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 "memory/resourceArea.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "opto/addnode.hpp"
36 #include "opto/arraycopynode.hpp"
37 #include "opto/c2compiler.hpp"
38 #include "opto/callGenerator.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/intrinsicnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/movenode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/nativeLookup.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/objectMonitor.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "utilities/macros.hpp"
59
60
61 class LibraryIntrinsic : public InlineCallGenerator {
62 // Extend the set of intrinsics known to the runtime:
63 public:
64 private:
65 bool _is_virtual;
66 bool _does_virtual_dispatch;
67 int8_t _predicates_count; // Intrinsic is predicated by several conditions
68 int8_t _last_predicate; // Last generated predicate
69 vmIntrinsics::ID _intrinsic_id;
70
71 public:
72 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
73 : InlineCallGenerator(m),
74 _is_virtual(is_virtual),
75 _does_virtual_dispatch(does_virtual_dispatch),
76 _predicates_count((int8_t)predicates_count),
77 _last_predicate((int8_t)-1),
78 _intrinsic_id(id)
79 {
80 }
81 virtual bool is_intrinsic() const { return true; }
82 virtual bool is_virtual() const { return _is_virtual; }
83 virtual bool is_predicated() const { return _predicates_count > 0; }
84 virtual int predicates_count() const { return _predicates_count; }
85 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
86 virtual JVMState* generate(JVMState* jvms);
87 virtual Node* generate_predicate(JVMState* jvms, int predicate);
88 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
89 };
90
91
92 // Local helper class for LibraryIntrinsic:
93 class LibraryCallKit : public GraphKit {
94 private:
95 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
96 Node* _result; // the result node, if any
97 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
98
99 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
100
101 public:
102 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
103 : GraphKit(jvms),
104 _intrinsic(intrinsic),
105 _result(NULL)
106 {
107 // Check if this is a root compile. In that case we don't have a caller.
108 if (!jvms->has_method()) {
109 _reexecute_sp = sp();
110 } else {
111 // Find out how many arguments the interpreter needs when deoptimizing
112 // and save the stack pointer value so it can used by uncommon_trap.
113 // We find the argument count by looking at the declared signature.
114 bool ignored_will_link;
115 ciSignature* declared_signature = NULL;
116 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
117 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
118 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
119 }
120 }
121
122 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
123
124 ciMethod* caller() const { return jvms()->method(); }
125 int bci() const { return jvms()->bci(); }
126 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
127 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
128 ciMethod* callee() const { return _intrinsic->method(); }
129
130 bool try_to_inline(int predicate);
131 Node* try_to_predicate(int predicate);
132
133 void push_result() {
134 // Push the result onto the stack.
135 if (!stopped() && result() != NULL) {
136 BasicType bt = result()->bottom_type()->basic_type();
137 push_node(bt, result());
138 }
139 }
140
141 private:
142 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
143 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
144 }
145
146 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
147 void set_result(RegionNode* region, PhiNode* value);
148 Node* result() { return _result; }
149
150 virtual int reexecute_sp() { return _reexecute_sp; }
151
152 // Helper functions to inline natives
153 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
154 Node* generate_slow_guard(Node* test, RegionNode* region);
155 Node* generate_fair_guard(Node* test, RegionNode* region);
156 Node* generate_negative_guard(Node* index, RegionNode* region,
157 // resulting CastII of index:
158 Node* *pos_index = NULL);
159 Node* generate_limit_guard(Node* offset, Node* subseq_length,
160 Node* array_length,
161 RegionNode* region);
162 void generate_string_range_check(Node* array, Node* offset,
163 Node* length, bool char_count);
164 Node* generate_current_thread(Node* &tls_output);
165 Node* load_mirror_from_klass(Node* klass);
166 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
167 RegionNode* region, int null_path,
168 int offset);
169 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
170 RegionNode* region, int null_path) {
171 int offset = java_lang_Class::klass_offset_in_bytes();
172 return load_klass_from_mirror_common(mirror, never_see_null,
173 region, null_path,
174 offset);
175 }
176 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
177 RegionNode* region, int null_path) {
178 int offset = java_lang_Class::array_klass_offset_in_bytes();
179 return load_klass_from_mirror_common(mirror, never_see_null,
180 region, null_path,
181 offset);
182 }
183 Node* generate_access_flags_guard(Node* kls,
184 int modifier_mask, int modifier_bits,
185 RegionNode* region);
186 Node* generate_interface_guard(Node* kls, RegionNode* region);
187 Node* generate_array_guard(Node* kls, RegionNode* region) {
188 return generate_array_guard_common(kls, region, false, false);
189 }
190 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
191 return generate_array_guard_common(kls, region, false, true);
192 }
193 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
194 return generate_array_guard_common(kls, region, true, false);
195 }
196 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
197 return generate_array_guard_common(kls, region, true, true);
198 }
199 Node* generate_array_guard_common(Node* kls, RegionNode* region,
200 bool obj_array, bool not_array);
201 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
202 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
203 bool is_virtual = false, bool is_static = false);
204 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
205 return generate_method_call(method_id, false, true);
206 }
207 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
208 return generate_method_call(method_id, true, false);
209 }
210 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
211 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
212
213 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
214 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
215 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
217 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
218 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
219 bool inline_string_indexOfChar();
220 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
221 bool inline_string_toBytesU();
222 bool inline_string_getCharsU();
223 bool inline_string_copy(bool compress);
224 bool inline_string_char_access(bool is_store);
225 Node* round_double_node(Node* n);
226 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
227 bool inline_math_native(vmIntrinsics::ID id);
228 bool inline_math(vmIntrinsics::ID id);
229 template <typename OverflowOp>
230 bool inline_math_overflow(Node* arg1, Node* arg2);
231 void inline_math_mathExact(Node* math, Node* test);
232 bool inline_math_addExactI(bool is_increment);
233 bool inline_math_addExactL(bool is_increment);
234 bool inline_math_multiplyExactI();
235 bool inline_math_multiplyExactL();
236 bool inline_math_multiplyHigh();
237 bool inline_math_negateExactI();
238 bool inline_math_negateExactL();
239 bool inline_math_subtractExactI(bool is_decrement);
240 bool inline_math_subtractExactL(bool is_decrement);
241 bool inline_min_max(vmIntrinsics::ID id);
242 bool inline_notify(vmIntrinsics::ID id);
243 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
244 // This returns Type::AnyPtr, RawPtr, or OopPtr.
245 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
246 Node* make_unsafe_address(Node*& base, Node* offset, BasicType type = T_ILLEGAL, bool can_cast = false);
247
248 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
249 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
250 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
251 static bool klass_needs_init_guard(Node* kls);
252 bool inline_unsafe_allocate();
253 bool inline_unsafe_newArray(bool uninitialized);
254 bool inline_unsafe_copyMemory();
255 bool inline_native_currentThread();
256
257 bool inline_native_time_funcs(address method, const char* funcName);
258 #ifdef JFR_HAVE_INTRINSICS
259 bool inline_native_classID();
260 bool inline_native_getEventWriter();
261 #endif
262 bool inline_native_isInterrupted();
263 bool inline_native_Class_query(vmIntrinsics::ID id);
264 bool inline_native_subtype_check();
265 bool inline_native_getLength();
266 bool inline_array_copyOf(bool is_copyOfRange);
267 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
268 bool inline_preconditions_checkIndex();
269 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
270 bool inline_native_clone(bool is_virtual);
271 bool inline_native_Reflection_getCallerClass();
272 // Helper function for inlining native object hash method
273 bool inline_native_hashcode(bool is_virtual, bool is_static);
274 bool inline_native_getClass();
275
276 // Helper functions for inlining arraycopy
277 bool inline_arraycopy();
278 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
279 RegionNode* slow_region);
280 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
281 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
282 uint new_idx);
283
284 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
285 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
286 bool inline_unsafe_fence(vmIntrinsics::ID id);
287 bool inline_onspinwait();
288 bool inline_fp_conversions(vmIntrinsics::ID id);
289 bool inline_number_methods(vmIntrinsics::ID id);
290 bool inline_reference_get();
291 bool inline_Class_cast();
292 bool inline_aescrypt_Block(vmIntrinsics::ID id);
293 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
294 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
295 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
296 Node* inline_counterMode_AESCrypt_predicate();
297 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
298 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
299 bool inline_ghash_processBlocks();
300 bool inline_sha_implCompress(vmIntrinsics::ID id);
301 bool inline_digestBase_implCompressMB(int predicate);
302 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
303 bool long_state, address stubAddr, const char *stubName,
304 Node* src_start, Node* ofs, Node* limit);
305 Node* get_state_from_sha_object(Node *sha_object);
306 Node* get_state_from_sha5_object(Node *sha_object);
307 Node* inline_digestBase_implCompressMB_predicate(int predicate);
308 bool inline_encodeISOArray();
309 bool inline_updateCRC32();
310 bool inline_updateBytesCRC32();
311 bool inline_updateByteBufferCRC32();
312 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
313 bool inline_updateBytesCRC32C();
314 bool inline_updateDirectByteBufferCRC32C();
315 bool inline_updateBytesAdler32();
316 bool inline_updateByteBufferAdler32();
317 bool inline_multiplyToLen();
318 bool inline_hasNegatives();
319 bool inline_squareToLen();
320 bool inline_mulAdd();
321 bool inline_montgomeryMultiply();
322 bool inline_montgomerySquare();
323 bool inline_vectorizedMismatch();
324 bool inline_fma(vmIntrinsics::ID id);
325
326 bool inline_profileBoolean();
327 bool inline_isCompileConstant();
328 bool inline_continuation(vmIntrinsics::ID id);
329 bool inline_get_frame_pointer();
330 bool inline_continuation_do_yield();
331 bool inline_continuation_runLevel();
332 bool inline_continuation_do_continue();
333 void clear_upper_avx() {
334 #ifdef X86
335 if (UseAVX >= 2) {
336 C->set_clear_upper_avx(true);
337 }
338 #endif
339 }
340 };
341
342 //---------------------------make_vm_intrinsic----------------------------
343 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
344 vmIntrinsics::ID id = m->intrinsic_id();
345 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
346
347 if (!m->is_loaded()) {
348 // Do not attempt to inline unloaded methods.
349 return NULL;
350 }
351
352 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
353 bool is_available = false;
354
355 {
356 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
357 // the compiler must transition to '_thread_in_vm' state because both
358 // methods access VM-internal data.
359 VM_ENTRY_MARK;
360 methodHandle mh(THREAD, m->get_Method());
361 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
362 !C->directive()->is_intrinsic_disabled(mh) &&
363 !vmIntrinsics::is_disabled_by_flags(mh);
364
365 }
366
367 if (is_available) {
368 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
369 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
370 return new LibraryIntrinsic(m, is_virtual,
371 vmIntrinsics::predicates_needed(id),
372 vmIntrinsics::does_virtual_dispatch(id),
373 (vmIntrinsics::ID) id);
374 } else {
375 return NULL;
376 }
377 }
378
379 //----------------------register_library_intrinsics-----------------------
380 // Initialize this file's data structures, for each Compile instance.
381 void Compile::register_library_intrinsics() {
382 // Nothing to do here.
383 }
384
385 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
386 LibraryCallKit kit(jvms, this);
387 Compile* C = kit.C;
388 int nodes = C->unique();
389 #ifndef PRODUCT
390 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
391 char buf[1000];
392 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
393 tty->print_cr("Intrinsic %s", str);
394 }
395 #endif
396 ciMethod* callee = kit.callee();
397 const int bci = kit.bci();
398
399 // Try to inline the intrinsic.
400 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
401 kit.try_to_inline(_last_predicate)) {
402 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
403 : "(intrinsic)";
404 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
405 if (C->print_intrinsics() || C->print_inlining()) {
406 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
407 }
408 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
409 if (C->log()) {
410 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
411 vmIntrinsics::name_at(intrinsic_id()),
412 (is_virtual() ? " virtual='1'" : ""),
413 C->unique() - nodes);
414 }
415 // Push the result from the inlined method onto the stack.
416 kit.push_result();
417 C->print_inlining_update(this);
418 return kit.transfer_exceptions_into_jvms();
419 }
420
421 // The intrinsic bailed out
422 if (jvms->has_method()) {
423 // Not a root compile.
424 const char* msg;
425 if (callee->intrinsic_candidate()) {
426 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
427 } else {
428 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
429 : "failed to inline (intrinsic), method not annotated";
430 }
431 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
432 if (C->print_intrinsics() || C->print_inlining()) {
433 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
434 }
435 } else {
436 // Root compile
437 ResourceMark rm;
438 stringStream msg_stream;
439 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
440 vmIntrinsics::name_at(intrinsic_id()),
441 is_virtual() ? " (virtual)" : "", bci);
442 const char *msg = msg_stream.as_string();
443 log_debug(jit, inlining)("%s", msg);
444 if (C->print_intrinsics() || C->print_inlining()) {
445 tty->print("%s", msg);
446 }
447 }
448 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
449 C->print_inlining_update(this);
450 return NULL;
451 }
452
453 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
454 LibraryCallKit kit(jvms, this);
455 Compile* C = kit.C;
456 int nodes = C->unique();
457 _last_predicate = predicate;
458 #ifndef PRODUCT
459 assert(is_predicated() && predicate < predicates_count(), "sanity");
460 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
461 char buf[1000];
462 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
463 tty->print_cr("Predicate for intrinsic %s", str);
464 }
465 #endif
466 ciMethod* callee = kit.callee();
467 const int bci = kit.bci();
468
469 Node* slow_ctl = kit.try_to_predicate(predicate);
470 if (!kit.failing()) {
471 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
472 : "(intrinsic, predicate)";
473 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
474 if (C->print_intrinsics() || C->print_inlining()) {
475 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
476 }
477 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
478 if (C->log()) {
479 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
480 vmIntrinsics::name_at(intrinsic_id()),
481 (is_virtual() ? " virtual='1'" : ""),
482 C->unique() - nodes);
483 }
484 return slow_ctl; // Could be NULL if the check folds.
485 }
486
487 // The intrinsic bailed out
488 if (jvms->has_method()) {
489 // Not a root compile.
490 const char* msg = "failed to generate predicate for intrinsic";
491 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
492 if (C->print_intrinsics() || C->print_inlining()) {
493 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
494 }
495 } else {
496 // Root compile
497 ResourceMark rm;
498 stringStream msg_stream;
499 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
500 vmIntrinsics::name_at(intrinsic_id()),
501 is_virtual() ? " (virtual)" : "", bci);
502 const char *msg = msg_stream.as_string();
503 log_debug(jit, inlining)("%s", msg);
504 if (C->print_intrinsics() || C->print_inlining()) {
505 C->print_inlining_stream()->print("%s", msg);
506 }
507 }
508 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
509 return NULL;
510 }
511
512 bool LibraryCallKit::try_to_inline(int predicate) {
513 // Handle symbolic names for otherwise undistinguished boolean switches:
514 const bool is_store = true;
515 const bool is_compress = true;
516 const bool is_static = true;
517 const bool is_volatile = true;
518
519 if (!jvms()->has_method()) {
520 // Root JVMState has a null method.
521 assert(map()->memory()->Opcode() == Op_Parm, "");
522 // Insert the memory aliasing node
523 set_all_memory(reset_memory());
524 }
525 assert(merged_memory(), "");
526
527
528 switch (intrinsic_id()) {
529 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
530 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
531 case vmIntrinsics::_getClass: return inline_native_getClass();
532
533 case vmIntrinsics::_dsin:
534 case vmIntrinsics::_dcos:
535 case vmIntrinsics::_dtan:
536 case vmIntrinsics::_dabs:
537 case vmIntrinsics::_datan2:
538 case vmIntrinsics::_dsqrt:
539 case vmIntrinsics::_dexp:
540 case vmIntrinsics::_dlog:
541 case vmIntrinsics::_dlog10:
542 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
543
544 case vmIntrinsics::_min:
545 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
546
547 case vmIntrinsics::_notify:
548 case vmIntrinsics::_notifyAll:
549 if (ObjectMonitor::Knob_InlineNotify) {
550 return inline_notify(intrinsic_id());
551 }
552 return false;
553
554 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
555 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
556 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
557 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
558 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
559 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
560 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
561 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
562 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
563 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
564 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
565 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
566 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
567
568 case vmIntrinsics::_arraycopy: return inline_arraycopy();
569
570 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
571 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
572 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
573 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
574
575 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
576 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
577 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
578 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
579 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
580 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
581 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
582
583 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
584 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
585
586 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
587 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
588 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
589 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
590
591 case vmIntrinsics::_compressStringC:
592 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
593 case vmIntrinsics::_inflateStringC:
594 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
595
596 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
597 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
598 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
599 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
600 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
601 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
602 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
603 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
604 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
605
606 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
607 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
608 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
609 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
610 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
611 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
612 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
613 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
614 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
615
616 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
617 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
618 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
619 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
620 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
621 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
622 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
623 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
624 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
625
626 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
627 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
628 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
629 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
630 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
631 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
632 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
633 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
634 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
635
636 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
637 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
638 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
639 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
640
641 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
642 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
643 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
644 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
645
646 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
647 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
648 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
649 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
650 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
651 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
652 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
653 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
654 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
655
656 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
657 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
658 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
659 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
660 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
661 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
662 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
663 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
664 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
665
666 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
667 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
668 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
669 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
670 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
671 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
672 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
673 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
674 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
675
676 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
677 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
678 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
679 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
680 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
681 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
682 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
683 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
684 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
685
686 case vmIntrinsics::_compareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
687 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
688 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
689 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
690 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
691
692 case vmIntrinsics::_weakCompareAndSetObjectPlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
693 case vmIntrinsics::_weakCompareAndSetObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
694 case vmIntrinsics::_weakCompareAndSetObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
695 case vmIntrinsics::_weakCompareAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
696 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
697 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
698 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
699 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
700 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
701 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
702 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
703 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
704 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
705 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
706 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
707 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
708 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
709 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
710 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
711 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
712
713 case vmIntrinsics::_compareAndExchangeObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
714 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
715 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
716 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
717 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
718 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
719 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
720 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
721 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
722 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
723 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
724 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
725 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
726 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
727 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
728
729 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
730 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
731 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
732 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
733
734 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
735 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
736 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
737 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
738 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
739
740 case vmIntrinsics::_loadFence:
741 case vmIntrinsics::_storeFence:
742 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
743
744 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
745
746 case vmIntrinsics::_currentThread: return inline_native_currentThread();
747 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
748
749 #ifdef JFR_HAVE_INTRINSICS
750 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
751 case vmIntrinsics::_getClassId: return inline_native_classID();
752 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
753 #endif
754 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
755 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
756 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
757 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
758 case vmIntrinsics::_getLength: return inline_native_getLength();
759 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
760 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
761 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
762 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
763 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
764 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
765
766 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
767 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
768
769 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
770
771 case vmIntrinsics::_isInstance:
772 case vmIntrinsics::_getModifiers:
773 case vmIntrinsics::_isInterface:
774 case vmIntrinsics::_isArray:
775 case vmIntrinsics::_isPrimitive:
776 case vmIntrinsics::_getSuperclass:
777 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
778
779 case vmIntrinsics::_floatToRawIntBits:
780 case vmIntrinsics::_floatToIntBits:
781 case vmIntrinsics::_intBitsToFloat:
782 case vmIntrinsics::_doubleToRawLongBits:
783 case vmIntrinsics::_doubleToLongBits:
784 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
785
786 case vmIntrinsics::_numberOfLeadingZeros_i:
787 case vmIntrinsics::_numberOfLeadingZeros_l:
788 case vmIntrinsics::_numberOfTrailingZeros_i:
789 case vmIntrinsics::_numberOfTrailingZeros_l:
790 case vmIntrinsics::_bitCount_i:
791 case vmIntrinsics::_bitCount_l:
792 case vmIntrinsics::_reverseBytes_i:
793 case vmIntrinsics::_reverseBytes_l:
794 case vmIntrinsics::_reverseBytes_s:
795 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
796
797 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
798
799 case vmIntrinsics::_Reference_get: return inline_reference_get();
800
801 case vmIntrinsics::_Class_cast: return inline_Class_cast();
802
803 case vmIntrinsics::_aescrypt_encryptBlock:
804 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
805
806 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
807 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
809
810 case vmIntrinsics::_counterMode_AESCrypt:
811 return inline_counterMode_AESCrypt(intrinsic_id());
812
813 case vmIntrinsics::_sha_implCompress:
814 case vmIntrinsics::_sha2_implCompress:
815 case vmIntrinsics::_sha5_implCompress:
816 return inline_sha_implCompress(intrinsic_id());
817
818 case vmIntrinsics::_digestBase_implCompressMB:
819 return inline_digestBase_implCompressMB(predicate);
820
821 case vmIntrinsics::_multiplyToLen:
822 return inline_multiplyToLen();
823
824 case vmIntrinsics::_squareToLen:
825 return inline_squareToLen();
826
827 case vmIntrinsics::_mulAdd:
828 return inline_mulAdd();
829
830 case vmIntrinsics::_montgomeryMultiply:
831 return inline_montgomeryMultiply();
832 case vmIntrinsics::_montgomerySquare:
833 return inline_montgomerySquare();
834
835 case vmIntrinsics::_vectorizedMismatch:
836 return inline_vectorizedMismatch();
837
838 case vmIntrinsics::_ghash_processBlocks:
839 return inline_ghash_processBlocks();
840
841 case vmIntrinsics::_encodeISOArray:
842 case vmIntrinsics::_encodeByteISOArray:
843 return inline_encodeISOArray();
844
845 case vmIntrinsics::_updateCRC32:
846 return inline_updateCRC32();
847 case vmIntrinsics::_updateBytesCRC32:
848 return inline_updateBytesCRC32();
849 case vmIntrinsics::_updateByteBufferCRC32:
850 return inline_updateByteBufferCRC32();
851
852 case vmIntrinsics::_updateBytesCRC32C:
853 return inline_updateBytesCRC32C();
854 case vmIntrinsics::_updateDirectByteBufferCRC32C:
855 return inline_updateDirectByteBufferCRC32C();
856
857 case vmIntrinsics::_updateBytesAdler32:
858 return inline_updateBytesAdler32();
859 case vmIntrinsics::_updateByteBufferAdler32:
860 return inline_updateByteBufferAdler32();
861
862 case vmIntrinsics::_profileBoolean:
863 return inline_profileBoolean();
864 case vmIntrinsics::_isCompileConstant:
865 return inline_isCompileConstant();
866
867 case vmIntrinsics::_hasNegatives:
868 return inline_hasNegatives();
869
870 case vmIntrinsics::_fmaD:
871 case vmIntrinsics::_fmaF:
872 return inline_fma(intrinsic_id());
873
874 case vmIntrinsics::_Continuation_getFP:
875 return inline_get_frame_pointer();
876
877 case vmIntrinsics::_Continuation_getSP:
878 case vmIntrinsics::_Continuation_getPC:
879 return inline_continuation(intrinsic_id());
880 case vmIntrinsics::_Continuation_doContinue:
881 return inline_continuation_do_continue();
882 case vmIntrinsics::_Continuation_doYield:
883 return inline_continuation_do_yield();
884 case vmIntrinsics::_Continuation_runLevel:
885 return inline_continuation_runLevel();
886
887 default:
888 // If you get here, it may be that someone has added a new intrinsic
889 // to the list in vmSymbols.hpp without implementing it here.
890 #ifndef PRODUCT
891 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
892 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
893 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
894 }
895 #endif
896 return false;
897 }
898 }
899
900 Node* LibraryCallKit::try_to_predicate(int predicate) {
901 if (!jvms()->has_method()) {
902 // Root JVMState has a null method.
903 assert(map()->memory()->Opcode() == Op_Parm, "");
904 // Insert the memory aliasing node
905 set_all_memory(reset_memory());
906 }
907 assert(merged_memory(), "");
908
909 switch (intrinsic_id()) {
910 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
911 return inline_cipherBlockChaining_AESCrypt_predicate(false);
912 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
913 return inline_cipherBlockChaining_AESCrypt_predicate(true);
914 case vmIntrinsics::_counterMode_AESCrypt:
915 return inline_counterMode_AESCrypt_predicate();
916 case vmIntrinsics::_digestBase_implCompressMB:
917 return inline_digestBase_implCompressMB_predicate(predicate);
918
919 default:
920 // If you get here, it may be that someone has added a new intrinsic
921 // to the list in vmSymbols.hpp without implementing it here.
922 #ifndef PRODUCT
923 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
924 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
925 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
926 }
927 #endif
928 Node* slow_ctl = control();
929 set_control(top()); // No fast path instrinsic
930 return slow_ctl;
931 }
932 }
933
934 //------------------------------set_result-------------------------------
935 // Helper function for finishing intrinsics.
936 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
937 record_for_igvn(region);
938 set_control(_gvn.transform(region));
939 set_result( _gvn.transform(value));
940 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
941 }
942
943 //------------------------------generate_guard---------------------------
944 // Helper function for generating guarded fast-slow graph structures.
