 if (stopped()) {
3475 return NULL;
3476 }
3477
3478 // If obj_array/non_array==false/false:
3479 // Branch around if the given klass is in fact an array (either obj or prim).
3480 // If obj_array/non_array==false/true:
3481 // Branch around if the given klass is not an array klass of any kind.
3482 // If obj_array/non_array==true/true:
3483 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3484 // If obj_array/non_array==true/false:
3485 // Branch around if the kls is an oop array (Object[] or subtype)
3486 //
3487 // Like generate_guard, adds a new path onto the region.
3488 jint layout_con = 0;
3489 Node* layout_val = get_layout_helper(kls, layout_con);
3490 if (layout_val == NULL) {
3491 bool query = (obj_array
3492 ? Klass::layout_helper_is_objArray(layout_con)
3493 : Klass::layout_helper_is_array(layout_con));
3494 if (query == not_array) {
3495 return NULL; // never a branch
3496 } else { // always a branch
3497 Node* always_branch = control();
3498 if (region != NULL)
3499 region->add_req(always_branch);
3500 set_control(top());
3501 return always_branch;
3502 }
3503 }
3504 // Now test the correct condition.
3505 jint nval = (obj_array
3506 ? (jint)(Klass::_lh_array_tag_type_value
3507 << Klass::_lh_array_tag_shift)
3508 : Klass::_lh_neutral_value);
3509 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3510 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3511 // invert the test if we are looking for a non-array
3512 if (not_array) btest = BoolTest(btest).negate();
3513 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3514 return generate_fair_guard(bol, region);
3515 }
3516
3517
3518 //-----------------------inline_native_newArray--------------------------
3519 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3520 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3521 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3522 Node* mirror;
3523 Node* count_val;
3524 if (uninitialized) {
3525 mirror = argument(1);
3526 count_val = argument(2);
3527 } else {
3528 mirror = argument(0);
3529 count_val = argument(1);
3530 }
3531
3532 mirror = null_check(mirror);
3533 // If mirror or obj is dead, only null-path is taken.
3534 if (stopped()) return true;
3535
3536 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3537 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3538 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3539 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3540 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3541
3542 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3543 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3544 result_reg, _slow_path);
3545 Node* normal_ctl = control();
3546 Node* no_array_ctl = result_reg->in(_slow_path);
3547
3548 // Generate code for the slow case. We make a call to newArray().
3549 set_control(no_array_ctl);
3550 if (!stopped()) {
3551 // Either the input type is void.class, or else the
3552 // array klass has not yet been cached. Either the
3553 // ensuing call will throw an exception, or else it
3554 // will cache the array klass for next time.
3555 PreserveJVMState pjvms(this);
3556 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3557 Node* slow_result = set_results_for_java_call(slow_call);
3558 // this->control() comes from set_results_for_java_call
3559 result_reg->set_req(_slow_path, control());
3560 result_val->set_req(_slow_path, slow_result);
3561 result_io ->set_req(_slow_path, i_o());
3562 result_mem->set_req(_slow_path, reset_memory());
3563 }
3564
3565 set_control(normal_ctl);
3566 if (!stopped()) {
3567 // Normal case: The array type has been cached in the java.lang.Class.
3568 // The following call works fine even if the array type is polymorphic.
3569 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3570 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3571 result_reg->init_req(_normal_path, control());
3572 result_val->init_req(_normal_path, obj);
3573 result_io ->init_req(_normal_path, i_o());
3574 result_mem->init_req(_normal_path, reset_memory());
3575
3576 if (uninitialized) {
3577 // Mark the allocation so that zeroing is skipped
3578 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3579 alloc->maybe_set_complete(&_gvn);
3580 }
3581 }
3582
3583 // Return the combined state.
3584 set_i_o( _gvn.transform(result_io) );
3585 set_all_memory( _gvn.transform(result_mem));
3586
3587 C->set_has_split_ifs(true); // Has chance for split-if optimization
3588 set_result(result_reg, result_val);
3589 return true;
3590 }
3591
3592 //----------------------inline_native_getLength--------------------------
3593 // public static native int java.lang.reflect.Array.getLength(Object array);
3594 bool LibraryCallKit::inline_native_getLength() {
3595 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3596
3597 Node* array = null_check(argument(0));
3598 // If array is dead, only null-path is taken.
3599 if (stopped()) return true;
3600
3601 // Deoptimize if it is a non-array.
3602 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3603
3604 if (non_array != NULL) {
3605 PreserveJVMState pjvms(this);
3606 set_control(non_array);
3607 uncommon_trap(Deoptimization::Reason_intrinsic,
3608 Deoptimization::Action_maybe_recompile);
3609 }
3610
3611 // If control is dead, only non-array-path is taken.
3612 if (stopped()) return true;
3613
3614 // The works fine even if the array type is polymorphic.
3615 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3616 Node* result = load_array_length(array);
3617
3618 C->set_has_split_ifs(true); // Has chance for split-if optimization
3619 set_result(result);
3620 return true;
3621 }
3622
3623 //------------------------inline_array_copyOf----------------------------
3624 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3625 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3626 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3627 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3628
3629 // Get the arguments.
3630 Node* original = argument(0);
3631 Node* start = is_copyOfRange? argument(1): intcon(0);
3632 Node* end = is_copyOfRange? argument(2): argument(1);
3633 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3634
3635 Node* newcopy = NULL;
3636
3637 // Set the original stack and the reexecute bit for the interpreter to reexecute
3638 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3639 { PreserveReexecuteState preexecs(this);
3640 jvms()->set_should_reexecute(true);
3641
3642 array_type_mirror = null_check(array_type_mirror);
3643 original = null_check(original);
3644
3645 // Check if a null path was taken unconditionally.
3646 if (stopped()) return true;
3647
3648 Node* orig_length = load_array_length(original);
3649
3650 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3651 klass_node = null_check(klass_node);
3652
3653 RegionNode* bailout = new RegionNode(1);
3654 record_for_igvn(bailout);
3655
3656 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3657 // Bail out if that is so.
3658 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3659 if (not_objArray != NULL) {
3660 // Improve the klass node's type from the new optimistic assumption:
3661 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3662 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3663 Node* cast = new CastPPNode(klass_node, akls);
3664 cast->init_req(0, control());
3665 klass_node = _gvn.transform(cast);
3666 }
3667
3668 // Bail out if either start or end is negative.
3669 generate_negative_guard(start, bailout, &start);
3670 generate_negative_guard(end, bailout, &end);
3671
3672 Node* length = end;
3673 if (_gvn.type(start) != TypeInt::ZERO) {
3674 length = _gvn.transform(new SubINode(end, start));
3675 }
3676
3677 // Bail out if length is negative.
3678 // Without this the new_array would throw
3679 // NegativeArraySizeException but IllegalArgumentException is what
3680 // should be thrown
3681 generate_negative_guard(length, bailout, &length);
3682
3683 if (bailout->req() > 1) {
3684 PreserveJVMState pjvms(this);
3685 set_control(_gvn.transform(bailout));
3686 uncommon_trap(Deoptimization::Reason_intrinsic,
3687 Deoptimization::Action_maybe_recompile);
3688 }
3689
3690 if (!stopped()) {
3691 // How many elements will we copy from the original?
3692 // The answer is MinI(orig_length - start, length).
3693 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3694 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3695
3696 // Generate a direct call to the right arraycopy function(s).
3697 // We know the copy is disjoint but we might not know if the
3698 // oop stores need checking.
3699 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3700 // This will fail a store-check if x contains any non-nulls.
3701
3702 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3703 // loads/stores but it is legal only if we're sure the
3704 // Arrays.copyOf would succeed. So we need all input arguments
3705 // to the copyOf to be validated, including that the copy to the
3706 // new array won't trigger an ArrayStoreException. That subtype
3707 // check can be optimized if we know something on the type of
3708 // the input array from type speculation.
3709 if (_gvn.type(klass_node)->singleton()) {
3710 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3711 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3712
3713 int test = C->static_subtype_check(superk, subk);
3714 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3715 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3716 if (t_original->speculative_type() != NULL) {
3717 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3718 }
3719 }
3720 }
3721
3722 bool validated = false;
3723 // Reason_class_check rather than Reason_intrinsic because we
3724 // want to intrinsify even if this traps.
3725 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3726 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3727 klass_node);
3728
3729 if (not_subtype_ctrl != top()) {
3730 PreserveJVMState pjvms(this);
3731 set_control(not_subtype_ctrl);
3732 uncommon_trap(Deoptimization::Reason_class_check,
3733 Deoptimization::Action_make_not_entrant);
3734 assert(stopped(), "Should be stopped");
3735 }
3736 validated = true;
3737 }
3738
3739 if (!stopped()) {
3740 newcopy = new_array(klass_node, length, 0); // no arguments to push
3741
3742 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3743 load_object_klass(original), klass_node);
3744 if (!is_copyOfRange) {
3745 ac->set_copyof(validated);
3746 } else {
3747 ac->set_copyofrange(validated);
3748 }
3749 Node* n = _gvn.transform(ac);
3750 if (n == ac) {
3751 ac->connect_outputs(this);
3752 } else {
3753 assert(validated, "shouldn't transform if all arguments not validated");
3754 set_all_memory(n);
3755 }
3756 }
3757 }
3758 } // original reexecute is set back here
3759
3760 C->set_has_split_ifs(true); // Has chance for split-if optimization
3761 if (!stopped()) {
3762 set_result(newcopy);
3763 }
3764 return true;
3765 }
3766
3767
3768 //----------------------generate_virtual_guard---------------------------
3769 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3770 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3771 RegionNode* slow_region) {
3772 ciMethod* method = callee();
3773 int vtable_index = method->vtable_index();
3774 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3775 "bad index %d", vtable_index);
3776 // Get the Method* out of the appropriate vtable entry.
3777 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3778 vtable_index*vtableEntry::size_in_bytes() +
3779 vtableEntry::method_offset_in_bytes();
3780 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3781 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3782
3783 // Compare the target method with the expected method (e.g., Object.hashCode).
3784 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3785
3786 Node* native_call = makecon(native_call_addr);
3787 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3788 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3789
3790 return generate_slow_guard(test_native, slow_region);
3791 }
3792
3793 //-----------------------generate_method_call----------------------------
3794 // Use generate_method_call to make a slow-call to the real
3795 // method if the fast path fails. An alternative would be to
3796 // use a stub like OptoRuntime::slow_arraycopy_Java.
3797 // This only works for expanding the current library call,
3798 // not another intrinsic. (E.g., don't use this for making an
3799 // arraycopy call inside of the copyOf intrinsic.)
3800 CallJavaNode*
3801 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3802 // When compiling the intrinsic method itself, do not use this technique.
3803 guarantee(callee() != C->method(), "cannot make slow-call to self");
3804
3805 ciMethod* method = callee();
3806 // ensure the JVMS we have will be correct for this call
3807 guarantee(method_id == method->intrinsic_id(), "must match");
3808
3809 const TypeFunc* tf = TypeFunc::make(method);
3810 CallJavaNode* slow_call;
3811 if (is_static) {
3812 assert(!is_virtual, "");
3813 slow_call = new CallStaticJavaNode(C, tf,
3814 SharedRuntime::get_resolve_static_call_stub(),
3815 method, bci());
3816 } else if (is_virtual) {
3817 null_check_receiver();
3818 int vtable_index = Method::invalid_vtable_index;
3819 if (UseInlineCaches) {
3820 // Suppress the vtable call
3821 } else {
3822 // hashCode and clone are not a miranda methods,
3823 // so the vtable index is fixed.
3824 // No need to use the linkResolver to get it.
3825 vtable_index = method->vtable_index();
3826 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3827 "bad index %d", vtable_index);
3828 }
3829 slow_call = new CallDynamicJavaNode(tf,
3830 SharedRuntime::get_resolve_virtual_call_stub(),
3831 method, vtable_index, bci());
3832 } else { // neither virtual nor static: opt_virtual
3833 null_check_receiver();
3834 slow_call = new CallStaticJavaNode(C, tf,
3835 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3836 method, bci());
3837 slow_call->set_optimized_virtual(true);
3838 }
3839 set_arguments_for_java_call(slow_call);
3840 set_edges_for_java_call(slow_call);
3841 return slow_call;
3842 }
3843
3844
3845 /**
3846 * Build special case code for calls to hashCode on an object. This call may
3847 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3848 * slightly different code.
