1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/callGenerator.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/intrinsicnode.hpp"
45 #include "opto/idealKit.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/movenode.hpp"
48 #include "opto/mulnode.hpp"
49 #include "opto/narrowptrnode.hpp"
50 #include "opto/opaquenode.hpp"
51 #include "opto/parse.hpp"
52 #include "opto/runtime.hpp"
53 #include "opto/rootnode.hpp"
54 #include "opto/subnode.hpp"
55 #include "opto/valuetypenode.hpp"
56 #include "prims/nativeLookup.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "utilities/macros.hpp"
61
62
63 class LibraryIntrinsic : public InlineCallGenerator {
64 // Extend the set of intrinsics known to the runtime:
65 public:
66 private:
67 bool _is_virtual;
68 bool _does_virtual_dispatch;
69 int8_t _predicates_count; // Intrinsic is predicated by several conditions
70 int8_t _last_predicate; // Last generated predicate
71 vmIntrinsics::ID _intrinsic_id;
72
73 public:
74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
75 : InlineCallGenerator(m),
76 _is_virtual(is_virtual),
77 _does_virtual_dispatch(does_virtual_dispatch),
78 _predicates_count((int8_t)predicates_count),
79 _last_predicate((int8_t)-1),
80 _intrinsic_id(id)
81 {
82 }
83 virtual bool is_intrinsic() const { return true; }
84 virtual bool is_virtual() const { return _is_virtual; }
85 virtual bool is_predicated() const { return _predicates_count > 0; }
86 virtual int predicates_count() const { return _predicates_count; }
87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
88 virtual JVMState* generate(JVMState* jvms);
89 virtual Node* generate_predicate(JVMState* jvms, int predicate);
90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
91 };
92
93
94 // Local helper class for LibraryIntrinsic:
95 class LibraryCallKit : public GraphKit {
96 private:
97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
98 Node* _result; // the result node, if any
99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
100
101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
102
103 public:
104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
105 : GraphKit(jvms),
106 _intrinsic(intrinsic),
107 _result(NULL)
108 {
109 // Check if this is a root compile. In that case we don't have a caller.
110 if (!jvms->has_method()) {
111 _reexecute_sp = sp();
112 } else {
113 // Find out how many arguments the interpreter needs when deoptimizing
114 // and save the stack pointer value so it can used by uncommon_trap.
115 // We find the argument count by looking at the declared signature.
116 bool ignored_will_link;
117 ciSignature* declared_signature = NULL;
118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
121 }
122 }
123
124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
125
126 ciMethod* caller() const { return jvms()->method(); }
127 int bci() const { return jvms()->bci(); }
128 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
130 ciMethod* callee() const { return _intrinsic->method(); }
131
132 bool try_to_inline(int predicate);
133 Node* try_to_predicate(int predicate);
134
135 void push_result() {
136 // Push the result onto the stack.
137 if (!stopped() && result() != NULL) {
138 BasicType bt = result()->bottom_type()->basic_type();
139 push_node(bt, result());
140 }
141 }
142
143 private:
144 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
146 }
147
148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
149 void set_result(RegionNode* region, PhiNode* value);
150 Node* result() { return _result; }
151
152 virtual int reexecute_sp() { return _reexecute_sp; }
153
154 // Helper functions to inline natives
155 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
156 Node* generate_slow_guard(Node* test, RegionNode* region);
157 Node* generate_fair_guard(Node* test, RegionNode* region);
158 Node* generate_negative_guard(Node* index, RegionNode* region,
159 // resulting CastII of index:
160 Node* *pos_index = NULL);
161 Node* generate_limit_guard(Node* offset, Node* subseq_length,
162 Node* array_length,
163 RegionNode* region);
164 void generate_string_range_check(Node* array, Node* offset,
165 Node* length, bool char_count);
166 Node* generate_current_thread(Node* &tls_output);
167 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path,
169 int offset);
170 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
171 RegionNode* region, int null_path) {
172 int offset = java_lang_Class::klass_offset_in_bytes();
173 return load_klass_from_mirror_common(mirror, never_see_null,
174 region, null_path,
175 offset);
176 }
177 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
178 RegionNode* region, int null_path) {
179 int offset = java_lang_Class::array_klass_offset_in_bytes();
180 return load_klass_from_mirror_common(mirror, never_see_null,
181 region, null_path,
182 offset);
183 }
184 Node* generate_access_flags_guard(Node* kls,
185 int modifier_mask, int modifier_bits,
186 RegionNode* region);
187 Node* generate_interface_guard(Node* kls, RegionNode* region);
188 Node* generate_value_guard(Node* kls, RegionNode* region);
189
190 enum ArrayKind {
191 AnyArray,
192 NonArray,
193 ObjectArray,
194 NonObjectArray,
195 TypeArray,
196 ValueArray
197 };
198
199 Node* generate_array_guard(Node* kls, RegionNode* region) {
200 return generate_array_guard_common(kls, region, AnyArray);
201 }
202 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
203 return generate_array_guard_common(kls, region, NonArray);
204 }
205 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
206 return generate_array_guard_common(kls, region, ObjectArray);
207 }
208 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
209 return generate_array_guard_common(kls, region, NonObjectArray);
210 }
211 Node* generate_typeArray_guard(Node* kls, RegionNode* region) {
212 return generate_array_guard_common(kls, region, TypeArray);
213 }
214 Node* generate_valueArray_guard(Node* kls, RegionNode* region) {
215 return generate_array_guard_common(kls, region, ValueArray);
216 }
217 Node* generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind);
218 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
219 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
220 bool is_virtual = false, bool is_static = false);
221 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
222 return generate_method_call(method_id, false, true);
223 }
224 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
225 return generate_method_call(method_id, true, false);
226 }
227 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
228 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
229
230 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
231 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
232 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
233 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
234 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
235 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
236 bool inline_string_indexOfChar();
237 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
238 bool inline_string_toBytesU();
239 bool inline_string_getCharsU();
240 bool inline_string_copy(bool compress);
241 bool inline_string_char_access(bool is_store);
242 Node* round_double_node(Node* n);
243 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
244 bool inline_math_native(vmIntrinsics::ID id);
245 bool inline_math(vmIntrinsics::ID id);
246 template <typename OverflowOp>
247 bool inline_math_overflow(Node* arg1, Node* arg2);
248 void inline_math_mathExact(Node* math, Node* test);
249 bool inline_math_addExactI(bool is_increment);
250 bool inline_math_addExactL(bool is_increment);
251 bool inline_math_multiplyExactI();
252 bool inline_math_multiplyExactL();
253 bool inline_math_multiplyHigh();
254 bool inline_math_negateExactI();
255 bool inline_math_negateExactL();
256 bool inline_math_subtractExactI(bool is_decrement);
257 bool inline_math_subtractExactL(bool is_decrement);
258 bool inline_min_max(vmIntrinsics::ID id);
259 bool inline_notify(vmIntrinsics::ID id);
260 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
261 // This returns Type::AnyPtr, RawPtr, or OopPtr.