945 // The given 'test', if true, guards a slow path. If the test fails
946 // then a fast path can be taken. (We generally hope it fails.)
947 // In all cases, GraphKit::control() is updated to the fast path.
948 // The returned value represents the control for the slow path.
949 // The return value is never 'top'; it is either a valid control
950 // or NULL if it is obvious that the slow path can never be taken.
951 // Also, if region and the slow control are not NULL, the slow edge
952 // is appended to the region.
953 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
954 if (stopped()) {
955 // Already short circuited.
956 return NULL;
957 }
958
959 // Build an if node and its projections.
960 // If test is true we take the slow path, which we assume is uncommon.
961 if (_gvn.type(test) == TypeInt::ZERO) {
962 // The slow branch is never taken. No need to build this guard.
963 return NULL;
964 }
965
966 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
967
968 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
969 if (if_slow == top()) {
970 // The slow branch is never taken. No need to build this guard.
971 return NULL;
972 }
973
974 if (region != NULL)
975 region->add_req(if_slow);
976
977 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
978 set_control(if_fast);
979
980 return if_slow;
981 }
982
983 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
984 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
985 }
986 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
987 return generate_guard(test, region, PROB_FAIR);
988 }
989
990 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
991 Node* *pos_index) {
992 if (stopped())
993 return NULL; // already stopped
994 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
995 return NULL; // index is already adequately typed
996 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
997 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
998 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
999 if (is_neg != NULL && pos_index != NULL) {
1000 // Emulate effect of Parse::adjust_map_after_if.
1001 Node* ccast = new CastIINode(index, TypeInt::POS);
1002 ccast->set_req(0, control());
1003 (*pos_index) = _gvn.transform(ccast);
1004 }
1005 return is_neg;
1006 }
1007
1008 // Make sure that 'position' is a valid limit index, in [0..length].
1009 // There are two equivalent plans for checking this:
1010 // A. (offset + copyLength) unsigned<= arrayLength
1011 // B. offset <= (arrayLength - copyLength)
1012 // We require that all of the values above, except for the sum and
1013 // difference, are already known to be non-negative.
1014 // Plan A is robust in the face of overflow, if offset and copyLength
1015 // are both hugely positive.
1016 //
1017 // Plan B is less direct and intuitive, but it does not overflow at
1018 // all, since the difference of two non-negatives is always
1019 // representable. Whenever Java methods must perform the equivalent
1020 // check they generally use Plan B instead of Plan A.
1021 // For the moment we use Plan A.
1022 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1023 Node* subseq_length,
1024 Node* array_length,
1025 RegionNode* region) {
1026 if (stopped())
1027 return NULL; // already stopped
1028 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1029 if (zero_offset && subseq_length->eqv_uncast(array_length))
1030 return NULL; // common case of whole-array copy
1031 Node* last = subseq_length;
1032 if (!zero_offset) // last += offset
1033 last = _gvn.transform(new AddINode(last, offset));
1034 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1035 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1036 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1037 return is_over;
1038 }
1039
1040 // Emit range checks for the given String.value byte array
1041 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1042 if (stopped()) {
1043 return; // already stopped
1044 }
1045 RegionNode* bailout = new RegionNode(1);
1046 record_for_igvn(bailout);
1047 if (char_count) {
1048 // Convert char count to byte count
1049 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1050 }
1051
1052 // Offset and count must not be negative
1053 generate_negative_guard(offset, bailout);
1054 generate_negative_guard(count, bailout);
1055 // Offset + count must not exceed length of array
1056 generate_limit_guard(offset, count, load_array_length(array), bailout);
1057
1058 if (bailout->req() > 1) {
1059 PreserveJVMState pjvms(this);
1060 set_control(_gvn.transform(bailout));
1061 uncommon_trap(Deoptimization::Reason_intrinsic,
1062 Deoptimization::Action_maybe_recompile);
1063 }
1064 }
1065
1066 //--------------------------generate_current_thread--------------------
1067 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1068 ciKlass* thread_klass = env()->Thread_klass();
1069 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1070 Node* thread = _gvn.transform(new ThreadLocalNode());
1071 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1072 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1073 tls_output = thread;
1074 return threadObj;
1075 }
1076
1077
1078 //------------------------------make_string_method_node------------------------
1079 // Helper method for String intrinsic functions. This version is called with
1080 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1081 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1082 // containing the lengths of str1 and str2.
1083 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1084 Node* result = NULL;
1085 switch (opcode) {
1086 case Op_StrIndexOf:
1087 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1088 str1_start, cnt1, str2_start, cnt2, ae);
1089 break;
1090 case Op_StrComp:
1091 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1092 str1_start, cnt1, str2_start, cnt2, ae);
1093 break;
1094 case Op_StrEquals:
1095 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1096 // Use the constant length if there is one because optimized match rule may exist.
1097 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1098 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1099 break;
1100 default:
1101 ShouldNotReachHere();
1102 return NULL;
1103 }
1104
1105 // All these intrinsics have checks.
1106 C->set_has_split_ifs(true); // Has chance for split-if optimization
1107 clear_upper_avx();
1108
1109 return _gvn.transform(result);
1110 }
1111
1112 //------------------------------inline_string_compareTo------------------------
1113 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1114 Node* arg1 = argument(0);
1115 Node* arg2 = argument(1);
1116
1117 // Get start addr and length of first argument
1118 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1119 Node* arg1_cnt = load_array_length(arg1);
1120
1121 // Get start addr and length of second argument
1122 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1123 Node* arg2_cnt = load_array_length(arg2);
1124
1125 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1126 set_result(result);
1127 return true;
1128 }
1129
1130 //------------------------------inline_string_equals------------------------
1131 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1132 Node* arg1 = argument(0);
1133 Node* arg2 = argument(1);
1134
1135 // paths (plus control) merge
1136 RegionNode* region = new RegionNode(3);
1137 Node* phi = new PhiNode(region, TypeInt::BOOL);
1138
1139 if (!stopped()) {
1140 // Get start addr and length of first argument
1141 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1142 Node* arg1_cnt = load_array_length(arg1);
1143
1144 // Get start addr and length of second argument
1145 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1146 Node* arg2_cnt = load_array_length(arg2);
1147
1148 // Check for arg1_cnt != arg2_cnt
1149 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1150 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1151 Node* if_ne = generate_slow_guard(bol, NULL);
1152 if (if_ne != NULL) {
1153 phi->init_req(2, intcon(0));
1154 region->init_req(2, if_ne);
1155 }
1156
1157 // Check for count == 0 is done by assembler code for StrEquals.
1158
1159 if (!stopped()) {
1160 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1161 phi->init_req(1, equals);
1162 region->init_req(1, control());
1163 }
1164 }
1165
1166 // post merge
1167 set_control(_gvn.transform(region));
1168 record_for_igvn(region);
1169
1170 set_result(_gvn.transform(phi));
1171 return true;
1172 }
1173
1174 //------------------------------inline_array_equals----------------------------
1175 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1176 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1177 Node* arg1 = argument(0);
1178 Node* arg2 = argument(1);
1179
1180 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1181 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1182 clear_upper_avx();
1183
1184 return true;
1185 }
1186
1187 //------------------------------inline_hasNegatives------------------------------
1188 bool LibraryCallKit::inline_hasNegatives() {
1189 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1190 return false;
1191 }
1192
1193 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1194 // no receiver since it is static method
1195 Node* ba = argument(0);
1196 Node* offset = argument(1);
1197 Node* len = argument(2);
1198
1199 // Range checks
1200 generate_string_range_check(ba, offset, len, false);
1201 if (stopped()) {
1202 return true;
1203 }
1204 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1205 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1206 set_result(_gvn.transform(result));
1207 return true;
1208 }
1209
1210 bool LibraryCallKit::inline_preconditions_checkIndex() {
1211 Node* index = argument(0);
1212 Node* length = argument(1);
1213 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1214 return false;
1215 }
1216
1217 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1218 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1219
1220 {
1221 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1222 uncommon_trap(Deoptimization::Reason_intrinsic,
1223 Deoptimization::Action_make_not_entrant);
1224 }
1225
1226 if (stopped()) {
1227 return false;
1228 }
1229
1230 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1231 BoolTest::mask btest = BoolTest::lt;
1232 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1233 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1234 _gvn.set_type(rc, rc->Value(&_gvn));
1235 if (!rc_bool->is_Con()) {
1236 record_for_igvn(rc);
1237 }
1238 set_control(_gvn.transform(new IfTrueNode(rc)));
1239 {
1240 PreserveJVMState pjvms(this);
1241 set_control(_gvn.transform(new IfFalseNode(rc)));
1242 uncommon_trap(Deoptimization::Reason_range_check,
1243 Deoptimization::Action_make_not_entrant);
1244 }
1245
1246 if (stopped()) {
1247 return false;
1248 }
1249
1250 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1251 result->set_req(0, control());
1252 result = _gvn.transform(result);
1253 set_result(result);
1254 replace_in_map(index, result);
1255 clear_upper_avx();
1256 return true;
1257 }
1258
1259 //------------------------------inline_string_indexOf------------------------
1260 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1261 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1262 return false;
1263 }
1264 Node* src = argument(0);
1265 Node* tgt = argument(1);
1266
1267 // Make the merge point
1268 RegionNode* result_rgn = new RegionNode(4);
1269 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1270
1271 // Get start addr and length of source string
1272 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1273 Node* src_count = load_array_length(src);
1274
1275 // Get start addr and length of substring
1276 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1277 Node* tgt_count = load_array_length(tgt);
1278
1279 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1280 // Divide src size by 2 if String is UTF16 encoded
1281 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1282 }
1283 if (ae == StrIntrinsicNode::UU) {
1284 // Divide substring size by 2 if String is UTF16 encoded
1285 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1286 }
1287
1288 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1289 if (result != NULL) {
1290 result_phi->init_req(3, result);
1291 result_rgn->init_req(3, control());
1292 }
1293 set_control(_gvn.transform(result_rgn));
1294 record_for_igvn(result_rgn);
1295 set_result(_gvn.transform(result_phi));
1296
1297 return true;
1298 }
1299
1300 //-----------------------------inline_string_indexOf-----------------------
1301 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1302 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1303 return false;
1304 }
1305 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1306 return false;
1307 }
1308 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1309 Node* src = argument(0); // byte[]
1310 Node* src_count = argument(1); // char count
1311 Node* tgt = argument(2); // byte[]
1312 Node* tgt_count = argument(3); // char count
1313 Node* from_index = argument(4); // char index
1314
1315 // Multiply byte array index by 2 if String is UTF16 encoded
1316 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1317 src_count = _gvn.transform(new SubINode(src_count, from_index));
1318 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1319 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1320
1321 // Range checks
1322 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1323 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1324 if (stopped()) {
1325 return true;
1326 }
1327
1328 RegionNode* region = new RegionNode(5);
1329 Node* phi = new PhiNode(region, TypeInt::INT);
1330
1331 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1332 if (result != NULL) {
1333 // The result is index relative to from_index if substring was found, -1 otherwise.
1334 // Generate code which will fold into cmove.
1335 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1336 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1337
1338 Node* if_lt = generate_slow_guard(bol, NULL);
1339 if (if_lt != NULL) {
1340 // result == -1
1341 phi->init_req(3, result);
1342 region->init_req(3, if_lt);
1343 }
1344 if (!stopped()) {
1345 result = _gvn.transform(new AddINode(result, from_index));
1346 phi->init_req(4, result);
1347 region->init_req(4, control());
1348 }
1349 }
1350
1351 set_control(_gvn.transform(region));
1352 record_for_igvn(region);
1353 set_result(_gvn.transform(phi));
1354 clear_upper_avx();
1355
1356 return true;
1357 }
1358
1359 // Create StrIndexOfNode with fast path checks
1360 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1361 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1362 // Check for substr count > string count
1363 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1364 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1365 Node* if_gt = generate_slow_guard(bol, NULL);
1366 if (if_gt != NULL) {
1367 phi->init_req(1, intcon(-1));
1368 region->init_req(1, if_gt);
1369 }
1370 if (!stopped()) {
1371 // Check for substr count == 0
1372 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1373 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1374 Node* if_zero = generate_slow_guard(bol, NULL);
1375 if (if_zero != NULL) {
1376 phi->init_req(2, intcon(0));
1377 region->init_req(2, if_zero);
1378 }
1379 }
1380 if (!stopped()) {
1381 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1382 }
1383 return NULL;
1384 }
1385
1386 //-----------------------------inline_string_indexOfChar-----------------------
1387 bool LibraryCallKit::inline_string_indexOfChar() {
1388 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1389 return false;
1390 }
1391 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1392 return false;
1393 }
1394 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1395 Node* src = argument(0); // byte[]
1396 Node* tgt = argument(1); // tgt is int ch
1397 Node* from_index = argument(2);
1398 Node* max = argument(3);
1399
1400 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1401 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1402 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1403
1404 // Range checks
1405 generate_string_range_check(src, src_offset, src_count, true);
1406 if (stopped()) {
1407 return true;
1408 }
1409
1410 RegionNode* region = new RegionNode(3);
1411 Node* phi = new PhiNode(region, TypeInt::INT);
1412
1413 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1414 C->set_has_split_ifs(true); // Has chance for split-if optimization
1415 _gvn.transform(result);
1416
1417 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1418 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1419
1420 Node* if_lt = generate_slow_guard(bol, NULL);
1421 if (if_lt != NULL) {
1422 // result == -1
1423 phi->init_req(2, result);
1424 region->init_req(2, if_lt);
1425 }
1426 if (!stopped()) {
1427 result = _gvn.transform(new AddINode(result, from_index));
1428 phi->init_req(1, result);
1429 region->init_req(1, control());
1430 }
1431 set_control(_gvn.transform(region));
1432 record_for_igvn(region);
1433 set_result(_gvn.transform(phi));
1434
1435 return true;
1436 }
1437 //---------------------------inline_string_copy---------------------
1438 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1439 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1440 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1441 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1442 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1443 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1444 bool LibraryCallKit::inline_string_copy(bool compress) {
1445 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1446 return false;
1447 }
1448 int nargs = 5; // 2 oops, 3 ints
1449 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1450
1451 Node* src = argument(0);
1452 Node* src_offset = argument(1);
1453 Node* dst = argument(2);
1454 Node* dst_offset = argument(3);
1455 Node* length = argument(4);
1456
1457 // Check for allocation before we add nodes that would confuse
1458 // tightly_coupled_allocation()
1459 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1460
1461 // Figure out the size and type of the elements we will be copying.
1462 const Type* src_type = src->Value(&_gvn);
1463 const Type* dst_type = dst->Value(&_gvn);
1464 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1465 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1466 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1467 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1468 "Unsupported array types for inline_string_copy");
1469
1470 // Convert char[] offsets to byte[] offsets
1471 bool convert_src = (compress && src_elem == T_BYTE);
1472 bool convert_dst = (!compress && dst_elem == T_BYTE);
1473 if (convert_src) {
1474 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1475 } else if (convert_dst) {
1476 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1477 }
1478
1479 // Range checks
1480 generate_string_range_check(src, src_offset, length, convert_src);
1481 generate_string_range_check(dst, dst_offset, length, convert_dst);
1482 if (stopped()) {
1483 return true;
1484 }
1485
1486 Node* src_start = array_element_address(src, src_offset, src_elem);
1487 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1488 // 'src_start' points to src array + scaled offset
1489 // 'dst_start' points to dst array + scaled offset
1490 Node* count = NULL;
1491 if (compress) {
1492 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1493 } else {
1494 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1495 }
1496
1497 if (alloc != NULL) {
1498 if (alloc->maybe_set_complete(&_gvn)) {
1499 // "You break it, you buy it."
1500 InitializeNode* init = alloc->initialization();
1501 assert(init->is_complete(), "we just did this");
1502 init->set_complete_with_arraycopy();
1503 assert(dst->is_CheckCastPP(), "sanity");
1504 assert(dst->in(0)->in(0) == init, "dest pinned");
1505 }
1506 // Do not let stores that initialize this object be reordered with
1507 // a subsequent store that would make this object accessible by
1508 // other threads.
1509 // Record what AllocateNode this StoreStore protects so that
1510 // escape analysis can go from the MemBarStoreStoreNode to the
1511 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1512 // based on the escape status of the AllocateNode.
1513 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1514 }
1515 if (compress) {
1516 set_result(_gvn.transform(count));
1517 }
1518 clear_upper_avx();
1519
1520 return true;
1521 }
1522
1523 #ifdef _LP64
1524 #define XTOP ,top() /*additional argument*/
1525 #else //_LP64
1526 #define XTOP /*no additional argument*/
1527 #endif //_LP64
1528
1529 //------------------------inline_string_toBytesU--------------------------
1530 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1531 bool LibraryCallKit::inline_string_toBytesU() {
1532 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1533 return false;
1534 }
1535 // Get the arguments.
1536 Node* value = argument(0);
1537 Node* offset = argument(1);
1538 Node* length = argument(2);
1539
1540 Node* newcopy = NULL;
1541
1542 // Set the original stack and the reexecute bit for the interpreter to reexecute
1543 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1544 { PreserveReexecuteState preexecs(this);
1545 jvms()->set_should_reexecute(true);
1546
1547 // Check if a null path was taken unconditionally.
1548 value = null_check(value);
1549
1550 RegionNode* bailout = new RegionNode(1);
1551 record_for_igvn(bailout);
1552
1553 // Range checks
1554 generate_negative_guard(offset, bailout);
1555 generate_negative_guard(length, bailout);
1556 generate_limit_guard(offset, length, load_array_length(value), bailout);
1557 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1558 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1559
1560 if (bailout->req() > 1) {
1561 PreserveJVMState pjvms(this);
1562 set_control(_gvn.transform(bailout));
1563 uncommon_trap(Deoptimization::Reason_intrinsic,
1564 Deoptimization::Action_maybe_recompile);
1565 }
1566 if (stopped()) {
1567 return true;
1568 }
1569
1570 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1571 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1572 newcopy = new_array(klass_node, size, 0); // no arguments to push
1573 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1574
1575 // Calculate starting addresses.
1576 Node* src_start = array_element_address(value, offset, T_CHAR);
1577 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1578
1579 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1580 const TypeInt* toffset = gvn().type(offset)->is_int();
1581 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1582
1583 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1584 const char* copyfunc_name = "arraycopy";
1585 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1586 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1587 OptoRuntime::fast_arraycopy_Type(),
1588 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1589 src_start, dst_start, ConvI2X(length) XTOP);
1590 // Do not let reads from the cloned object float above the arraycopy.
1591 if (alloc != NULL) {
1592 if (alloc->maybe_set_complete(&_gvn)) {
1593 // "You break it, you buy it."
1594 InitializeNode* init = alloc->initialization();
1595 assert(init->is_complete(), "we just did this");
1596 init->set_complete_with_arraycopy();
1597 assert(newcopy->is_CheckCastPP(), "sanity");
1598 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1599 }
1600 // Do not let stores that initialize this object be reordered with
1601 // a subsequent store that would make this object accessible by
1602 // other threads.
1603 // Record what AllocateNode this StoreStore protects so that
1604 // escape analysis can go from the MemBarStoreStoreNode to the
1605 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1606 // based on the escape status of the AllocateNode.
1607 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1608 } else {
1609 insert_mem_bar(Op_MemBarCPUOrder);
1610 }
1611 } // original reexecute is set back here
1612
1613 C->set_has_split_ifs(true); // Has chance for split-if optimization
1614 if (!stopped()) {
1615 set_result(newcopy);
1616 }
1617 clear_upper_avx();
1618
1619 return true;
1620 }
1621
1622 //------------------------inline_string_getCharsU--------------------------
1623 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1624 bool LibraryCallKit::inline_string_getCharsU() {
1625 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1626 return false;
1627 }
1628
1629 // Get the arguments.
1630 Node* src = argument(0);
1631 Node* src_begin = argument(1);
1632 Node* src_end = argument(2); // exclusive offset (i < src_end)
1633 Node* dst = argument(3);
1634 Node* dst_begin = argument(4);
1635
1636 // Check for allocation before we add nodes that would confuse
1637 // tightly_coupled_allocation()
1638 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1639
1640 // Check if a null path was taken unconditionally.
1641 src = null_check(src);
1642 dst = null_check(dst);
1643 if (stopped()) {
1644 return true;
1645 }
1646
1647 // Get length and convert char[] offset to byte[] offset
1648 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1649 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1650
1651 // Range checks
1652 generate_string_range_check(src, src_begin, length, true);
1653 generate_string_range_check(dst, dst_begin, length, false);
1654 if (stopped()) {
1655 return true;
1656 }
1657
1658 if (!stopped()) {
1659 // Calculate starting addresses.
1660 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1661 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1662
1663 // Check if array addresses are aligned to HeapWordSize
1664 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1665 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1666 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1667 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1668
1669 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1670 const char* copyfunc_name = "arraycopy";
1671 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1672 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1673 OptoRuntime::fast_arraycopy_Type(),
1674 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1675 src_start, dst_start, ConvI2X(length) XTOP);
1676 // Do not let reads from the cloned object float above the arraycopy.
1677 if (alloc != NULL) {
1678 if (alloc->maybe_set_complete(&_gvn)) {
1679 // "You break it, you buy it."
1680 InitializeNode* init = alloc->initialization();
1681 assert(init->is_complete(), "we just did this");
1682 init->set_complete_with_arraycopy();
1683 assert(dst->is_CheckCastPP(), "sanity");
1684 assert(dst->in(0)->in(0) == init, "dest pinned");
1685 }
1686 // Do not let stores that initialize this object be reordered with
1687 // a subsequent store that would make this object accessible by
1688 // other threads.
1689 // Record what AllocateNode this StoreStore protects so that
1690 // escape analysis can go from the MemBarStoreStoreNode to the
1691 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1692 // based on the escape status of the AllocateNode.
1693 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1694 } else {
1695 insert_mem_bar(Op_MemBarCPUOrder);
1696 }
1697 }
1698
1699 C->set_has_split_ifs(true); // Has chance for split-if optimization
1700 return true;
1701 }
1702
1703 //----------------------inline_string_char_access----------------------------
1704 // Store/Load char to/from byte[] array.
1705 // static void StringUTF16.putChar(byte[] val, int index, int c)
1706 // static char StringUTF16.getChar(byte[] val, int index)
1707 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1708 Node* value = argument(0);
1709 Node* index = argument(1);
1710 Node* ch = is_store ? argument(2) : NULL;
1711
1712 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1713 // correctly requires matched array shapes.
1714 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1715 "sanity: byte[] and char[] bases agree");
1716 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1717 "sanity: byte[] and char[] scales agree");
1718
1719 // Bail when getChar over constants is requested: constant folding would
1720 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1721 // Java method would constant fold nicely instead.
1722 if (!is_store && value->is_Con() && index->is_Con()) {
1723 return false;
1724 }
1725
1726 Node* adr = array_element_address(value, index, T_CHAR);
1727 if (adr->is_top()) {
1728 return false;
1729 }
1730 if (is_store) {
1731 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1732 false, false, true /* mismatched */);
1733 } else {
1734 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1735 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1736 set_result(ch);
1737 }
1738 return true;
1739 }
1740
1741 //--------------------------round_double_node--------------------------------
1742 // Round a double node if necessary.
1743 Node* LibraryCallKit::round_double_node(Node* n) {
1744 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1745 n = _gvn.transform(new RoundDoubleNode(0, n));
1746 return n;
1747 }
1748
1749 //------------------------------inline_math-----------------------------------
1750 // public static double Math.abs(double)
1751 // public static double Math.sqrt(double)
1752 // public static double Math.log(double)
1753 // public static double Math.log10(double)
1754 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1755 Node* arg = round_double_node(argument(0));
1756 Node* n = NULL;
1757 switch (id) {
1758 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1759 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1760 default: fatal_unexpected_iid(id); break;
1761 }
1762 set_result(_gvn.transform(n));
1763 return true;
1764 }
1765
1766 //------------------------------runtime_math-----------------------------
1767 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1768 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1769 "must be (DD)D or (D)D type");
1770
1771 // Inputs
1772 Node* a = round_double_node(argument(0));
1773 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1774
1775 const TypePtr* no_memory_effects = NULL;
1776 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1777 no_memory_effects,
1778 a, top(), b, b ? top() : NULL);
1779 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1780 #ifdef ASSERT
1781 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1782 assert(value_top == top(), "second value must be top");
1783 #endif
1784
1785 set_result(value);
1786 return true;
1787 }
1788
1789 //------------------------------inline_math_native-----------------------------
1790 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1791 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1792 switch (id) {
1793 // These intrinsics are not properly supported on all hardware
1794 case vmIntrinsics::_dsin:
1795 return StubRoutines::dsin() != NULL ?
1796 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1797 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1798 case vmIntrinsics::_dcos:
1799 return StubRoutines::dcos() != NULL ?