3849 */
3850 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3851 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3852 assert(!(is_virtual && is_static), "either virtual, special, or static");
3853
3854 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3855
3856 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3857 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3858 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3859 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3860 Node* obj = NULL;
3861 if (!is_static) {
3862 // Check for hashing null object
3863 obj = null_check_receiver();
3864 if (stopped()) return true; // unconditionally null
3865 result_reg->init_req(_null_path, top());
3866 result_val->init_req(_null_path, top());
3867 } else {
3868 // Do a null check, and return zero if null.
3869 // System.identityHashCode(null) == 0
3870 obj = argument(0);
3871 Node* null_ctl = top();
3872 obj = null_check_oop(obj, &null_ctl);
3873 result_reg->init_req(_null_path, null_ctl);
3874 result_val->init_req(_null_path, _gvn.intcon(0));
3875 }
3876
3877 // Unconditionally null? Then return right away.
3878 if (stopped()) {
3879 set_control( result_reg->in(_null_path));
3880 if (!stopped())
3881 set_result(result_val->in(_null_path));
3882 return true;
3883 }
3884
3885 // We only go to the fast case code if we pass a number of guards. The
3886 // paths which do not pass are accumulated in the slow_region.
3887 RegionNode* slow_region = new RegionNode(1);
3888 record_for_igvn(slow_region);
3889
3890 // If this is a virtual call, we generate a funny guard. We pull out
3891 // the vtable entry corresponding to hashCode() from the target object.
3892 // If the target method which we are calling happens to be the native
3893 // Object hashCode() method, we pass the guard. We do not need this
3894 // guard for non-virtual calls -- the caller is known to be the native
3895 // Object hashCode().
3896 if (is_virtual) {
3897 // After null check, get the object's klass.
3898 Node* obj_klass = load_object_klass(obj);
3899 generate_virtual_guard(obj_klass, slow_region);
3900 }
3901
3902 // Get the header out of the object, use LoadMarkNode when available
3903 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3904 // The control of the load must be NULL. Otherwise, the load can move before
3905 // the null check after castPP removal.
3906 Node* no_ctrl = NULL;
3907 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3908
3909 // Test the header to see if it is unlocked.
3910 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
3911 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3912 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
3913 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3914 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3915
3916 generate_slow_guard(test_unlocked, slow_region);
3917
3918 // Get the hash value and check to see that it has been properly assigned.
3919 // We depend on hash_mask being at most 32 bits and avoid the use of
3920 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3921 // vm: see markOop.hpp.
3922 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
3923 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
3924 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3925 // This hack lets the hash bits live anywhere in the mark object now, as long
3926 // as the shift drops the relevant bits into the low 32 bits. Note that
3927 // Java spec says that HashCode is an int so there's no point in capturing
3928 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3929 hshifted_header = ConvX2I(hshifted_header);
3930 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3931
3932 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
3933 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3934 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3935
3936 generate_slow_guard(test_assigned, slow_region);
3937
3938 Node* init_mem = reset_memory();
3939 // fill in the rest of the null path:
3940 result_io ->init_req(_null_path, i_o());
3941 result_mem->init_req(_null_path, init_mem);
3942
3943 result_val->init_req(_fast_path, hash_val);
3944 result_reg->init_req(_fast_path, control());
3945 result_io ->init_req(_fast_path, i_o());
3946 result_mem->init_req(_fast_path, init_mem);
3947
3948 // Generate code for the slow case. We make a call to hashCode().
3949 set_control(_gvn.transform(slow_region));
3950 if (!stopped()) {
3951 // No need for PreserveJVMState, because we're using up the present state.
3952 set_all_memory(init_mem);
3953 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3954 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3955 Node* slow_result = set_results_for_java_call(slow_call);
3956 // this->control() comes from set_results_for_java_call
3957 result_reg->init_req(_slow_path, control());
3958 result_val->init_req(_slow_path, slow_result);
3959 result_io ->set_req(_slow_path, i_o());
3960 result_mem ->set_req(_slow_path, reset_memory());
3961 }
3962
3963 // Return the combined state.
3964 set_i_o( _gvn.transform(result_io) );
3965 set_all_memory( _gvn.transform(result_mem));
3966
3967 set_result(result_reg, result_val);
3968 return true;
3969 }
3970
3971 //---------------------------inline_native_getClass----------------------------
3972 // public final native Class<?> java.lang.Object.getClass();
3973 //
3974 // Build special case code for calls to getClass on an object.
3975 bool LibraryCallKit::inline_native_getClass() {
3976 Node* obj = null_check_receiver();
3977 if (stopped()) return true;
3978 set_result(load_mirror_from_klass(load_object_klass(obj)));
3979 return true;
3980 }
3981
3982 //-----------------inline_native_Reflection_getCallerClass---------------------
3983 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3984 //
3985 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3986 //
3987 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3988 // in that it must skip particular security frames and checks for
3989 // caller sensitive methods.
3990 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3991 #ifndef PRODUCT
3992 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3993 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3994 }
3995 #endif
3996
3997 if (!jvms()->has_method()) {
3998 #ifndef PRODUCT
3999 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4000 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4001 }
4002 #endif
4003 return false;
4004 }
4005
4006 // Walk back up the JVM state to find the caller at the required
4007 // depth.
4008 JVMState* caller_jvms = jvms();
4009
4010 // Cf. JVM_GetCallerClass
4011 // NOTE: Start the loop at depth 1 because the current JVM state does
4012 // not include the Reflection.getCallerClass() frame.
4013 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4014 ciMethod* m = caller_jvms->method();
4015 switch (n) {
4016 case 0:
4017 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4018 break;
4019 case 1:
4020 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4021 if (!m->caller_sensitive()) {
4022 #ifndef PRODUCT
4023 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4024 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4025 }
4026 #endif
4027 return false; // bail-out; let JVM_GetCallerClass do the work
4028 }
4029 break;
4030 default:
4031 if (!m->is_ignored_by_security_stack_walk()) {
4032 // We have reached the desired frame; return the holder class.
4033 // Acquire method holder as java.lang.Class and push as constant.
4034 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4035 ciInstance* caller_mirror = caller_klass->java_mirror();
4036 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4037
4038 #ifndef PRODUCT
4039 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4040 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());
4041 tty->print_cr(" JVM state at this point:");
4042 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4043 ciMethod* m = jvms()->of_depth(i)->method();
4044 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4045 }
4046 }
4047 #endif
4048 return true;
4049 }
4050 break;
4051 }
4052 }
4053
4054 #ifndef PRODUCT
4055 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4056 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4057 tty->print_cr(" JVM state at this point:");
4058 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4059 ciMethod* m = jvms()->of_depth(i)->method();
4060 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4061 }
4062 }
4063 #endif
4064
4065 return false; // bail-out; let JVM_GetCallerClass do the work
4066 }
4067
4068 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4069 Node* arg = argument(0);
4070 Node* result = NULL;
4071
4072 switch (id) {
4073 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4074 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4075 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4076 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4077
4078 case vmIntrinsics::_doubleToLongBits: {
4079 // two paths (plus control) merge in a wood
4080 RegionNode *r = new RegionNode(3);
4081 Node *phi = new PhiNode(r, TypeLong::LONG);
4082
4083 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4084 // Build the boolean node
4085 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4086
4087 // Branch either way.
4088 // NaN case is less traveled, which makes all the difference.
4089 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4090 Node *opt_isnan = _gvn.transform(ifisnan);
4091 assert( opt_isnan->is_If(), "Expect an IfNode");
4092 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4093 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4094
4095 set_control(iftrue);
4096
4097 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4098 Node *slow_result = longcon(nan_bits); // return NaN
4099 phi->init_req(1, _gvn.transform( slow_result ));
4100 r->init_req(1, iftrue);
4101
4102 // Else fall through
4103 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4104 set_control(iffalse);
4105
4106 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4107 r->init_req(2, iffalse);
4108
4109 // Post merge
4110 set_control(_gvn.transform(r));
4111 record_for_igvn(r);
4112
4113 C->set_has_split_ifs(true); // Has chance for split-if optimization
4114 result = phi;
4115 assert(result->bottom_type()->isa_long(), "must be");
4116 break;
4117 }
4118
4119 case vmIntrinsics::_floatToIntBits: {
4120 // two paths (plus control) merge in a wood
4121 RegionNode *r = new RegionNode(3);
4122 Node *phi = new PhiNode(r, TypeInt::INT);
4123
4124 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4125 // Build the boolean node
4126 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4127
4128 // Branch either way.
4129 // NaN case is less traveled, which makes all the difference.
4130 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4131 Node *opt_isnan = _gvn.transform(ifisnan);
4132 assert( opt_isnan->is_If(), "Expect an IfNode");
4133 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4134 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4135
4136 set_control(iftrue);
4137
4138 static const jint nan_bits = 0x7fc00000;
4139 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4140 phi->init_req(1, _gvn.transform( slow_result ));
4141 r->init_req(1, iftrue);
4142
4143 // Else fall through
4144 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4145 set_control(iffalse);
4146
4147 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4148 r->init_req(2, iffalse);
4149
4150 // Post merge
4151 set_control(_gvn.transform(r));
4152 record_for_igvn(r);
4153
4154 C->set_has_split_ifs(true); // Has chance for split-if optimization
4155 result = phi;
4156 assert(result->bottom_type()->isa_int(), "must be");
4157 break;
4158 }
4159
4160 default:
4161 fatal_unexpected_iid(id);
4162 break;
4163 }
4164 set_result(_gvn.transform(result));
4165 return true;
4166 }
4167
4168 //----------------------inline_unsafe_copyMemory-------------------------
4169 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4170 bool LibraryCallKit::inline_unsafe_copyMemory() {
4171 if (callee()->is_static()) return false; // caller must have the capability!
4172 null_check_receiver(); // null-check receiver
4173 if (stopped()) return true;
4174
4175 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4176
4177 Node* src_ptr = argument(1); // type: oop
4178 Node* src_off = ConvL2X(argument(2)); // type: long
4179 Node* dst_ptr = argument(4); // type: oop
4180 Node* dst_off = ConvL2X(argument(5)); // type: long
4181 Node* size = ConvL2X(argument(7)); // type: long
4182
4183 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4184 "fieldOffset must be byte-scaled");
4185
4186 Node* src = make_unsafe_address(src_ptr, src_off);
4187 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4188
4189 // Conservatively insert a memory barrier on all memory slices.
4190 // Do not let writes of the copy source or destination float below the copy.
4191 insert_mem_bar(Op_MemBarCPUOrder);
4192
4193 // Call it. Note that the length argument is not scaled.
4194 make_runtime_call(RC_LEAF|RC_NO_FP,
4195 OptoRuntime::fast_arraycopy_Type(),
4196 StubRoutines::unsafe_arraycopy(),
4197 "unsafe_arraycopy",
4198 TypeRawPtr::BOTTOM,
4199 src, dst, size XTOP);
4200
4201 // Do not let reads of the copy destination float above the copy.
4202 insert_mem_bar(Op_MemBarCPUOrder);
4203
4204 return true;
4205 }
4206
4207 //------------------------clone_coping-----------------------------------
4208 // Helper function for inline_native_clone.
4209 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4210 assert(obj_size != NULL, "");
4211 Node* raw_obj = alloc_obj->in(1);
4212 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4213
4214 AllocateNode* alloc = NULL;
4215 if (ReduceBulkZeroing) {
4216 // We will be completely responsible for initializing this object -
4217 // mark Initialize node as complete.
4218 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4219 // The object was just allocated - there should be no any stores!
4220 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4221 // Mark as complete_with_arraycopy so that on AllocateNode
4222 // expansion, we know this AllocateNode is initialized by an array
4223 // copy and a StoreStore barrier exists after the array copy.
4224 alloc->initialization()->set_complete_with_arraycopy();
4225 }
4226
4227 // Copy the fastest available way.
4228 // TODO: generate fields copies for small objects instead.
4229 Node* size = _gvn.transform(obj_size);
4230
4231 access_clone(control(), obj, alloc_obj, size, is_array);
4232
4233 // Do not let reads from the cloned object float above the arraycopy.