262 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
263 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
264
265 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
266 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
267 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
268 static bool klass_needs_init_guard(Node* kls);
269 bool inline_unsafe_allocate();
270 bool inline_unsafe_newArray(bool uninitialized);
271 bool inline_unsafe_copyMemory();
272 bool inline_unsafe_make_private_buffer();
273 bool inline_unsafe_finish_private_buffer();
274 bool inline_native_currentThread();
275
276 bool inline_native_time_funcs(address method, const char* funcName);
277 #ifdef JFR_HAVE_INTRINSICS
278 bool inline_native_classID();
279 bool inline_native_getEventWriter();
280 #endif
281 bool inline_native_isInterrupted();
282 bool inline_native_Class_query(vmIntrinsics::ID id);
283 bool inline_native_subtype_check();
284 bool inline_native_getLength();
285 bool inline_array_copyOf(bool is_copyOfRange);
286 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
287 bool inline_preconditions_checkIndex();
288 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
289 bool inline_native_clone(bool is_virtual);
290 bool inline_native_Reflection_getCallerClass();
291 // Helper function for inlining native object hash method
292 bool inline_native_hashcode(bool is_virtual, bool is_static);
293 bool inline_native_getClass();
294
295 // Helper functions for inlining arraycopy
296 bool inline_arraycopy();
297 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
298 RegionNode* slow_region);
299 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
300 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
301 uint new_idx);
302
303 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
304 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
305 bool inline_unsafe_fence(vmIntrinsics::ID id);
306 bool inline_onspinwait();
307 bool inline_fp_conversions(vmIntrinsics::ID id);
308 bool inline_number_methods(vmIntrinsics::ID id);
309 bool inline_reference_get();
310 bool inline_Class_cast();
311 bool inline_aescrypt_Block(vmIntrinsics::ID id);
312 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
313 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
314 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
315 Node* inline_counterMode_AESCrypt_predicate();
316 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
317 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
318 bool inline_ghash_processBlocks();
319 bool inline_base64_encodeBlock();
320 bool inline_sha_implCompress(vmIntrinsics::ID id);
321 bool inline_digestBase_implCompressMB(int predicate);
322 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
323 bool long_state, address stubAddr, const char *stubName,
324 Node* src_start, Node* ofs, Node* limit);
325 Node* get_state_from_sha_object(Node *sha_object);
326 Node* get_state_from_sha5_object(Node *sha_object);
327 Node* inline_digestBase_implCompressMB_predicate(int predicate);
328 bool inline_encodeISOArray();
329 bool inline_updateCRC32();
330 bool inline_updateBytesCRC32();
331 bool inline_updateByteBufferCRC32();
332 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
333 bool inline_updateBytesCRC32C();
334 bool inline_updateDirectByteBufferCRC32C();
335 bool inline_updateBytesAdler32();
336 bool inline_updateByteBufferAdler32();
337 bool inline_multiplyToLen();
338 bool inline_hasNegatives();
339 bool inline_squareToLen();
340 bool inline_mulAdd();
341 bool inline_montgomeryMultiply();
342 bool inline_montgomerySquare();
343 bool inline_vectorizedMismatch();
344 bool inline_fma(vmIntrinsics::ID id);
345 bool inline_character_compare(vmIntrinsics::ID id);
346 bool inline_fp_min_max(vmIntrinsics::ID id);
347
348 bool inline_profileBoolean();
349 bool inline_isCompileConstant();
350 void clear_upper_avx() {
351 #ifdef X86
352 if (UseAVX >= 2) {
353 C->set_clear_upper_avx(true);
354 }
355 #endif
356 }
357 };
358
359 //---------------------------make_vm_intrinsic----------------------------
360 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
361 vmIntrinsics::ID id = m->intrinsic_id();
362 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
363
364 if (!m->is_loaded()) {
365 // Do not attempt to inline unloaded methods.
366 return NULL;
367 }
368
369 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
370 bool is_available = false;
371
372 {
373 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
374 // the compiler must transition to '_thread_in_vm' state because both
375 // methods access VM-internal data.
376 VM_ENTRY_MARK;
377 methodHandle mh(THREAD, m->get_Method());
378 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
379 !C->directive()->is_intrinsic_disabled(mh) &&
380 !vmIntrinsics::is_disabled_by_flags(mh);
381
382 }
383
384 if (is_available) {
385 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
386 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
387 return new LibraryIntrinsic(m, is_virtual,
388 vmIntrinsics::predicates_needed(id),
389 vmIntrinsics::does_virtual_dispatch(id),
390 (vmIntrinsics::ID) id);
391 } else {
392 return NULL;
393 }
394 }
395
396 //----------------------register_library_intrinsics-----------------------
397 // Initialize this file's data structures, for each Compile instance.
398 void Compile::register_library_intrinsics() {
399 // Nothing to do here.
400 }
401
402 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
403 LibraryCallKit kit(jvms, this);
404 Compile* C = kit.C;
405 int nodes = C->unique();
406 #ifndef PRODUCT
407 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
408 char buf[1000];
409 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
410 tty->print_cr("Intrinsic %s", str);
411 }
412 #endif
413 ciMethod* callee = kit.callee();
414 const int bci = kit.bci();
415
416 // Try to inline the intrinsic.
417 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
418 kit.try_to_inline(_last_predicate)) {
419 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
420 : "(intrinsic)";
421 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
422 if (C->print_intrinsics() || C->print_inlining()) {
423 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
424 }
425 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
426 if (C->log()) {
427 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
428 vmIntrinsics::name_at(intrinsic_id()),
429 (is_virtual() ? " virtual='1'" : ""),
430 C->unique() - nodes);
431 }
432 // Push the result from the inlined method onto the stack.
433 kit.push_result();
434 C->print_inlining_update(this);
435 return kit.transfer_exceptions_into_jvms();
436 }
437
438 // The intrinsic bailed out
439 if (jvms->has_method()) {
440 // Not a root compile.
441 const char* msg;
442 if (callee->intrinsic_candidate()) {
443 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
444 } else {
445 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
446 : "failed to inline (intrinsic), method not annotated";
447 }
448 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
449 if (C->print_intrinsics() || C->print_inlining()) {
450 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
451 }
452 } else {
453 // Root compile
454 ResourceMark rm;
455 stringStream msg_stream;
456 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
457 vmIntrinsics::name_at(intrinsic_id()),
458 is_virtual() ? " (virtual)" : "", bci);
459 const char *msg = msg_stream.as_string();
460 log_debug(jit, inlining)("%s", msg);
461 if (C->print_intrinsics() || C->print_inlining()) {
462 tty->print("%s", msg);
463 }
464 }
465 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
466 C->print_inlining_update(this);
467 return NULL;
468 }
469
470 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
471 LibraryCallKit kit(jvms, this);
472 Compile* C = kit.C;
473 int nodes = C->unique();
474 _last_predicate = predicate;
475 #ifndef PRODUCT
476 assert(is_predicated() && predicate < predicates_count(), "sanity");
477 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
478 char buf[1000];
479 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
480 tty->print_cr("Predicate for intrinsic %s", str);
481 }
482 #endif
483 ciMethod* callee = kit.callee();
484 const int bci = kit.bci();
485
486 Node* slow_ctl = kit.try_to_predicate(predicate);
487 if (!kit.failing()) {
488 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
489 : "(intrinsic, predicate)";
490 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
491 if (C->print_intrinsics() || C->print_inlining()) {
492 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
493 }
494 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
495 if (C->log()) {
496 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
497 vmIntrinsics::name_at(intrinsic_id()),
498 (is_virtual() ? " virtual='1'" : ""),
499 C->unique() - nodes);
500 }
501 return slow_ctl; // Could be NULL if the check folds.