1800 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1801 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1802 case vmIntrinsics::_dtan:
1803 return StubRoutines::dtan() != NULL ?
1804 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1805 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1806 case vmIntrinsics::_dlog:
1807 return StubRoutines::dlog() != NULL ?
1808 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1809 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1810 case vmIntrinsics::_dlog10:
1811 return StubRoutines::dlog10() != NULL ?
1812 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1813 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1814
1815 // These intrinsics are supported on all hardware
1816 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1817 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1818
1819 case vmIntrinsics::_dexp:
1820 return StubRoutines::dexp() != NULL ?
1821 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1822 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1823 case vmIntrinsics::_dpow: {
1824 Node* exp = round_double_node(argument(2));
1825 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1826 if (d != NULL && d->getd() == 2.0) {
1827 // Special case: pow(x, 2.0) => x * x
1828 Node* base = round_double_node(argument(0));
1829 set_result(_gvn.transform(new MulDNode(base, base)));
1830 return true;
1831 }
1832 return StubRoutines::dexp() != NULL ?
1833 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1834 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1835 }
1836 #undef FN_PTR
1837
1838 // These intrinsics are not yet correctly implemented
1839 case vmIntrinsics::_datan2:
1840 return false;
1841
1842 default:
1843 fatal_unexpected_iid(id);
1844 return false;
1845 }
1846 }
1847
1848 static bool is_simple_name(Node* n) {
1849 return (n->req() == 1 // constant
1850 || (n->is_Type() && n->as_Type()->type()->singleton())
1851 || n->is_Proj() // parameter or return value
1852 || n->is_Phi() // local of some sort
1853 );
1854 }
1855
1856 //----------------------------inline_notify-----------------------------------*
1857 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1858 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1859 address func;
1860 if (id == vmIntrinsics::_notify) {
1861 func = OptoRuntime::monitor_notify_Java();
1862 } else {
1863 func = OptoRuntime::monitor_notifyAll_Java();
1864 }
1865 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1866 make_slow_call_ex(call, env()->Throwable_klass(), false);
1867 return true;
1868 }
1869
1870
1871 //----------------------------inline_min_max-----------------------------------
1872 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1873 set_result(generate_min_max(id, argument(0), argument(1)));
1874 return true;
1875 }
1876
1877 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1878 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1879 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1880 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1881 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1882
1883 {
1884 PreserveJVMState pjvms(this);
1885 PreserveReexecuteState preexecs(this);
1886 jvms()->set_should_reexecute(true);
1887
1888 set_control(slow_path);
1889 set_i_o(i_o());
1890
1891 uncommon_trap(Deoptimization::Reason_intrinsic,
1892 Deoptimization::Action_none);
1893 }
1894
1895 set_control(fast_path);
1896 set_result(math);
1897 }
1898
1899 template <typename OverflowOp>
1900 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1901 typedef typename OverflowOp::MathOp MathOp;
1902
1903 MathOp* mathOp = new MathOp(arg1, arg2);
1904 Node* operation = _gvn.transform( mathOp );
1905 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1906 inline_math_mathExact(operation, ofcheck);
1907 return true;
1908 }
1909
1910 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1911 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1912 }
1913
1914 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1915 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1916 }
1917
1918 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1919 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1920 }
1921
1922 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1923 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1924 }
1925
1926 bool LibraryCallKit::inline_math_negateExactI() {
1927 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1928 }
1929
1930 bool LibraryCallKit::inline_math_negateExactL() {
1931 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1932 }
1933
1934 bool LibraryCallKit::inline_math_multiplyExactI() {
1935 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1936 }
1937
1938 bool LibraryCallKit::inline_math_multiplyExactL() {
1939 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1940 }
1941
1942 bool LibraryCallKit::inline_math_multiplyHigh() {
1943 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1944 return true;
1945 }
1946
1947 Node*
1948 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1949 // These are the candidate return value:
1950 Node* xvalue = x0;
1951 Node* yvalue = y0;
1952
1953 if (xvalue == yvalue) {
1954 return xvalue;
1955 }
1956
1957 bool want_max = (id == vmIntrinsics::_max);
1958
1959 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1960 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1961 if (txvalue == NULL || tyvalue == NULL) return top();
1962 // This is not really necessary, but it is consistent with a
1963 // hypothetical MaxINode::Value method:
1964 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1965
1966 // %%% This folding logic should (ideally) be in a different place.
1967 // Some should be inside IfNode, and there to be a more reliable
1968 // transformation of ?: style patterns into cmoves. We also want
1969 // more powerful optimizations around cmove and min/max.
1970
1971 // Try to find a dominating comparison of these guys.
1972 // It can simplify the index computation for Arrays.copyOf
1973 // and similar uses of System.arraycopy.
1974 // First, compute the normalized version of CmpI(x, y).
1975 int cmp_op = Op_CmpI;
1976 Node* xkey = xvalue;
1977 Node* ykey = yvalue;
1978 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1979 if (ideal_cmpxy->is_Cmp()) {
1980 // E.g., if we have CmpI(length - offset, count),
1981 // it might idealize to CmpI(length, count + offset)
1982 cmp_op = ideal_cmpxy->Opcode();
1983 xkey = ideal_cmpxy->in(1);
1984 ykey = ideal_cmpxy->in(2);
1985 }
1986
1987 // Start by locating any relevant comparisons.
1988 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1989 Node* cmpxy = NULL;
1990 Node* cmpyx = NULL;
1991 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1992 Node* cmp = start_from->fast_out(k);
1993 if (cmp->outcnt() > 0 && // must have prior uses
1994 cmp->in(0) == NULL && // must be context-independent
1995 cmp->Opcode() == cmp_op) { // right kind of compare
1996 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1997 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1998 }
1999 }
2000
2001 const int NCMPS = 2;
2002 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2003 int cmpn;
2004 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2005 if (cmps[cmpn] != NULL) break; // find a result
2006 }
2007 if (cmpn < NCMPS) {
2008 // Look for a dominating test that tells us the min and max.
2009 int depth = 0; // Limit search depth for speed
2010 Node* dom = control();
2011 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2012 if (++depth >= 100) break;
2013 Node* ifproj = dom;
2014 if (!ifproj->is_Proj()) continue;
2015 Node* iff = ifproj->in(0);
2016 if (!iff->is_If()) continue;
2017 Node* bol = iff->in(1);
2018 if (!bol->is_Bool()) continue;
2019 Node* cmp = bol->in(1);
2020 if (cmp == NULL) continue;
2021 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2022 if (cmps[cmpn] == cmp) break;
2023 if (cmpn == NCMPS) continue;
2024 BoolTest::mask btest = bol->as_Bool()->_test._test;
2025 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2026 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2027 // At this point, we know that 'x btest y' is true.
2028 switch (btest) {
2029 case BoolTest::eq:
2030 // They are proven equal, so we can collapse the min/max.
2031 // Either value is the answer. Choose the simpler.
2032 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2033 return yvalue;
2034 return xvalue;
2035 case BoolTest::lt: // x < y
2036 case BoolTest::le: // x <= y
2037 return (want_max ? yvalue : xvalue);
2038 case BoolTest::gt: // x > y
2039 case BoolTest::ge: // x >= y
2040 return (want_max ? xvalue : yvalue);
2041 default:
2042 break;
2043 }
2044 }
2045 }
2046
2047 // We failed to find a dominating test.
2048 // Let's pick a test that might GVN with prior tests.
2049 Node* best_bol = NULL;
2050 BoolTest::mask best_btest = BoolTest::illegal;
2051 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2052 Node* cmp = cmps[cmpn];
2053 if (cmp == NULL) continue;
2054 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2055 Node* bol = cmp->fast_out(j);
2056 if (!bol->is_Bool()) continue;
2057 BoolTest::mask btest = bol->as_Bool()->_test._test;
2058 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2059 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2060 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2061 best_bol = bol->as_Bool();
2062 best_btest = btest;
2063 }
2064 }
2065 }
2066
2067 Node* answer_if_true = NULL;
2068 Node* answer_if_false = NULL;
2069 switch (best_btest) {
2070 default:
2071 if (cmpxy == NULL)
2072 cmpxy = ideal_cmpxy;
2073 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2074 // and fall through:
2075 case BoolTest::lt: // x < y
2076 case BoolTest::le: // x <= y
2077 answer_if_true = (want_max ? yvalue : xvalue);
2078 answer_if_false = (want_max ? xvalue : yvalue);
2079 break;
2080 case BoolTest::gt: // x > y
2081 case BoolTest::ge: // x >= y
2082 answer_if_true = (want_max ? xvalue : yvalue);
2083 answer_if_false = (want_max ? yvalue : xvalue);
2084 break;
2085 }
2086
2087 jint hi, lo;
2088 if (want_max) {
2089 // We can sharpen the minimum.
2090 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2091 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2092 } else {
2093 // We can sharpen the maximum.
2094 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2095 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2096 }
2097
2098 // Use a flow-free graph structure, to avoid creating excess control edges
2099 // which could hinder other optimizations.
2100 // Since Math.min/max is often used with arraycopy, we want
2101 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2102 Node* cmov = CMoveNode::make(NULL, best_bol,
2103 answer_if_false, answer_if_true,
2104 TypeInt::make(lo, hi, widen));
2105
2106 return _gvn.transform(cmov);
2107
2108 /*
2109 // This is not as desirable as it may seem, since Min and Max
2110 // nodes do not have a full set of optimizations.
2111 // And they would interfere, anyway, with 'if' optimizations
2112 // and with CMoveI canonical forms.
2113 switch (id) {
2114 case vmIntrinsics::_min:
2115 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2116 case vmIntrinsics::_max:
2117 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2118 default:
2119 ShouldNotReachHere();
2120 }
2121 */
2122 }
2123
2124 inline int
2125 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2126 const TypePtr* base_type = TypePtr::NULL_PTR;
2127 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2128 if (base_type == NULL) {
2129 // Unknown type.
2130 return Type::AnyPtr;
2131 } else if (base_type == TypePtr::NULL_PTR) {
2132 // Since this is a NULL+long form, we have to switch to a rawptr.
2133 base = _gvn.transform(new CastX2PNode(offset));
2134 offset = MakeConX(0);
2135 return Type::RawPtr;
2136 } else if (base_type->base() == Type::RawPtr) {
2137 return Type::RawPtr;
2138 } else if (base_type->isa_oopptr()) {
2139 // Base is never null => always a heap address.
2140 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2141 return Type::OopPtr;
2142 }
2143 // Offset is small => always a heap address.
2144 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2145 if (offset_type != NULL &&
2146 base_type->offset() == 0 && // (should always be?)
2147 offset_type->_lo >= 0 &&
2148 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2149 return Type::OopPtr;
2150 } else if (type == T_OBJECT) {
2151 // off heap access to an oop doesn't make any sense. Has to be on
2152 // heap.
2153 return Type::OopPtr;
2154 }
2155 // Otherwise, it might either be oop+off or NULL+addr.
2156 return Type::AnyPtr;
2157 } else {
2158 // No information:
2159 return Type::AnyPtr;
2160 }
2161 }
2162
2163 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2164 Node* uncasted_base = base;
2165 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2166 if (kind == Type::RawPtr) {
2167 return basic_plus_adr(top(), uncasted_base, offset);
2168 } else if (kind == Type::AnyPtr) {
2169 assert(base == uncasted_base, "unexpected base change");
2170 if (can_cast) {
2171 if (!_gvn.type(base)->speculative_maybe_null() &&
2172 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2173 // According to profiling, this access is always on
2174 // heap. Casting the base to not null and thus avoiding membars
2175 // around the access should allow better optimizations
2176 Node* null_ctl = top();
2177 base = null_check_oop(base, &null_ctl, true, true, true);
2178 assert(null_ctl->is_top(), "no null control here");
2179 return basic_plus_adr(base, offset);
2180 } else if (_gvn.type(base)->speculative_always_null() &&
2181 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2182 // According to profiling, this access is always off
2183 // heap.
2184 base = null_assert(base);
2185 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2186 offset = MakeConX(0);
2187 return basic_plus_adr(top(), raw_base, offset);
2188 }
2189 }
2190 // We don't know if it's an on heap or off heap access. Fall back
2191 // to raw memory access.
2192 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2193 return basic_plus_adr(top(), raw, offset);
2194 } else {
2195 assert(base == uncasted_base, "unexpected base change");
2196 // We know it's an on heap access so base can't be null
2197 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2198 base = must_be_not_null(base, true);
2199 }
2200 return basic_plus_adr(base, offset);
2201 }
2202 }
2203
2204 //--------------------------inline_number_methods-----------------------------
2205 // inline int Integer.numberOfLeadingZeros(int)
2206 // inline int Long.numberOfLeadingZeros(long)
2207 //
2208 // inline int Integer.numberOfTrailingZeros(int)
2209 // inline int Long.numberOfTrailingZeros(long)
2210 //
2211 // inline int Integer.bitCount(int)
2212 // inline int Long.bitCount(long)
2213 //
2214 // inline char Character.reverseBytes(char)
2215 // inline short Short.reverseBytes(short)
2216 // inline int Integer.reverseBytes(int)
2217 // inline long Long.reverseBytes(long)
2218 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2219 Node* arg = argument(0);
2220 Node* n = NULL;
2221 switch (id) {
2222 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2223 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2224 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2225 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2226 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2227 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2228 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2229 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2230 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2231 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2232 default: fatal_unexpected_iid(id); break;
2233 }
2234 set_result(_gvn.transform(n));
2235 return true;
2236 }
2237
2238 //----------------------------inline_unsafe_access----------------------------
2239
2240 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2241 // Attempt to infer a sharper value type from the offset and base type.
2242 ciKlass* sharpened_klass = NULL;
2243
2244 // See if it is an instance field, with an object type.
2245 if (alias_type->field() != NULL) {
2246 if (alias_type->field()->type()->is_klass()) {
2247 sharpened_klass = alias_type->field()->type()->as_klass();
2248 }
2249 }
2250
2251 // See if it is a narrow oop array.
2252 if (adr_type->isa_aryptr()) {
2253 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2254 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2255 if (elem_type != NULL) {
2256 sharpened_klass = elem_type->klass();
2257 }
2258 }
2259 }
2260
2261 // The sharpened class might be unloaded if there is no class loader
2262 // contraint in place.
2263 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2264 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2265
2266 #ifndef PRODUCT
2267 if (C->print_intrinsics() || C->print_inlining()) {
2268 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2269 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2270 }
2271 #endif
2272 // Sharpen the value type.
2273 return tjp;
2274 }
2275 return NULL;
2276 }
2277
2278 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2279 switch (kind) {
2280 case Relaxed:
2281 return MO_UNORDERED;
2282 case Opaque:
2283 return MO_RELAXED;
2284 case Acquire:
2285 return MO_ACQUIRE;
2286 case Release:
2287 return MO_RELEASE;
2288 case Volatile:
2289 return MO_SEQ_CST;
2290 default:
2291 ShouldNotReachHere();
2292 return 0;
2293 }
2294 }
2295
2296 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2297 if (callee()->is_static()) return false; // caller must have the capability!
2298 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2299 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2300 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2301 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2302
2303 if (type == T_OBJECT || type == T_ARRAY) {
2304 decorators |= ON_UNKNOWN_OOP_REF;
2305 }
2306
2307 if (unaligned) {
2308 decorators |= C2_UNALIGNED;
2309 }
2310
2311 #ifndef PRODUCT
2312 {
2313 ResourceMark rm;
2314 // Check the signatures.
2315 ciSignature* sig = callee()->signature();
2316 #ifdef ASSERT
2317 if (!is_store) {
2318 // Object getObject(Object base, int/long offset), etc.
2319 BasicType rtype = sig->return_type()->basic_type();
2320 assert(rtype == type, "getter must return the expected value");
2321 assert(sig->count() == 2, "oop getter has 2 arguments");
2322 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2323 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2324 } else {
2325 // void putObject(Object base, int/long offset, Object x), etc.
2326 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2327 assert(sig->count() == 3, "oop putter has 3 arguments");
2328 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2329 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2330 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2331 assert(vtype == type, "putter must accept the expected value");
2332 }
2333 #endif // ASSERT
2334 }
2335 #endif //PRODUCT
2336
2337 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2338
2339 Node* receiver = argument(0); // type: oop
2340
2341 // Build address expression.
2342 Node* adr;
2343 Node* heap_base_oop = top();
2344 Node* offset = top();
2345 Node* val;
2346
2347 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2348 Node* base = argument(1); // type: oop
2349 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2350 offset = argument(2); // type: long
2351 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2352 // to be plain byte offsets, which are also the same as those accepted
2353 // by oopDesc::field_addr.
2354 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2355 "fieldOffset must be byte-scaled");
2356 // 32-bit machines ignore the high half!
2357 offset = ConvL2X(offset);
2358 adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2359
2360 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2361 heap_base_oop = base;
2362 } else if (type == T_OBJECT) {
2363 return false; // off-heap oop accesses are not supported
2364 }
2365
2366 // Can base be NULL? Otherwise, always on-heap access.
2367 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2368
2369 if (!can_access_non_heap) {
2370 decorators |= IN_HEAP;
2371 }
2372
2373 val = is_store ? argument(4) : NULL;
2374
2375 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2376
2377 // Try to categorize the address.
2378 Compile::AliasType* alias_type = C->alias_type(adr_type);
2379 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2380
2381 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2382 alias_type->adr_type() == TypeAryPtr::RANGE) {
2383 return false; // not supported
2384 }
2385
2386 bool mismatched = false;
2387 BasicType bt = alias_type->basic_type();
2388 if (bt != T_ILLEGAL) {
2389 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2390 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2391 // Alias type doesn't differentiate between byte[] and boolean[]).
2392 // Use address type to get the element type.
2393 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2394 }
2395 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2396 // accessing an array field with getObject is not a mismatch
2397 bt = T_OBJECT;
2398 }
2399 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2400 // Don't intrinsify mismatched object accesses
2401 return false;
2402 }
2403 mismatched = (bt != type);
2404 } else if (alias_type->adr_type()->isa_oopptr()) {
2405 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2406 }
2407
2408 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2409
2410 if (mismatched) {
2411 decorators |= C2_MISMATCHED;
2412 }
2413
2414 // First guess at the value type.
2415 const Type *value_type = Type::get_const_basic_type(type);
2416
2417 // Figure out the memory ordering.
2418 decorators |= mo_decorator_for_access_kind(kind);
2419
2420 if (!is_store && type == T_OBJECT) {
2421 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2422 if (tjp != NULL) {
2423 value_type = tjp;
2424 }
2425 }
2426
2427 receiver = null_check(receiver);
2428 if (stopped()) {
2429 return true;
2430 }
2431 // Heap pointers get a null-check from the interpreter,
2432 // as a courtesy. However, this is not guaranteed by Unsafe,
2433 // and it is not possible to fully distinguish unintended nulls
2434 // from intended ones in this API.
2435
2436 if (!is_store) {
2437 Node* p = NULL;
2438 // Try to constant fold a load from a constant field
2439 ciField* field = alias_type->field();
2440 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2441 // final or stable field
2442 p = make_constant_from_field(field, heap_base_oop);
2443 }
2444
2445 if (p == NULL) { // Could not constant fold the load
2446 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2447 // Normalize the value returned by getBoolean in the following cases
2448 if (type == T_BOOLEAN &&
2449 (mismatched ||
2450 heap_base_oop == top() || // - heap_base_oop is NULL or
2451 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2452 // and the unsafe access is made to large offset
2453 // (i.e., larger than the maximum offset necessary for any
2454 // field access)
2455 ) {
2456 IdealKit ideal = IdealKit(this);
2457 #define __ ideal.
2458 IdealVariable normalized_result(ideal);
2459 __ declarations_done();
2460 __ set(normalized_result, p);
2461 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2462 __ set(normalized_result, ideal.ConI(1));
2463 ideal.end_if();
2464 final_sync(ideal);
2465 p = __ value(normalized_result);
2466 #undef __
2467 }
2468 }
2469 if (type == T_ADDRESS) {
2470 p = gvn().transform(new CastP2XNode(NULL, p));
2471 p = ConvX2UL(p);
2472 }
2473 // The load node has the control of the preceding MemBarCPUOrder. All
2474 // following nodes will have the control of the MemBarCPUOrder inserted at
2475 // the end of this method. So, pushing the load onto the stack at a later
2476 // point is fine.
2477 set_result(p);
2478 } else {
2479 if (bt == T_ADDRESS) {
2480 // Repackage the long as a pointer.
2481 val = ConvL2X(val);
2482 val = gvn().transform(new CastX2PNode(val));
2483 }
2484 access_store_at(control(), heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2485 }
2486
2487 return true;
2488 }
2489
2490 //----------------------------inline_unsafe_load_store----------------------------
2491 // This method serves a couple of different customers (depending on LoadStoreKind):
2492 //
2493 // LS_cmp_swap:
2494 //
2495 // boolean compareAndSetObject(Object o, long offset, Object expected, Object x);
2496 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2497 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2498 //
2499 // LS_cmp_swap_weak:
2500 //
2501 // boolean weakCompareAndSetObject( Object o, long offset, Object expected, Object x);
2502 // boolean weakCompareAndSetObjectPlain( Object o, long offset, Object expected, Object x);
2503 // boolean weakCompareAndSetObjectAcquire(Object o, long offset, Object expected, Object x);
2504 // boolean weakCompareAndSetObjectRelease(Object o, long offset, Object expected, Object x);
2505 //
2506 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2507 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2508 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2509 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2510 //
2511 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2512 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2513 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2514 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2515 //
2516 // LS_cmp_exchange:
2517 //
2518 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2519 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2520 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2521 //
2522 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2523 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2524 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2525 //
2526 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2527 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2528 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2529 //
2530 // LS_get_add:
2531 //
2532 // int getAndAddInt( Object o, long offset, int delta)
2533 // long getAndAddLong(Object o, long offset, long delta)
2534 //
2535 // LS_get_set:
2536 //
2537 // int getAndSet(Object o, long offset, int newValue)
2538 // long getAndSet(Object o, long offset, long newValue)
2539 // Object getAndSet(Object o, long offset, Object newValue)
2540 //
2541 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2542 // This basic scheme here is the same as inline_unsafe_access, but
2543 // differs in enough details that combining them would make the code
2544 // overly confusing. (This is a true fact! I originally combined
2545 // them, but even I was confused by it!) As much code/comments as
2546 // possible are retained from inline_unsafe_access though to make
2547 // the correspondences clearer. - dl
2548
2549 if (callee()->is_static()) return false; // caller must have the capability!
2550
2551 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2552 decorators |= mo_decorator_for_access_kind(access_kind);
2553
2554 #ifndef PRODUCT
2555 BasicType rtype;
2556 {
2557 ResourceMark rm;
2558 // Check the signatures.
2559 ciSignature* sig = callee()->signature();
2560 rtype = sig->return_type()->basic_type();
2561 switch(kind) {
2562 case LS_get_add:
2563 case LS_get_set: {
2564 // Check the signatures.
2565 #ifdef ASSERT
2566 assert(rtype == type, "get and set must return the expected type");
2567 assert(sig->count() == 3, "get and set has 3 arguments");
2568 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2569 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2570 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2571 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2572 #endif // ASSERT
2573 break;
2574 }
2575 case LS_cmp_swap:
2576 case LS_cmp_swap_weak: {
2577 // Check the signatures.
2578 #ifdef ASSERT
2579 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2580 assert(sig->count() == 4, "CAS has 4 arguments");
2581 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2582 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2583 #endif // ASSERT
2584 break;
2585 }
2586 case LS_cmp_exchange: {
2587 // Check the signatures.
2588 #ifdef ASSERT
2589 assert(rtype == type, "CAS must return the expected type");
2590 assert(sig->count() == 4, "CAS has 4 arguments");
2591 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2592 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2593 #endif // ASSERT
2594 break;
2595 }
2596 default:
2597 ShouldNotReachHere();
2598 }
2599 }
2600 #endif //PRODUCT
2601
2602 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2603
2604 // Get arguments:
2605 Node* receiver = NULL;
2606 Node* base = NULL;
2607 Node* offset = NULL;
2608 Node* oldval = NULL;
2609 Node* newval = NULL;
2610 switch(kind) {
2611 case LS_cmp_swap:
2612 case LS_cmp_swap_weak:
2613 case LS_cmp_exchange: {
2614 const bool two_slot_type = type2size[type] == 2;
2615 receiver = argument(0); // type: oop
2616 base = argument(1); // type: oop
2617 offset = argument(2); // type: long
2618 oldval = argument(4); // type: oop, int, or long
2619 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2620 break;
2621 }
2622 case LS_get_add:
2623 case LS_get_set: {
2624 receiver = argument(0); // type: oop
2625 base = argument(1); // type: oop
2626 offset = argument(2); // type: long
2627 oldval = NULL;
2628 newval = argument(4); // type: oop, int, or long
2629 break;
2630 }
2631 default:
2632 ShouldNotReachHere();
2633 }
2634
2635 // Build field offset expression.