4234 if (alloc != NULL) {
4235 // Do not let stores that initialize this object be reordered with
4236 // a subsequent store that would make this object accessible by
4237 // other threads.
4238 // Record what AllocateNode this StoreStore protects so that
4239 // escape analysis can go from the MemBarStoreStoreNode to the
4240 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4241 // based on the escape status of the AllocateNode.
4242 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4243 } else {
4244 insert_mem_bar(Op_MemBarCPUOrder);
4245 }
4246 }
4247
4248 //------------------------inline_native_clone----------------------------
4249 // protected native Object java.lang.Object.clone();
4250 //
4251 // Here are the simple edge cases:
4252 // null receiver => normal trap
4253 // virtual and clone was overridden => slow path to out-of-line clone
4254 // not cloneable or finalizer => slow path to out-of-line Object.clone
4255 //
4256 // The general case has two steps, allocation and copying.
4257 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4258 //
4259 // Copying also has two cases, oop arrays and everything else.
4260 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4261 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4262 //
4263 // These steps fold up nicely if and when the cloned object's klass
4264 // can be sharply typed as an object array, a type array, or an instance.
4265 //
4266 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4267 PhiNode* result_val;
4268
4269 // Set the reexecute bit for the interpreter to reexecute
4270 // the bytecode that invokes Object.clone if deoptimization happens.
4271 { PreserveReexecuteState preexecs(this);
4272 jvms()->set_should_reexecute(true);
4273
4274 Node* obj = null_check_receiver();
4275 if (stopped()) return true;
4276
4277 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4278
4279 // If we are going to clone an instance, we need its exact type to
4280 // know the number and types of fields to convert the clone to
4281 // loads/stores. Maybe a speculative type can help us.
4282 if (!obj_type->klass_is_exact() &&
4283 obj_type->speculative_type() != NULL &&
4284 obj_type->speculative_type()->is_instance_klass()) {
4285 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4286 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4287 !spec_ik->has_injected_fields()) {
4288 ciKlass* k = obj_type->klass();
4289 if (!k->is_instance_klass() ||
4290 k->as_instance_klass()->is_interface() ||
4291 k->as_instance_klass()->has_subklass()) {
4292 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4293 }
4294 }
4295 }
4296
4297 Node* obj_klass = load_object_klass(obj);
4298 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4299 const TypeOopPtr* toop = ((tklass != NULL)
4300 ? tklass->as_instance_type()
4301 : TypeInstPtr::NOTNULL);
4302
4303 // Conservatively insert a memory barrier on all memory slices.
4304 // Do not let writes into the original float below the clone.
4305 insert_mem_bar(Op_MemBarCPUOrder);
4306
4307 // paths into result_reg:
4308 enum {
4309 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4310 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4311 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4312 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4313 PATH_LIMIT
4314 };
4315 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4316 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4317 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4318 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4319 record_for_igvn(result_reg);
4320
4321 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4322 if (array_ctl != NULL) {
4323 // It's an array.
4324 PreserveJVMState pjvms(this);
4325 set_control(array_ctl);
4326 Node* obj_length = load_array_length(obj);
4327 Node* obj_size = NULL;
4328 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4329
4330 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4331 if (bs->array_copy_requires_gc_barriers(T_OBJECT)) {
4332 // If it is an oop array, it requires very special treatment,
4333 // because gc barriers are required when accessing the array.
4334 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4335 if (is_obja != NULL) {
4336 PreserveJVMState pjvms2(this);
4337 set_control(is_obja);
4338 // Generate a direct call to the right arraycopy function(s).
4339 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4340 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4341 ac->set_cloneoop();
4342 Node* n = _gvn.transform(ac);
4343 assert(n == ac, "cannot disappear");
4344 ac->connect_outputs(this);
4345
4346 result_reg->init_req(_objArray_path, control());
4347 result_val->init_req(_objArray_path, alloc_obj);
4348 result_i_o ->set_req(_objArray_path, i_o());
4349 result_mem ->set_req(_objArray_path, reset_memory());
4350 }
4351 }
4352 // Otherwise, there are no barriers to worry about.
4353 // (We can dispense with card marks if we know the allocation
4354 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4355 // causes the non-eden paths to take compensating steps to
4356 // simulate a fresh allocation, so that no further
4357 // card marks are required in compiled code to initialize
4358 // the object.)
4359
4360 if (!stopped()) {
4361 copy_to_clone(obj, alloc_obj, obj_size, true);
4362
4363 // Present the results of the copy.
4364 result_reg->init_req(_array_path, control());
4365 result_val->init_req(_array_path, alloc_obj);
4366 result_i_o ->set_req(_array_path, i_o());
4367 result_mem ->set_req(_array_path, reset_memory());
4368 }
4369 }
4370
4371 // We only go to the instance fast case code if we pass a number of guards.
4372 // The paths which do not pass are accumulated in the slow_region.
4373 RegionNode* slow_region = new RegionNode(1);
4374 record_for_igvn(slow_region);
4375 if (!stopped()) {
4376 // It's an instance (we did array above). Make the slow-path tests.
4377 // If this is a virtual call, we generate a funny guard. We grab
4378 // the vtable entry corresponding to clone() from the target object.
4379 // If the target method which we are calling happens to be the
4380 // Object clone() method, we pass the guard. We do not need this
4381 // guard for non-virtual calls; the caller is known to be the native
4382 // Object clone().
4383 if (is_virtual) {
4384 generate_virtual_guard(obj_klass, slow_region);
4385 }
4386
4387 // The object must be easily cloneable and must not have a finalizer.
4388 // Both of these conditions may be checked in a single test.
4389 // We could optimize the test further, but we don't care.
4390 generate_access_flags_guard(obj_klass,
4391 // Test both conditions:
4392 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4393 // Must be cloneable but not finalizer:
4394 JVM_ACC_IS_CLONEABLE_FAST,
4395 slow_region);
4396 }
4397
4398 if (!stopped()) {
4399 // It's an instance, and it passed the slow-path tests.
4400 PreserveJVMState pjvms(this);
4401 Node* obj_size = NULL;
4402 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4403 // is reexecuted if deoptimization occurs and there could be problems when merging
4404 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4405 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4406
4407 copy_to_clone(obj, alloc_obj, obj_size, false);
4408
4409 // Present the results of the slow call.
4410 result_reg->init_req(_instance_path, control());
4411 result_val->init_req(_instance_path, alloc_obj);
4412 result_i_o ->set_req(_instance_path, i_o());
4413 result_mem ->set_req(_instance_path, reset_memory());
4414 }
4415
4416 // Generate code for the slow case. We make a call to clone().
4417 set_control(_gvn.transform(slow_region));
4418 if (!stopped()) {
4419 PreserveJVMState pjvms(this);
4420 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4421 Node* slow_result = set_results_for_java_call(slow_call);
4422 // this->control() comes from set_results_for_java_call
4423 result_reg->init_req(_slow_path, control());
4424 result_val->init_req(_slow_path, slow_result);
4425 result_i_o ->set_req(_slow_path, i_o());
4426 result_mem ->set_req(_slow_path, reset_memory());
4427 }
4428
4429 // Return the combined state.
4430 set_control( _gvn.transform(result_reg));
4431 set_i_o( _gvn.transform(result_i_o));
4432 set_all_memory( _gvn.transform(result_mem));
4433 } // original reexecute is set back here
4434
4435 set_result(_gvn.transform(result_val));
4436 return true;
4437 }
4438
4439 // If we have a tighly coupled allocation, the arraycopy may take care
4440 // of the array initialization. If one of the guards we insert between
4441 // the allocation and the arraycopy causes a deoptimization, an
4442 // unitialized array will escape the compiled method. To prevent that
4443 // we set the JVM state for uncommon traps between the allocation and
4444 // the arraycopy to the state before the allocation so, in case of
4445 // deoptimization, we'll reexecute the allocation and the
4446 // initialization.
4447 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4448 if (alloc != NULL) {
4449 ciMethod* trap_method = alloc->jvms()->method();
4450 int trap_bci = alloc->jvms()->bci();
4451
4452 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4453 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4454 // Make sure there's no store between the allocation and the
4455 // arraycopy otherwise visible side effects could be rexecuted
4456 // in case of deoptimization and cause incorrect execution.
4457 bool no_interfering_store = true;
4458 Node* mem = alloc->in(TypeFunc::Memory);
4459 if (mem->is_MergeMem()) {
4460 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4461 Node* n = mms.memory();
4462 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4463 assert(n->is_Store(), "what else?");
4464 no_interfering_store = false;
4465 break;
4466 }
4467 }
4468 } else {
4469 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4470 Node* n = mms.memory();
4471 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4472 assert(n->is_Store(), "what else?");
4473 no_interfering_store = false;
4474 break;
4475 }
4476 }
4477 }
4478
4479 if (no_interfering_store) {
4480 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4481 uint size = alloc->req();
4482 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4483 old_jvms->set_map(sfpt);
4484 for (uint i = 0; i < size; i++) {
4485 sfpt->init_req(i, alloc->in(i));
4486 }
4487 // re-push array length for deoptimization
4488 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4489 old_jvms->set_sp(old_jvms->sp()+1);
4490 old_jvms->set_monoff(old_jvms->monoff()+1);
4491 old_jvms->set_scloff(old_jvms->scloff()+1);
4492 old_jvms->set_endoff(old_jvms->endoff()+1);
4493 old_jvms->set_should_reexecute(true);
4494
4495 sfpt->set_i_o(map()->i_o());
4496 sfpt->set_memory(map()->memory());
4497 sfpt->set_control(map()->control());
4498
4499 JVMState* saved_jvms = jvms();
4500 saved_reexecute_sp = _reexecute_sp;
4501
4502 set_jvms(sfpt->jvms());
4503 _reexecute_sp = jvms()->sp();
4504
4505 return saved_jvms;
4506 }
4507 }
4508 }
4509 return NULL;
4510 }
4511
4512 // In case of a deoptimization, we restart execution at the
4513 // allocation, allocating a new array. We would leave an uninitialized
4514 // array in the heap that GCs wouldn't expect. Move the allocation
4515 // after the traps so we don't allocate the array if we
4516 // deoptimize. This is possible because tightly_coupled_allocation()
4517 // guarantees there's no observer of the allocated array at this point
4518 // and the control flow is simple enough.
4519 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4520 int saved_reexecute_sp, uint new_idx) {
4521 if (saved_jvms != NULL && !stopped()) {
4522 assert(alloc != NULL, "only with a tightly coupled allocation");
4523 // restore JVM state to the state at the arraycopy
4524 saved_jvms->map()->set_control(map()->control());
4525 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4526 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4527 // If we've improved the types of some nodes (null check) while
4528 // emitting the guards, propagate them to the current state
4529 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4530 set_jvms(saved_jvms);
4531 _reexecute_sp = saved_reexecute_sp;
4532
4533 // Remove the allocation from above the guards
4534 CallProjections callprojs;
4535 alloc->extract_projections(&callprojs, true);
4536 InitializeNode* init = alloc->initialization();
4537 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4538 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4539 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4540 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4541
4542 // move the allocation here (after the guards)
4543 _gvn.hash_delete(alloc);
4544 alloc->set_req(TypeFunc::Control, control());
4545 alloc->set_req(TypeFunc::I_O, i_o());
4546 Node *mem = reset_memory();
4547 set_all_memory(mem);
4548 alloc->set_req(TypeFunc::Memory, mem);
4549 set_control(init->proj_out_or_null(TypeFunc::Control));
4550 set_i_o(callprojs.fallthrough_ioproj);
4551
4552 // Update memory as done in GraphKit::set_output_for_allocation()
4553 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4554 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4555 if (ary_type->isa_aryptr() && length_type != NULL) {
4556 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4557 }
4558 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4559 int elemidx = C->get_alias_index(telemref);
4560 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4561 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4562
4563 Node* allocx = _gvn.transform(alloc);
4564 assert(allocx == alloc, "where has the allocation gone?");
4565 assert(dest->is_CheckCastPP(), "not an allocation result?");
4566
4567 _gvn.hash_delete(dest);
4568 dest->set_req(0, control());
4569 Node* destx = _gvn.transform(dest);
4570 assert(destx == dest, "where has the allocation result gone?");
4571 }
4572 }
4573
4574
4575 //------------------------------inline_arraycopy-----------------------
4576 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4577 // Object dest, int destPos,
4578 // int length);
4579 bool LibraryCallKit::inline_arraycopy() {
4580 // Get the arguments.