502 }
503
504 // The intrinsic bailed out
505 if (jvms->has_method()) {
506 // Not a root compile.
507 const char* msg = "failed to generate predicate for intrinsic";
508 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
509 if (C->print_intrinsics() || C->print_inlining()) {
510 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
511 }
512 } else {
513 // Root compile
514 ResourceMark rm;
515 stringStream msg_stream;
516 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
517 vmIntrinsics::name_at(intrinsic_id()),
518 is_virtual() ? " (virtual)" : "", bci);
519 const char *msg = msg_stream.as_string();
520 log_debug(jit, inlining)("%s", msg);
521 if (C->print_intrinsics() || C->print_inlining()) {
522 C->print_inlining_stream()->print("%s", msg);
523 }
524 }
525 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
526 return NULL;
527 }
528
529 bool LibraryCallKit::try_to_inline(int predicate) {
530 // Handle symbolic names for otherwise undistinguished boolean switches:
531 const bool is_store = true;
532 const bool is_compress = true;
533 const bool is_static = true;
534 const bool is_volatile = true;
535
536 if (!jvms()->has_method()) {
537 // Root JVMState has a null method.
538 assert(map()->memory()->Opcode() == Op_Parm, "");
539 // Insert the memory aliasing node
540 set_all_memory(reset_memory());
541 }
542 assert(merged_memory(), "");
543
544
545 switch (intrinsic_id()) {
546 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
547 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
548 case vmIntrinsics::_getClass: return inline_native_getClass();
549
550 case vmIntrinsics::_dsin:
551 case vmIntrinsics::_dcos:
552 case vmIntrinsics::_dtan:
553 case vmIntrinsics::_dabs:
554 case vmIntrinsics::_datan2:
555 case vmIntrinsics::_dsqrt:
556 case vmIntrinsics::_dexp:
557 case vmIntrinsics::_dlog:
558 case vmIntrinsics::_dlog10:
559 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
560
561 case vmIntrinsics::_min:
562 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
563
564 case vmIntrinsics::_notify:
565 case vmIntrinsics::_notifyAll:
566 return inline_notify(intrinsic_id());
567
568 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
569 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
570 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
571 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
572 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
573 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
574 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
575 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
576 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
577 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
578 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
579 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
580 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
581
582 case vmIntrinsics::_arraycopy: return inline_arraycopy();
583
584 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
585 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
586 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
587 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
588
589 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
590 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
591 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
592 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
593 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
594 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
595 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
596
597 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
598 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
599
600 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
601 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
602 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
603 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
604
605 case vmIntrinsics::_compressStringC:
606 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
607 case vmIntrinsics::_inflateStringC:
608 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
609
610 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
611 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
612 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
613 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
614 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
615 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
616 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
617 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
618 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
619 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
620 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
621 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_VALUETYPE,Relaxed, false);
622
623 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
624 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
625 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
626 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
627 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
628 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
629 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
630 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
631 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
632 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_VALUETYPE,Relaxed, false);
633
634 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
635 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
636 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
637 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
638 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
639 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
640 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
641 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
642 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
643
644 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
645 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
646 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
647 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
648 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
649 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
650 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
651 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
652 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
653
654 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
655 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
656 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
657 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
658
659 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
660 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
661 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
662 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
663
664 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
665 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
666 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
667 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
668 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
669 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
670 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
671 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
672 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
673
674 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
675 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
676 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
677 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
678 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
679 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
680 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
681 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
682 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
683
684 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
685 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
686 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
687 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
688 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
689 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
690 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
691 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
692 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
693
694 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
695 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
696 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
697 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
698 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
699 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
700 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
701 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
702 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
703
704 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
705 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
706 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
707 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
708 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
709
710 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
711 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
712 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
713 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
714 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
715 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
716 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
717 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
718 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
719 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
720 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
721 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
722 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
723 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
724 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
725 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
726 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
727 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
728 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
729 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
730
731 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
732 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
733 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
734 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
735 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
736 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
737 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
738 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
739 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