2636 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2637 // to be plain byte offsets, which are also the same as those accepted
2638 // by oopDesc::field_addr.
2639 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2640 // 32-bit machines ignore the high half of long offsets
2641 offset = ConvL2X(offset);
2642 Node* adr = make_unsafe_address(base, offset, type, false);
2643 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2644
2645 Compile::AliasType* alias_type = C->alias_type(adr_type);
2646 BasicType bt = alias_type->basic_type();
2647 if (bt != T_ILLEGAL &&
2648 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2649 // Don't intrinsify mismatched object accesses.
2650 return false;
2651 }
2652
2653 // For CAS, unlike inline_unsafe_access, there seems no point in
2654 // trying to refine types. Just use the coarse types here.
2655 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2656 const Type *value_type = Type::get_const_basic_type(type);
2657
2658 switch (kind) {
2659 case LS_get_set:
2660 case LS_cmp_exchange: {
2661 if (type == T_OBJECT) {
2662 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2663 if (tjp != NULL) {
2664 value_type = tjp;
2665 }
2666 }
2667 break;
2668 }
2669 case LS_cmp_swap:
2670 case LS_cmp_swap_weak:
2671 case LS_get_add:
2672 break;
2673 default:
2674 ShouldNotReachHere();
2675 }
2676
2677 // Null check receiver.
2678 receiver = null_check(receiver);
2679 if (stopped()) {
2680 return true;
2681 }
2682
2683 int alias_idx = C->get_alias_index(adr_type);
2684
2685 if (type == T_OBJECT || type == T_ARRAY) {
2686 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2687
2688 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2689 // could be delayed during Parse (for example, in adjust_map_after_if()).
2690 // Execute transformation here to avoid barrier generation in such case.
2691 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2692 newval = _gvn.makecon(TypePtr::NULL_PTR);
2693
2694 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2695 // Refine the value to a null constant, when it is known to be null
2696 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2697 }
2698 }
2699
2700 Node* result = NULL;
2701 switch (kind) {
2702 case LS_cmp_exchange: {
2703 result = access_atomic_cmpxchg_val_at(control(), base, adr, adr_type, alias_idx,
2704 oldval, newval, value_type, type, decorators);
2705 break;
2706 }
2707 case LS_cmp_swap_weak:
2708 decorators |= C2_WEAK_CMPXCHG;
2709 case LS_cmp_swap: {
2710 result = access_atomic_cmpxchg_bool_at(control(), base, adr, adr_type, alias_idx,
2711 oldval, newval, value_type, type, decorators);
2712 break;
2713 }
2714 case LS_get_set: {
2715 result = access_atomic_xchg_at(control(), base, adr, adr_type, alias_idx,
2716 newval, value_type, type, decorators);
2717 break;
2718 }
2719 case LS_get_add: {
2720 result = access_atomic_add_at(control(), base, adr, adr_type, alias_idx,
2721 newval, value_type, type, decorators);
2722 break;
2723 }
2724 default:
2725 ShouldNotReachHere();
2726 }
2727
2728 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2729 set_result(result);
2730 return true;
2731 }
2732
2733 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2734 // Regardless of form, don't allow previous ld/st to move down,
2735 // then issue acquire, release, or volatile mem_bar.
2736 insert_mem_bar(Op_MemBarCPUOrder);
2737 switch(id) {
2738 case vmIntrinsics::_loadFence:
2739 insert_mem_bar(Op_LoadFence);
2740 return true;
2741 case vmIntrinsics::_storeFence:
2742 insert_mem_bar(Op_StoreFence);
2743 return true;
2744 case vmIntrinsics::_fullFence:
2745 insert_mem_bar(Op_MemBarVolatile);
2746 return true;
2747 default:
2748 fatal_unexpected_iid(id);
2749 return false;
2750 }
2751 }
2752
2753 bool LibraryCallKit::inline_onspinwait() {
2754 insert_mem_bar(Op_OnSpinWait);
2755 return true;
2756 }
2757
2758 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2759 if (!kls->is_Con()) {
2760 return true;
2761 }
2762 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2763 if (klsptr == NULL) {
2764 return true;
2765 }
2766 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2767 // don't need a guard for a klass that is already initialized
2768 return !ik->is_initialized();
2769 }
2770
2771 //----------------------------inline_unsafe_allocate---------------------------
2772 // public native Object Unsafe.allocateInstance(Class<?> cls);
2773 bool LibraryCallKit::inline_unsafe_allocate() {
2774 if (callee()->is_static()) return false; // caller must have the capability!
2775
2776 null_check_receiver(); // null-check, then ignore
2777 Node* cls = null_check(argument(1));
2778 if (stopped()) return true;
2779
2780 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2781 kls = null_check(kls);
2782 if (stopped()) return true; // argument was like int.class
2783
2784 Node* test = NULL;
2785 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2786 // Note: The argument might still be an illegal value like
2787 // Serializable.class or Object[].class. The runtime will handle it.
2788 // But we must make an explicit check for initialization.
2789 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2790 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2791 // can generate code to load it as unsigned byte.
2792 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2793 Node* bits = intcon(InstanceKlass::fully_initialized);
2794 test = _gvn.transform(new SubINode(inst, bits));
2795 // The 'test' is non-zero if we need to take a slow path.
2796 }
2797
2798 Node* obj = new_instance(kls, test);
2799 set_result(obj);
2800 return true;
2801 }
2802
2803 //------------------------inline_native_time_funcs--------------
2804 // inline code for System.currentTimeMillis() and System.nanoTime()
2805 // these have the same type and signature
2806 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2807 const TypeFunc* tf = OptoRuntime::void_long_Type();
2808 const TypePtr* no_memory_effects = NULL;
2809 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2810 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2811 #ifdef ASSERT
2812 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2813 assert(value_top == top(), "second value must be top");
2814 #endif
2815 set_result(value);
2816 return true;
2817 }
2818
2819 #ifdef JFR_HAVE_INTRINSICS
2820
2821 /*
2822 * oop -> myklass
2823 * myklass->trace_id |= USED
2824 * return myklass->trace_id & ~0x3
2825 */
2826 bool LibraryCallKit::inline_native_classID() {
2827 Node* cls = null_check(argument(0), T_OBJECT);
2828 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2829 kls = null_check(kls, T_OBJECT);
2830
2831 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2832 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2833 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2834
2835 Node* clsused = longcon(0x01l); // set the class bit
2836 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2837 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2838 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2839
2840 #ifdef TRACE_ID_META_BITS
2841 Node* mbits = longcon(~TRACE_ID_META_BITS);
2842 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2843 #endif
2844 #ifdef TRACE_ID_SHIFT
2845 Node* cbits = intcon(TRACE_ID_SHIFT);
2846 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2847 #endif
2848
2849 set_result(tvalue);
2850 return true;
2851
2852 }
2853
2854 bool LibraryCallKit::inline_native_getEventWriter() {
2855 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2856
2857 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2858 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)
2859 );
2860
2861 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2862
2863 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2864 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2865
2866 IfNode* iff_jobj_null =
2867 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2868
2869 enum { _normal_path = 1,
2870 _null_path = 2,
2871 PATH_LIMIT };
2872
2873 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2874 PhiNode* result_val = new PhiNode(result_rgn, TypePtr::BOTTOM);
2875
2876 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2877 result_rgn->init_req(_null_path, jobj_is_null);
2878 result_val->init_req(_null_path, null());
2879
2880 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2881 result_rgn->init_req(_normal_path, jobj_is_not_null);
2882
2883 Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
2884 result_val->init_req(_normal_path, res);
2885
2886 set_result(result_rgn, result_val);
2887
2888 return true;
2889 }
2890
2891 #endif // JFR_HAVE_INTRINSICS
2892
2893 //------------------------inline_native_currentThread------------------
2894 bool LibraryCallKit::inline_native_currentThread() {
2895 Node* junk = NULL;
2896 set_result(generate_current_thread(junk));
2897 return true;
2898 }
2899
2900 //------------------------inline_native_isInterrupted------------------
2901 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
2902 bool LibraryCallKit::inline_native_isInterrupted() {
2903 // Add a fast path to t.isInterrupted(clear_int):
2904 // (t == Thread.current() &&
2905 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
2906 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
2907 // So, in the common case that the interrupt bit is false,
2908 // we avoid making a call into the VM. Even if the interrupt bit
2909 // is true, if the clear_int argument is false, we avoid the VM call.
2910 // However, if the receiver is not currentThread, we must call the VM,
2911 // because there must be some locking done around the operation.
2912
2913 // We only go to the fast case code if we pass two guards.
2914 // Paths which do not pass are accumulated in the slow_region.
2915
2916 enum {
2917 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
2918 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
2919 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
2920 PATH_LIMIT
2921 };
2922
2923 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
2924 // out of the function.
2925 insert_mem_bar(Op_MemBarCPUOrder);
2926
2927 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2928 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
2929
2930 RegionNode* slow_region = new RegionNode(1);
2931 record_for_igvn(slow_region);
2932
2933 // (a) Receiving thread must be the current thread.
2934 Node* rec_thr = argument(0);
2935 Node* tls_ptr = NULL;
2936 Node* cur_thr = generate_current_thread(tls_ptr);
2937 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
2938 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
2939
2940 generate_slow_guard(bol_thr, slow_region);
2941
2942 // (b) Interrupt bit on TLS must be false.
2943 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
2944 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
2945 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
2946
2947 // Set the control input on the field _interrupted read to prevent it floating up.
2948 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
2949 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
2950 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
2951
2952 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2953
2954 // First fast path: if (!TLS._interrupted) return false;
2955 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
2956 result_rgn->init_req(no_int_result_path, false_bit);
2957 result_val->init_req(no_int_result_path, intcon(0));
2958
2959 // drop through to next case
2960 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
2961
2962 #ifndef _WINDOWS
2963 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
2964 Node* clr_arg = argument(1);
2965 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
2966 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
2967 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
2968
2969 // Second fast path: ... else if (!clear_int) return true;
2970 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
2971 result_rgn->init_req(no_clear_result_path, false_arg);
2972 result_val->init_req(no_clear_result_path, intcon(1));
2973
2974 // drop through to next case
2975 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
2976 #else
2977 // To return true on Windows you must read the _interrupted field
2978 // and check the event state i.e. take the slow path.
2979 #endif // _WINDOWS
2980
2981 // (d) Otherwise, go to the slow path.
2982 slow_region->add_req(control());
2983 set_control( _gvn.transform(slow_region));
2984
2985 if (stopped()) {
2986 // There is no slow path.
2987 result_rgn->init_req(slow_result_path, top());
2988 result_val->init_req(slow_result_path, top());
2989 } else {
2990 // non-virtual because it is a private non-static
2991 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
2992
2993 Node* slow_val = set_results_for_java_call(slow_call);
2994 // this->control() comes from set_results_for_java_call
2995
2996 Node* fast_io = slow_call->in(TypeFunc::I_O);
2997 Node* fast_mem = slow_call->in(TypeFunc::Memory);
2998
2999 // These two phis are pre-filled with copies of of the fast IO and Memory
3000 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3001 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3002
3003 result_rgn->init_req(slow_result_path, control());
3004 result_io ->init_req(slow_result_path, i_o());
3005 result_mem->init_req(slow_result_path, reset_memory());
3006 result_val->init_req(slow_result_path, slow_val);
3007
3008 set_all_memory(_gvn.transform(result_mem));
3009 set_i_o( _gvn.transform(result_io));
3010 }
3011
3012 C->set_has_split_ifs(true); // Has chance for split-if optimization
3013 set_result(result_rgn, result_val);
3014 return true;
3015 }
3016
3017 //---------------------------load_mirror_from_klass----------------------------
3018 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3019 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3020 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3021 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3022 // mirror = ((OopHandle)mirror)->resolve();
3023 return make_load(NULL, load, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3024 }
3025
3026 //-----------------------load_klass_from_mirror_common-------------------------
3027 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3028 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3029 // and branch to the given path on the region.
3030 // If never_see_null, take an uncommon trap on null, so we can optimistically
3031 // compile for the non-null case.
3032 // If the region is NULL, force never_see_null = true.
3033 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3034 bool never_see_null,
3035 RegionNode* region,
3036 int null_path,
3037 int offset) {
3038 if (region == NULL) never_see_null = true;
3039 Node* p = basic_plus_adr(mirror, offset);
3040 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3041 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3042 Node* null_ctl = top();
3043 kls = null_check_oop(kls, &null_ctl, never_see_null);
3044 if (region != NULL) {
3045 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3046 region->init_req(null_path, null_ctl);
3047 } else {
3048 assert(null_ctl == top(), "no loose ends");
3049 }
3050 return kls;
3051 }
3052
3053 //--------------------(inline_native_Class_query helpers)---------------------
3054 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3055 // Fall through if (mods & mask) == bits, take the guard otherwise.
3056 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3057 // Branch around if the given klass has the given modifier bit set.
3058 // Like generate_guard, adds a new path onto the region.
3059 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3060 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3061 Node* mask = intcon(modifier_mask);
3062 Node* bits = intcon(modifier_bits);
3063 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3064 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3065 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3066 return generate_fair_guard(bol, region);
3067 }
3068 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3069 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3070 }
3071
3072 //-------------------------inline_native_Class_query-------------------
3073 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3074 const Type* return_type = TypeInt::BOOL;
3075 Node* prim_return_value = top(); // what happens if it's a primitive class?
3076 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3077 bool expect_prim = false; // most of these guys expect to work on refs
3078
3079 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3080
3081 Node* mirror = argument(0);
3082 Node* obj = top();
3083
3084 switch (id) {
3085 case vmIntrinsics::_isInstance:
3086 // nothing is an instance of a primitive type
3087 prim_return_value = intcon(0);
3088 obj = argument(1);
3089 break;
3090 case vmIntrinsics::_getModifiers:
3091 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3092 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3093 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3094 break;
3095 case vmIntrinsics::_isInterface:
3096 prim_return_value = intcon(0);
3097 break;
3098 case vmIntrinsics::_isArray:
3099 prim_return_value = intcon(0);
3100 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3101 break;
3102 case vmIntrinsics::_isPrimitive:
3103 prim_return_value = intcon(1);
3104 expect_prim = true; // obviously
3105 break;
3106 case vmIntrinsics::_getSuperclass:
3107 prim_return_value = null();
3108 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3109 break;
3110 case vmIntrinsics::_getClassAccessFlags:
3111 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3112 return_type = TypeInt::INT; // not bool! 6297094
3113 break;
3114 default:
3115 fatal_unexpected_iid(id);
3116 break;
3117 }
3118
3119 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3120 if (mirror_con == NULL) return false; // cannot happen?
3121
3122 #ifndef PRODUCT
3123 if (C->print_intrinsics() || C->print_inlining()) {
3124 ciType* k = mirror_con->java_mirror_type();
3125 if (k) {
3126 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3127 k->print_name();
3128 tty->cr();
3129 }
3130 }
3131 #endif
3132
3133 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3134 RegionNode* region = new RegionNode(PATH_LIMIT);
3135 record_for_igvn(region);
3136 PhiNode* phi = new PhiNode(region, return_type);
3137
3138 // The mirror will never be null of Reflection.getClassAccessFlags, however
3139 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3140 // if it is. See bug 4774291.
3141
3142 // For Reflection.getClassAccessFlags(), the null check occurs in
3143 // the wrong place; see inline_unsafe_access(), above, for a similar
3144 // situation.
3145 mirror = null_check(mirror);
3146 // If mirror or obj is dead, only null-path is taken.
3147 if (stopped()) return true;
3148
3149 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3150
3151 // Now load the mirror's klass metaobject, and null-check it.
3152 // Side-effects region with the control path if the klass is null.
3153 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3154 // If kls is null, we have a primitive mirror.
3155 phi->init_req(_prim_path, prim_return_value);
3156 if (stopped()) { set_result(region, phi); return true; }
3157 bool safe_for_replace = (region->in(_prim_path) == top());
3158
3159 Node* p; // handy temp
3160 Node* null_ctl;
3161
3162 // Now that we have the non-null klass, we can perform the real query.
3163 // For constant classes, the query will constant-fold in LoadNode::Value.
3164 Node* query_value = top();
3165 switch (id) {
3166 case vmIntrinsics::_isInstance:
3167 // nothing is an instance of a primitive type
3168 query_value = gen_instanceof(obj, kls, safe_for_replace);
3169 break;
3170
3171 case vmIntrinsics::_getModifiers:
3172 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3173 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3174 break;
3175
3176 case vmIntrinsics::_isInterface:
3177 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3178 if (generate_interface_guard(kls, region) != NULL)
3179 // A guard was added. If the guard is taken, it was an interface.
3180 phi->add_req(intcon(1));
3181 // If we fall through, it's a plain class.
3182 query_value = intcon(0);
3183 break;
3184
3185 case vmIntrinsics::_isArray:
3186 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3187 if (generate_array_guard(kls, region) != NULL)
3188 // A guard was added. If the guard is taken, it was an array.
3189 phi->add_req(intcon(1));
3190 // If we fall through, it's a plain class.
3191 query_value = intcon(0);
3192 break;
3193
3194 case vmIntrinsics::_isPrimitive:
3195 query_value = intcon(0); // "normal" path produces false
3196 break;
3197
3198 case vmIntrinsics::_getSuperclass:
3199 // The rules here are somewhat unfortunate, but we can still do better
3200 // with random logic than with a JNI call.
3201 // Interfaces store null or Object as _super, but must report null.
3202 // Arrays store an intermediate super as _super, but must report Object.
3203 // Other types can report the actual _super.
3204 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3205 if (generate_interface_guard(kls, region) != NULL)
3206 // A guard was added. If the guard is taken, it was an interface.
3207 phi->add_req(null());
3208 if (generate_array_guard(kls, region) != NULL)
3209 // A guard was added. If the guard is taken, it was an array.
3210 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3211 // If we fall through, it's a plain class. Get its _super.
3212 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3213 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3214 null_ctl = top();
3215 kls = null_check_oop(kls, &null_ctl);
3216 if (null_ctl != top()) {
3217 // If the guard is taken, Object.superClass is null (both klass and mirror).
3218 region->add_req(null_ctl);
3219 phi ->add_req(null());
3220 }
3221 if (!stopped()) {
3222 query_value = load_mirror_from_klass(kls);
3223 }
3224 break;
3225
3226 case vmIntrinsics::_getClassAccessFlags:
3227 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3228 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3229 break;
3230
3231 default:
3232 fatal_unexpected_iid(id);
3233 break;
3234 }
3235
3236 // Fall-through is the normal case of a query to a real class.
3237 phi->init_req(1, query_value);
3238 region->init_req(1, control());
3239
3240 C->set_has_split_ifs(true); // Has chance for split-if optimization
3241 set_result(region, phi);
3242 return true;
3243 }
3244
3245 //-------------------------inline_Class_cast-------------------
3246 bool LibraryCallKit::inline_Class_cast() {
3247 Node* mirror = argument(0); // Class
3248 Node* obj = argument(1);
3249 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3250 if (mirror_con == NULL) {
3251 return false; // dead path (mirror->is_top()).
3252 }
3253 if (obj == NULL || obj->is_top()) {
3254 return false; // dead path
3255 }
3256 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3257
3258 // First, see if Class.cast() can be folded statically.
3259 // java_mirror_type() returns non-null for compile-time Class constants.
3260 ciType* tm = mirror_con->java_mirror_type();
3261 if (tm != NULL && tm->is_klass() &&
3262 tp != NULL && tp->klass() != NULL) {
3263 if (!tp->klass()->is_loaded()) {
3264 // Don't use intrinsic when class is not loaded.
3265 return false;
3266 } else {
3267 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3268 if (static_res == Compile::SSC_always_true) {
3269 // isInstance() is true - fold the code.
3270 set_result(obj);
3271 return true;
3272 } else if (static_res == Compile::SSC_always_false) {
3273 // Don't use intrinsic, have to throw ClassCastException.
3274 // If the reference is null, the non-intrinsic bytecode will
3275 // be optimized appropriately.
3276 return false;
3277 }
3278 }
3279 }
3280
3281 // Bailout intrinsic and do normal inlining if exception path is frequent.
3282 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3283 return false;
3284 }
3285
3286 // Generate dynamic checks.
3287 // Class.cast() is java implementation of _checkcast bytecode.
3288 // Do checkcast (Parse::do_checkcast()) optimizations here.
3289
3290 mirror = null_check(mirror);
3291 // If mirror is dead, only null-path is taken.
3292 if (stopped()) {
3293 return true;
3294 }
3295
3296 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3297 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3298 RegionNode* region = new RegionNode(PATH_LIMIT);
3299 record_for_igvn(region);
3300
3301 // Now load the mirror's klass metaobject, and null-check it.
3302 // If kls is null, we have a primitive mirror and
3303 // nothing is an instance of a primitive type.
3304 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3305
3306 Node* res = top();
3307 if (!stopped()) {
3308 Node* bad_type_ctrl = top();
3309 // Do checkcast optimizations.
3310 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3311 region->init_req(_bad_type_path, bad_type_ctrl);
3312 }
3313 if (region->in(_prim_path) != top() ||
3314 region->in(_bad_type_path) != top()) {
3315 // Let Interpreter throw ClassCastException.
3316 PreserveJVMState pjvms(this);
3317 set_control(_gvn.transform(region));
3318 uncommon_trap(Deoptimization::Reason_intrinsic,
3319 Deoptimization::Action_maybe_recompile);
3320 }
3321 if (!stopped()) {
3322 set_result(res);
3323 }
3324 return true;
3325 }
3326
3327
3328 //--------------------------inline_native_subtype_check------------------------
3329 // This intrinsic takes the JNI calls out of the heart of
3330 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3331 bool LibraryCallKit::inline_native_subtype_check() {
3332 // Pull both arguments off the stack.
3333 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3334 args[0] = argument(0);
3335 args[1] = argument(1);
3336 Node* klasses[2]; // corresponding Klasses: superk, subk
3337 klasses[0] = klasses[1] = top();
3338
3339 enum {
3340 // A full decision tree on {superc is prim, subc is prim}:
3341 _prim_0_path = 1, // {P,N} => false
3342 // {P,P} & superc!=subc => false
3343 _prim_same_path, // {P,P} & superc==subc => true
3344 _prim_1_path, // {N,P} => false
3345 _ref_subtype_path, // {N,N} & subtype check wins => true
3346 _both_ref_path, // {N,N} & subtype check loses => false
3347 PATH_LIMIT
3348 };
3349
3350 RegionNode* region = new RegionNode(PATH_LIMIT);
3351 Node* phi = new PhiNode(region, TypeInt::BOOL);
3352 record_for_igvn(region);
3353
3354 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3355 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3356 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3357
3358 // First null-check both mirrors and load each mirror's klass metaobject.
3359 int which_arg;
3360 for (which_arg = 0; which_arg <= 1; which_arg++) {
3361 Node* arg = args[which_arg];
3362 arg = null_check(arg);
3363 if (stopped()) break;
3364 args[which_arg] = arg;
3365
3366 Node* p = basic_plus_adr(arg, class_klass_offset);
3367 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3368 klasses[which_arg] = _gvn.transform(kls);
3369 }
3370
3371 // Having loaded both klasses, test each for null.
3372 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3373 for (which_arg = 0; which_arg <= 1; which_arg++) {
3374 Node* kls = klasses[which_arg];
3375 Node* null_ctl = top();
3376 kls = null_check_oop(kls, &null_ctl, never_see_null);
3377 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3378 region->init_req(prim_path, null_ctl);
3379 if (stopped()) break;
3380 klasses[which_arg] = kls;
3381 }
3382
3383 if (!stopped()) {
3384 // now we have two reference types, in klasses[0..1]
3385 Node* subk = klasses[1]; // the argument to isAssignableFrom
3386 Node* superk = klasses[0]; // the receiver
3387 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3388 // now we have a successful reference subtype check
3389 region->set_req(_ref_subtype_path, control());
3390 }
3391
3392 // If both operands are primitive (both klasses null), then
3393 // we must return true when they are identical primitives.
3394 // It is convenient to test this after the first null klass check.
3395 set_control(region->in(_prim_0_path)); // go back to first null check
3396 if (!stopped()) {
3397 // Since superc is primitive, make a guard for the superc==subc case.
3398 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3399 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3400 generate_guard(bol_eq, region, PROB_FAIR);
3401 if (region->req() == PATH_LIMIT+1) {
3402 // A guard was added. If the added guard is taken, superc==subc.
3403 region->swap_edges(PATH_LIMIT, _prim_same_path);
3404 region->del_req(PATH_LIMIT);
3405 }
3406 region->set_req(_prim_0_path, control()); // Not equal after all.