4581 Node* src = argument(0); // type: oop
4582 Node* src_offset = argument(1); // type: int
4583 Node* dest = argument(2); // type: oop
4584 Node* dest_offset = argument(3); // type: int
4585 Node* length = argument(4); // type: int
4586
4587 uint new_idx = C->unique();
4588
4589 // Check for allocation before we add nodes that would confuse
4590 // tightly_coupled_allocation()
4591 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4592
4593 int saved_reexecute_sp = -1;
4594 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4595 // See arraycopy_restore_alloc_state() comment
4596 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4597 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4598 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4599 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4600
4601 // The following tests must be performed
4602 // (1) src and dest are arrays.
4603 // (2) src and dest arrays must have elements of the same BasicType
4604 // (3) src and dest must not be null.
4605 // (4) src_offset must not be negative.
4606 // (5) dest_offset must not be negative.
4607 // (6) length must not be negative.
4608 // (7) src_offset + length must not exceed length of src.
4609 // (8) dest_offset + length must not exceed length of dest.
4610 // (9) each element of an oop array must be assignable
4611
4612 // (3) src and dest must not be null.
4613 // always do this here because we need the JVM state for uncommon traps
4614 Node* null_ctl = top();
4615 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4616 assert(null_ctl->is_top(), "no null control here");
4617 dest = null_check(dest, T_ARRAY);
4618
4619 if (!can_emit_guards) {
4620 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4621 // guards but the arraycopy node could still take advantage of a
4622 // tightly allocated allocation. tightly_coupled_allocation() is
4623 // called again to make sure it takes the null check above into
4624 // account: the null check is mandatory and if it caused an
4625 // uncommon trap to be emitted then the allocation can't be
4626 // considered tightly coupled in this context.
4627 alloc = tightly_coupled_allocation(dest, NULL);
4628 }
4629
4630 bool validated = false;
4631
4632 const Type* src_type = _gvn.type(src);
4633 const Type* dest_type = _gvn.type(dest);
4634 const TypeAryPtr* top_src = src_type->isa_aryptr();
4635 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4636
4637 // Do we have the type of src?
4638 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4639 // Do we have the type of dest?
4640 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4641 // Is the type for src from speculation?
4642 bool src_spec = false;
4643 // Is the type for dest from speculation?
4644 bool dest_spec = false;
4645
4646 if ((!has_src || !has_dest) && can_emit_guards) {
4647 // We don't have sufficient type information, let's see if
4648 // speculative types can help. We need to have types for both src
4649 // and dest so that it pays off.
4650
4651 // Do we already have or could we have type information for src
4652 bool could_have_src = has_src;
4653 // Do we already have or could we have type information for dest
4654 bool could_have_dest = has_dest;
4655
4656 ciKlass* src_k = NULL;
4657 if (!has_src) {
4658 src_k = src_type->speculative_type_not_null();
4659 if (src_k != NULL && src_k->is_array_klass()) {
4660 could_have_src = true;
4661 }
4662 }
4663
4664 ciKlass* dest_k = NULL;
4665 if (!has_dest) {
4666 dest_k = dest_type->speculative_type_not_null();
4667 if (dest_k != NULL && dest_k->is_array_klass()) {
4668 could_have_dest = true;
4669 }
4670 }
4671
4672 if (could_have_src && could_have_dest) {
4673 // This is going to pay off so emit the required guards
4674 if (!has_src) {
4675 src = maybe_cast_profiled_obj(src, src_k, true);
4676 src_type = _gvn.type(src);
4677 top_src = src_type->isa_aryptr();
4678 has_src = (top_src != NULL && top_src->klass() != NULL);
4679 src_spec = true;
4680 }
4681 if (!has_dest) {
4682 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4683 dest_type = _gvn.type(dest);
4684 top_dest = dest_type->isa_aryptr();
4685 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4686 dest_spec = true;
4687 }
4688 }
4689 }
4690
4691 if (has_src && has_dest && can_emit_guards) {
4692 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4693 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4694 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
4695 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
4696
4697 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4698 // If both arrays are object arrays then having the exact types
4699 // for both will remove the need for a subtype check at runtime
4700 // before the call and may make it possible to pick a faster copy
4701 // routine (without a subtype check on every element)
4702 // Do we have the exact type of src?
4703 bool could_have_src = src_spec;
4704 // Do we have the exact type of dest?
4705 bool could_have_dest = dest_spec;
4706 ciKlass* src_k = top_src->klass();
4707 ciKlass* dest_k = top_dest->klass();
4708 if (!src_spec) {
4709 src_k = src_type->speculative_type_not_null();
4710 if (src_k != NULL && src_k->is_array_klass()) {
4711 could_have_src = true;
4712 }
4713 }
4714 if (!dest_spec) {
4715 dest_k = dest_type->speculative_type_not_null();
4716 if (dest_k != NULL && dest_k->is_array_klass()) {
4717 could_have_dest = true;
4718 }
4719 }
4720 if (could_have_src && could_have_dest) {
4721 // If we can have both exact types, emit the missing guards
4722 if (could_have_src && !src_spec) {
4723 src = maybe_cast_profiled_obj(src, src_k, true);
4724 }
4725 if (could_have_dest && !dest_spec) {
4726 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4727 }
4728 }
4729 }
4730 }
4731
4732 ciMethod* trap_method = method();
4733 int trap_bci = bci();
4734 if (saved_jvms != NULL) {
4735 trap_method = alloc->jvms()->method();
4736 trap_bci = alloc->jvms()->bci();
4737 }
4738
4739 bool negative_length_guard_generated = false;
4740
4741 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4742 can_emit_guards &&
4743 !src->is_top() && !dest->is_top()) {
4744 // validate arguments: enables transformation the ArrayCopyNode
4745 validated = true;
4746
4747 RegionNode* slow_region = new RegionNode(1);
4748 record_for_igvn(slow_region);
4749
4750 // (1) src and dest are arrays.
4751 generate_non_array_guard(load_object_klass(src), slow_region);
4752 generate_non_array_guard(load_object_klass(dest), slow_region);
4753
4754 // (2) src and dest arrays must have elements of the same BasicType
4755 // done at macro expansion or at Ideal transformation time
4756
4757 // (4) src_offset must not be negative.
4758 generate_negative_guard(src_offset, slow_region);
4759
4760 // (5) dest_offset must not be negative.
4761 generate_negative_guard(dest_offset, slow_region);
4762
4763 // (7) src_offset + length must not exceed length of src.
4764 generate_limit_guard(src_offset, length,
4765 load_array_length(src),
4766 slow_region);
4767
4768 // (8) dest_offset + length must not exceed length of dest.
4769 generate_limit_guard(dest_offset, length,
4770 load_array_length(dest),
4771 slow_region);
4772
4773 // (6) length must not be negative.
4774 // This is also checked in generate_arraycopy() during macro expansion, but
4775 // we also have to check it here for the case where the ArrayCopyNode will
4776 // be eliminated by Escape Analysis.
4777 if (EliminateAllocations) {
4778 generate_negative_guard(length, slow_region);
4779 negative_length_guard_generated = true;
4780 }
4781
4782 // (9) each element of an oop array must be assignable
4783 Node* src_klass = load_object_klass(src);
4784 Node* dest_klass = load_object_klass(dest);
4785 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4786
4787 if (not_subtype_ctrl != top()) {
4788 PreserveJVMState pjvms(this);
4789 set_control(not_subtype_ctrl);
4790 uncommon_trap(Deoptimization::Reason_intrinsic,
4791 Deoptimization::Action_make_not_entrant);
4792 assert(stopped(), "Should be stopped");
4793 }
4794 {
4795 PreserveJVMState pjvms(this);
4796 set_control(_gvn.transform(slow_region));
4797 uncommon_trap(Deoptimization::Reason_intrinsic,
4798 Deoptimization::Action_make_not_entrant);
4799 assert(stopped(), "Should be stopped");
4800 }
4801
4802 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4803 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4804 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4805 }
4806
4807 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4808
4809 if (stopped()) {
4810 return true;
4811 }
4812
4813 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4814 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4815 // so the compiler has a chance to eliminate them: during macro expansion,
4816 // we have to set their control (CastPP nodes are eliminated).
4817 load_object_klass(src), load_object_klass(dest),
4818 load_array_length(src), load_array_length(dest));
4819
4820 ac->set_arraycopy(validated);
4821
4822 Node* n = _gvn.transform(ac);
4823 if (n == ac) {
4824 ac->connect_outputs(this);
4825 } else {
4826 assert(validated, "shouldn't transform if all arguments not validated");
4827 set_all_memory(n);
4828 }
4829 clear_upper_avx();
4830
4831
4832 return true;
4833 }
4834
4835
4836 // Helper function which determines if an arraycopy immediately follows
4837 // an allocation, with no intervening tests or other escapes for the object.
4838 AllocateArrayNode*
4839 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4840 RegionNode* slow_region) {
4841 if (stopped()) return NULL; // no fast path
4842 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4843
4844 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4845 if (alloc == NULL) return NULL;
4846
4847 Node* rawmem = memory(Compile::AliasIdxRaw);
4848 // Is the allocation's memory state untouched?
4849 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4850 // Bail out if there have been raw-memory effects since the allocation.
4851 // (Example: There might have been a call or safepoint.)
4852 return NULL;
4853 }
4854 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4855 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4856 return NULL;
4857 }
4858
4859 // There must be no unexpected observers of this allocation.
4860 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4861 Node* obs = ptr->fast_out(i);
4862 if (obs != this->map()) {
4863 return NULL;
4864 }
4865 }
4866
4867 // This arraycopy must unconditionally follow the allocation of the ptr.
4868 Node* alloc_ctl = ptr->in(0);
4869 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4870
4871 Node* ctl = control();
4872 while (ctl != alloc_ctl) {
4873 // There may be guards which feed into the slow_region.
4874 // Any other control flow means that we might not get a chance
4875 // to finish initializing the allocated object.
4876 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4877 IfNode* iff = ctl->in(0)->as_If();
4878 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4879 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4880 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4881 ctl = iff->in(0); // This test feeds the known slow_region.
4882 continue;
4883 }
4884 // One more try: Various low-level checks bottom out in
4885 // uncommon traps. If the debug-info of the trap omits
4886 // any reference to the allocation, as we've already
4887 // observed, then there can be no objection to the trap.
4888 bool found_trap = false;
4889 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4890 Node* obs = not_ctl->fast_out(j);
4891 if (obs->in(0) == not_ctl && obs->is_Call() &&
4892 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4893 found_trap = true; break;
4894 }
4895 }
4896 if (found_trap) {
4897 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4898 continue;
4899 }
4900 }
4901 return NULL;
4902 }
4903
4904 // If we get this far, we have an allocation which immediately
4905 // precedes the arraycopy, and we can take over zeroing the new object.
4906 // The arraycopy will finish the initialization, and provide
4907 // a new control state to which we will anchor the destination pointer.
4908
4909 return alloc;
4910 }
4911
4912 //-------------inline_encodeISOArray-----------------------------------
4913 // encode char[] to byte[] in ISO_8859_1
4914 bool LibraryCallKit::inline_encodeISOArray() {
4915 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4916 // no receiver since it is static method
4917 Node *src = argument(0);
4918 Node *src_offset = argument(1);
4919 Node *dst = argument(2);
4920 Node *dst_offset = argument(3);
4921 Node *length = argument(4);
4922
4923 const Type* src_type = src->Value(&_gvn);
4924 const Type* dst_type = dst->Value(&_gvn);
4925 const TypeAryPtr* top_src = src_type->isa_aryptr();
4926 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4927 if (top_src == NULL || top_src->klass() == NULL ||
4928 top_dest == NULL || top_dest->klass() == NULL) {
4929 // failed array check
4930 return false;
4931 }
4932
4933 // Figure out the size and type of the elements we will be copying.
