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