3407 }
3408
3409 // these are the only paths that produce 'true':
3410 phi->set_req(_prim_same_path, intcon(1));
3411 phi->set_req(_ref_subtype_path, intcon(1));
3412
3413 // pull together the cases:
3414 assert(region->req() == PATH_LIMIT, "sane region");
3415 for (uint i = 1; i < region->req(); i++) {
3416 Node* ctl = region->in(i);
3417 if (ctl == NULL || ctl == top()) {
3418 region->set_req(i, top());
3419 phi ->set_req(i, top());
3420 } else if (phi->in(i) == NULL) {
3421 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3422 }
3423 }
3424
3425 set_control(_gvn.transform(region));
3426 set_result(_gvn.transform(phi));
3427 return true;
3428 }
3429
3430 //---------------------generate_array_guard_common------------------------
3431 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3432 bool obj_array, bool not_array) {
3433
3434 if (stopped()) {
3435 return NULL;
3436 }
3437
3438 // If obj_array/non_array==false/false:
3439 // Branch around if the given klass is in fact an array (either obj or prim).
3440 // If obj_array/non_array==false/true:
3441 // Branch around if the given klass is not an array klass of any kind.
3442 // If obj_array/non_array==true/true:
3443 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3444 // If obj_array/non_array==true/false:
3445 // Branch around if the kls is an oop array (Object[] or subtype)
3446 //
3447 // Like generate_guard, adds a new path onto the region.
3448 jint layout_con = 0;
3449 Node* layout_val = get_layout_helper(kls, layout_con);
3450 if (layout_val == NULL) {
3451 bool query = (obj_array
3452 ? Klass::layout_helper_is_objArray(layout_con)
3453 : Klass::layout_helper_is_array(layout_con));
3454 if (query == not_array) {
3455 return NULL; // never a branch
3456 } else { // always a branch
3457 Node* always_branch = control();
3458 if (region != NULL)
3459 region->add_req(always_branch);
3460 set_control(top());
3461 return always_branch;
3462 }
3463 }
3464 // Now test the correct condition.
3465 jint nval = (obj_array
3466 ? (jint)(Klass::_lh_array_tag_type_value
3467 << Klass::_lh_array_tag_shift)
3468 : Klass::_lh_neutral_value);
3469 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3470 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3471 // invert the test if we are looking for a non-array
3472 if (not_array) btest = BoolTest(btest).negate();
3473 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3474 return generate_fair_guard(bol, region);
3475 }
3476
3477
3478 //-----------------------inline_native_newArray--------------------------
3479 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3480 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3481 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3482 Node* mirror;
3483 Node* count_val;
3484 if (uninitialized) {
3485 mirror = argument(1);
3486 count_val = argument(2);
3487 } else {
3488 mirror = argument(0);
3489 count_val = argument(1);
3490 }
3491
3492 mirror = null_check(mirror);
3493 // If mirror or obj is dead, only null-path is taken.
3494 if (stopped()) return true;
3495
3496 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3497 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3498 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3499 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3500 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3501
3502 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3503 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3504 result_reg, _slow_path);
3505 Node* normal_ctl = control();
3506 Node* no_array_ctl = result_reg->in(_slow_path);
3507
3508 // Generate code for the slow case. We make a call to newArray().
3509 set_control(no_array_ctl);
3510 if (!stopped()) {
3511 // Either the input type is void.class, or else the
3512 // array klass has not yet been cached. Either the
3513 // ensuing call will throw an exception, or else it
3514 // will cache the array klass for next time.
3515 PreserveJVMState pjvms(this);
3516 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3517 Node* slow_result = set_results_for_java_call(slow_call);
3518 // this->control() comes from set_results_for_java_call
3519 result_reg->set_req(_slow_path, control());
3520 result_val->set_req(_slow_path, slow_result);
3521 result_io ->set_req(_slow_path, i_o());
3522 result_mem->set_req(_slow_path, reset_memory());
3523 }
3524
3525 set_control(normal_ctl);
3526 if (!stopped()) {
3527 // Normal case: The array type has been cached in the java.lang.Class.
3528 // The following call works fine even if the array type is polymorphic.
3529 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3530 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3531 result_reg->init_req(_normal_path, control());
3532 result_val->init_req(_normal_path, obj);
3533 result_io ->init_req(_normal_path, i_o());
3534 result_mem->init_req(_normal_path, reset_memory());
3535
3536 if (uninitialized) {
3537 // Mark the allocation so that zeroing is skipped
3538 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3539 alloc->maybe_set_complete(&_gvn);
3540 }
3541 }
3542
3543 // Return the combined state.
3544 set_i_o( _gvn.transform(result_io) );
3545 set_all_memory( _gvn.transform(result_mem));
3546
3547 C->set_has_split_ifs(true); // Has chance for split-if optimization
3548 set_result(result_reg, result_val);
3549 return true;
3550 }
3551
3552 //----------------------inline_native_getLength--------------------------
3553 // public static native int java.lang.reflect.Array.getLength(Object array);
3554 bool LibraryCallKit::inline_native_getLength() {
3555 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3556
3557 Node* array = null_check(argument(0));
3558 // If array is dead, only null-path is taken.
3559 if (stopped()) return true;
3560
3561 // Deoptimize if it is a non-array.
3562 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3563
3564 if (non_array != NULL) {
3565 PreserveJVMState pjvms(this);
3566 set_control(non_array);
3567 uncommon_trap(Deoptimization::Reason_intrinsic,
3568 Deoptimization::Action_maybe_recompile);
3569 }
3570
3571 // If control is dead, only non-array-path is taken.
3572 if (stopped()) return true;
3573
3574 // The works fine even if the array type is polymorphic.
3575 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3576 Node* result = load_array_length(array);
3577
3578 C->set_has_split_ifs(true); // Has chance for split-if optimization
3579 set_result(result);
3580 return true;
3581 }
3582
3583 //------------------------inline_array_copyOf----------------------------
3584 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3585 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3586 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3587 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3588
3589 // Get the arguments.
3590 Node* original = argument(0);
3591 Node* start = is_copyOfRange? argument(1): intcon(0);
3592 Node* end = is_copyOfRange? argument(2): argument(1);
3593 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3594
3595 Node* newcopy = NULL;
3596
3597 // Set the original stack and the reexecute bit for the interpreter to reexecute
3598 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3599 { PreserveReexecuteState preexecs(this);
3600 jvms()->set_should_reexecute(true);
3601
3602 array_type_mirror = null_check(array_type_mirror);
3603 original = null_check(original);
3604
3605 // Check if a null path was taken unconditionally.
3606 if (stopped()) return true;
3607
3608 Node* orig_length = load_array_length(original);
3609
3610 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3611 klass_node = null_check(klass_node);
3612
3613 RegionNode* bailout = new RegionNode(1);
3614 record_for_igvn(bailout);
3615
3616 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3617 // Bail out if that is so.
3618 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3619 if (not_objArray != NULL) {
3620 // Improve the klass node's type from the new optimistic assumption:
3621 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3622 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3623 Node* cast = new CastPPNode(klass_node, akls);
3624 cast->init_req(0, control());
3625 klass_node = _gvn.transform(cast);
3626 }
3627
3628 // Bail out if either start or end is negative.
3629 generate_negative_guard(start, bailout, &start);
3630 generate_negative_guard(end, bailout, &end);
3631
3632 Node* length = end;
3633 if (_gvn.type(start) != TypeInt::ZERO) {
3634 length = _gvn.transform(new SubINode(end, start));
3635 }
3636
3637 // Bail out if length is negative.
3638 // Without this the new_array would throw
3639 // NegativeArraySizeException but IllegalArgumentException is what
3640 // should be thrown
3641 generate_negative_guard(length, bailout, &length);
3642
3643 if (bailout->req() > 1) {
3644 PreserveJVMState pjvms(this);
3645 set_control(_gvn.transform(bailout));
3646 uncommon_trap(Deoptimization::Reason_intrinsic,
3647 Deoptimization::Action_maybe_recompile);
3648 }
3649
3650 if (!stopped()) {
3651 // How many elements will we copy from the original?
3652 // The answer is MinI(orig_length - start, length).
3653 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3654 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3655
3656 // Generate a direct call to the right arraycopy function(s).
3657 // We know the copy is disjoint but we might not know if the
3658 // oop stores need checking.
3659 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3660 // This will fail a store-check if x contains any non-nulls.
3661
3662 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3663 // loads/stores but it is legal only if we're sure the
3664 // Arrays.copyOf would succeed. So we need all input arguments
3665 // to the copyOf to be validated, including that the copy to the
3666 // new array won't trigger an ArrayStoreException. That subtype
3667 // check can be optimized if we know something on the type of
3668 // the input array from type speculation.
3669 if (_gvn.type(klass_node)->singleton()) {
3670 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3671 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3672
3673 int test = C->static_subtype_check(superk, subk);
3674 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3675 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3676 if (t_original->speculative_type() != NULL) {
3677 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3678 }
3679 }
3680 }
3681
3682 bool validated = false;
3683 // Reason_class_check rather than Reason_intrinsic because we
3684 // want to intrinsify even if this traps.
3685 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3686 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3687 klass_node);
3688
3689 if (not_subtype_ctrl != top()) {
3690 PreserveJVMState pjvms(this);
3691 set_control(not_subtype_ctrl);
3692 uncommon_trap(Deoptimization::Reason_class_check,
3693 Deoptimization::Action_make_not_entrant);
3694 assert(stopped(), "Should be stopped");
3695 }
3696 validated = true;
3697 }
3698
3699 if (!stopped()) {
3700 newcopy = new_array(klass_node, length, 0); // no arguments to push
3701
3702 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3703 load_object_klass(original), klass_node);
3704 if (!is_copyOfRange) {
3705 ac->set_copyof(validated);
3706 } else {
3707 ac->set_copyofrange(validated);
3708 }
3709 Node* n = _gvn.transform(ac);
3710 if (n == ac) {
3711 ac->connect_outputs(this);
3712 } else {
3713 assert(validated, "shouldn't transform if all arguments not validated");
3714 set_all_memory(n);
3715 }
3716 }
3717 }
3718 } // original reexecute is set back here
3719
3720 C->set_has_split_ifs(true); // Has chance for split-if optimization
3721 if (!stopped()) {
3722 set_result(newcopy);
3723 }
3724 return true;
3725 }
3726
3727
3728 //----------------------generate_virtual_guard---------------------------
3729 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3730 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3731 RegionNode* slow_region) {
3732 ciMethod* method = callee();
3733 int vtable_index = method->vtable_index();
3734 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3735 "bad index %d", vtable_index);
3736 // Get the Method* out of the appropriate vtable entry.
3737 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3738 vtable_index*vtableEntry::size_in_bytes() +
3739 vtableEntry::method_offset_in_bytes();
3740 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3741 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3742
3743 // Compare the target method with the expected method (e.g., Object.hashCode).
3744 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3745
3746 Node* native_call = makecon(native_call_addr);
3747 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3748 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3749
3750 return generate_slow_guard(test_native, slow_region);
3751 }
3752
3753 //-----------------------generate_method_call----------------------------
3754 // Use generate_method_call to make a slow-call to the real
3755 // method if the fast path fails. An alternative would be to
3756 // use a stub like OptoRuntime::slow_arraycopy_Java.
3757 // This only works for expanding the current library call,
3758 // not another intrinsic. (E.g., don't use this for making an
3759 // arraycopy call inside of the copyOf intrinsic.)
3760 CallJavaNode*
3761 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3762 // When compiling the intrinsic method itself, do not use this technique.
3763 guarantee(callee() != C->method(), "cannot make slow-call to self");
3764
3765 ciMethod* method = callee();
3766 // ensure the JVMS we have will be correct for this call
3767 guarantee(method_id == method->intrinsic_id(), "must match");
3768
3769 const TypeFunc* tf = TypeFunc::make(method);
3770 CallJavaNode* slow_call;
3771 if (is_static) {
3772 assert(!is_virtual, "");
3773 slow_call = new CallStaticJavaNode(C, tf,
3774 SharedRuntime::get_resolve_static_call_stub(),
3775 method, bci());
3776 } else if (is_virtual) {
3777 null_check_receiver();
3778 int vtable_index = Method::invalid_vtable_index;
3779 if (UseInlineCaches) {
3780 // Suppress the vtable call
3781 } else {
3782 // hashCode and clone are not a miranda methods,
3783 // so the vtable index is fixed.
3784 // No need to use the linkResolver to get it.
3785 vtable_index = method->vtable_index();
3786 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3787 "bad index %d", vtable_index);
3788 }
3789 slow_call = new CallDynamicJavaNode(tf,
3790 SharedRuntime::get_resolve_virtual_call_stub(),
3791 method, vtable_index, bci());
3792 } else { // neither virtual nor static: opt_virtual
3793 null_check_receiver();
3794 slow_call = new CallStaticJavaNode(C, tf,
3795 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3796 method, bci());
3797 slow_call->set_optimized_virtual(true);
3798 }
3799 set_arguments_for_java_call(slow_call);
3800 set_edges_for_java_call(slow_call);
3801 return slow_call;
3802 }
3803
3804
3805 /**
3806 * Build special case code for calls to hashCode on an object. This call may
3807 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3808 * slightly different code.
3809 */
3810 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3811 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3812 assert(!(is_virtual && is_static), "either virtual, special, or static");
3813
3814 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3815
3816 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3817 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3818 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3819 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3820 Node* obj = NULL;
3821 if (!is_static) {
3822 // Check for hashing null object
3823 obj = null_check_receiver();
3824 if (stopped()) return true; // unconditionally null
3825 result_reg->init_req(_null_path, top());
3826 result_val->init_req(_null_path, top());
3827 } else {
3828 // Do a null check, and return zero if null.
3829 // System.identityHashCode(null) == 0
3830 obj = argument(0);
3831 Node* null_ctl = top();
3832 obj = null_check_oop(obj, &null_ctl);
3833 result_reg->init_req(_null_path, null_ctl);
3834 result_val->init_req(_null_path, _gvn.intcon(0));
3835 }
3836
3837 // Unconditionally null? Then return right away.
3838 if (stopped()) {
3839 set_control( result_reg->in(_null_path));
3840 if (!stopped())
3841 set_result(result_val->in(_null_path));
3842 return true;
3843 }
3844
3845 // We only go to the fast case code if we pass a number of guards. The
3846 // paths which do not pass are accumulated in the slow_region.
3847 RegionNode* slow_region = new RegionNode(1);
3848 record_for_igvn(slow_region);
3849
3850 // If this is a virtual call, we generate a funny guard. We pull out
3851 // the vtable entry corresponding to hashCode() from the target object.
3852 // If the target method which we are calling happens to be the native
3853 // Object hashCode() method, we pass the guard. We do not need this
3854 // guard for non-virtual calls -- the caller is known to be the native
3855 // Object hashCode().
3856 if (is_virtual) {
3857 // After null check, get the object's klass.
3858 Node* obj_klass = load_object_klass(obj);
3859 generate_virtual_guard(obj_klass, slow_region);
3860 }
3861
3862 // Get the header out of the object, use LoadMarkNode when available
3863 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3864 // The control of the load must be NULL. Otherwise, the load can move before
3865 // the null check after castPP removal.
3866 Node* no_ctrl = NULL;
3867 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3868
3869 // Test the header to see if it is unlocked.
3870 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
3871 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3872 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
3873 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3874 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3875
3876 generate_slow_guard(test_unlocked, slow_region);
3877
3878 // Get the hash value and check to see that it has been properly assigned.
3879 // We depend on hash_mask being at most 32 bits and avoid the use of
3880 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3881 // vm: see markOop.hpp.
3882 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
3883 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
3884 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3885 // This hack lets the hash bits live anywhere in the mark object now, as long
3886 // as the shift drops the relevant bits into the low 32 bits. Note that
3887 // Java spec says that HashCode is an int so there's no point in capturing
3888 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3889 hshifted_header = ConvX2I(hshifted_header);
3890 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3891
3892 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
3893 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3894 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3895
3896 generate_slow_guard(test_assigned, slow_region);
3897
3898 Node* init_mem = reset_memory();
3899 // fill in the rest of the null path:
3900 result_io ->init_req(_null_path, i_o());
3901 result_mem->init_req(_null_path, init_mem);
3902
3903 result_val->init_req(_fast_path, hash_val);
3904 result_reg->init_req(_fast_path, control());
3905 result_io ->init_req(_fast_path, i_o());
3906 result_mem->init_req(_fast_path, init_mem);
3907
3908 // Generate code for the slow case. We make a call to hashCode().
3909 set_control(_gvn.transform(slow_region));
3910 if (!stopped()) {
3911 // No need for PreserveJVMState, because we're using up the present state.
3912 set_all_memory(init_mem);
3913 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3914 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3915 Node* slow_result = set_results_for_java_call(slow_call);
3916 // this->control() comes from set_results_for_java_call
3917 result_reg->init_req(_slow_path, control());
3918 result_val->init_req(_slow_path, slow_result);
3919 result_io ->set_req(_slow_path, i_o());
3920 result_mem ->set_req(_slow_path, reset_memory());
3921 }
3922
3923 // Return the combined state.
3924 set_i_o( _gvn.transform(result_io) );
3925 set_all_memory( _gvn.transform(result_mem));
3926
3927 set_result(result_reg, result_val);
3928 return true;
3929 }
3930
3931 //---------------------------inline_native_getClass----------------------------
3932 // public final native Class<?> java.lang.Object.getClass();
3933 //
3934 // Build special case code for calls to getClass on an object.
3935 bool LibraryCallKit::inline_native_getClass() {
3936 Node* obj = null_check_receiver();
3937 if (stopped()) return true;
3938 set_result(load_mirror_from_klass(load_object_klass(obj)));
3939 return true;
3940 }
3941
3942 //-----------------inline_native_Reflection_getCallerClass---------------------
3943 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3944 //
3945 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3946 //
3947 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3948 // in that it must skip particular security frames and checks for
3949 // caller sensitive methods.
3950 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3951 #ifndef PRODUCT
3952 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3953 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3954 }
3955 #endif
3956
3957 if (!jvms()->has_method()) {
3958 #ifndef PRODUCT
3959 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3960 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
3961 }
3962 #endif
3963 return false;
3964 }
3965
3966 // Walk back up the JVM state to find the caller at the required
3967 // depth.
3968 JVMState* caller_jvms = jvms();
3969
3970 // Cf. JVM_GetCallerClass
3971 // NOTE: Start the loop at depth 1 because the current JVM state does
3972 // not include the Reflection.getCallerClass() frame.
3973 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
3974 ciMethod* m = caller_jvms->method();
3975 switch (n) {
3976 case 0:
3977 fatal("current JVM state does not include the Reflection.getCallerClass frame");
3978 break;
3979 case 1:
3980 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
3981 if (!m->caller_sensitive()) {
3982 #ifndef PRODUCT
3983 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3984 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
3985 }
3986 #endif
3987 return false; // bail-out; let JVM_GetCallerClass do the work
3988 }
3989 break;
3990 default:
3991 if (!m->is_ignored_by_security_stack_walk()) {
3992 // We have reached the desired frame; return the holder class.
3993 // Acquire method holder as java.lang.Class and push as constant.
3994 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
3995 ciInstance* caller_mirror = caller_klass->java_mirror();
3996 set_result(makecon(TypeInstPtr::make(caller_mirror)));
3997
3998 #ifndef PRODUCT
3999 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4000 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());
4001 tty->print_cr(" JVM state at this point:");
4002 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4003 ciMethod* m = jvms()->of_depth(i)->method();
4004 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4005 }
4006 }
4007 #endif
4008 return true;
4009 }
4010 break;
4011 }
4012 }
4013
4014 #ifndef PRODUCT
4015 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4016 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4017 tty->print_cr(" JVM state at this point:");
4018 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4019 ciMethod* m = jvms()->of_depth(i)->method();
4020 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4021 }
4022 }
4023 #endif
4024
4025 return false; // bail-out; let JVM_GetCallerClass do the work
4026 }
4027
4028 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4029 Node* arg = argument(0);
4030 Node* result = NULL;
4031
4032 switch (id) {
4033 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4034 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4035 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4036 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4037
4038 case vmIntrinsics::_doubleToLongBits: {
4039 // two paths (plus control) merge in a wood
4040 RegionNode *r = new RegionNode(3);
4041 Node *phi = new PhiNode(r, TypeLong::LONG);
4042
4043 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4044 // Build the boolean node
4045 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4046
4047 // Branch either way.
4048 // NaN case is less traveled, which makes all the difference.
4049 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4050 Node *opt_isnan = _gvn.transform(ifisnan);
4051 assert( opt_isnan->is_If(), "Expect an IfNode");
4052 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4053 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4054
4055 set_control(iftrue);
4056
4057 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4058 Node *slow_result = longcon(nan_bits); // return NaN
4059 phi->init_req(1, _gvn.transform( slow_result ));
4060 r->init_req(1, iftrue);
4061
4062 // Else fall through
4063 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4064 set_control(iffalse);
4065
4066 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4067 r->init_req(2, iffalse);
4068
4069 // Post merge
4070 set_control(_gvn.transform(r));
4071 record_for_igvn(r);
4072
4073 C->set_has_split_ifs(true); // Has chance for split-if optimization
4074 result = phi;
4075 assert(result->bottom_type()->isa_long(), "must be");
4076 break;
4077 }
4078
4079 case vmIntrinsics::_floatToIntBits: {
4080 // two paths (plus control) merge in a wood
4081 RegionNode *r = new RegionNode(3);
4082 Node *phi = new PhiNode(r, TypeInt::INT);
4083
4084 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4085 // Build the boolean node
4086 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4087
4088 // Branch either way.
4089 // NaN case is less traveled, which makes all the difference.
4090 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4091 Node *opt_isnan = _gvn.transform(ifisnan);
4092 assert( opt_isnan->is_If(), "Expect an IfNode");
4093 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4094 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4095
4096 set_control(iftrue);
4097
4098 static const jint nan_bits = 0x7fc00000;
4099 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4100 phi->init_req(1, _gvn.transform( slow_result ));
4101 r->init_req(1, iftrue);
4102
4103 // Else fall through
4104 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4105 set_control(iffalse);
4106
4107 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4108 r->init_req(2, iffalse);
4109
4110 // Post merge
4111 set_control(_gvn.transform(r));
4112 record_for_igvn(r);
4113
4114 C->set_has_split_ifs(true); // Has chance for split-if optimization
4115 result = phi;
4116 assert(result->bottom_type()->isa_int(), "must be");
4117 break;
4118 }
4119
4120 default:
4121 fatal_unexpected_iid(id);
4122 break;
4123 }
4124 set_result(_gvn.transform(result));
4125 return true;
4126 }
4127
4128 //----------------------inline_unsafe_copyMemory-------------------------
4129 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4130 bool LibraryCallKit::inline_unsafe_copyMemory() {
4131 if (callee()->is_static()) return false; // caller must have the capability!
4132 null_check_receiver(); // null-check receiver
4133 if (stopped()) return true;
4134
4135 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4136
4137 Node* src_ptr = argument(1); // type: oop
4138 Node* src_off = ConvL2X(argument(2)); // type: long
4139 Node* dst_ptr = argument(4); // type: oop
4140 Node* dst_off = ConvL2X(argument(5)); // type: long
4141 Node* size = ConvL2X(argument(7)); // type: long
4142
4143 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4144 "fieldOffset must be byte-scaled");
4145
4146 Node* src = make_unsafe_address(src_ptr, src_off);
4147 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4148
4149 // Conservatively insert a memory barrier on all memory slices.
4150 // Do not let writes of the copy source or destination float below the copy.
4151 insert_mem_bar(Op_MemBarCPUOrder);
4152
4153 // Call it. Note that the length argument is not scaled.
4154 make_runtime_call(RC_LEAF|RC_NO_FP,
4155 OptoRuntime::fast_arraycopy_Type(),
4156 StubRoutines::unsafe_arraycopy(),
4157 "unsafe_arraycopy",
4158 TypeRawPtr::BOTTOM,
4159 src, dst, size XTOP);
4160
4161 // Do not let reads of the copy destination float above the copy.
4162 insert_mem_bar(Op_MemBarCPUOrder);
4163
4164 return true;
4165 }
4166
4167 //------------------------clone_coping-----------------------------------
4168 // Helper function for inline_native_clone.
4169 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4170 assert(obj_size != NULL, "");
4171 Node* raw_obj = alloc_obj->in(1);
4172 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4173
4174 AllocateNode* alloc = NULL;
4175 if (ReduceBulkZeroing) {
4176 // We will be completely responsible for initializing this object -
4177 // mark Initialize node as complete.
4178 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4179 // The object was just allocated - there should be no any stores!
4180 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4181 // Mark as complete_with_arraycopy so that on AllocateNode
4182 // expansion, we know this AllocateNode is initialized by an array
4183 // copy and a StoreStore barrier exists after the array copy.
4184 alloc->initialization()->set_complete_with_arraycopy();
4185 }
4186
4187 // Copy the fastest available way.
4188 // TODO: generate fields copies for small objects instead.