4934 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4935 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4936 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4937 return false;
4938 }
4939
4940 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4941 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4942 // 'src_start' points to src array + scaled offset
4943 // 'dst_start' points to dst array + scaled offset
4944
4945 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4946 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4947 enc = _gvn.transform(enc);
4948 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4949 set_memory(res_mem, mtype);
4950 set_result(enc);
4951 clear_upper_avx();
4952
4953 return true;
4954 }
4955
4956 //-------------inline_multiplyToLen-----------------------------------
4957 bool LibraryCallKit::inline_multiplyToLen() {
4958 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4959
4960 address stubAddr = StubRoutines::multiplyToLen();
4961 if (stubAddr == NULL) {
4962 return false; // Intrinsic's stub is not implemented on this platform
4963 }
4964 const char* stubName = "multiplyToLen";
4965
4966 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4967
4968 // no receiver because it is a static method
4969 Node* x = argument(0);
4970 Node* xlen = argument(1);
4971 Node* y = argument(2);
4972 Node* ylen = argument(3);
4973 Node* z = argument(4);
4974
4975 const Type* x_type = x->Value(&_gvn);
4976 const Type* y_type = y->Value(&_gvn);
4977 const TypeAryPtr* top_x = x_type->isa_aryptr();
4978 const TypeAryPtr* top_y = y_type->isa_aryptr();
4979 if (top_x == NULL || top_x->klass() == NULL ||
4980 top_y == NULL || top_y->klass() == NULL) {
4981 // failed array check
4982 return false;
4983 }
4984
4985 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4986 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4987 if (x_elem != T_INT || y_elem != T_INT) {
4988 return false;
4989 }
4990
4991 // Set the original stack and the reexecute bit for the interpreter to reexecute
4992 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
4993 // on the return from z array allocation in runtime.
4994 { PreserveReexecuteState preexecs(this);
4995 jvms()->set_should_reexecute(true);
4996
4997 Node* x_start = array_element_address(x, intcon(0), x_elem);
4998 Node* y_start = array_element_address(y, intcon(0), y_elem);
4999 // 'x_start' points to x array + scaled xlen
5000 // 'y_start' points to y array + scaled ylen
5001
5002 // Allocate the result array
5003 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5004 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5005 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5006
5007 IdealKit ideal(this);
5008
5009 #define __ ideal.
5010 Node* one = __ ConI(1);
5011 Node* zero = __ ConI(0);
5012 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5013 __ set(need_alloc, zero);
5014 __ set(z_alloc, z);
5015 __ if_then(z, BoolTest::eq, null()); {
5016 __ increment (need_alloc, one);
5017 } __ else_(); {
5018 // Update graphKit memory and control from IdealKit.
5019 sync_kit(ideal);
5020 Node* zlen_arg = load_array_length(z);
5021 // Update IdealKit memory and control from graphKit.
5022 __ sync_kit(this);
5023 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5024 __ increment (need_alloc, one);
5025 } __ end_if();
5026 } __ end_if();
5027
5028 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5029 // Update graphKit memory and control from IdealKit.
5030 sync_kit(ideal);
5031 Node * narr = new_array(klass_node, zlen, 1);
5032 // Update IdealKit memory and control from graphKit.
5033 __ sync_kit(this);
5034 __ set(z_alloc, narr);
5035 } __ end_if();
5036
5037 sync_kit(ideal);
5038 z = __ value(z_alloc);
5039 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5040 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5041 // Final sync IdealKit and GraphKit.
5042 final_sync(ideal);
5043 #undef __
5044
5045 Node* z_start = array_element_address(z, intcon(0), T_INT);
5046
5047 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5048 OptoRuntime::multiplyToLen_Type(),
5049 stubAddr, stubName, TypePtr::BOTTOM,
5050 x_start, xlen, y_start, ylen, z_start, zlen);
5051 } // original reexecute is set back here
5052
5053 C->set_has_split_ifs(true); // Has chance for split-if optimization
5054 set_result(z);
5055 return true;
5056 }
5057
5058 //-------------inline_squareToLen------------------------------------
5059 bool LibraryCallKit::inline_squareToLen() {
5060 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5061
5062 address stubAddr = StubRoutines::squareToLen();
5063 if (stubAddr == NULL) {
5064 return false; // Intrinsic's stub is not implemented on this platform
5065 }
5066 const char* stubName = "squareToLen";
5067
5068 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5069
5070 Node* x = argument(0);
5071 Node* len = argument(1);
5072 Node* z = argument(2);
5073 Node* zlen = argument(3);
5074
5075 const Type* x_type = x->Value(&_gvn);
5076 const Type* z_type = z->Value(&_gvn);
5077 const TypeAryPtr* top_x = x_type->isa_aryptr();
5078 const TypeAryPtr* top_z = z_type->isa_aryptr();
5079 if (top_x == NULL || top_x->klass() == NULL ||
5080 top_z == NULL || top_z->klass() == NULL) {
5081 // failed array check
5082 return false;
5083 }
5084
5085 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5086 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5087 if (x_elem != T_INT || z_elem != T_INT) {
5088 return false;
5089 }
5090
5091
5092 Node* x_start = array_element_address(x, intcon(0), x_elem);
5093 Node* z_start = array_element_address(z, intcon(0), z_elem);
5094
5095 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5096 OptoRuntime::squareToLen_Type(),
5097 stubAddr, stubName, TypePtr::BOTTOM,
5098 x_start, len, z_start, zlen);
5099
5100 set_result(z);
5101 return true;
5102 }
5103
5104 //-------------inline_mulAdd------------------------------------------
5105 bool LibraryCallKit::inline_mulAdd() {
5106 assert(UseMulAddIntrinsic, "not implemented on this platform");
5107
5108 address stubAddr = StubRoutines::mulAdd();
5109 if (stubAddr == NULL) {
5110 return false; // Intrinsic's stub is not implemented on this platform
5111 }
5112 const char* stubName = "mulAdd";
5113
5114 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5115
5116 Node* out = argument(0);
5117 Node* in = argument(1);
5118 Node* offset = argument(2);
5119 Node* len = argument(3);
5120 Node* k = argument(4);
5121
5122 const Type* out_type = out->Value(&_gvn);
5123 const Type* in_type = in->Value(&_gvn);
5124 const TypeAryPtr* top_out = out_type->isa_aryptr();
5125 const TypeAryPtr* top_in = in_type->isa_aryptr();
5126 if (top_out == NULL || top_out->klass() == NULL ||
5127 top_in == NULL || top_in->klass() == NULL) {
5128 // failed array check
5129 return false;
5130 }
5131
5132 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5133 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5134 if (out_elem != T_INT || in_elem != T_INT) {
5135 return false;
5136 }
5137
5138 Node* outlen = load_array_length(out);
5139 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5140 Node* out_start = array_element_address(out, intcon(0), out_elem);
5141 Node* in_start = array_element_address(in, intcon(0), in_elem);
5142
5143 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5144 OptoRuntime::mulAdd_Type(),
5145 stubAddr, stubName, TypePtr::BOTTOM,
5146 out_start,in_start, new_offset, len, k);
5147 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5148 set_result(result);
5149 return true;
5150 }
5151
5152 //-------------inline_montgomeryMultiply-----------------------------------
5153 bool LibraryCallKit::inline_montgomeryMultiply() {
5154 address stubAddr = StubRoutines::montgomeryMultiply();
5155 if (stubAddr == NULL) {
5156 return false; // Intrinsic's stub is not implemented on this platform
5157 }
5158
5159 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5160 const char* stubName = "montgomery_multiply";
5161
5162 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5163
5164 Node* a = argument(0);
5165 Node* b = argument(1);
5166 Node* n = argument(2);
5167 Node* len = argument(3);
5168 Node* inv = argument(4);
5169 Node* m = argument(6);
5170
5171 const Type* a_type = a->Value(&_gvn);
5172 const TypeAryPtr* top_a = a_type->isa_aryptr();
5173 const Type* b_type = b->Value(&_gvn);
5174 const TypeAryPtr* top_b = b_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_b == NULL || top_b->klass() == NULL ||
5181 top_n == NULL || top_n->klass() == NULL ||
5182 top_m == NULL || top_m->klass() == NULL) {
5183 // failed array check
5184 return false;
5185 }
5186
5187 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5188 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5189 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5190 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5191 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5192 return false;
5193 }
5194
5195 // Make the call
5196 {
5197 Node* a_start = array_element_address(a, intcon(0), a_elem);
5198 Node* b_start = array_element_address(b, intcon(0), b_elem);
5199 Node* n_start = array_element_address(n, intcon(0), n_elem);
5200 Node* m_start = array_element_address(m, intcon(0), m_elem);
5201
5202 Node* call = make_runtime_call(RC_LEAF,
5203 OptoRuntime::montgomeryMultiply_Type(),
5204 stubAddr, stubName, TypePtr::BOTTOM,
5205 a_start, b_start, n_start, len, inv, top(),
5206 m_start);
5207 set_result(m);
5208 }
5209
5210 return true;
5211 }
5212
5213 bool LibraryCallKit::inline_montgomerySquare() {
5214 address stubAddr = StubRoutines::montgomerySquare();
5215 if (stubAddr == NULL) {
5216 return false; // Intrinsic's stub is not implemented on this platform
5217 }
5218
5219 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5220 const char* stubName = "montgomery_square";
5221
5222 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5223
5224 Node* a = argument(0);
5225 Node* n = argument(1);
5226 Node* len = argument(2);
5227 Node* inv = argument(3);
5228 Node* m = argument(5);
5229
5230 const Type* a_type = a->Value(&_gvn);
5231 const TypeAryPtr* top_a = a_type->isa_aryptr();
5232 const Type* n_type = a->Value(&_gvn);
5233 const TypeAryPtr* top_n = n_type->isa_aryptr();
5234 const Type* m_type = a->Value(&_gvn);
5235 const TypeAryPtr* top_m = m_type->isa_aryptr();
5236 if (top_a == NULL || top_a->klass() == NULL ||
5237 top_n == NULL || top_n->klass() == NULL ||
5238 top_m == NULL || top_m->klass() == NULL) {
5239 // failed array check
5240 return false;
5241 }
5242
5243 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5244 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5245 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5246 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5247 return false;
5248 }
5249
5250 // Make the call
5251 {
5252 Node* a_start = array_element_address(a, intcon(0), a_elem);
5253 Node* n_start = array_element_address(n, intcon(0), n_elem);
5254 Node* m_start = array_element_address(m, intcon(0), m_elem);
5255
5256 Node* call = make_runtime_call(RC_LEAF,
5257 OptoRuntime::montgomerySquare_Type(),
5258 stubAddr, stubName, TypePtr::BOTTOM,
5259 a_start, n_start, len, inv, top(),
5260 m_start);
5261 set_result(m);
5262 }
5263
5264 return true;
5265 }
5266
5267 //-------------inline_vectorizedMismatch------------------------------
5268 bool LibraryCallKit::inline_vectorizedMismatch() {
5269 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5270
5271 address stubAddr = StubRoutines::vectorizedMismatch();
5272 if (stubAddr == NULL) {
5273 return false; // Intrinsic's stub is not implemented on this platform
5274 }
5275 const char* stubName = "vectorizedMismatch";
5276 int size_l = callee()->signature()->size();
5277 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5278
5279 Node* obja = argument(0);
5280 Node* aoffset = argument(1);
5281 Node* objb = argument(3);
5282 Node* boffset = argument(4);
5283 Node* length = argument(6);
5284 Node* scale = argument(7);
5285
5286 const Type* a_type = obja->Value(&_gvn);
5287 const Type* b_type = objb->Value(&_gvn);
5288 const TypeAryPtr* top_a = a_type->isa_aryptr();
5289 const TypeAryPtr* top_b = b_type->isa_aryptr();
5290 if (top_a == NULL || top_a->klass() == NULL ||
5291 top_b == NULL || top_b->klass() == NULL) {
5292 // failed array check
5293 return false;
5294 }
5295
5296 Node* call;
5297 jvms()->set_should_reexecute(true);
5298
5299 Node* obja_adr = make_unsafe_address(obja, aoffset);
5300 Node* objb_adr = make_unsafe_address(objb, boffset);
5301
5302 call = make_runtime_call(RC_LEAF,
5303 OptoRuntime::vectorizedMismatch_Type(),
5304 stubAddr, stubName, TypePtr::BOTTOM,
5305 obja_adr, objb_adr, length, scale);
5306
5307 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5308 set_result(result);
5309 return true;
5310 }
5311
5312 /**
5313 * Calculate CRC32 for byte.