740 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
741 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
742 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
743 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
744 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
745 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
746
747 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
748 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
749 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
750 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
751
752 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
753 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
754 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
755 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
756 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
757
758 case vmIntrinsics::_loadFence:
759 case vmIntrinsics::_storeFence:
760 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
761
762 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
763
764 case vmIntrinsics::_currentThread: return inline_native_currentThread();
765 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
766
767 #ifdef JFR_HAVE_INTRINSICS
768 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
769 case vmIntrinsics::_getClassId: return inline_native_classID();
770 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
771 #endif
772 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
773 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
774 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
775 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
776 case vmIntrinsics::_getLength: return inline_native_getLength();
777 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
778 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
779 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
780 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
781 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
782 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
783
784 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
785 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
786
787 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
788
789 case vmIntrinsics::_isInstance:
790 case vmIntrinsics::_getModifiers:
791 case vmIntrinsics::_isInterface:
792 case vmIntrinsics::_isArray:
793 case vmIntrinsics::_isPrimitive:
794 case vmIntrinsics::_getSuperclass:
795 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
796
797 case vmIntrinsics::_floatToRawIntBits:
798 case vmIntrinsics::_floatToIntBits:
799 case vmIntrinsics::_intBitsToFloat:
800 case vmIntrinsics::_doubleToRawLongBits:
801 case vmIntrinsics::_doubleToLongBits:
802 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
803
804 case vmIntrinsics::_numberOfLeadingZeros_i:
805 case vmIntrinsics::_numberOfLeadingZeros_l:
806 case vmIntrinsics::_numberOfTrailingZeros_i:
807 case vmIntrinsics::_numberOfTrailingZeros_l:
808 case vmIntrinsics::_bitCount_i:
809 case vmIntrinsics::_bitCount_l:
810 case vmIntrinsics::_reverseBytes_i:
811 case vmIntrinsics::_reverseBytes_l:
812 case vmIntrinsics::_reverseBytes_s:
813 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
814
815 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
816
817 case vmIntrinsics::_Reference_get: return inline_reference_get();
818
819 case vmIntrinsics::_Class_cast: return inline_Class_cast();
820
821 case vmIntrinsics::_aescrypt_encryptBlock:
822 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
823
824 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
825 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
826 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
827
828 case vmIntrinsics::_counterMode_AESCrypt:
829 return inline_counterMode_AESCrypt(intrinsic_id());
830
831 case vmIntrinsics::_sha_implCompress:
832 case vmIntrinsics::_sha2_implCompress:
833 case vmIntrinsics::_sha5_implCompress:
834 return inline_sha_implCompress(intrinsic_id());
835
836 case vmIntrinsics::_digestBase_implCompressMB:
837 return inline_digestBase_implCompressMB(predicate);
838
839 case vmIntrinsics::_multiplyToLen:
840 return inline_multiplyToLen();
841
842 case vmIntrinsics::_squareToLen:
843 return inline_squareToLen();
844
845 case vmIntrinsics::_mulAdd:
846 return inline_mulAdd();
847
848 case vmIntrinsics::_montgomeryMultiply:
849 return inline_montgomeryMultiply();
850 case vmIntrinsics::_montgomerySquare:
851 return inline_montgomerySquare();
852
853 case vmIntrinsics::_vectorizedMismatch:
854 return inline_vectorizedMismatch();
855
856 case vmIntrinsics::_ghash_processBlocks:
857 return inline_ghash_processBlocks();
858 case vmIntrinsics::_base64_encodeBlock:
859 return inline_base64_encodeBlock();
860
861 case vmIntrinsics::_encodeISOArray:
862 case vmIntrinsics::_encodeByteISOArray:
863 return inline_encodeISOArray();
864
865 case vmIntrinsics::_updateCRC32:
866 return inline_updateCRC32();
867 case vmIntrinsics::_updateBytesCRC32:
868 return inline_updateBytesCRC32();
869 case vmIntrinsics::_updateByteBufferCRC32:
870 return inline_updateByteBufferCRC32();
871
872 case vmIntrinsics::_updateBytesCRC32C:
873 return inline_updateBytesCRC32C();
874 case vmIntrinsics::_updateDirectByteBufferCRC32C:
875 return inline_updateDirectByteBufferCRC32C();
876
877 case vmIntrinsics::_updateBytesAdler32:
878 return inline_updateBytesAdler32();
879 case vmIntrinsics::_updateByteBufferAdler32:
880 return inline_updateByteBufferAdler32();
881
882 case vmIntrinsics::_profileBoolean:
883 return inline_profileBoolean();
884 case vmIntrinsics::_isCompileConstant:
885 return inline_isCompileConstant();
886
887 case vmIntrinsics::_hasNegatives:
888 return inline_hasNegatives();
889
890 case vmIntrinsics::_fmaD:
891 case vmIntrinsics::_fmaF:
892 return inline_fma(intrinsic_id());
893
894 case vmIntrinsics::_isDigit:
895 case vmIntrinsics::_isLowerCase:
896 case vmIntrinsics::_isUpperCase:
897 case vmIntrinsics::_isWhitespace:
898 return inline_character_compare(intrinsic_id());
899
900 case vmIntrinsics::_maxF:
901 case vmIntrinsics::_minF:
902 case vmIntrinsics::_maxD:
903 case vmIntrinsics::_minD:
904 return inline_fp_min_max(intrinsic_id());
905
906 default:
907 // If you get here, it may be that someone has added a new intrinsic
908 // to the list in vmSymbols.hpp without implementing it here.
909 #ifndef PRODUCT
910 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
911 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
912 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
913 }
914 #endif
915 return false;
916 }
917 }
918
919 Node* LibraryCallKit::try_to_predicate(int predicate) {
920 if (!jvms()->has_method()) {
921 // Root JVMState has a null method.
922 assert(map()->memory()->Opcode() == Op_Parm, "");
923 // Insert the memory aliasing node
924 set_all_memory(reset_memory());
925 }
926 assert(merged_memory(), "");
927
928 switch (intrinsic_id()) {
929 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
930 return inline_cipherBlockChaining_AESCrypt_predicate(false);
931 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
932 return inline_cipherBlockChaining_AESCrypt_predicate(true);
933 case vmIntrinsics::_counterMode_AESCrypt:
934 return inline_counterMode_AESCrypt_predicate();
935 case vmIntrinsics::_digestBase_implCompressMB:
936 return inline_digestBase_implCompressMB_predicate(predicate);
937
938 default:
939 // If you get here, it may be that someone has added a new intrinsic
940 // to the list in vmSymbols.hpp without implementing it here.
941 #ifndef PRODUCT
942 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
943 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
944 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
945 }
946 #endif
947 Node* slow_ctl = control();
948 set_control(top()); // No fast path instrinsic
949 return slow_ctl;
950 }
951 }
952
953 //------------------------------set_result-------------------------------
954 // Helper function for finishing intrinsics.
955 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
956 record_for_igvn(region);
957 set_control(_gvn.transform(region));
958 set_result( _gvn.transform(value));
959 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
960 }
961
962 //------------------------------generate_guard---------------------------
963 // Helper function for generating guarded fast-slow graph structures.
964 // The given 'test', if true, guards a slow path. If the test fails
965 // then a fast path can be taken. (We generally hope it fails.)
966 // In all cases, GraphKit::control() is updated to the fast path.
967 // The returned value represents the control for the slow path.
968 // The return value is never 'top'; it is either a valid control
969 // or NULL if it is obvious that the slow path can never be taken.
970 // Also, if region and the slow control are not NULL, the slow edge
971 // is appended to the region.
972 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
973 if (stopped()) {
974 // Already short circuited.
975 return NULL;
976 }
977
978 // Build an if node and its projections.
979 // If test is true we take the slow path, which we assume is uncommon.
980 if (_gvn.type(test) == TypeInt::ZERO) {
981 // The slow branch is never taken. No need to build this guard.
982 return NULL;
983 }
984
985 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
986
987 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
988 if (if_slow == top()) {
989 // The slow branch is never taken. No need to build this guard.
990 return NULL;
991 }
992
993 if (region != NULL)
994 region->add_req(if_slow);
995
996 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
997 set_control(if_fast);
998
999 return if_slow;
1000 }
1001
1002 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1003 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1004 }
1005 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1006 return generate_guard(test, region, PROB_FAIR);
1007 }
1008
1009 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1010 Node* *pos_index) {
1011 if (stopped())
1012 return NULL; // already stopped
1013 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1014 return NULL; // index is already adequately typed
1015 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1016 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1017 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1018 if (is_neg != NULL && pos_index != NULL) {
1019 // Emulate effect of Parse::adjust_map_after_if.
1020 Node* ccast = new CastIINode(index, TypeInt::POS);
1021 ccast->set_req(0, control());
1022 (*pos_index) = _gvn.transform(ccast);
1023 }
1024 return is_neg;
1025 }
1026
1027 // Make sure that 'position' is a valid limit index, in [0..length].
1028 // There are two equivalent plans for checking this:
1029 // A. (offset + copyLength) unsigned<= arrayLength
1030 // B. offset <= (arrayLength - copyLength)
1031 // We require that all of the values above, except for the sum and
1032 // difference, are already known to be non-negative.