4189 Node* size = _gvn.transform(obj_size);
4190
4191 access_clone(control(), obj, alloc_obj, size, is_array);
4192
4193 // Do not let reads from the cloned object float above the arraycopy.
4194 if (alloc != NULL) {
4195 // Do not let stores that initialize this object be reordered with
4196 // a subsequent store that would make this object accessible by
4197 // other threads.
4198 // Record what AllocateNode this StoreStore protects so that
4199 // escape analysis can go from the MemBarStoreStoreNode to the
4200 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4201 // based on the escape status of the AllocateNode.
4202 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4203 } else {
4204 insert_mem_bar(Op_MemBarCPUOrder);
4205 }
4206 }
4207
4208 //------------------------inline_native_clone----------------------------
4209 // protected native Object java.lang.Object.clone();
4210 //
4211 // Here are the simple edge cases:
4212 // null receiver => normal trap
4213 // virtual and clone was overridden => slow path to out-of-line clone
4214 // not cloneable or finalizer => slow path to out-of-line Object.clone
4215 //
4216 // The general case has two steps, allocation and copying.
4217 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4218 //
4219 // Copying also has two cases, oop arrays and everything else.
4220 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4221 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4222 //
4223 // These steps fold up nicely if and when the cloned object's klass
4224 // can be sharply typed as an object array, a type array, or an instance.
4225 //
4226 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4227 PhiNode* result_val;
4228
4229 // Set the reexecute bit for the interpreter to reexecute
4230 // the bytecode that invokes Object.clone if deoptimization happens.
4231 { PreserveReexecuteState preexecs(this);
4232 jvms()->set_should_reexecute(true);
4233
4234 Node* obj = null_check_receiver();
4235 if (stopped()) return true;
4236
4237 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4238
4239 // If we are going to clone an instance, we need its exact type to
4240 // know the number and types of fields to convert the clone to
4241 // loads/stores. Maybe a speculative type can help us.
4242 if (!obj_type->klass_is_exact() &&
4243 obj_type->speculative_type() != NULL &&
4244 obj_type->speculative_type()->is_instance_klass()) {
4245 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4246 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4247 !spec_ik->has_injected_fields()) {
4248 ciKlass* k = obj_type->klass();
4249 if (!k->is_instance_klass() ||
4250 k->as_instance_klass()->is_interface() ||
4251 k->as_instance_klass()->has_subklass()) {
4252 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4253 }
4254 }
4255 }
4256
4257 Node* obj_klass = load_object_klass(obj);
4258 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4259 const TypeOopPtr* toop = ((tklass != NULL)
4260 ? tklass->as_instance_type()
4261 : TypeInstPtr::NOTNULL);
4262
4263 // Conservatively insert a memory barrier on all memory slices.
4264 // Do not let writes into the original float below the clone.
4265 insert_mem_bar(Op_MemBarCPUOrder);
4266
4267 // paths into result_reg:
4268 enum {
4269 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4270 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4271 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4272 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4273 PATH_LIMIT
4274 };
4275 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4276 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4277 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4278 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4279 record_for_igvn(result_reg);
4280
4281 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4282 if (array_ctl != NULL) {
4283 // It's an array.
4284 PreserveJVMState pjvms(this);
4285 set_control(array_ctl);
4286 Node* obj_length = load_array_length(obj);
4287 Node* obj_size = NULL;
4288 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4289
4290 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4291 if (bs->array_copy_requires_gc_barriers(T_OBJECT)) {
4292 // If it is an oop array, it requires very special treatment,
4293 // because gc barriers are required when accessing the array.
4294 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4295 if (is_obja != NULL) {
4296 PreserveJVMState pjvms2(this);
4297 set_control(is_obja);
4298 // Generate a direct call to the right arraycopy function(s).
4299 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4300 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4301 ac->set_cloneoop();
4302 Node* n = _gvn.transform(ac);
4303 assert(n == ac, "cannot disappear");
4304 ac->connect_outputs(this);
4305
4306 result_reg->init_req(_objArray_path, control());
4307 result_val->init_req(_objArray_path, alloc_obj);
4308 result_i_o ->set_req(_objArray_path, i_o());
4309 result_mem ->set_req(_objArray_path, reset_memory());
4310 }
4311 }
4312 // Otherwise, there are no barriers to worry about.
4313 // (We can dispense with card marks if we know the allocation
4314 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4315 // causes the non-eden paths to take compensating steps to
4316 // simulate a fresh allocation, so that no further
4317 // card marks are required in compiled code to initialize
4318 // the object.)
4319
4320 if (!stopped()) {
4321 copy_to_clone(obj, alloc_obj, obj_size, true);
4322
4323 // Present the results of the copy.
4324 result_reg->init_req(_array_path, control());
4325 result_val->init_req(_array_path, alloc_obj);
4326 result_i_o ->set_req(_array_path, i_o());
4327 result_mem ->set_req(_array_path, reset_memory());
4328 }
4329 }
4330
4331 // We only go to the instance fast case code if we pass a number of guards.
4332 // The paths which do not pass are accumulated in the slow_region.
4333 RegionNode* slow_region = new RegionNode(1);
4334 record_for_igvn(slow_region);
4335 if (!stopped()) {
4336 // It's an instance (we did array above). Make the slow-path tests.
4337 // If this is a virtual call, we generate a funny guard. We grab
4338 // the vtable entry corresponding to clone() from the target object.
4339 // If the target method which we are calling happens to be the
4340 // Object clone() method, we pass the guard. We do not need this
4341 // guard for non-virtual calls; the caller is known to be the native
4342 // Object clone().
4343 if (is_virtual) {
4344 generate_virtual_guard(obj_klass, slow_region);
4345 }
4346
4347 // The object must be easily cloneable and must not have a finalizer.
4348 // Both of these conditions may be checked in a single test.
4349 // We could optimize the test further, but we don't care.
4350 generate_access_flags_guard(obj_klass,
4351 // Test both conditions:
4352 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4353 // Must be cloneable but not finalizer:
4354 JVM_ACC_IS_CLONEABLE_FAST,
4355 slow_region);
4356 }
4357
4358 if (!stopped()) {
4359 // It's an instance, and it passed the slow-path tests.
4360 PreserveJVMState pjvms(this);
4361 Node* obj_size = NULL;
4362 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4363 // is reexecuted if deoptimization occurs and there could be problems when merging
4364 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4365 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4366
4367 copy_to_clone(obj, alloc_obj, obj_size, false);
4368
4369 // Present the results of the slow call.
4370 result_reg->init_req(_instance_path, control());
4371 result_val->init_req(_instance_path, alloc_obj);
4372 result_i_o ->set_req(_instance_path, i_o());
4373 result_mem ->set_req(_instance_path, reset_memory());
4374 }
4375
4376 // Generate code for the slow case. We make a call to clone().
4377 set_control(_gvn.transform(slow_region));
4378 if (!stopped()) {
4379 PreserveJVMState pjvms(this);
4380 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4381 Node* slow_result = set_results_for_java_call(slow_call);
4382 // this->control() comes from set_results_for_java_call
4383 result_reg->init_req(_slow_path, control());
4384 result_val->init_req(_slow_path, slow_result);
4385 result_i_o ->set_req(_slow_path, i_o());
4386 result_mem ->set_req(_slow_path, reset_memory());
4387 }
4388
4389 // Return the combined state.
4390 set_control( _gvn.transform(result_reg));
4391 set_i_o( _gvn.transform(result_i_o));
4392 set_all_memory( _gvn.transform(result_mem));
4393 } // original reexecute is set back here
4394
4395 set_result(_gvn.transform(result_val));
4396 return true;
4397 }
4398
4399 // If we have a tighly coupled allocation, the arraycopy may take care
4400 // of the array initialization. If one of the guards we insert between
4401 // the allocation and the arraycopy causes a deoptimization, an
4402 // unitialized array will escape the compiled method. To prevent that
4403 // we set the JVM state for uncommon traps between the allocation and
4404 // the arraycopy to the state before the allocation so, in case of
4405 // deoptimization, we'll reexecute the allocation and the
4406 // initialization.
4407 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4408 if (alloc != NULL) {
4409 ciMethod* trap_method = alloc->jvms()->method();
4410 int trap_bci = alloc->jvms()->bci();
4411
4412 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4413 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4414 // Make sure there's no store between the allocation and the
4415 // arraycopy otherwise visible side effects could be rexecuted
4416 // in case of deoptimization and cause incorrect execution.
4417 bool no_interfering_store = true;
4418 Node* mem = alloc->in(TypeFunc::Memory);
4419 if (mem->is_MergeMem()) {
4420 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4421 Node* n = mms.memory();
4422 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4423 assert(n->is_Store(), "what else?");
4424 no_interfering_store = false;
4425 break;
4426 }
4427 }
4428 } else {
4429 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4430 Node* n = mms.memory();
4431 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4432 assert(n->is_Store(), "what else?");
4433 no_interfering_store = false;
4434 break;
4435 }
4436 }
4437 }
4438
4439 if (no_interfering_store) {
4440 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4441 uint size = alloc->req();
4442 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4443 old_jvms->set_map(sfpt);
4444 for (uint i = 0; i < size; i++) {
4445 sfpt->init_req(i, alloc->in(i));
4446 }
4447 // re-push array length for deoptimization
4448 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4449 old_jvms->set_sp(old_jvms->sp()+1);
4450 old_jvms->set_monoff(old_jvms->monoff()+1);
4451 old_jvms->set_scloff(old_jvms->scloff()+1);
4452 old_jvms->set_endoff(old_jvms->endoff()+1);
4453 old_jvms->set_should_reexecute(true);
4454
4455 sfpt->set_i_o(map()->i_o());
4456 sfpt->set_memory(map()->memory());
4457 sfpt->set_control(map()->control());
4458
4459 JVMState* saved_jvms = jvms();
4460 saved_reexecute_sp = _reexecute_sp;
4461
4462 set_jvms(sfpt->jvms());
4463 _reexecute_sp = jvms()->sp();
4464
4465 return saved_jvms;
4466 }
4467 }
4468 }
4469 return NULL;
4470 }
4471
4472 // In case of a deoptimization, we restart execution at the
4473 // allocation, allocating a new array. We would leave an uninitialized
4474 // array in the heap that GCs wouldn't expect. Move the allocation
4475 // after the traps so we don't allocate the array if we
4476 // deoptimize. This is possible because tightly_coupled_allocation()
4477 // guarantees there's no observer of the allocated array at this point
4478 // and the control flow is simple enough.
4479 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4480 int saved_reexecute_sp, uint new_idx) {
4481 if (saved_jvms != NULL && !stopped()) {
4482 assert(alloc != NULL, "only with a tightly coupled allocation");
4483 // restore JVM state to the state at the arraycopy
4484 saved_jvms->map()->set_control(map()->control());
4485 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4486 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4487 // If we've improved the types of some nodes (null check) while
4488 // emitting the guards, propagate them to the current state
4489 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4490 set_jvms(saved_jvms);
4491 _reexecute_sp = saved_reexecute_sp;
4492
4493 // Remove the allocation from above the guards
4494 CallProjections callprojs;
4495 alloc->extract_projections(&callprojs, true);
4496 InitializeNode* init = alloc->initialization();
4497 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4498 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4499 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4500 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4501
4502 // move the allocation here (after the guards)
4503 _gvn.hash_delete(alloc);
4504 alloc->set_req(TypeFunc::Control, control());
4505 alloc->set_req(TypeFunc::I_O, i_o());
4506 Node *mem = reset_memory();
4507 set_all_memory(mem);
4508 alloc->set_req(TypeFunc::Memory, mem);
4509 set_control(init->proj_out_or_null(TypeFunc::Control));
4510 set_i_o(callprojs.fallthrough_ioproj);
4511
4512 // Update memory as done in GraphKit::set_output_for_allocation()
4513 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4514 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4515 if (ary_type->isa_aryptr() && length_type != NULL) {
4516 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4517 }
4518 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4519 int elemidx = C->get_alias_index(telemref);
4520 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4521 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4522
4523 Node* allocx = _gvn.transform(alloc);
4524 assert(allocx == alloc, "where has the allocation gone?");
4525 assert(dest->is_CheckCastPP(), "not an allocation result?");
4526
4527 _gvn.hash_delete(dest);
4528 dest->set_req(0, control());
4529 Node* destx = _gvn.transform(dest);
4530 assert(destx == dest, "where has the allocation result gone?");
4531 }
4532 }
4533
4534
4535 //------------------------------inline_arraycopy-----------------------
4536 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4537 // Object dest, int destPos,
4538 // int length);
4539 bool LibraryCallKit::inline_arraycopy() {
4540 // Get the arguments.
4541 Node* src = argument(0); // type: oop
4542 Node* src_offset = argument(1); // type: int
4543 Node* dest = argument(2); // type: oop
4544 Node* dest_offset = argument(3); // type: int
4545 Node* length = argument(4); // type: int
4546
4547 uint new_idx = C->unique();
4548
4549 // Check for allocation before we add nodes that would confuse
4550 // tightly_coupled_allocation()
4551 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4552
4553 int saved_reexecute_sp = -1;
4554 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4555 // See arraycopy_restore_alloc_state() comment
4556 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4557 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4558 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4559 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4560
4561 // The following tests must be performed
4562 // (1) src and dest are arrays.
4563 // (2) src and dest arrays must have elements of the same BasicType
4564 // (3) src and dest must not be null.
4565 // (4) src_offset must not be negative.
4566 // (5) dest_offset must not be negative.
4567 // (6) length must not be negative.
4568 // (7) src_offset + length must not exceed length of src.
4569 // (8) dest_offset + length must not exceed length of dest.
4570 // (9) each element of an oop array must be assignable
4571
4572 // (3) src and dest must not be null.
4573 // always do this here because we need the JVM state for uncommon traps
4574 Node* null_ctl = top();
4575 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4576 assert(null_ctl->is_top(), "no null control here");
4577 dest = null_check(dest, T_ARRAY);
4578
4579 if (!can_emit_guards) {
4580 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4581 // guards but the arraycopy node could still take advantage of a
4582 // tightly allocated allocation. tightly_coupled_allocation() is
4583 // called again to make sure it takes the null check above into
4584 // account: the null check is mandatory and if it caused an
4585 // uncommon trap to be emitted then the allocation can't be
4586 // considered tightly coupled in this context.
4587 alloc = tightly_coupled_allocation(dest, NULL);
4588 }
4589
4590 bool validated = false;
4591
4592 const Type* src_type = _gvn.type(src);
4593 const Type* dest_type = _gvn.type(dest);
4594 const TypeAryPtr* top_src = src_type->isa_aryptr();
4595 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4596
4597 // Do we have the type of src?
4598 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4599 // Do we have the type of dest?
4600 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4601 // Is the type for src from speculation?
4602 bool src_spec = false;
4603 // Is the type for dest from speculation?
4604 bool dest_spec = false;
4605
4606 if ((!has_src || !has_dest) && can_emit_guards) {
4607 // We don't have sufficient type information, let's see if
4608 // speculative types can help. We need to have types for both src
4609 // and dest so that it pays off.
4610
4611 // Do we already have or could we have type information for src
4612 bool could_have_src = has_src;
4613 // Do we already have or could we have type information for dest
4614 bool could_have_dest = has_dest;
4615
4616 ciKlass* src_k = NULL;
4617 if (!has_src) {
4618 src_k = src_type->speculative_type_not_null();
4619 if (src_k != NULL && src_k->is_array_klass()) {
4620 could_have_src = true;
4621 }
4622 }
4623
4624 ciKlass* dest_k = NULL;
4625 if (!has_dest) {
4626 dest_k = dest_type->speculative_type_not_null();
4627 if (dest_k != NULL && dest_k->is_array_klass()) {
4628 could_have_dest = true;
4629 }
4630 }
4631
4632 if (could_have_src && could_have_dest) {
4633 // This is going to pay off so emit the required guards
4634 if (!has_src) {
4635 src = maybe_cast_profiled_obj(src, src_k, true);
4636 src_type = _gvn.type(src);
4637 top_src = src_type->isa_aryptr();
4638 has_src = (top_src != NULL && top_src->klass() != NULL);
4639 src_spec = true;
4640 }
4641 if (!has_dest) {
4642 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4643 dest_type = _gvn.type(dest);
4644 top_dest = dest_type->isa_aryptr();
4645 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4646 dest_spec = true;
4647 }
4648 }
4649 }
4650
4651 if (has_src && has_dest && can_emit_guards) {
4652 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4653 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4654 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
4655 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
4656
4657 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4658 // If both arrays are object arrays then having the exact types
4659 // for both will remove the need for a subtype check at runtime
4660 // before the call and may make it possible to pick a faster copy
4661 // routine (without a subtype check on every element)
4662 // Do we have the exact type of src?
4663 bool could_have_src = src_spec;
4664 // Do we have the exact type of dest?
4665 bool could_have_dest = dest_spec;
4666 ciKlass* src_k = top_src->klass();
4667 ciKlass* dest_k = top_dest->klass();
4668 if (!src_spec) {
4669 src_k = src_type->speculative_type_not_null();
4670 if (src_k != NULL && src_k->is_array_klass()) {
4671 could_have_src = true;
4672 }
4673 }
4674 if (!dest_spec) {
4675 dest_k = dest_type->speculative_type_not_null();
4676 if (dest_k != NULL && dest_k->is_array_klass()) {
4677 could_have_dest = true;
4678 }
4679 }
4680 if (could_have_src && could_have_dest) {
4681 // If we can have both exact types, emit the missing guards
4682 if (could_have_src && !src_spec) {
4683 src = maybe_cast_profiled_obj(src, src_k, true);
4684 }
4685 if (could_have_dest && !dest_spec) {
4686 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4687 }
4688 }
4689 }
4690 }
4691
4692 ciMethod* trap_method = method();
4693 int trap_bci = bci();
4694 if (saved_jvms != NULL) {
4695 trap_method = alloc->jvms()->method();
4696 trap_bci = alloc->jvms()->bci();
4697 }
4698
4699 bool negative_length_guard_generated = false;
4700
4701 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4702 can_emit_guards &&
4703 !src->is_top() && !dest->is_top()) {
4704 // validate arguments: enables transformation the ArrayCopyNode
4705 validated = true;
4706
4707 RegionNode* slow_region = new RegionNode(1);
4708 record_for_igvn(slow_region);
4709
4710 // (1) src and dest are arrays.
4711 generate_non_array_guard(load_object_klass(src), slow_region);
4712 generate_non_array_guard(load_object_klass(dest), slow_region);
4713
4714 // (2) src and dest arrays must have elements of the same BasicType
4715 // done at macro expansion or at Ideal transformation time
4716
4717 // (4) src_offset must not be negative.
4718 generate_negative_guard(src_offset, slow_region);
4719
4720 // (5) dest_offset must not be negative.
4721 generate_negative_guard(dest_offset, slow_region);
4722
4723 // (7) src_offset + length must not exceed length of src.
4724 generate_limit_guard(src_offset, length,
4725 load_array_length(src),
4726 slow_region);
4727
4728 // (8) dest_offset + length must not exceed length of dest.
4729 generate_limit_guard(dest_offset, length,
4730 load_array_length(dest),
4731 slow_region);
4732
4733 // (6) length must not be negative.
4734 // This is also checked in generate_arraycopy() during macro expansion, but
4735 // we also have to check it here for the case where the ArrayCopyNode will
4736 // be eliminated by Escape Analysis.
4737 if (EliminateAllocations) {
4738 generate_negative_guard(length, slow_region);
4739 negative_length_guard_generated = true;
4740 }
4741
4742 // (9) each element of an oop array must be assignable
4743 Node* src_klass = load_object_klass(src);
4744 Node* dest_klass = load_object_klass(dest);
4745 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4746
4747 if (not_subtype_ctrl != top()) {
4748 PreserveJVMState pjvms(this);
4749 set_control(not_subtype_ctrl);
4750 uncommon_trap(Deoptimization::Reason_intrinsic,
4751 Deoptimization::Action_make_not_entrant);
4752 assert(stopped(), "Should be stopped");
4753 }
4754 {
4755 PreserveJVMState pjvms(this);
4756 set_control(_gvn.transform(slow_region));
4757 uncommon_trap(Deoptimization::Reason_intrinsic,
4758 Deoptimization::Action_make_not_entrant);
4759 assert(stopped(), "Should be stopped");
4760 }
4761
4762 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4763 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4764 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4765 }
4766
4767 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4768
4769 if (stopped()) {
4770 return true;
4771 }
4772
4773 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4774 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4775 // so the compiler has a chance to eliminate them: during macro expansion,
4776 // we have to set their control (CastPP nodes are eliminated).
4777 load_object_klass(src), load_object_klass(dest),
4778 load_array_length(src), load_array_length(dest));
4779
4780 ac->set_arraycopy(validated);
4781
4782 Node* n = _gvn.transform(ac);
4783 if (n == ac) {
4784 ac->connect_outputs(this);
4785 } else {
4786 assert(validated, "shouldn't transform if all arguments not validated");
4787 set_all_memory(n);
4788 }
4789 clear_upper_avx();
4790
4791
4792 return true;
4793 }
4794
4795
4796 // Helper function which determines if an arraycopy immediately follows
4797 // an allocation, with no intervening tests or other escapes for the object.
4798 AllocateArrayNode*
4799 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4800 RegionNode* slow_region) {
4801 if (stopped()) return NULL; // no fast path
4802 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4803
4804 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4805 if (alloc == NULL) return NULL;
4806
4807 Node* rawmem = memory(Compile::AliasIdxRaw);
4808 // Is the allocation's memory state untouched?
4809 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4810 // Bail out if there have been raw-memory effects since the allocation.
4811 // (Example: There might have been a call or safepoint.)
4812 return NULL;
4813 }
4814 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4815 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4816 return NULL;
4817 }
4818
4819 // There must be no unexpected observers of this allocation.
4820 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4821 Node* obs = ptr->fast_out(i);
4822 if (obs != this->map()) {
4823 return NULL;
4824 }
4825 }
4826
4827 // This arraycopy must unconditionally follow the allocation of the ptr.
4828 Node* alloc_ctl = ptr->in(0);
4829 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4830
4831 Node* ctl = control();
4832 while (ctl != alloc_ctl) {
4833 // There may be guards which feed into the slow_region.
4834 // Any other control flow means that we might not get a chance
4835 // to finish initializing the allocated object.
4836 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4837 IfNode* iff = ctl->in(0)->as_If();
4838 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4839 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4840 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4841 ctl = iff->in(0); // This test feeds the known slow_region.
4842 continue;
4843 }
4844 // One more try: Various low-level checks bottom out in
4845 // uncommon traps. If the debug-info of the trap omits
4846 // any reference to the allocation, as we've already
4847 // observed, then there can be no objection to the trap.
4848 bool found_trap = false;
4849 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4850 Node* obs = not_ctl->fast_out(j);
4851 if (obs->in(0) == not_ctl && obs->is_Call() &&
4852 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4853 found_trap = true; break;
4854 }
4855 }
4856 if (found_trap) {
4857 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4858 continue;
4859 }
4860 }
4861 return NULL;
4862 }
4863
4864 // If we get this far, we have an allocation which immediately
4865 // precedes the arraycopy, and we can take over zeroing the new object.
4866 // The arraycopy will finish the initialization, and provide
4867 // a new control state to which we will anchor the destination pointer.
4868
4869 return alloc;
4870 }
4871
4872 //-------------inline_encodeISOArray-----------------------------------
4873 // encode char[] to byte[] in ISO_8859_1
4874 bool LibraryCallKit::inline_encodeISOArray() {
4875 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4876 // no receiver since it is static method
4877 Node *src = argument(0);
4878 Node *src_offset = argument(1);
4879 Node *dst = argument(2);
4880 Node *dst_offset = argument(3);
4881 Node *length = argument(4);
4882
4883 const Type* src_type = src->Value(&_gvn);
4884 const Type* dst_type = dst->Value(&_gvn);
4885 const TypeAryPtr* top_src = src_type->isa_aryptr();
4886 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4887 if (top_src == NULL || top_src->klass() == NULL ||
4888 top_dest == NULL || top_dest->klass() == NULL) {
4889 // failed array check
4890 return false;
4891 }
4892
4893 // Figure out the size and type of the elements we will be copying.