5314 * int java.util.zip.CRC32.update(int crc, int b)
5315 */
5316 bool LibraryCallKit::inline_updateCRC32() {
5317 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5318 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5319 // no receiver since it is static method
5320 Node* crc = argument(0); // type: int
5321 Node* b = argument(1); // type: int
5322
5323 /*
5324 * int c = ~ crc;
5325 * b = timesXtoThe32[(b ^ c) & 0xFF];
5326 * b = b ^ (c >>> 8);
5327 * crc = ~b;
5328 */
5329
5330 Node* M1 = intcon(-1);
5331 crc = _gvn.transform(new XorINode(crc, M1));
5332 Node* result = _gvn.transform(new XorINode(crc, b));
5333 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5334
5335 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5336 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5337 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5338 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5339
5340 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5341 result = _gvn.transform(new XorINode(crc, result));
5342 result = _gvn.transform(new XorINode(result, M1));
5343 set_result(result);
5344 return true;
5345 }
5346
5347 /**
5348 * Calculate CRC32 for byte[] array.
5349 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5350 */
5351 bool LibraryCallKit::inline_updateBytesCRC32() {
5352 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5353 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5354 // no receiver since it is static method
5355 Node* crc = argument(0); // type: int
5356 Node* src = argument(1); // type: oop
5357 Node* offset = argument(2); // type: int
5358 Node* length = argument(3); // type: int
5359
5360 const Type* src_type = src->Value(&_gvn);
5361 const TypeAryPtr* top_src = src_type->isa_aryptr();
5362 if (top_src == NULL || top_src->klass() == NULL) {
5363 // failed array check
5364 return false;
5365 }
5366
5367 // Figure out the size and type of the elements we will be copying.
5368 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5369 if (src_elem != T_BYTE) {
5370 return false;
5371 }
5372
5373 // 'src_start' points to src array + scaled offset
5374 Node* src_start = array_element_address(src, offset, src_elem);
5375
5376 // We assume that range check is done by caller.
5377 // TODO: generate range check (offset+length < src.length) in debug VM.
5378
5379 // Call the stub.
5380 address stubAddr = StubRoutines::updateBytesCRC32();
5381 const char *stubName = "updateBytesCRC32";
5382
5383 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5384 stubAddr, stubName, TypePtr::BOTTOM,
5385 crc, src_start, length);
5386 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5387 set_result(result);
5388 return true;
5389 }
5390
5391 /**
5392 * Calculate CRC32 for ByteBuffer.
5393 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5394 */
5395 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5396 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5397 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5398 // no receiver since it is static method
5399 Node* crc = argument(0); // type: int
5400 Node* src = argument(1); // type: long
5401 Node* offset = argument(3); // type: int
5402 Node* length = argument(4); // type: int
5403
5404 src = ConvL2X(src); // adjust Java long to machine word
5405 Node* base = _gvn.transform(new CastX2PNode(src));
5406 offset = ConvI2X(offset);
5407
5408 // 'src_start' points to src array + scaled offset
5409 Node* src_start = basic_plus_adr(top(), base, offset);
5410
5411 // Call the stub.
5412 address stubAddr = StubRoutines::updateBytesCRC32();
5413 const char *stubName = "updateBytesCRC32";
5414
5415 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5416 stubAddr, stubName, TypePtr::BOTTOM,
5417 crc, src_start, length);
5418 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5419 set_result(result);
5420 return true;
5421 }
5422
5423 //------------------------------get_table_from_crc32c_class-----------------------
5424 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5425 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5426 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5427
5428 return table;
5429 }
5430
5431 //------------------------------inline_updateBytesCRC32C-----------------------
5432 //
5433 // Calculate CRC32C for byte[] array.
5434 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5435 //
5436 bool LibraryCallKit::inline_updateBytesCRC32C() {
5437 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5438 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5439 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5440 // no receiver since it is a static method
5441 Node* crc = argument(0); // type: int
5442 Node* src = argument(1); // type: oop
5443 Node* offset = argument(2); // type: int
5444 Node* end = argument(3); // type: int
5445
5446 Node* length = _gvn.transform(new SubINode(end, offset));
5447
5448 const Type* src_type = src->Value(&_gvn);
5449 const TypeAryPtr* top_src = src_type->isa_aryptr();
5450 if (top_src == NULL || top_src->klass() == NULL) {
5451 // failed array check
5452 return false;
5453 }
5454
5455 // Figure out the size and type of the elements we will be copying.
5456 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5457 if (src_elem != T_BYTE) {
5458 return false;
5459 }
5460
5461 // 'src_start' points to src array + scaled offset
5462 Node* src_start = array_element_address(src, offset, src_elem);
5463
5464 // static final int[] byteTable in class CRC32C
5465 Node* table = get_table_from_crc32c_class(callee()->holder());
5466 Node* table_start = array_element_address(table, intcon(0), T_INT);
5467
5468 // We assume that range check is done by caller.
5469 // TODO: generate range check (offset+length < src.length) in debug VM.
5470
5471 // Call the stub.
5472 address stubAddr = StubRoutines::updateBytesCRC32C();
5473 const char *stubName = "updateBytesCRC32C";
5474
5475 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5476 stubAddr, stubName, TypePtr::BOTTOM,
5477 crc, src_start, length, table_start);
5478 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5479 set_result(result);
5480 return true;
5481 }
5482
5483 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5484 //
5485 // Calculate CRC32C for DirectByteBuffer.
5486 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5487 //
5488 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5489 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5490 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5491 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5492 // no receiver since it is a static method
5493 Node* crc = argument(0); // type: int
5494 Node* src = argument(1); // type: long
5495 Node* offset = argument(3); // type: int
5496 Node* end = argument(4); // type: int
5497
5498 Node* length = _gvn.transform(new SubINode(end, offset));
5499
5500 src = ConvL2X(src); // adjust Java long to machine word
5501 Node* base = _gvn.transform(new CastX2PNode(src));
5502 offset = ConvI2X(offset);
5503
5504 // 'src_start' points to src array + scaled offset
5505 Node* src_start = basic_plus_adr(top(), base, offset);
5506
5507 // static final int[] byteTable in class CRC32C
5508 Node* table = get_table_from_crc32c_class(callee()->holder());
5509 Node* table_start = array_element_address(table, intcon(0), T_INT);
5510
5511 // Call the stub.
5512 address stubAddr = StubRoutines::updateBytesCRC32C();
5513 const char *stubName = "updateBytesCRC32C";
5514
5515 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5516 stubAddr, stubName, TypePtr::BOTTOM,
5517 crc, src_start, length, table_start);
5518 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5519 set_result(result);
5520 return true;
5521 }
5522
5523 //------------------------------inline_updateBytesAdler32----------------------
5524 //
5525 // Calculate Adler32 checksum for byte[] array.
5526 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5527 //
5528 bool LibraryCallKit::inline_updateBytesAdler32() {
5529 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5530 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5531 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5532 // no receiver since it is static method
5533 Node* crc = argument(0); // type: int
5534 Node* src = argument(1); // type: oop
5535 Node* offset = argument(2); // type: int
5536 Node* length = argument(3); // type: int
5537
5538 const Type* src_type = src->Value(&_gvn);
5539 const TypeAryPtr* top_src = src_type->isa_aryptr();
5540 if (top_src == NULL || top_src->klass() == NULL) {
5541 // failed array check
5542 return false;
5543 }
5544
5545 // Figure out the size and type of the elements we will be copying.
5546 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5547 if (src_elem != T_BYTE) {
5548 return false;
5549 }
5550
5551 // 'src_start' points to src array + scaled offset
5552 Node* src_start = array_element_address(src, offset, src_elem);
5553
5554 // We assume that range check is done by caller.
5555 // TODO: generate range check (offset+length < src.length) in debug VM.
5556
5557 // Call the stub.
5558 address stubAddr = StubRoutines::updateBytesAdler32();
5559 const char *stubName = "updateBytesAdler32";
5560
5561 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5562 stubAddr, stubName, TypePtr::BOTTOM,
5563 crc, src_start, length);
5564 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5565 set_result(result);
5566 return true;
5567 }
5568
5569 //------------------------------inline_updateByteBufferAdler32---------------
5570 //
5571 // Calculate Adler32 checksum for DirectByteBuffer.
5572 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5573 //
5574 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5575 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5576 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5577 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5578 // no receiver since it is static method
5579 Node* crc = argument(0); // type: int
5580 Node* src = argument(1); // type: long
5581 Node* offset = argument(3); // type: int
5582 Node* length = argument(4); // type: int
5583
5584 src = ConvL2X(src); // adjust Java long to machine word
5585 Node* base = _gvn.transform(new CastX2PNode(src));
5586 offset = ConvI2X(offset);
5587
5588 // 'src_start' points to src array + scaled offset
5589 Node* src_start = basic_plus_adr(top(), base, offset);
5590
5591 // Call the stub.
5592 address stubAddr = StubRoutines::updateBytesAdler32();
5593 const char *stubName = "updateBytesAdler32";
5594
5595 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5596 stubAddr, stubName, TypePtr::BOTTOM,
5597 crc, src_start, length);
5598
5599 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5600 set_result(result);
5601 return true;
5602 }
5603
5604 //----------------------------inline_reference_get----------------------------
5605 // public T java.lang.ref.Reference.get();
5606 bool LibraryCallKit::inline_reference_get() {
5607 const int referent_offset = java_lang_ref_Reference::referent_offset;
5608 guarantee(referent_offset > 0, "should have already been set");
5609
5610 // Get the argument:
5611 Node* reference_obj = null_check_receiver();
5612 if (stopped()) return true;
5613
5614 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5615 assert(tinst != NULL, "obj is null");
5616 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5617 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5618 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5619 ciSymbol::make("Ljava/lang/Object;"),
5620 false);
5621 assert (field != NULL, "undefined field");
5622
5623 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5624 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5625
5626 ciInstanceKlass* klass = env()->Object_klass();
5627 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5628
5629 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5630 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5631 // Add memory barrier to prevent commoning reads from this field
5632 // across safepoint since GC can change its value.
5633 insert_mem_bar(Op_MemBarCPUOrder);
5634
5635 set_result(result);
5636 return true;
5637 }
5638
5639
5640 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5641 bool is_exact=true, bool is_static=false,
5642 ciInstanceKlass * fromKls=NULL) {
5643 if (fromKls == NULL) {
5644 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5645 assert(tinst != NULL, "obj is null");
5646 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5647 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5648 fromKls = tinst->klass()->as_instance_klass();
5649 } else {
5650 assert(is_static, "only for static field access");
5651 }
5652 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5653 ciSymbol::make(fieldTypeString),
5654 is_static);
5655
5656 assert (field != NULL, "undefined field");
5657 if (field == NULL) return (Node *) NULL;
5658
5659 if (is_static) {
5660 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5661 fromObj = makecon(tip);
5662 }
5663
5664 // Next code copied from Parse::do_get_xxx():
5665
5666 // Compute address and memory type.
5667 int offset = field->offset_in_bytes();
5668 bool is_vol = field->is_volatile();
5669 ciType* field_klass = field->type();
5670 assert(field_klass->is_loaded(), "should be loaded");
5671 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5672 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5673 BasicType bt = field->layout_type();
5674
5675 // Build the resultant type of the load
5676 const Type *type;
5677 if (bt == T_OBJECT) {
5678 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5679 } else {
5680 type = Type::get_const_basic_type(bt);
5681 }
5682
5683 DecoratorSet decorators = IN_HEAP;
5684
5685 if (is_vol) {
5686 decorators |= MO_SEQ_CST;
5687 }
5688
5689 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5690 }
5691
5692 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5693 bool is_exact = true, bool is_static = false,
5694 ciInstanceKlass * fromKls = NULL) {
5695 if (fromKls == NULL) {
5696 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5697 assert(tinst != NULL, "obj is null");
5698 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5699 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5700 fromKls = tinst->klass()->as_instance_klass();
5701 }
5702 else {
5703 assert(is_static, "only for static field access");
5704 }
5705 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5706 ciSymbol::make(fieldTypeString),
5707 is_static);
5708
5709 assert(field != NULL, "undefined field");
5710 assert(!field->is_volatile(), "not defined for volatile fields");
5711
5712 if (is_static) {
5713 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5714 fromObj = makecon(tip);
5715 }
5716
5717 // Next code copied from Parse::do_get_xxx():
5718
5719 // Compute address and memory type.