1033 // Plan A is robust in the face of overflow, if offset and copyLength
1034 // are both hugely positive.
1035 //
1036 // Plan B is less direct and intuitive, but it does not overflow at
1037 // all, since the difference of two non-negatives is always
1038 // representable. Whenever Java methods must perform the equivalent
1039 // check they generally use Plan B instead of Plan A.
1040 // For the moment we use Plan A.
1041 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1042 Node* subseq_length,
1043 Node* array_length,
1044 RegionNode* region) {
1045 if (stopped())
1046 return NULL; // already stopped
1047 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1048 if (zero_offset && subseq_length->eqv_uncast(array_length))
1049 return NULL; // common case of whole-array copy
1050 Node* last = subseq_length;
1051 if (!zero_offset) // last += offset
1052 last = _gvn.transform(new AddINode(last, offset));
1053 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1054 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1055 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1056 return is_over;
1057 }
1058
1059 // Emit range checks for the given String.value byte array
1060 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1061 if (stopped()) {
1062 return; // already stopped
1063 }
1064 RegionNode* bailout = new RegionNode(1);
1065 record_for_igvn(bailout);
1066 if (char_count) {
1067 // Convert char count to byte count
1068 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1069 }
1070
1071 // Offset and count must not be negative
1072 generate_negative_guard(offset, bailout);
1073 generate_negative_guard(count, bailout);
1074 // Offset + count must not exceed length of array
1075 generate_limit_guard(offset, count, load_array_length(array), bailout);
1076
1077 if (bailout->req() > 1) {
1078 PreserveJVMState pjvms(this);
1079 set_control(_gvn.transform(bailout));
1080 uncommon_trap(Deoptimization::Reason_intrinsic,
1081 Deoptimization::Action_maybe_recompile);
1082 }
1083 }
1084
1085 //--------------------------generate_current_thread--------------------
1086 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1087 ciKlass* thread_klass = env()->Thread_klass();
1088 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1089 Node* thread = _gvn.transform(new ThreadLocalNode());
1090 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1091 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1092 tls_output = thread;
1093 return threadObj;
1094 }
1095
1096
1097 //------------------------------make_string_method_node------------------------
1098 // Helper method for String intrinsic functions. This version is called with
1099 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1100 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1101 // containing the lengths of str1 and str2.
1102 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1103 Node* result = NULL;
1104 switch (opcode) {
1105 case Op_StrIndexOf:
1106 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1107 str1_start, cnt1, str2_start, cnt2, ae);
1108 break;
1109 case Op_StrComp:
1110 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1111 str1_start, cnt1, str2_start, cnt2, ae);
1112 break;
1113 case Op_StrEquals:
1114 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1115 // Use the constant length if there is one because optimized match rule may exist.
1116 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1117 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1118 break;
1119 default:
1120 ShouldNotReachHere();
1121 return NULL;
1122 }
1123
1124 // All these intrinsics have checks.
1125 C->set_has_split_ifs(true); // Has chance for split-if optimization
1126 clear_upper_avx();
1127
1128 return _gvn.transform(result);
1129 }
1130
1131 //------------------------------inline_string_compareTo------------------------
1132 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1133 Node* arg1 = argument(0);
1134 Node* arg2 = argument(1);
1135
1136 arg1 = must_be_not_null(arg1, true);
1137 arg2 = must_be_not_null(arg2, true);
1138
1139 arg1 = access_resolve(arg1, ACCESS_READ);
1140 arg2 = access_resolve(arg2, ACCESS_READ);
1141
1142 // Get start addr and length of first argument
1143 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1144 Node* arg1_cnt = load_array_length(arg1);
1145
1146 // Get start addr and length of second argument
1147 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1148 Node* arg2_cnt = load_array_length(arg2);
1149
1150 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1151 set_result(result);
1152 return true;
1153 }
1154
1155 //------------------------------inline_string_equals------------------------
1156 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1157 Node* arg1 = argument(0);
1158 Node* arg2 = argument(1);
1159
1160 // paths (plus control) merge
1161 RegionNode* region = new RegionNode(3);
1162 Node* phi = new PhiNode(region, TypeInt::BOOL);
1163
1164 if (!stopped()) {
1165
1166 arg1 = must_be_not_null(arg1, true);
1167 arg2 = must_be_not_null(arg2, true);
1168
1169 arg1 = access_resolve(arg1, ACCESS_READ);
1170 arg2 = access_resolve(arg2, ACCESS_READ);
1171
1172 // Get start addr and length of first argument
1173 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1174 Node* arg1_cnt = load_array_length(arg1);
1175
1176 // Get start addr and length of second argument
1177 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1178 Node* arg2_cnt = load_array_length(arg2);
1179
1180 // Check for arg1_cnt != arg2_cnt
1181 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1182 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1183 Node* if_ne = generate_slow_guard(bol, NULL);
1184 if (if_ne != NULL) {
1185 phi->init_req(2, intcon(0));
1186 region->init_req(2, if_ne);
1187 }
1188
1189 // Check for count == 0 is done by assembler code for StrEquals.