4894 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4895 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4896 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4897 return false;
4898 }
4899
4900 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4901 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4902 // 'src_start' points to src array + scaled offset
4903 // 'dst_start' points to dst array + scaled offset
4904
4905 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4906 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4907 enc = _gvn.transform(enc);
4908 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4909 set_memory(res_mem, mtype);
4910 set_result(enc);
4911 clear_upper_avx();
4912
4913 return true;
4914 }
4915
4916 //-------------inline_multiplyToLen-----------------------------------
4917 bool LibraryCallKit::inline_multiplyToLen() {
4918 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4919
4920 address stubAddr = StubRoutines::multiplyToLen();
4921 if (stubAddr == NULL) {
4922 return false; // Intrinsic's stub is not implemented on this platform
4923 }
4924 const char* stubName = "multiplyToLen";
4925
4926 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4927
4928 // no receiver because it is a static method
4929 Node* x = argument(0);
4930 Node* xlen = argument(1);
4931 Node* y = argument(2);
4932 Node* ylen = argument(3);
4933 Node* z = argument(4);
4934
4935 const Type* x_type = x->Value(&_gvn);
4936 const Type* y_type = y->Value(&_gvn);
4937 const TypeAryPtr* top_x = x_type->isa_aryptr();
4938 const TypeAryPtr* top_y = y_type->isa_aryptr();
4939 if (top_x == NULL || top_x->klass() == NULL ||
4940 top_y == NULL || top_y->klass() == NULL) {
4941 // failed array check
4942 return false;
4943 }
4944
4945 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4946 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4947 if (x_elem != T_INT || y_elem != T_INT) {
4948 return false;
4949 }
4950
4951 // Set the original stack and the reexecute bit for the interpreter to reexecute
4952 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
4953 // on the return from z array allocation in runtime.
4954 { PreserveReexecuteState preexecs(this);
4955 jvms()->set_should_reexecute(true);
4956
4957 Node* x_start = array_element_address(x, intcon(0), x_elem);
4958 Node* y_start = array_element_address(y, intcon(0), y_elem);
4959 // 'x_start' points to x array + scaled xlen
4960 // 'y_start' points to y array + scaled ylen
4961
4962 // Allocate the result array
4963 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
4964 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
4965 Node* klass_node = makecon(TypeKlassPtr::make(klass));
4966
4967 IdealKit ideal(this);
4968
4969 #define __ ideal.
4970 Node* one = __ ConI(1);
4971 Node* zero = __ ConI(0);
4972 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
4973 __ set(need_alloc, zero);
4974 __ set(z_alloc, z);
4975 __ if_then(z, BoolTest::eq, null()); {
4976 __ increment (need_alloc, one);
4977 } __ else_(); {
4978 // Update graphKit memory and control from IdealKit.
4979 sync_kit(ideal);
4980 Node* zlen_arg = load_array_length(z);
4981 // Update IdealKit memory and control from graphKit.
4982 __ sync_kit(this);
4983 __ if_then(zlen_arg, BoolTest::lt, zlen); {
4984 __ increment (need_alloc, one);
4985 } __ end_if();
4986 } __ end_if();
4987
4988 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
4989 // Update graphKit memory and control from IdealKit.
4990 sync_kit(ideal);
4991 Node * narr = new_array(klass_node, zlen, 1);
4992 // Update IdealKit memory and control from graphKit.
4993 __ sync_kit(this);
4994 __ set(z_alloc, narr);
4995 } __ end_if();
4996
4997 sync_kit(ideal);
4998 z = __ value(z_alloc);
4999 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5000 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5001 // Final sync IdealKit and GraphKit.
5002 final_sync(ideal);
5003 #undef __
5004
5005 Node* z_start = array_element_address(z, intcon(0), T_INT);
5006
5007 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5008 OptoRuntime::multiplyToLen_Type(),
5009 stubAddr, stubName, TypePtr::BOTTOM,
5010 x_start, xlen, y_start, ylen, z_start, zlen);
5011 } // original reexecute is set back here
5012
5013 C->set_has_split_ifs(true); // Has chance for split-if optimization
5014 set_result(z);
5015 return true;
5016 }
5017
5018 //-------------inline_squareToLen------------------------------------
5019 bool LibraryCallKit::inline_squareToLen() {
5020 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5021
5022 address stubAddr = StubRoutines::squareToLen();
5023 if (stubAddr == NULL) {
5024 return false; // Intrinsic's stub is not implemented on this platform
5025 }
5026 const char* stubName = "squareToLen";
5027
5028 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5029
5030 Node* x = argument(0);
5031 Node* len = argument(1);
5032 Node* z = argument(2);
5033 Node* zlen = argument(3);
5034
5035 const Type* x_type = x->Value(&_gvn);
5036 const Type* z_type = z->Value(&_gvn);
5037 const TypeAryPtr* top_x = x_type->isa_aryptr();
5038 const TypeAryPtr* top_z = z_type->isa_aryptr();
5039 if (top_x == NULL || top_x->klass() == NULL ||
5040 top_z == NULL || top_z->klass() == NULL) {
5041 // failed array check
5042 return false;
5043 }
5044
5045 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5046 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5047 if (x_elem != T_INT || z_elem != T_INT) {
5048 return false;
5049 }
5050
5051
5052 Node* x_start = array_element_address(x, intcon(0), x_elem);
5053 Node* z_start = array_element_address(z, intcon(0), z_elem);
5054
5055 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5056 OptoRuntime::squareToLen_Type(),
5057 stubAddr, stubName, TypePtr::BOTTOM,
5058 x_start, len, z_start, zlen);
5059
5060 set_result(z);
5061 return true;
5062 }
5063
5064 //-------------inline_mulAdd------------------------------------------
5065 bool LibraryCallKit::inline_mulAdd() {
5066 assert(UseMulAddIntrinsic, "not implemented on this platform");
5067
5068 address stubAddr = StubRoutines::mulAdd();
5069 if (stubAddr == NULL) {
5070 return false; // Intrinsic's stub is not implemented on this platform
5071 }
5072 const char* stubName = "mulAdd";
5073
5074 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5075
5076 Node* out = argument(0);
5077 Node* in = argument(1);
5078 Node* offset = argument(2);
5079 Node* len = argument(3);
5080 Node* k = argument(4);
5081
5082 const Type* out_type = out->Value(&_gvn);
5083 const Type* in_type = in->Value(&_gvn);
5084 const TypeAryPtr* top_out = out_type->isa_aryptr();
5085 const TypeAryPtr* top_in = in_type->isa_aryptr();
5086 if (top_out == NULL || top_out->klass() == NULL ||
5087 top_in == NULL || top_in->klass() == NULL) {
5088 // failed array check
5089 return false;
5090 }
5091
5092 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5093 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5094 if (out_elem != T_INT || in_elem != T_INT) {
5095 return false;
5096 }
5097
5098 Node* outlen = load_array_length(out);
5099 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5100 Node* out_start = array_element_address(out, intcon(0), out_elem);
5101 Node* in_start = array_element_address(in, intcon(0), in_elem);
5102
5103 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5104 OptoRuntime::mulAdd_Type(),
5105 stubAddr, stubName, TypePtr::BOTTOM,
5106 out_start,in_start, new_offset, len, k);
5107 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5108 set_result(result);
5109 return true;
5110 }
5111
5112 //-------------inline_montgomeryMultiply-----------------------------------
5113 bool LibraryCallKit::inline_montgomeryMultiply() {
5114 address stubAddr = StubRoutines::montgomeryMultiply();
5115 if (stubAddr == NULL) {
5116 return false; // Intrinsic's stub is not implemented on this platform
5117 }
5118
5119 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5120 const char* stubName = "montgomery_multiply";
5121
5122 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5123
5124 Node* a = argument(0);
5125 Node* b = argument(1);
5126 Node* n = argument(2);
5127 Node* len = argument(3);
5128 Node* inv = argument(4);
5129 Node* m = argument(6);
5130
5131 const Type* a_type = a->Value(&_gvn);
5132 const TypeAryPtr* top_a = a_type->isa_aryptr();
5133 const Type* b_type = b->Value(&_gvn);
5134 const TypeAryPtr* top_b = b_type->isa_aryptr();
5135 const Type* n_type = a->Value(&_gvn);
5136 const TypeAryPtr* top_n = n_type->isa_aryptr();
5137 const Type* m_type = a->Value(&_gvn);
5138 const TypeAryPtr* top_m = m_type->isa_aryptr();
5139 if (top_a == NULL || top_a->klass() == NULL ||
5140 top_b == NULL || top_b->klass() == NULL ||
5141 top_n == NULL || top_n->klass() == NULL ||
5142 top_m == NULL || top_m->klass() == NULL) {
5143 // failed array check
5144 return false;
5145 }
5146
5147 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5148 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5149 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5150 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5151 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5152 return false;
5153 }
5154
5155 // Make the call
5156 {
5157 Node* a_start = array_element_address(a, intcon(0), a_elem);
5158 Node* b_start = array_element_address(b, intcon(0), b_elem);
5159 Node* n_start = array_element_address(n, intcon(0), n_elem);
5160 Node* m_start = array_element_address(m, intcon(0), m_elem);
5161
5162 Node* call = make_runtime_call(RC_LEAF,
5163 OptoRuntime::montgomeryMultiply_Type(),
5164 stubAddr, stubName, TypePtr::BOTTOM,
5165 a_start, b_start, n_start, len, inv, top(),
5166 m_start);
5167 set_result(m);
5168 }
5169
5170 return true;
5171 }
5172
5173 bool LibraryCallKit::inline_montgomerySquare() {
5174 address stubAddr = StubRoutines::montgomerySquare();
5175 if (stubAddr == NULL) {
5176 return false; // Intrinsic's stub is not implemented on this platform
5177 }
5178
5179 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5180 const char* stubName = "montgomery_square";
5181
5182 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5183
5184 Node* a = argument(0);
5185 Node* n = argument(1);
5186 Node* len = argument(2);
5187 Node* inv = argument(3);
5188 Node* m = argument(5);
5189
5190 const Type* a_type = a->Value(&_gvn);
5191 const TypeAryPtr* top_a = a_type->isa_aryptr();
5192 const Type* n_type = a->Value(&_gvn);
5193 const TypeAryPtr* top_n = n_type->isa_aryptr();
5194 const Type* m_type = a->Value(&_gvn);
5195 const TypeAryPtr* top_m = m_type->isa_aryptr();
5196 if (top_a == NULL || top_a->klass() == NULL ||
5197 top_n == NULL || top_n->klass() == NULL ||
5198 top_m == NULL || top_m->klass() == NULL) {
5199 // failed array check
5200 return false;
5201 }
5202
5203 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5204 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5205 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5206 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5207 return false;
5208 }
5209
5210 // Make the call
5211 {
5212 Node* a_start = array_element_address(a, intcon(0), a_elem);
5213 Node* n_start = array_element_address(n, intcon(0), n_elem);
5214 Node* m_start = array_element_address(m, intcon(0), m_elem);
5215
5216 Node* call = make_runtime_call(RC_LEAF,
5217 OptoRuntime::montgomerySquare_Type(),
5218 stubAddr, stubName, TypePtr::BOTTOM,
5219 a_start, n_start, len, inv, top(),
5220 m_start);
5221 set_result(m);
5222 }
5223
5224 return true;
5225 }
5226
5227 //-------------inline_vectorizedMismatch------------------------------
5228 bool LibraryCallKit::inline_vectorizedMismatch() {
5229 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5230
5231 address stubAddr = StubRoutines::vectorizedMismatch();
5232 if (stubAddr == NULL) {
5233 return false; // Intrinsic's stub is not implemented on this platform
5234 }
5235 const char* stubName = "vectorizedMismatch";
5236 int size_l = callee()->signature()->size();
5237 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5238
5239 Node* obja = argument(0);
5240 Node* aoffset = argument(1);
5241 Node* objb = argument(3);
5242 Node* boffset = argument(4);
5243 Node* length = argument(6);
5244 Node* scale = argument(7);
5245
5246 const Type* a_type = obja->Value(&_gvn);
5247 const Type* b_type = objb->Value(&_gvn);
5248 const TypeAryPtr* top_a = a_type->isa_aryptr();
5249 const TypeAryPtr* top_b = b_type->isa_aryptr();
5250 if (top_a == NULL || top_a->klass() == NULL ||
5251 top_b == NULL || top_b->klass() == NULL) {
5252 // failed array check
5253 return false;
5254 }
5255
5256 Node* call;
5257 jvms()->set_should_reexecute(true);
5258
5259 Node* obja_adr = make_unsafe_address(obja, aoffset);
5260 Node* objb_adr = make_unsafe_address(objb, boffset);
5261
5262 call = make_runtime_call(RC_LEAF,
5263 OptoRuntime::vectorizedMismatch_Type(),
5264 stubAddr, stubName, TypePtr::BOTTOM,
5265 obja_adr, objb_adr, length, scale);
5266
5267 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5268 set_result(result);
5269 return true;
5270 }
5271
5272 /**
5273 * Calculate CRC32 for byte.
5274 * int java.util.zip.CRC32.update(int crc, int b)
5275 */
5276 bool LibraryCallKit::inline_updateCRC32() {
5277 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5278 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5279 // no receiver since it is static method
5280 Node* crc = argument(0); // type: int
5281 Node* b = argument(1); // type: int
5282
5283 /*
5284 * int c = ~ crc;
5285 * b = timesXtoThe32[(b ^ c) & 0xFF];
5286 * b = b ^ (c >>> 8);
5287 * crc = ~b;
5288 */
5289
5290 Node* M1 = intcon(-1);
5291 crc = _gvn.transform(new XorINode(crc, M1));
5292 Node* result = _gvn.transform(new XorINode(crc, b));
5293 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5294
5295 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5296 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5297 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5298 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5299
5300 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5301 result = _gvn.transform(new XorINode(crc, result));
5302 result = _gvn.transform(new XorINode(result, M1));
5303 set_result(result);
5304 return true;
5305 }
5306
5307 /**
5308 * Calculate CRC32 for byte[] array.
5309 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5310 */
5311 bool LibraryCallKit::inline_updateBytesCRC32() {
5312 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5313 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5314 // no receiver since it is static method
5315 Node* crc = argument(0); // type: int
5316 Node* src = argument(1); // type: oop
5317 Node* offset = argument(2); // type: int
5318 Node* length = argument(3); // type: int
5319
5320 const Type* src_type = src->Value(&_gvn);
5321 const TypeAryPtr* top_src = src_type->isa_aryptr();
5322 if (top_src == NULL || top_src->klass() == NULL) {
5323 // failed array check
5324 return false;
5325 }
5326
5327 // Figure out the size and type of the elements we will be copying.
5328 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5329 if (src_elem != T_BYTE) {
5330 return false;
5331 }
5332
5333 // 'src_start' points to src array + scaled offset
5334 Node* src_start = array_element_address(src, offset, src_elem);
5335
5336 // We assume that range check is done by caller.
5337 // TODO: generate range check (offset+length < src.length) in debug VM.
5338
5339 // Call the stub.
5340 address stubAddr = StubRoutines::updateBytesCRC32();
5341 const char *stubName = "updateBytesCRC32";
5342
5343 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5344 stubAddr, stubName, TypePtr::BOTTOM,
5345 crc, src_start, length);
5346 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5347 set_result(result);
5348 return true;
5349 }
5350
5351 /**
5352 * Calculate CRC32 for ByteBuffer.
5353 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5354 */
5355 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5356 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5357 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5358 // no receiver since it is static method
5359 Node* crc = argument(0); // type: int
5360 Node* src = argument(1); // type: long
5361 Node* offset = argument(3); // type: int
5362 Node* length = argument(4); // type: int
5363
5364 src = ConvL2X(src); // adjust Java long to machine word
5365 Node* base = _gvn.transform(new CastX2PNode(src));
5366 offset = ConvI2X(offset);
5367
5368 // 'src_start' points to src array + scaled offset
5369 Node* src_start = basic_plus_adr(top(), base, offset);
5370
5371 // Call the stub.
5372 address stubAddr = StubRoutines::updateBytesCRC32();
5373 const char *stubName = "updateBytesCRC32";
5374
5375 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5376 stubAddr, stubName, TypePtr::BOTTOM,
5377 crc, src_start, length);
5378 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5379 set_result(result);
5380 return true;
5381 }
5382
5383 //------------------------------get_table_from_crc32c_class-----------------------
5384 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5385 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5386 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5387
5388 return table;
5389 }
5390
5391 //------------------------------inline_updateBytesCRC32C-----------------------
5392 //
5393 // Calculate CRC32C for byte[] array.
5394 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5395 //
5396 bool LibraryCallKit::inline_updateBytesCRC32C() {
5397 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5398 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5399 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5400 // no receiver since it is a static method
5401 Node* crc = argument(0); // type: int
5402 Node* src = argument(1); // type: oop
5403 Node* offset = argument(2); // type: int
5404 Node* end = argument(3); // type: int
5405
5406 Node* length = _gvn.transform(new SubINode(end, offset));
5407
5408 const Type* src_type = src->Value(&_gvn);
5409 const TypeAryPtr* top_src = src_type->isa_aryptr();
5410 if (top_src == NULL || top_src->klass() == NULL) {
5411 // failed array check
5412 return false;
5413 }
5414
5415 // Figure out the size and type of the elements we will be copying.
5416 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5417 if (src_elem != T_BYTE) {
5418 return false;
5419 }
5420
5421 // 'src_start' points to src array + scaled offset
5422 Node* src_start = array_element_address(src, offset, src_elem);
5423
5424 // static final int[] byteTable in class CRC32C
5425 Node* table = get_table_from_crc32c_class(callee()->holder());
5426 Node* table_start = array_element_address(table, intcon(0), T_INT);
5427
5428 // We assume that range check is done by caller.
5429 // TODO: generate range check (offset+length < src.length) in debug VM.
5430
5431 // Call the stub.
5432 address stubAddr = StubRoutines::updateBytesCRC32C();
5433 const char *stubName = "updateBytesCRC32C";
5434
5435 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5436 stubAddr, stubName, TypePtr::BOTTOM,
5437 crc, src_start, length, table_start);
5438 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5439 set_result(result);
5440 return true;
5441 }
5442
5443 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5444 //
5445 // Calculate CRC32C for DirectByteBuffer.
5446 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5447 //
5448 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5449 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5450 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5451 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5452 // no receiver since it is a static method
5453 Node* crc = argument(0); // type: int
5454 Node* src = argument(1); // type: long
5455 Node* offset = argument(3); // type: int
5456 Node* end = argument(4); // type: int
5457
5458 Node* length = _gvn.transform(new SubINode(end, offset));
5459
5460 src = ConvL2X(src); // adjust Java long to machine word
5461 Node* base = _gvn.transform(new CastX2PNode(src));
5462 offset = ConvI2X(offset);
5463
5464 // 'src_start' points to src array + scaled offset
5465 Node* src_start = basic_plus_adr(top(), base, offset);
5466
5467 // static final int[] byteTable in class CRC32C
5468 Node* table = get_table_from_crc32c_class(callee()->holder());
5469 Node* table_start = array_element_address(table, intcon(0), T_INT);
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_updateBytesAdler32----------------------
5484 //
5485 // Calculate Adler32 checksum for byte[] array.
5486 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5487 //
5488 bool LibraryCallKit::inline_updateBytesAdler32() {
5489 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5490 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5491 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5492 // no receiver since it is static method
5493 Node* crc = argument(0); // type: int
5494 Node* src = argument(1); // type: oop
5495 Node* offset = argument(2); // type: int
5496 Node* length = argument(3); // type: int
5497
5498 const Type* src_type = src->Value(&_gvn);
5499 const TypeAryPtr* top_src = src_type->isa_aryptr();
5500 if (top_src == NULL || top_src->klass() == NULL) {
5501 // failed array check
5502 return false;
5503 }
5504
5505 // Figure out the size and type of the elements we will be copying.
5506 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5507 if (src_elem != T_BYTE) {
5508 return false;
5509 }
5510
5511 // 'src_start' points to src array + scaled offset
5512 Node* src_start = array_element_address(src, offset, src_elem);
5513
5514 // We assume that range check is done by caller.
5515 // TODO: generate range check (offset+length < src.length) in debug VM.
5516
5517 // Call the stub.
5518 address stubAddr = StubRoutines::updateBytesAdler32();
5519 const char *stubName = "updateBytesAdler32";
5520
5521 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5522 stubAddr, stubName, TypePtr::BOTTOM,
5523 crc, src_start, length);
5524 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5525 set_result(result);
5526 return true;
5527 }
5528
5529 //------------------------------inline_updateByteBufferAdler32---------------
5530 //
5531 // Calculate Adler32 checksum for DirectByteBuffer.
5532 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5533 //
5534 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5535 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5536 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5537 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5538 // no receiver since it is static method
5539 Node* crc = argument(0); // type: int
5540 Node* src = argument(1); // type: long
5541 Node* offset = argument(3); // type: int
5542 Node* length = argument(4); // type: int
5543
5544 src = ConvL2X(src); // adjust Java long to machine word
5545 Node* base = _gvn.transform(new CastX2PNode(src));
5546 offset = ConvI2X(offset);
5547
5548 // 'src_start' points to src array + scaled offset
5549 Node* src_start = basic_plus_adr(top(), base, offset);
5550
5551 // Call the stub.
5552 address stubAddr = StubRoutines::updateBytesAdler32();
5553 const char *stubName = "updateBytesAdler32";
5554
5555 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5556 stubAddr, stubName, TypePtr::BOTTOM,
5557 crc, src_start, length);
5558
5559 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5560 set_result(result);
5561 return true;
5562 }
5563
5564 //----------------------------inline_reference_get----------------------------
5565 // public T java.lang.ref.Reference.get();
5566 bool LibraryCallKit::inline_reference_get() {
5567 const int referent_offset = java_lang_ref_Reference::referent_offset;
5568 guarantee(referent_offset > 0, "should have already been set");
5569
5570 // Get the argument:
5571 Node* reference_obj = null_check_receiver();
5572 if (stopped()) return true;
5573
5574 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5575 assert(tinst != NULL, "obj is null");
5576 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5577 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5578 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5579 ciSymbol::make("Ljava/lang/Object;"),
5580 false);
5581 assert (field != NULL, "undefined field");
5582
5583 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5584 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5585
5586 ciInstanceKlass* klass = env()->Object_klass();
5587 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5588
5589 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5590 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5591 // Add memory barrier to prevent commoning reads from this field
5592 // across safepoint since GC can change its value.
5593 insert_mem_bar(Op_MemBarCPUOrder);
5594
5595 set_result(result);
5596 return true;
5597 }
5598
5599
5600 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5601 bool is_exact=true, bool is_static=false,
5602 ciInstanceKlass * fromKls=NULL) {
5603 if (fromKls == NULL) {
5604 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5605 assert(tinst != NULL, "obj is null");
5606 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5607 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5608 fromKls = tinst->klass()->as_instance_klass();
5609 } else {
5610 assert(is_static, "only for static field access");
5611 }
5612 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5613 ciSymbol::make(fieldTypeString),
5614 is_static);
5615
5616 assert (field != NULL, "undefined field");
5617 if (field == NULL) return (Node *) NULL;
5618
5619 if (is_static) {
5620 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5621 fromObj = makecon(tip);
5622 }
5623
5624 // Next code copied from Parse::do_get_xxx():
5625
5626 // Compute address and memory type.
5627 int offset = field->offset_in_bytes();
5628 bool is_vol = field->is_volatile();
5629 ciType* field_klass = field->type();
5630 assert(field_klass->is_loaded(), "should be loaded");
5631 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5632 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5633 BasicType bt = field->layout_type();
5634
5635 // Build the resultant type of the load
5636 const Type *type;
5637 if (bt == T_OBJECT) {
5638 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5639 } else {
5640 type = Type::get_const_basic_type(bt);
5641 }
5642
5643 DecoratorSet decorators = IN_HEAP;
5644
5645 if (is_vol) {
5646 decorators |= MO_SEQ_CST;
5647 }
5648
5649 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5650 }
5651
5652 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5653 bool is_exact = true, bool is_static = false,
5654 ciInstanceKlass * fromKls = NULL) {
5655 if (fromKls == NULL) {
5656 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5657 assert(tinst != NULL, "obj is null");
5658 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5659 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5660 fromKls = tinst->klass()->as_instance_klass();
5661 }
5662 else {
5663 assert(is_static, "only for static field access");
5664 }
5665 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5666 ciSymbol::make(fieldTypeString),
5667 is_static);
5668
5669 assert(field != NULL, "undefined field");
5670 assert(!field->is_volatile(), "not defined for volatile fields");
5671
5672 if (is_static) {
5673 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5674 fromObj = makecon(tip);
5675 }
5676
5677 // Next code copied from Parse::do_get_xxx():
5678
5679 // Compute address and memory type.
5680 int offset = field->offset_in_bytes();
5681 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5682
5683 return adr;
5684 }
5685
5686 //------------------------------inline_aescrypt_Block-----------------------
5687 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5688 address stubAddr = NULL;
5689 const char *stubName;
5690 assert(UseAES, "need AES instruction support");
5691
5692 switch(id) {
5693 case vmIntrinsics::_aescrypt_encryptBlock:
5694 stubAddr = StubRoutines::aescrypt_encryptBlock();
5695 stubName = "aescrypt_encryptBlock";
5696 break;
5697 case vmIntrinsics::_aescrypt_decryptBlock:
5698 stubAddr = StubRoutines::aescrypt_decryptBlock();
5699 stubName = "aescrypt_decryptBlock";
5700 break;
5701 default:
5702 break;
5703 }
5704 if (stubAddr == NULL) return false;
5705
5706 Node* aescrypt_object = argument(0);
5707 Node* src = argument(1);
5708 Node* src_offset = argument(2);
5709 Node* dest = argument(3);
5710 Node* dest_offset = argument(4);
5711
5712 // (1) src and dest are arrays.