5720 int offset = field->offset_in_bytes();
5721 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5722
5723 return adr;
5724 }
5725
5726 //------------------------------inline_aescrypt_Block-----------------------
5727 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5728 address stubAddr = NULL;
5729 const char *stubName;
5730 assert(UseAES, "need AES instruction support");
5731
5732 switch(id) {
5733 case vmIntrinsics::_aescrypt_encryptBlock:
5734 stubAddr = StubRoutines::aescrypt_encryptBlock();
5735 stubName = "aescrypt_encryptBlock";
5736 break;
5737 case vmIntrinsics::_aescrypt_decryptBlock:
5738 stubAddr = StubRoutines::aescrypt_decryptBlock();
5739 stubName = "aescrypt_decryptBlock";
5740 break;
5741 default:
5742 break;
5743 }
5744 if (stubAddr == NULL) return false;
5745
5746 Node* aescrypt_object = argument(0);
5747 Node* src = argument(1);
5748 Node* src_offset = argument(2);
5749 Node* dest = argument(3);
5750 Node* dest_offset = argument(4);
5751
5752 // (1) src and dest are arrays.
5753 const Type* src_type = src->Value(&_gvn);
5754 const Type* dest_type = dest->Value(&_gvn);
5755 const TypeAryPtr* top_src = src_type->isa_aryptr();
5756 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5757 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5758
5759 // for the quick and dirty code we will skip all the checks.
5760 // we are just trying to get the call to be generated.
5761 Node* src_start = src;
5762 Node* dest_start = dest;
5763 if (src_offset != NULL || dest_offset != NULL) {
5764 assert(src_offset != NULL && dest_offset != NULL, "");
5765 src_start = array_element_address(src, src_offset, T_BYTE);
5766 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5767 }
5768
5769 // now need to get the start of its expanded key array
5770 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5771 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5772 if (k_start == NULL) return false;
5773
5774 if (Matcher::pass_original_key_for_aes()) {
5775 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5776 // compatibility issues between Java key expansion and SPARC crypto instructions
5777 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5778 if (original_k_start == NULL) return false;
5779
5780 // Call the stub.
5781 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5782 stubAddr, stubName, TypePtr::BOTTOM,
5783 src_start, dest_start, k_start, original_k_start);
5784 } else {
5785 // Call the stub.
5786 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5787 stubAddr, stubName, TypePtr::BOTTOM,
5788 src_start, dest_start, k_start);
5789 }
5790
5791 return true;
5792 }
5793
5794 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5795 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5796 address stubAddr = NULL;
5797 const char *stubName = NULL;
5798
5799 assert(UseAES, "need AES instruction support");
5800
5801 switch(id) {
5802 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5803 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5804 stubName = "cipherBlockChaining_encryptAESCrypt";
5805 break;
5806 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5807 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5808 stubName = "cipherBlockChaining_decryptAESCrypt";
5809 break;
5810 default:
5811 break;
5812 }
5813 if (stubAddr == NULL) return false;
5814
5815 Node* cipherBlockChaining_object = argument(0);
5816 Node* src = argument(1);
5817 Node* src_offset = argument(2);
5818 Node* len = argument(3);
5819 Node* dest = argument(4);
5820 Node* dest_offset = argument(5);
5821
5822 // (1) src and dest are arrays.
5823 const Type* src_type = src->Value(&_gvn);
5824 const Type* dest_type = dest->Value(&_gvn);
5825 const TypeAryPtr* top_src = src_type->isa_aryptr();
5826 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5827 assert (top_src != NULL && top_src->klass() != NULL
5828 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5829
5830 // checks are the responsibility of the caller
5831 Node* src_start = src;
5832 Node* dest_start = dest;
5833 if (src_offset != NULL || dest_offset != NULL) {
5834 assert(src_offset != NULL && dest_offset != NULL, "");
5835 src_start = array_element_address(src, src_offset, T_BYTE);
5836 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5837 }
5838
5839 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5840 // (because of the predicated logic executed earlier).
5841 // so we cast it here safely.
5842 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5843
5844 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5845 if (embeddedCipherObj == NULL) return false;
5846
5847 // cast it to what we know it will be at runtime
5848 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5849 assert(tinst != NULL, "CBC obj is null");
5850 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5851 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5852 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5853
5854 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5855 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5856 const TypeOopPtr* xtype = aklass->as_instance_type();
5857 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5858 aescrypt_object = _gvn.transform(aescrypt_object);
5859
5860 // we need to get the start of the aescrypt_object's expanded key array
5861 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5862 if (k_start == NULL) return false;
5863
5864 // similarly, get the start address of the r vector
5865 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5866 if (objRvec == NULL) return false;
5867 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5868
5869 Node* cbcCrypt;
5870 if (Matcher::pass_original_key_for_aes()) {
5871 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5872 // compatibility issues between Java key expansion and SPARC crypto instructions
5873 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5874 if (original_k_start == NULL) return false;
5875
5876 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5877 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5878 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5879 stubAddr, stubName, TypePtr::BOTTOM,
5880 src_start, dest_start, k_start, r_start, len, original_k_start);
5881 } else {
5882 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5883 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5884 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5885 stubAddr, stubName, TypePtr::BOTTOM,
5886 src_start, dest_start, k_start, r_start, len);
5887 }
5888
5889 // return cipher length (int)
5890 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5891 set_result(retvalue);
5892 return true;
5893 }
5894
5895 //------------------------------inline_counterMode_AESCrypt-----------------------
5896 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
5897 assert(UseAES, "need AES instruction support");
5898 if (!UseAESCTRIntrinsics) return false;
5899
5900 address stubAddr = NULL;
5901 const char *stubName = NULL;
5902 if (id == vmIntrinsics::_counterMode_AESCrypt) {
5903 stubAddr = StubRoutines::counterMode_AESCrypt();
5904 stubName = "counterMode_AESCrypt";
5905 }
5906 if (stubAddr == NULL) return false;
5907
5908 Node* counterMode_object = argument(0);
5909 Node* src = argument(1);
5910 Node* src_offset = argument(2);
5911 Node* len = argument(3);
5912 Node* dest = argument(4);
5913 Node* dest_offset = argument(5);
5914
5915 // (1) src and dest are arrays.
5916 const Type* src_type = src->Value(&_gvn);
5917 const Type* dest_type = dest->Value(&_gvn);
5918 const TypeAryPtr* top_src = src_type->isa_aryptr();
5919 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5920 assert(top_src != NULL && top_src->klass() != NULL &&
5921 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5922
5923 // checks are the responsibility of the caller
5924 Node* src_start = src;
5925 Node* dest_start = dest;
5926 if (src_offset != NULL || dest_offset != NULL) {
5927 assert(src_offset != NULL && dest_offset != NULL, "");
5928 src_start = array_element_address(src, src_offset, T_BYTE);
5929 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5930 }
5931
5932 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5933 // (because of the predicated logic executed earlier).
5934 // so we cast it here safely.
5935 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5936 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5937 if (embeddedCipherObj == NULL) return false;
5938 // cast it to what we know it will be at runtime
5939 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
5940 assert(tinst != NULL, "CTR obj is null");
5941 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
5942 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5943 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5944 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5945 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5946 const TypeOopPtr* xtype = aklass->as_instance_type();
5947 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5948 aescrypt_object = _gvn.transform(aescrypt_object);
5949 // we need to get the start of the aescrypt_object's expanded key array
5950 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5951 if (k_start == NULL) return false;
5952 // similarly, get the start address of the r vector
5953 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
5954 if (obj_counter == NULL) return false;
5955 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
5956
5957 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
5958 if (saved_encCounter == NULL) return false;
5959 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
5960 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
5961
5962 Node* ctrCrypt;
5963 if (Matcher::pass_original_key_for_aes()) {
5964 // no SPARC version for AES/CTR intrinsics now.
5965 return false;
5966 }
5967 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5968 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5969 OptoRuntime::counterMode_aescrypt_Type(),
5970 stubAddr, stubName, TypePtr::BOTTOM,
5971 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
5972
5973 // return cipher length (int)
5974 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
5975 set_result(retvalue);
5976 return true;
5977 }
5978
5979 //------------------------------get_key_start_from_aescrypt_object-----------------------
5980 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
5981 #if defined(PPC64) || defined(S390)
5982 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
5983 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
5984 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
5985 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
5986 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
5987 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5988 if (objSessionK == NULL) {
5989 return (Node *) NULL;
5990 }
5991 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
5992 #else
5993 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
5994 #endif // PPC64
5995 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5996 if (objAESCryptKey == NULL) return (Node *) NULL;
5997
5998 // now have the array, need to get the start address of the K array
5999 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6000 return k_start;
6001 }
6002
6003 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6004 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6005 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6006 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6007 if (objAESCryptKey == NULL) return (Node *) NULL;
6008
6009 // now have the array, need to get the start address of the lastKey array
6010 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6011 return original_k_start;
6012 }
6013
6014 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6015 // Return node representing slow path of predicate check.
6016 // the pseudo code we want to emulate with this predicate is:
6017 // for encryption:
6018 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6019 // for decryption:
6020 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6021 // note cipher==plain is more conservative than the original java code but that's OK
6022 //
6023 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6024 // The receiver was checked for NULL already.
6025 Node* objCBC = argument(0);
6026
6027 // Load embeddedCipher field of CipherBlockChaining object.
6028 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6029
6030 // get AESCrypt klass for instanceOf check
6031 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6032 // will have same classloader as CipherBlockChaining object
6033 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6034 assert(tinst != NULL, "CBCobj is null");
6035 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6036
6037 // we want to do an instanceof comparison against the AESCrypt class
6038 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6039 if (!klass_AESCrypt->is_loaded()) {
6040 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6041 Node* ctrl = control();
6042 set_control(top()); // no regular fast path
6043 return ctrl;
6044 }
6045 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6046
6047 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6048 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6049 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6050
6051 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6052
6053 // for encryption, we are done
6054 if (!decrypting)
6055 return instof_false; // even if it is NULL
6056
6057 // for decryption, we need to add a further check to avoid
6058 // taking the intrinsic path when cipher and plain are the same
6059 // see the original java code for why.
6060 RegionNode* region = new RegionNode(3);
6061 region->init_req(1, instof_false);
6062 Node* src = argument(1);
6063 Node* dest = argument(4);
6064 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6065 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6066 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6067 region->init_req(2, src_dest_conjoint);
6068
6069 record_for_igvn(region);
6070 return _gvn.transform(region);
6071 }
6072
6073 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6074 // Return node representing slow path of predicate check.
6075 // the pseudo code we want to emulate with this predicate is:
6076 // for encryption:
6077 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6078 // for decryption:
6079 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6080 // note cipher==plain is more conservative than the original java code but that's OK
6081 //
6082
6083 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6084 // The receiver was checked for NULL already.
6085 Node* objCTR = argument(0);
6086
6087 // Load embeddedCipher field of CipherBlockChaining object.