1190
1191 if (!stopped()) {
1192 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1193 phi->init_req(1, equals);
1194 region->init_req(1, control());
1195 }
1196 }
1197
1198 // post merge
1199 set_control(_gvn.transform(region));
1200 record_for_igvn(region);
1201
1202 set_result(_gvn.transform(phi));
1203 return true;
1204 }
1205
1206 //------------------------------inline_array_equals----------------------------
1207 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1208 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1209 Node* arg1 = argument(0);
1210 Node* arg2 = argument(1);
1211
1212 arg1 = access_resolve(arg1, ACCESS_READ);
1213 arg2 = access_resolve(arg2, ACCESS_READ);
1214
1215 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1216 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1217 clear_upper_avx();
1218
1219 return true;
1220 }
1221
1222 //------------------------------inline_hasNegatives------------------------------
1223 bool LibraryCallKit::inline_hasNegatives() {
1224 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1225 return false;
1226 }
1227
1228 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1229 // no receiver since it is static method
1230 Node* ba = argument(0);
1231 Node* offset = argument(1);
1232 Node* len = argument(2);
1233
1234 ba = must_be_not_null(ba, true);
1235
1236 // Range checks
1237 generate_string_range_check(ba, offset, len, false);
1238 if (stopped()) {
1239 return true;
1240 }
1241 ba = access_resolve(ba, ACCESS_READ);
1242 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1243 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1244 set_result(_gvn.transform(result));
1245 return true;
1246 }
1247
1248 bool LibraryCallKit::inline_preconditions_checkIndex() {
1249 Node* index = argument(0);
1250 Node* length = argument(1);
1251 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1252 return false;
1253 }
1254
1255 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1256 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1257
1258 {
1259 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1260 uncommon_trap(Deoptimization::Reason_intrinsic,
1261 Deoptimization::Action_make_not_entrant);
1262 }
1263
1264 if (stopped()) {
1265 return false;
1266 }
1267
1268 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1269 BoolTest::mask btest = BoolTest::lt;
1270 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1271 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1272 _gvn.set_type(rc, rc->Value(&_gvn));
1273 if (!rc_bool->is_Con()) {
1274 record_for_igvn(rc);
1275 }
1276 set_control(_gvn.transform(new IfTrueNode(rc)));
1277 {
1278 PreserveJVMState pjvms(this);
1279 set_control(_gvn.transform(new IfFalseNode(rc)));
1280 uncommon_trap(Deoptimization::Reason_range_check,
1281 Deoptimization::Action_make_not_entrant);
1282 }
1283
1284 if (stopped()) {
1285 return false;
1286 }
1287
1288 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1289 result->set_req(0, control());
1290 result = _gvn.transform(result);
1291 set_result(result);
1292 replace_in_map(index, result);
1293 clear_upper_avx();
1294 return true;
1295 }
1296
1297 //------------------------------inline_string_indexOf------------------------
1298 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1299 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1300 return false;
1301 }
1302 Node* src = argument(0);
1303 Node* tgt = argument(1);
1304
1305 // Make the merge point
1306 RegionNode* result_rgn = new RegionNode(4);
1307 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1308
1309 src = must_be_not_null(src, true);
1310 tgt = must_be_not_null(tgt, true);
1311
1312 src = access_resolve(src, ACCESS_READ);
1313 tgt = access_resolve(tgt, ACCESS_READ);
1314
1315 // Get start addr and length of source string
1316 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1317 Node* src_count = load_array_length(src);
1318
1319 // Get start addr and length of substring
1320 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1321 Node* tgt_count = load_array_length(tgt);
1322
1323 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1324 // Divide src size by 2 if String is UTF16 encoded
1325 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1326 }
1327 if (ae == StrIntrinsicNode::UU) {
1328 // Divide substring size by 2 if String is UTF16 encoded
1329 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1330 }
1331
1332 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1333 if (result != NULL) {
1334 result_phi->init_req(3, result);
1335 result_rgn->init_req(3, control());
1336 }
1337 set_control(_gvn.transform(result_rgn));
1338 record_for_igvn(result_rgn);
1339 set_result(_gvn.transform(result_phi));
1340
1341 return true;
1342 }
1343
1344 //-----------------------------inline_string_indexOf-----------------------
1345 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1346 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1347 return false;
1348 }
1349 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1350 return false;
1351 }
1352 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1353 Node* src = argument(0); // byte[]
1354 Node* src_count = argument(1); // char count
1355 Node* tgt = argument(2); // byte[]
1356 Node* tgt_count = argument(3); // char count
1357 Node* from_index = argument(4); // char index
1358
1359 src = must_be_not_null(src, true);
1360 tgt = must_be_not_null(tgt, true);
1361
1362 src = access_resolve(src, ACCESS_READ);
1363 tgt = access_resolve(tgt, ACCESS_READ);
1364
1365 // Multiply byte array index by 2 if String is UTF16 encoded
1366 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1367 src_count = _gvn.transform(new SubINode(src_count, from_index));
1368 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1369 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1370
1371 // Range checks
1372 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1373 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1374 if (stopped()) {
1375 return true;
1376 }
1377
1378 RegionNode* region = new RegionNode(5);
1379 Node* phi = new PhiNode(region, TypeInt::INT);
1380
1381 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1382 if (result != NULL) {
1383 // The result is index relative to from_index if substring was found, -1 otherwise.
1384 // Generate code which will fold into cmove.
1385 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1386 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1387
1388 Node* if_lt = generate_slow_guard(bol, NULL);
1389 if (if_lt != NULL) {
1390 // result == -1
1391 phi->init_req(3, result);
1392 region->init_req(3, if_lt);
1393 }
1394 if (!stopped()) {
1395 result = _gvn.transform(new AddINode(result, from_index));
1396 phi->init_req(4, result);
1397 region->init_req(4, control());
1398 }
1399 }
1400
1401 set_control(_gvn.transform(region));
1402 record_for_igvn(region);
1403 set_result(_gvn.transform(phi));
1404 clear_upper_avx();
1405
1406 return true;
1407 }
1408
1409 // Create StrIndexOfNode with fast path checks
1410 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1411 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1412 // Check for substr count > string count
1413 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1414 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1415 Node* if_gt = generate_slow_guard(bol, NULL);
1416 if (if_gt != NULL) {
1417 phi->init_req(1, intcon(-1));
1418 region->init_req(1, if_gt);
1419 }
1420 if (!stopped()) {
1421 // Check for substr count == 0
1422 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1423 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1424 Node* if_zero = generate_slow_guard(bol, NULL);
1425 if (if_zero != NULL) {
1426 phi->init_req(2, intcon(0));
1427 region->init_req(2, if_zero);
1428 }
1429 }
1430 if (!stopped()) {
1431 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1432 }
1433 return NULL;
1434 }
1435
1436 //-----------------------------inline_string_indexOfChar-----------------------
1437 bool LibraryCallKit::inline_string_indexOfChar() {
1438 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1439 return false;
1440 }
1441 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1442 return false;
1443 }
1444 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1445 Node* src = argument(0); // byte[]
1446 Node* tgt = argument(1); // tgt is int ch
1447 Node* from_index = argument(2);
1448 Node* max = argument(3);
1449
1450 src = must_be_not_null(src, true);
1451 src = access_resolve(src, ACCESS_READ);
1452
1453 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1454 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1455 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1456
1457 // Range checks
1458 generate_string_range_check(src, src_offset, src_count, true);
1459 if (stopped()) {
1460 return true;
1461 }
1462
1463 RegionNode* region = new RegionNode(3);
1464 Node* phi = new PhiNode(region, TypeInt::INT);
1465
1466 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1467 C->set_has_split_ifs(true); // Has chance for split-if optimization
1468 _gvn.transform(result);
1469
1470 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1471 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1472
1473 Node* if_lt = generate_slow_guard(bol, NULL);
1474 if (if_lt != NULL) {
1475 // result == -1
1476 phi->init_req(2, result);
1477 region->init_req(2, if_lt);
1478 }
1479 if (!stopped()) {
1480 result = _gvn.transform(new AddINode(result, from_index));
1481 phi->init_req(1, result);
1482 region->init_req(1, control());
1483 }
1484 set_control(_gvn.transform(region));
1485 record_for_igvn(region);
1486 set_result(_gvn.transform(phi));
1487
1488 return true;
1489 }
1490 //---------------------------inline_string_copy---------------------
1491 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1492 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1493 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1494 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1495 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1496 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1497 bool LibraryCallKit::inline_string_copy(bool compress) {
1498 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1499 return false;
1500 }
1501 int nargs = 5; // 2 oops, 3 ints
1502 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1503
1504 Node* src = argument(0);
1505 Node* src_offset = argument(1);
1506 Node* dst = argument(2);
1507 Node* dst_offset = argument(3);
1508 Node* length = argument(4);
1509
1510 // Check for allocation before we add nodes that would confuse
1511 // tightly_coupled_allocation()
1512 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1513
1514 // Figure out the size and type of the elements we will be copying.