5713 const Type* src_type = src->Value(&_gvn);
5714 const Type* dest_type = dest->Value(&_gvn);
5715 const TypeAryPtr* top_src = src_type->isa_aryptr();
5716 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5717 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5718
5719 // for the quick and dirty code we will skip all the checks.
5720 // we are just trying to get the call to be generated.
5721 Node* src_start = src;
5722 Node* dest_start = dest;
5723 if (src_offset != NULL || dest_offset != NULL) {
5724 assert(src_offset != NULL && dest_offset != NULL, "");
5725 src_start = array_element_address(src, src_offset, T_BYTE);
5726 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5727 }
5728
5729 // now need to get the start of its expanded key array
5730 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5731 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5732 if (k_start == NULL) return false;
5733
5734 if (Matcher::pass_original_key_for_aes()) {
5735 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5736 // compatibility issues between Java key expansion and SPARC crypto instructions
5737 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5738 if (original_k_start == NULL) return false;
5739
5740 // Call the stub.
5741 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5742 stubAddr, stubName, TypePtr::BOTTOM,
5743 src_start, dest_start, k_start, original_k_start);
5744 } else {
5745 // Call the stub.
5746 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5747 stubAddr, stubName, TypePtr::BOTTOM,
5748 src_start, dest_start, k_start);
5749 }
5750
5751 return true;
5752 }
5753
5754 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5755 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5756 address stubAddr = NULL;
5757 const char *stubName = NULL;
5758
5759 assert(UseAES, "need AES instruction support");
5760
5761 switch(id) {
5762 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5763 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5764 stubName = "cipherBlockChaining_encryptAESCrypt";
5765 break;
5766 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5767 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5768 stubName = "cipherBlockChaining_decryptAESCrypt";
5769 break;
5770 default:
5771 break;
5772 }
5773 if (stubAddr == NULL) return false;
5774
5775 Node* cipherBlockChaining_object = argument(0);
5776 Node* src = argument(1);
5777 Node* src_offset = argument(2);
5778 Node* len = argument(3);
5779 Node* dest = argument(4);
5780 Node* dest_offset = argument(5);
5781
5782 // (1) src and dest are arrays.
5783 const Type* src_type = src->Value(&_gvn);
5784 const Type* dest_type = dest->Value(&_gvn);
5785 const TypeAryPtr* top_src = src_type->isa_aryptr();
5786 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5787 assert (top_src != NULL && top_src->klass() != NULL
5788 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5789
5790 // checks are the responsibility of the caller
5791 Node* src_start = src;
5792 Node* dest_start = dest;
5793 if (src_offset != NULL || dest_offset != NULL) {
5794 assert(src_offset != NULL && dest_offset != NULL, "");
5795 src_start = array_element_address(src, src_offset, T_BYTE);
5796 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5797 }
5798
5799 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5800 // (because of the predicated logic executed earlier).
5801 // so we cast it here safely.
5802 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5803
5804 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5805 if (embeddedCipherObj == NULL) return false;
5806
5807 // cast it to what we know it will be at runtime
5808 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5809 assert(tinst != NULL, "CBC obj is null");
5810 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5811 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5812 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5813
5814 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5815 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5816 const TypeOopPtr* xtype = aklass->as_instance_type();
5817 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5818 aescrypt_object = _gvn.transform(aescrypt_object);
5819
5820 // we need to get the start of the aescrypt_object's expanded key array
5821 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5822 if (k_start == NULL) return false;
5823
5824 // similarly, get the start address of the r vector
5825 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5826 if (objRvec == NULL) return false;
5827 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5828
5829 Node* cbcCrypt;
5830 if (Matcher::pass_original_key_for_aes()) {
5831 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5832 // compatibility issues between Java key expansion and SPARC crypto instructions
5833 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5834 if (original_k_start == NULL) return false;
5835
5836 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5837 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5838 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5839 stubAddr, stubName, TypePtr::BOTTOM,
5840 src_start, dest_start, k_start, r_start, len, original_k_start);
5841 } else {
5842 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5843 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5844 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5845 stubAddr, stubName, TypePtr::BOTTOM,
5846 src_start, dest_start, k_start, r_start, len);
5847 }
5848
5849 // return cipher length (int)
5850 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5851 set_result(retvalue);
5852 return true;
5853 }
5854
5855 //------------------------------inline_counterMode_AESCrypt-----------------------
5856 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
5857 assert(UseAES, "need AES instruction support");
5858 if (!UseAESCTRIntrinsics) return false;
5859
5860 address stubAddr = NULL;
5861 const char *stubName = NULL;
5862 if (id == vmIntrinsics::_counterMode_AESCrypt) {
5863 stubAddr = StubRoutines::counterMode_AESCrypt();
5864 stubName = "counterMode_AESCrypt";
5865 }
5866 if (stubAddr == NULL) return false;
5867
5868 Node* counterMode_object = argument(0);
5869 Node* src = argument(1);
5870 Node* src_offset = argument(2);
5871 Node* len = argument(3);
5872 Node* dest = argument(4);
5873 Node* dest_offset = argument(5);
5874
5875 // (1) src and dest are arrays.
5876 const Type* src_type = src->Value(&_gvn);
5877 const Type* dest_type = dest->Value(&_gvn);
5878 const TypeAryPtr* top_src = src_type->isa_aryptr();
5879 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5880 assert(top_src != NULL && top_src->klass() != NULL &&
5881 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5882
5883 // checks are the responsibility of the caller
5884 Node* src_start = src;
5885 Node* dest_start = dest;
5886 if (src_offset != NULL || dest_offset != NULL) {
5887 assert(src_offset != NULL && dest_offset != NULL, "");
5888 src_start = array_element_address(src, src_offset, T_BYTE);
5889 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5890 }
5891
5892 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5893 // (because of the predicated logic executed earlier).
5894 // so we cast it here safely.
5895 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5896 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5897 if (embeddedCipherObj == NULL) return false;
5898 // cast it to what we know it will be at runtime
5899 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
5900 assert(tinst != NULL, "CTR obj is null");
5901 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
5902 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5903 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5904 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5905 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5906 const TypeOopPtr* xtype = aklass->as_instance_type();
5907 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5908 aescrypt_object = _gvn.transform(aescrypt_object);
5909 // we need to get the start of the aescrypt_object's expanded key array
5910 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5911 if (k_start == NULL) return false;
5912 // similarly, get the start address of the r vector
5913 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
5914 if (obj_counter == NULL) return false;
5915 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
5916
5917 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
5918 if (saved_encCounter == NULL) return false;
5919 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
5920 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
5921
5922 Node* ctrCrypt;
5923 if (Matcher::pass_original_key_for_aes()) {
5924 // no SPARC version for AES/CTR intrinsics now.
5925 return false;
5926 }
5927 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5928 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5929 OptoRuntime::counterMode_aescrypt_Type(),
5930 stubAddr, stubName, TypePtr::BOTTOM,
5931 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
5932
5933 // return cipher length (int)
5934 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
5935 set_result(retvalue);
5936 return true;
5937 }
5938
5939 //------------------------------get_key_start_from_aescrypt_object-----------------------
5940 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
5941 #if defined(PPC64) || defined(S390)
5942 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
5943 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
5944 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
5945 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
5946 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
5947 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5948 if (objSessionK == NULL) {
5949 return (Node *) NULL;
5950 }
5951 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
5952 #else
5953 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
5954 #endif // PPC64
5955 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5956 if (objAESCryptKey == NULL) return (Node *) NULL;
5957
5958 // now have the array, need to get the start address of the K array
5959 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
5960 return k_start;
5961 }
5962
5963 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
5964 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
5965 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
5966 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
5967 if (objAESCryptKey == NULL) return (Node *) NULL;
5968
5969 // now have the array, need to get the start address of the lastKey array
5970 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
5971 return original_k_start;
5972 }
5973
5974 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
5975 // Return node representing slow path of predicate check.
5976 // the pseudo code we want to emulate with this predicate is:
5977 // for encryption:
5978 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
5979 // for decryption:
5980 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
5981 // note cipher==plain is more conservative than the original java code but that's OK
5982 //
5983 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
5984 // The receiver was checked for NULL already.
5985 Node* objCBC = argument(0);
5986
5987 // Load embeddedCipher field of CipherBlockChaining object.
5988 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5989
5990 // get AESCrypt klass for instanceOf check
5991 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
5992 // will have same classloader as CipherBlockChaining object
5993 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
5994 assert(tinst != NULL, "CBCobj is null");
5995 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
5996
5997 // we want to do an instanceof comparison against the AESCrypt class
5998 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5999 if (!klass_AESCrypt->is_loaded()) {
6000 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6001 Node* ctrl = control();
6002 set_control(top()); // no regular fast path
6003 return ctrl;
6004 }
6005 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6006
6007 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6008 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6009 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6010
6011 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6012
6013 // for encryption, we are done
6014 if (!decrypting)
6015 return instof_false; // even if it is NULL
6016
6017 // for decryption, we need to add a further check to avoid
6018 // taking the intrinsic path when cipher and plain are the same
6019 // see the original java code for why.
6020 RegionNode* region = new RegionNode(3);
6021 region->init_req(1, instof_false);
6022 Node* src = argument(1);
6023 Node* dest = argument(4);
6024 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6025 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6026 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6027 region->init_req(2, src_dest_conjoint);
6028
6029 record_for_igvn(region);
6030 return _gvn.transform(region);
6031 }
6032
6033 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6034 // Return node representing slow path of predicate check.
6035 // the pseudo code we want to emulate with this predicate is:
6036 // for encryption:
6037 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6038 // for decryption:
6039 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6040 // note cipher==plain is more conservative than the original java code but that's OK
6041 //
6042
6043 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6044 // The receiver was checked for NULL already.
6045 Node* objCTR = argument(0);
6046
6047 // Load embeddedCipher field of CipherBlockChaining object.
6048 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6049
6050 // get AESCrypt klass for instanceOf check
6051 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6052 // will have same classloader as CipherBlockChaining object
6053 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6054 assert(tinst != NULL, "CTRobj is null");
6055 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6056
6057 // we want to do an instanceof comparison against the AESCrypt class
6058 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6059 if (!klass_AESCrypt->is_loaded()) {
6060 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6061 Node* ctrl = control();
6062 set_control(top()); // no regular fast path
6063 return ctrl;
6064 }
6065
6066 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6067 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6068 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6069 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6070 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6071
6072 return instof_false; // even if it is NULL
6073 }
6074
6075 //------------------------------inline_ghash_processBlocks
6076 bool LibraryCallKit::inline_ghash_processBlocks() {
6077 address stubAddr;
6078 const char *stubName;
6079 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6080
6081 stubAddr = StubRoutines::ghash_processBlocks();
6082 stubName = "ghash_processBlocks";
6083
6084 Node* data = argument(0);
6085 Node* offset = argument(1);
6086 Node* len = argument(2);
6087 Node* state = argument(3);
6088 Node* subkeyH = argument(4);
6089
6090 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6091 assert(state_start, "state is NULL");
6092 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6093 assert(subkeyH_start, "subkeyH is NULL");
6094 Node* data_start = array_element_address(data, offset, T_BYTE);
6095 assert(data_start, "data is NULL");
6096
6097 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6098 OptoRuntime::ghash_processBlocks_Type(),
6099 stubAddr, stubName, TypePtr::BOTTOM,
6100 state_start, subkeyH_start, data_start, len);
6101 return true;
6102 }
6103
6104 //------------------------------inline_sha_implCompress-----------------------
6105 //
6106 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6107 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6108 //
6109 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6110 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6111 //
6112 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6113 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6114 //
6115 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6116 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6117
6118 Node* sha_obj = argument(0);
6119 Node* src = argument(1); // type oop
6120 Node* ofs = argument(2); // type int
6121
6122 const Type* src_type = src->Value(&_gvn);
6123 const TypeAryPtr* top_src = src_type->isa_aryptr();
6124 if (top_src == NULL || top_src->klass() == NULL) {
6125 // failed array check
6126 return false;
6127 }
6128 // Figure out the size and type of the elements we will be copying.
6129 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6130 if (src_elem != T_BYTE) {
6131 return false;
6132 }
6133 // 'src_start' points to src array + offset
6134 Node* src_start = array_element_address(src, ofs, src_elem);
6135 Node* state = NULL;
6136 address stubAddr;
6137 const char *stubName;
6138
6139 switch(id) {
6140 case vmIntrinsics::_sha_implCompress:
6141 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6142 state = get_state_from_sha_object(sha_obj);
6143 stubAddr = StubRoutines::sha1_implCompress();
6144 stubName = "sha1_implCompress";
6145 break;
6146 case vmIntrinsics::_sha2_implCompress:
6147 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6148 state = get_state_from_sha_object(sha_obj);
6149 stubAddr = StubRoutines::sha256_implCompress();
6150 stubName = "sha256_implCompress";
6151 break;
6152 case vmIntrinsics::_sha5_implCompress:
6153 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6154 state = get_state_from_sha5_object(sha_obj);
6155 stubAddr = StubRoutines::sha512_implCompress();
6156 stubName = "sha512_implCompress";
6157 break;
6158 default:
6159 fatal_unexpected_iid(id);
6160 return false;
6161 }
6162 if (state == NULL) return false;
6163
6164 // Call the stub.
6165 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6166 stubAddr, stubName, TypePtr::BOTTOM,
6167 src_start, state);
6168
6169 return true;
6170 }
6171
6172 //------------------------------inline_digestBase_implCompressMB-----------------------
6173 //
6174 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6175 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6176 //
6177 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6178 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6179 "need SHA1/SHA256/SHA512 instruction support");
6180 assert((uint)predicate < 3, "sanity");
6181 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6182
6183 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6184 Node* src = argument(1); // byte[] array
6185 Node* ofs = argument(2); // type int
6186 Node* limit = argument(3); // type int
6187
6188 const Type* src_type = src->Value(&_gvn);
6189 const TypeAryPtr* top_src = src_type->isa_aryptr();
6190 if (top_src == NULL || top_src->klass() == NULL) {
6191 // failed array check
6192 return false;
6193 }
6194 // Figure out the size and type of the elements we will be copying.
6195 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6196 if (src_elem != T_BYTE) {
6197 return false;
6198 }
6199 // 'src_start' points to src array + offset
6200 Node* src_start = array_element_address(src, ofs, src_elem);
6201
6202 const char* klass_SHA_name = NULL;
6203 const char* stub_name = NULL;
6204 address stub_addr = NULL;
6205 bool long_state = false;
6206
6207 switch (predicate) {
6208 case 0:
6209 if (UseSHA1Intrinsics) {
6210 klass_SHA_name = "sun/security/provider/SHA";
6211 stub_name = "sha1_implCompressMB";
6212 stub_addr = StubRoutines::sha1_implCompressMB();
6213 }
6214 break;
6215 case 1:
6216 if (UseSHA256Intrinsics) {
6217 klass_SHA_name = "sun/security/provider/SHA2";
6218 stub_name = "sha256_implCompressMB";
6219 stub_addr = StubRoutines::sha256_implCompressMB();
6220 }
6221 break;
6222 case 2:
6223 if (UseSHA512Intrinsics) {
6224 klass_SHA_name = "sun/security/provider/SHA5";
6225 stub_name = "sha512_implCompressMB";
6226 stub_addr = StubRoutines::sha512_implCompressMB();
6227 long_state = true;
6228 }
6229 break;
6230 default:
6231 fatal("unknown SHA intrinsic predicate: %d", predicate);
6232 }
6233 if (klass_SHA_name != NULL) {
6234 // get DigestBase klass to lookup for SHA klass
6235 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6236 assert(tinst != NULL, "digestBase_obj is not instance???");
6237 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6238
6239 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6240 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6241 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6242 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6243 }
6244 return false;
6245 }
6246 //------------------------------inline_sha_implCompressMB-----------------------
6247 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6248 bool long_state, address stubAddr, const char *stubName,
6249 Node* src_start, Node* ofs, Node* limit) {
6250 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6251 const TypeOopPtr* xtype = aklass->as_instance_type();
6252 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6253 sha_obj = _gvn.transform(sha_obj);
6254
6255 Node* state;
6256 if (long_state) {
6257 state = get_state_from_sha5_object(sha_obj);
6258 } else {
6259 state = get_state_from_sha_object(sha_obj);
6260 }
6261 if (state == NULL) return false;
6262
6263 // Call the stub.
6264 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6265 OptoRuntime::digestBase_implCompressMB_Type(),
6266 stubAddr, stubName, TypePtr::BOTTOM,
6267 src_start, state, ofs, limit);
6268 // return ofs (int)
6269 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6270 set_result(result);
6271
6272 return true;
6273 }
6274
6275 //------------------------------get_state_from_sha_object-----------------------
6276 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6277 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6278 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6279 if (sha_state == NULL) return (Node *) NULL;
6280
6281 // now have the array, need to get the start address of the state array
6282 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6283 return state;
6284 }
6285
6286 //------------------------------get_state_from_sha5_object-----------------------
6287 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6288 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6289 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6290 if (sha_state == NULL) return (Node *) NULL;
6291
6292 // now have the array, need to get the start address of the state array
6293 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6294 return state;
6295 }
6296
6297 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6298 // Return node representing slow path of predicate check.
6299 // the pseudo code we want to emulate with this predicate is:
6300 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6301 //
6302 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6303 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6304 "need SHA1/SHA256/SHA512 instruction support");
6305 assert((uint)predicate < 3, "sanity");
6306
6307 // The receiver was checked for NULL already.
6308 Node* digestBaseObj = argument(0);
6309
6310 // get DigestBase klass for instanceOf check
6311 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6312 assert(tinst != NULL, "digestBaseObj is null");
6313 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6314
6315 const char* klass_SHA_name = NULL;
6316 switch (predicate) {
6317 case 0:
6318 if (UseSHA1Intrinsics) {
6319 // we want to do an instanceof comparison against the SHA class
6320 klass_SHA_name = "sun/security/provider/SHA";
6321 }
6322 break;
6323 case 1:
6324 if (UseSHA256Intrinsics) {
6325 // we want to do an instanceof comparison against the SHA2 class
6326 klass_SHA_name = "sun/security/provider/SHA2";
6327 }
6328 break;
6329 case 2:
6330 if (UseSHA512Intrinsics) {
6331 // we want to do an instanceof comparison against the SHA5 class
6332 klass_SHA_name = "sun/security/provider/SHA5";
6333 }
6334 break;
6335 default:
6336 fatal("unknown SHA intrinsic predicate: %d", predicate);
6337 }
6338
6339 ciKlass* klass_SHA = NULL;
6340 if (klass_SHA_name != NULL) {
6341 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6342 }
6343 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6344 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6345 Node* ctrl = control();
6346 set_control(top()); // no intrinsic path
6347 return ctrl;
6348 }
6349 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6350
6351 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6352 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6353 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6354 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6355
6356 return instof_false; // even if it is NULL
6357 }
6358
6359 // long Continuations::getFP() ()J
6360 bool LibraryCallKit::inline_get_frame_pointer() {
6361 Node *frame = _gvn.transform(new GetFPNode(control()));
6362 set_result(frame);
6363 return true;
6364 }
6365
6366 bool LibraryCallKit::inline_continuation(vmIntrinsics::ID id) {
6367 address call_addr = NULL;
6368 const char *name = NULL;
6369
6370 switch (id) {
6371 case vmIntrinsics::_Continuation_getSP: call_addr = StubRoutines::cont_getSP(); name = "getSP"; break;
6372 case vmIntrinsics::_Continuation_getPC: call_addr = StubRoutines::cont_getPC(); name = "getPC"; break;
6373 case vmIntrinsics::_Continuation_getFP: call_addr = OptoRuntime::continuation_getFP_Java(); name = "getFP"; break;
6374 default: fatal("error"); return false;
6375 }
6376
6377 const TypeFunc* tf = OptoRuntime::void_long_Type();
6378 const TypePtr* no_memory_effects = NULL;
6379 Node* call = make_runtime_call(RC_LEAF, tf, call_addr, name, no_memory_effects);
6380 Node* value = _gvn.transform(new ProjNode(call, TypeFunc::Parms+0));
6381 #ifdef ASSERT
6382 Node* value_top = _gvn.transform(new ProjNode(call, TypeFunc::Parms+1));
6383 assert(value_top == top(), "second value must be top");
6384 #endif
6385 set_result(value);
6386 return true;
6387 }
6388
6389 bool LibraryCallKit::inline_continuation_runLevel() {
6390 set_result(intcon(2));
6391 return true;
6392 }
6393
6394 bool LibraryCallKit::inline_continuation_do_continue() {
6395 address call_addr = StubRoutines::cont_thaw();
6396 const TypeFunc* tf = OptoRuntime::void_void_Type();
6397 Node* call = make_runtime_call(RC_NO_LEAF, tf, call_addr, "doContinue", TypeRawPtr::BOTTOM);
6398 return true;
6399 }
6400
6401 bool LibraryCallKit::inline_continuation_do_yield() {
6402 address call_addr = StubRoutines::cont_doYield();
6403 Node* arg0 = argument(0); // type int
6404 const TypeFunc* tf = OptoRuntime::continuation_doYield_Type();
6405 Node* call = make_runtime_call(RC_NO_LEAF, tf, call_addr, "doYield", TypeRawPtr::BOTTOM, arg0);
6406 return true;
6407 }
6408
6409 //-------------inline_fma-----------------------------------
6410 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6411 Node *a = NULL;
6412 Node *b = NULL;
6413 Node *c = NULL;
6414 Node* result = NULL;
6415 switch (id) {
6416 case vmIntrinsics::_fmaD:
6417 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6418 // no receiver since it is static method
6419 a = round_double_node(argument(0));
6420 b = round_double_node(argument(2));
6421 c = round_double_node(argument(4));
6422 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6423 break;
6424 case vmIntrinsics::_fmaF:
6425 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6426 a = argument(0);
6427 b = argument(1);
6428 c = argument(2);
6429 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6430 break;
6431 default:
6432 fatal_unexpected_iid(id); break;
6433 }
6434 set_result(result);
6435 return true;
6436 }
6437
6438 bool LibraryCallKit::inline_profileBoolean() {
6439 Node* counts = argument(1);
6440 const TypeAryPtr* ary = NULL;
6441 ciArray* aobj = NULL;
6442 if (counts->is_Con()
6443 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6444 && (aobj = ary->const_oop()->as_array()) != NULL
6445 && (aobj->length() == 2)) {
6446 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6447 jint false_cnt = aobj->element_value(0).as_int();
6448 jint true_cnt = aobj->element_value(1).as_int();
6449
6450 if (C->log() != NULL) {
6451 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6452 false_cnt, true_cnt);
6453 }
6454
6455 if (false_cnt + true_cnt == 0) {
6456 // According to profile, never executed.
6457 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6458 Deoptimization::Action_reinterpret);
6459 return true;
6460 }
6461
6462 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6463 // is a number of each value occurrences.
6464 Node* result = argument(0);
6465 if (false_cnt == 0 || true_cnt == 0) {
6466 // According to profile, one value has been never seen.
6467 int expected_val = (false_cnt == 0) ? 1 : 0;
6468
6469 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6470 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6471
6472 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6473 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6474 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6475
6476 { // Slow path: uncommon trap for never seen value and then reexecute
6477 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6478 // the value has been seen at least once.
6479 PreserveJVMState pjvms(this);
6480 PreserveReexecuteState preexecs(this);
6481 jvms()->set_should_reexecute(true);
6482
6483 set_control(slow_path);
6484 set_i_o(i_o());
6485
6486 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6487 Deoptimization::Action_reinterpret);
6488 }
6489 // The guard for never seen value enables sharpening of the result and
6490 // returning a constant. It allows to eliminate branches on the same value
6491 // later on.
6492 set_control(fast_path);
6493 result = intcon(expected_val);
6494 }
6495 // Stop profiling.
6496 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6497 // By replacing method body with profile data (represented as ProfileBooleanNode
6498 // on IR level) we effectively disable profiling.
6499 // It enables full speed execution once optimized code is generated.
6500 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6501 C->record_for_igvn(profile);
6502 set_result(profile);
6503 return true;
6504 } else {
6505 // Continue profiling.
6506 // Profile data isn't available at the moment. So, execute method's bytecode version.
6507 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6508 // is compiled and counters aren't available since corresponding MethodHandle
6509 // isn't a compile-time constant.
6510 return false;
6511 }
6512 }
6513
6514 bool LibraryCallKit::inline_isCompileConstant() {
6515 Node* n = argument(0);
6516 set_result(n->is_Con() ? intcon(1) : intcon(0));
6517 return true;
6518 }