6088 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6089
6090 // get AESCrypt klass for instanceOf check
6091 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6092 // will have same classloader as CipherBlockChaining object
6093 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6094 assert(tinst != NULL, "CTRobj is null");
6095 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6096
6097 // we want to do an instanceof comparison against the AESCrypt class
6098 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6099 if (!klass_AESCrypt->is_loaded()) {
6100 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6101 Node* ctrl = control();
6102 set_control(top()); // no regular fast path
6103 return ctrl;
6104 }
6105
6106 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6107 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6108 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6109 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6110 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6111
6112 return instof_false; // even if it is NULL
6113 }
6114
6115 //------------------------------inline_ghash_processBlocks
6116 bool LibraryCallKit::inline_ghash_processBlocks() {
6117 address stubAddr;
6118 const char *stubName;
6119 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6120
6121 stubAddr = StubRoutines::ghash_processBlocks();
6122 stubName = "ghash_processBlocks";
6123
6124 Node* data = argument(0);
6125 Node* offset = argument(1);
6126 Node* len = argument(2);
6127 Node* state = argument(3);
6128 Node* subkeyH = argument(4);
6129
6130 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6131 assert(state_start, "state is NULL");
6132 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6133 assert(subkeyH_start, "subkeyH is NULL");
6134 Node* data_start = array_element_address(data, offset, T_BYTE);
6135 assert(data_start, "data is NULL");
6136
6137 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6138 OptoRuntime::ghash_processBlocks_Type(),
6139 stubAddr, stubName, TypePtr::BOTTOM,
6140 state_start, subkeyH_start, data_start, len);
6141 return true;
6142 }
6143
6144 bool LibraryCallKit::inline_base64_encodeBlock() {
6145 address stubAddr;
6146 const char *stubName;
6147 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6148 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6149 stubAddr = StubRoutines::base64_encodeBlock();
6150 stubName = "encodeBlock";
6151
6152 if (!stubAddr) return false;
6153 Node* base64obj = argument(0);
6154 Node* src = argument(1);
6155 Node* offset = argument(2);
6156 Node* len = argument(3);
6157 Node* dest = argument(4);
6158 Node* dp = argument(5);
6159 Node* isURL = argument(6);
6160
6161 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6162 assert(src_start, "source array is NULL");
6163 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6164 assert(dest_start, "destination array is NULL");
6165
6166 Node* base64 = make_runtime_call(RC_LEAF,
6167 OptoRuntime::base64_encodeBlock_Type(),
6168 stubAddr, stubName, TypePtr::BOTTOM,
6169 src_start, offset, len, dest_start, dp, isURL);
6170 return true;
6171 }
6172
6173 //------------------------------inline_sha_implCompress-----------------------
6174 //
6175 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6176 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6177 //
6178 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6179 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6180 //
6181 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6182 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6183 //
6184 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6185 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6186
6187 Node* sha_obj = argument(0);
6188 Node* src = argument(1); // type oop
6189 Node* ofs = argument(2); // type int
6190
6191 const Type* src_type = src->Value(&_gvn);
6192 const TypeAryPtr* top_src = src_type->isa_aryptr();
6193 if (top_src == NULL || top_src->klass() == NULL) {
6194 // failed array check
6195 return false;
6196 }
6197 // Figure out the size and type of the elements we will be copying.
6198 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6199 if (src_elem != T_BYTE) {
6200 return false;
6201 }
6202 // 'src_start' points to src array + offset
6203 Node* src_start = array_element_address(src, ofs, src_elem);
6204 Node* state = NULL;
6205 address stubAddr;
6206 const char *stubName;
6207
6208 switch(id) {
6209 case vmIntrinsics::_sha_implCompress:
6210 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6211 state = get_state_from_sha_object(sha_obj);
6212 stubAddr = StubRoutines::sha1_implCompress();
6213 stubName = "sha1_implCompress";
6214 break;
6215 case vmIntrinsics::_sha2_implCompress:
6216 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6217 state = get_state_from_sha_object(sha_obj);
6218 stubAddr = StubRoutines::sha256_implCompress();
6219 stubName = "sha256_implCompress";
6220 break;
6221 case vmIntrinsics::_sha5_implCompress:
6222 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6223 state = get_state_from_sha5_object(sha_obj);
6224 stubAddr = StubRoutines::sha512_implCompress();
6225 stubName = "sha512_implCompress";
6226 break;
6227 default:
6228 fatal_unexpected_iid(id);
6229 return false;
6230 }
6231 if (state == NULL) return false;
6232
6233 // Call the stub.
6234 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6235 stubAddr, stubName, TypePtr::BOTTOM,
6236 src_start, state);
6237
6238 return true;
6239 }
6240
6241 //------------------------------inline_digestBase_implCompressMB-----------------------
6242 //
6243 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6244 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6245 //
6246 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6247 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6248 "need SHA1/SHA256/SHA512 instruction support");
6249 assert((uint)predicate < 3, "sanity");
6250 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6251
6252 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6253 Node* src = argument(1); // byte[] array
6254 Node* ofs = argument(2); // type int
6255 Node* limit = argument(3); // type int
6256
6257 const Type* src_type = src->Value(&_gvn);
6258 const TypeAryPtr* top_src = src_type->isa_aryptr();
6259 if (top_src == NULL || top_src->klass() == NULL) {
6260 // failed array check
6261 return false;
6262 }
6263 // Figure out the size and type of the elements we will be copying.
6264 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6265 if (src_elem != T_BYTE) {
6266 return false;
6267 }
6268 // 'src_start' points to src array + offset
6269 Node* src_start = array_element_address(src, ofs, src_elem);
6270
6271 const char* klass_SHA_name = NULL;
6272 const char* stub_name = NULL;
6273 address stub_addr = NULL;
6274 bool long_state = false;
6275
6276 switch (predicate) {
6277 case 0:
6278 if (UseSHA1Intrinsics) {
6279 klass_SHA_name = "sun/security/provider/SHA";
6280 stub_name = "sha1_implCompressMB";
6281 stub_addr = StubRoutines::sha1_implCompressMB();
6282 }
6283 break;
6284 case 1:
6285 if (UseSHA256Intrinsics) {
6286 klass_SHA_name = "sun/security/provider/SHA2";
6287 stub_name = "sha256_implCompressMB";
6288 stub_addr = StubRoutines::sha256_implCompressMB();
6289 }
6290 break;
6291 case 2:
6292 if (UseSHA512Intrinsics) {
6293 klass_SHA_name = "sun/security/provider/SHA5";
6294 stub_name = "sha512_implCompressMB";
6295 stub_addr = StubRoutines::sha512_implCompressMB();
6296 long_state = true;
6297 }
6298 break;
6299 default:
6300 fatal("unknown SHA intrinsic predicate: %d", predicate);
6301 }
6302 if (klass_SHA_name != NULL) {
6303 // get DigestBase klass to lookup for SHA klass
6304 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6305 assert(tinst != NULL, "digestBase_obj is not instance???");
6306 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6307
6308 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6309 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6310 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6311 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6312 }
6313 return false;
6314 }
6315 //------------------------------inline_sha_implCompressMB-----------------------
6316 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6317 bool long_state, address stubAddr, const char *stubName,
6318 Node* src_start, Node* ofs, Node* limit) {
6319 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6320 const TypeOopPtr* xtype = aklass->as_instance_type();
6321 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6322 sha_obj = _gvn.transform(sha_obj);
6323
6324 Node* state;
6325 if (long_state) {
6326 state = get_state_from_sha5_object(sha_obj);
6327 } else {
6328 state = get_state_from_sha_object(sha_obj);
6329 }
6330 if (state == NULL) return false;
6331
6332 // Call the stub.
6333 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6334 OptoRuntime::digestBase_implCompressMB_Type(),
6335 stubAddr, stubName, TypePtr::BOTTOM,
6336 src_start, state, ofs, limit);
6337 // return ofs (int)
6338 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6339 set_result(result);
6340
6341 return true;
6342 }
6343
6344 //------------------------------get_state_from_sha_object-----------------------
6345 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6346 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6347 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6348 if (sha_state == NULL) return (Node *) NULL;
6349
6350 // now have the array, need to get the start address of the state array
6351 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6352 return state;
6353 }
6354
6355 //------------------------------get_state_from_sha5_object-----------------------
6356 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6357 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6358 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6359 if (sha_state == NULL) return (Node *) NULL;
6360
6361 // now have the array, need to get the start address of the state array
6362 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6363 return state;
6364 }
6365
6366 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6367 // Return node representing slow path of predicate check.
6368 // the pseudo code we want to emulate with this predicate is:
6369 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6370 //
6371 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6372 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6373 "need SHA1/SHA256/SHA512 instruction support");
6374 assert((uint)predicate < 3, "sanity");
6375
6376 // The receiver was checked for NULL already.
6377 Node* digestBaseObj = argument(0);
6378
6379 // get DigestBase klass for instanceOf check
6380 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6381 assert(tinst != NULL, "digestBaseObj is null");
6382 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6383
6384 const char* klass_SHA_name = NULL;
6385 switch (predicate) {
6386 case 0:
6387 if (UseSHA1Intrinsics) {
6388 // we want to do an instanceof comparison against the SHA class
6389 klass_SHA_name = "sun/security/provider/SHA";
6390 }
6391 break;
6392 case 1:
6393 if (UseSHA256Intrinsics) {
6394 // we want to do an instanceof comparison against the SHA2 class
6395 klass_SHA_name = "sun/security/provider/SHA2";
6396 }
6397 break;
6398 case 2:
6399 if (UseSHA512Intrinsics) {
6400 // we want to do an instanceof comparison against the SHA5 class
6401 klass_SHA_name = "sun/security/provider/SHA5";
6402 }
6403 break;
6404 default:
6405 fatal("unknown SHA intrinsic predicate: %d", predicate);
6406 }
6407
6408 ciKlass* klass_SHA = NULL;
6409 if (klass_SHA_name != NULL) {
6410 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6411 }
6412 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6413 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6414 Node* ctrl = control();
6415 set_control(top()); // no intrinsic path
6416 return ctrl;
6417 }
6418 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6419
6420 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6421 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6422 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6423 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6424
6425 return instof_false; // even if it is NULL
6426 }
6427
6428 //-------------inline_fma-----------------------------------
6429 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6430 Node *a = NULL;
6431 Node *b = NULL;
6432 Node *c = NULL;
6433 Node* result = NULL;
6434 switch (id) {
6435 case vmIntrinsics::_fmaD:
6436 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6437 // no receiver since it is static method
6438 a = round_double_node(argument(0));
6439 b = round_double_node(argument(2));
6440 c = round_double_node(argument(4));
6441 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6442 break;
6443 case vmIntrinsics::_fmaF:
6444 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6445 a = argument(0);
6446 b = argument(1);
6447 c = argument(2);
6448 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6449 break;
6450 default:
6451 fatal_unexpected_iid(id); break;
6452 }
6453 set_result(result);
6454 return true;
6455 }
6456
6457 bool LibraryCallKit::inline_profileBoolean() {
6458 Node* counts = argument(1);
6459 const TypeAryPtr* ary = NULL;
6460 ciArray* aobj = NULL;
6461 if (counts->is_Con()
6462 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6463 && (aobj = ary->const_oop()->as_array()) != NULL
6464 && (aobj->length() == 2)) {
6465 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6466 jint false_cnt = aobj->element_value(0).as_int();
6467 jint true_cnt = aobj->element_value(1).as_int();
6468
6469 if (C->log() != NULL) {
6470 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6471 false_cnt, true_cnt);
6472 }
6473
6474 if (false_cnt + true_cnt == 0) {
6475 // According to profile, never executed.
6476 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6477 Deoptimization::Action_reinterpret);
6478 return true;
6479 }
6480
6481 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6482 // is a number of each value occurrences.
6483 Node* result = argument(0);
6484 if (false_cnt == 0 || true_cnt == 0) {
6485 // According to profile, one value has been never seen.
6486 int expected_val = (false_cnt == 0) ? 1 : 0;
6487
6488 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6489 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6490
6491 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6492 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6493 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6494
6495 { // Slow path: uncommon trap for never seen value and then reexecute
6496 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6497 // the value has been seen at least once.
6498 PreserveJVMState pjvms(this);
6499 PreserveReexecuteState preexecs(this);
6500 jvms()->set_should_reexecute(true);
6501
6502 set_control(slow_path);
6503 set_i_o(i_o());
6504
6505 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6506 Deoptimization::Action_reinterpret);
6507 }
6508 // The guard for never seen value enables sharpening of the result and
6509 // returning a constant. It allows to eliminate branches on the same value
6510 // later on.
6511 set_control(fast_path);
6512 result = intcon(expected_val);
6513 }
6514 // Stop profiling.
6515 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6516 // By replacing method body with profile data (represented as ProfileBooleanNode
6517 // on IR level) we effectively disable profiling.
6518 // It enables full speed execution once optimized code is generated.
6519 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6520 C->record_for_igvn(profile);
6521 set_result(profile);
6522 return true;
6523 } else {
6524 // Continue profiling.
6525 // Profile data isn't available at the moment. So, execute method's bytecode version.
6526 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6527 // is compiled and counters aren't available since corresponding MethodHandle
6528 // isn't a compile-time constant.
6529 return false;
6530 }
6531 }
6532
6533 bool LibraryCallKit::inline_isCompileConstant() {
6534 Node* n = argument(0);
6535 set_result(n->is_Con() ? intcon(1) : intcon(0));
6536 return true;
6537 }