1515 const Type* src_type = src->Value(&_gvn);
1516 const Type* dst_type = dst->Value(&_gvn);
1517 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1518 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1519 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1520 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1521 "Unsupported array types for inline_string_copy");
1522
1523 src = must_be_not_null(src, true);
1524 dst = must_be_not_null(dst, true);
1525
1526 // Convert char[] offsets to byte[] offsets
1527 bool convert_src = (compress && src_elem == T_BYTE);
1528 bool convert_dst = (!compress && dst_elem == T_BYTE);
1529 if (convert_src) {
1530 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1531 } else if (convert_dst) {
1532 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1533 }
1534
1535 // Range checks
1536 generate_string_range_check(src, src_offset, length, convert_src);
1537 generate_string_range_check(dst, dst_offset, length, convert_dst);
1538 if (stopped()) {
1539 return true;
1540 }
1541
1542 src = access_resolve(src, ACCESS_READ);
1543 dst = access_resolve(dst, ACCESS_WRITE);
1544
1545 Node* src_start = array_element_address(src, src_offset, src_elem);
1546 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1547 // 'src_start' points to src array + scaled offset
1548 // 'dst_start' points to dst array + scaled offset
1549 Node* count = NULL;
1550 if (compress) {
1551 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1552 } else {
1553 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1554 }
1555
1556 if (alloc != NULL) {
1557 if (alloc->maybe_set_complete(&_gvn)) {
1558 // "You break it, you buy it."
1559 InitializeNode* init = alloc->initialization();
1560 assert(init->is_complete(), "we just did this");
1561 init->set_complete_with_arraycopy();
1562 assert(dst->is_CheckCastPP(), "sanity");
1563 assert(dst->in(0)->in(0) == init, "dest pinned");
1564 }
1565 // Do not let stores that initialize this object be reordered with
1566 // a subsequent store that would make this object accessible by
1567 // other threads.
1568 // Record what AllocateNode this StoreStore protects so that
1569 // escape analysis can go from the MemBarStoreStoreNode to the
1570 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1571 // based on the escape status of the AllocateNode.
1572 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1573 }
1574 if (compress) {
1575 set_result(_gvn.transform(count));
1576 }
1577 clear_upper_avx();
1578
1579 return true;
1580 }
1581
1582 #ifdef _LP64
1583 #define XTOP ,top() /*additional argument*/
1584 #else //_LP64
1585 #define XTOP /*no additional argument*/
1586 #endif //_LP64
1587
1588 //------------------------inline_string_toBytesU--------------------------
1589 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1590 bool LibraryCallKit::inline_string_toBytesU() {
1591 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1592 return false;
1593 }
1594 // Get the arguments.
1595 Node* value = argument(0);
1596 Node* offset = argument(1);
1597 Node* length = argument(2);
1598
1599 Node* newcopy = NULL;
1600
1601 // Set the original stack and the reexecute bit for the interpreter to reexecute
1602 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1603 { PreserveReexecuteState preexecs(this);
1604 jvms()->set_should_reexecute(true);
1605
1606 // Check if a null path was taken unconditionally.
1607 value = null_check(value);
1608
1609 RegionNode* bailout = new RegionNode(1);
1610 record_for_igvn(bailout);
1611
1612 // Range checks
1613 generate_negative_guard(offset, bailout);
1614 generate_negative_guard(length, bailout);
1615 generate_limit_guard(offset, length, load_array_length(value), bailout);
1616 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1617 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1618
1619 if (bailout->req() > 1) {
1620 PreserveJVMState pjvms(this);
1621 set_control(_gvn.transform(bailout));
1622 uncommon_trap(Deoptimization::Reason_intrinsic,
1623 Deoptimization::Action_maybe_recompile);
1624 }
1625 if (stopped()) {
1626 return true;
1627 }
1628
1629 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1630 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1631 newcopy = new_array(klass_node, size, 0); // no arguments to push
1632 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1633
1634 // Calculate starting addresses.
1635 value = access_resolve(value, ACCESS_READ);
1636 Node* src_start = array_element_address(value, offset, T_CHAR);
1637 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1638
1639 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1640 const TypeInt* toffset = gvn().type(offset)->is_int();
1641 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1642
1643 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1644 const char* copyfunc_name = "arraycopy";
1645 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1646 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1647 OptoRuntime::fast_arraycopy_Type(),
1648 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1649 src_start, dst_start, ConvI2X(length) XTOP);
1650 // Do not let reads from the cloned object float above the arraycopy.
1651 if (alloc != NULL) {
1652 if (alloc->maybe_set_complete(&_gvn)) {
1653 // "You break it, you buy it."
1654 InitializeNode* init = alloc->initialization();
1655 assert(init->is_complete(), "we just did this");
1656 init->set_complete_with_arraycopy();
1657 assert(newcopy->is_CheckCastPP(), "sanity");
1658 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1659 }
1660 // Do not let stores that initialize this object be reordered with
1661 // a subsequent store that would make this object accessible by
1662 // other threads.
1663 // Record what AllocateNode this StoreStore protects so that
1664 // escape analysis can go from the MemBarStoreStoreNode to the
1665 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1666 // based on the escape status of the AllocateNode.
1667 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1668 } else {
1669 insert_mem_bar(Op_MemBarCPUOrder);
1670 }
1671 } // original reexecute is set back here
1672
1673 C->set_has_split_ifs(true); // Has chance for split-if optimization
1674 if (!stopped()) {
1675 set_result(newcopy);
1676 }
1677 clear_upper_avx();
1678
1679 return true;
1680 }
1681
1682 //------------------------inline_string_getCharsU--------------------------
1683 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1684 bool LibraryCallKit::inline_string_getCharsU() {
1685 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1686 return false;
1687 }
1688
1689 // Get the arguments.
1690 Node* src = argument(0);
1691 Node* src_begin = argument(1);
1692 Node* src_end = argument(2); // exclusive offset (i < src_end)
1693 Node* dst = argument(3);
1694 Node* dst_begin = argument(4);
1695
1696 // Check for allocation before we add nodes that would confuse
1697 // tightly_coupled_allocation()
1698 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1699
1700 // Check if a null path was taken unconditionally.
1701 src = null_check(src);
1702 dst = null_check(dst);
1703 if (stopped()) {
1704 return true;
1705 }
1706
1707 // Get length and convert char[] offset to byte[] offset
1708 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1709 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1710
1711 // Range checks
1712 generate_string_range_check(src, src_begin, length, true);
1713 generate_string_range_check(dst, dst_begin, length, false);
1714 if (stopped()) {
1715 return true;
1716 }
1717
1718 if (!stopped()) {
1719 src = access_resolve(src, ACCESS_READ);
1720 dst = access_resolve(dst, ACCESS_WRITE);
1721
1722 // Calculate starting addresses.
1723 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1724 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1725
1726 // Check if array addresses are aligned to HeapWordSize
1727 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1728 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1729 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1730 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1731
1732 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1733 const char* copyfunc_name = "arraycopy";
1734 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1735 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1736 OptoRuntime::fast_arraycopy_Type(),
1737 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1738 src_start, dst_start, ConvI2X(length) XTOP);
1739 // Do not let reads from the cloned object float above the arraycopy.
1740 if (alloc != NULL) {
1741 if (alloc->maybe_set_complete(&_gvn)) {
1742 // "You break it, you buy it."
1743 InitializeNode* init = alloc->initialization();
1744 assert(init->is_complete(), "we just did this");
1745 init->set_complete_with_arraycopy();
1746 assert(dst->is_CheckCastPP(), "sanity");
1747 assert(dst->in(0)->in(0) == init, "dest pinned");
1748 }
1749 // Do not let stores that initialize this object be reordered with
1750 // a subsequent store that would make this object accessible by
1751 // other threads.
1752 // Record what AllocateNode this StoreStore protects so that
1753 // escape analysis can go from the MemBarStoreStoreNode to the
1754 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1755 // based on the escape status of the AllocateNode.
1756 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1757 } else {
1758 insert_mem_bar(Op_MemBarCPUOrder);
1759 }
1760 }
1761
1762 C->set_has_split_ifs(true); // Has chance for split-if optimization
1763 return true;
1764 }
1765
1766 //----------------------inline_string_char_access----------------------------
1767 // Store/Load char to/from byte[] array.
1768 // static void StringUTF16.putChar(byte[] val, int index, int c)
1769 // static char StringUTF16.getChar(byte[] val, int index)
1770 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1771 Node* value = argument(0);
1772 Node* index = argument(1);
1773 Node* ch = is_store ? argument(2) : NULL;
1774
1775 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1776 // correctly requires matched array shapes.
1777 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1778 "sanity: byte[] and char[] bases agree");
1779 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1780 "sanity: byte[] and char[] scales agree");
1781
1782 // Bail when getChar over constants is requested: constant folding would
1783 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1784 // Java method would constant fold nicely instead.
1785 if (!is_store && value->is_Con() && index->is_Con()) {
1786 return false;
1787 }
1788
1789 value = must_be_not_null(value, true);
1790 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1791
1792 Node* adr = array_element_address(value, index, T_CHAR);
1793 if (adr->is_top()) {
1794 return false;
1795 }
1796 if (is_store) {
1797 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1798 } else {
1799 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1800 set_result(ch);
1801 }
1802 return true;
1803 }
1804
1805 //--------------------------round_double_node--------------------------------
1806 // Round a double node if necessary.
1807 Node* LibraryCallKit::round_double_node(Node* n) {
1808 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1809 n = _gvn.transform(new RoundDoubleNode(0, n));
1810 return n;
1811 }
1812
1813 //------------------------------inline_math-----------------------------------
1814 // public static double Math.abs(double)
1815 // public static double Math.sqrt(double)
1816 // public static double Math.log(double)
1817 // public static double Math.log10(double)
1818 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1819 Node* arg = round_double_node(argument(0));
1820 Node* n = NULL;
1821 switch (id) {
1822 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1823 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1824 default: fatal_unexpected_iid(id); break;
1825 }
1826 set_result(_gvn.transform(n));
1827 return true;
1828 }
1829
1830 //------------------------------runtime_math-----------------------------
1831 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1832 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1833 "must be (DD)D or (D)D type");
1834
1835 // Inputs
1836 Node* a = round_double_node(argument(0));
1837 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1838
1839 const TypePtr* no_memory_effects = NULL;
1840 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1841 no_memory_effects,
1842 a, top(), b, b ? top() : NULL);
1843 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1844 #ifdef ASSERT
1845 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1846 assert(value_top == top(), "second value must be top");
1847 #endif
1848
1849 set_result(value);
1850 return true;
1851 }
1852
1853 //------------------------------inline_math_native-----------------------------
1854 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1855 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1856 switch (id) {
1857 // These intrinsics are not properly supported on all hardware
1858 case vmIntrinsics::_dsin:
1859 return StubRoutines::dsin() != NULL ?
1860 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1861 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1862 case vmIntrinsics::_dcos:
1863 return StubRoutines::dcos() != NULL ?
1864 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1865 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1866 case vmIntrinsics::_dtan:
1867 return StubRoutines::dtan() != NULL ?
1868 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1869 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1870 case vmIntrinsics::_dlog:
1871 return StubRoutines::dlog() != NULL ?
1872 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1873 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1874 case vmIntrinsics::_dlog10:
1875 return StubRoutines::dlog10() != NULL ?
1876 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1877 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1878
1879 // These intrinsics are supported on all hardware
1880 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1881 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1882
1883 case vmIntrinsics::_dexp:
1884 return StubRoutines::dexp() != NULL ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1887 case vmIntrinsics::_dpow: {
1888 Node* exp = round_double_node(argument(2));
1889 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1890 if (d != NULL && d->getd() == 2.0) {
1891 // Special case: pow(x, 2.0) => x * x
1892 Node* base = round_double_node(argument(0));
1893 set_result(_gvn.transform(new MulDNode(base, base)));
1894 return true;
1895 }
1896 return StubRoutines::dpow() != NULL ?
1897 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1898 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1899 }
1900 #undef FN_PTR
1901
1902 // These intrinsics are not yet correctly implemented
1903 case vmIntrinsics::_datan2:
1904 return false;
1905
1906 default:
1907 fatal_unexpected_iid(id);
1908 return false;
1909 }
1910 }
1911
1912 static bool is_simple_name(Node* n) {
1913 return (n->req() == 1 // constant
1914 || (n->is_Type() && n->as_Type()->type()->singleton())
1915 || n->is_Proj() // parameter or return value
1916 || n->is_Phi() // local of some sort
1917 );
1918 }
1919
1920 //----------------------------inline_notify-----------------------------------*
1921 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1922 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1923 address func;
1924 if (id == vmIntrinsics::_notify) {
1925 func = OptoRuntime::monitor_notify_Java();
1926 } else {
1927 func = OptoRuntime::monitor_notifyAll_Java();
1928 }
1929 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1930 make_slow_call_ex(call, env()->Throwable_klass(), false);
1931 return true;
1932 }
1933
1934
1935 //----------------------------inline_min_max-----------------------------------
1936 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1937 set_result(generate_min_max(id, argument(0), argument(1)));
1938 return true;
1939 }
1940
1941 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1942 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1943 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1944 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1945 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1946
1947 {
1948 PreserveJVMState pjvms(this);
1949 PreserveReexecuteState preexecs(this);
1950 jvms()->set_should_reexecute(true);
1951
1952 set_control(slow_path);
1953 set_i_o(i_o());
1954
1955 uncommon_trap(Deoptimization::Reason_intrinsic,
1956 Deoptimization::Action_none);
1957 }
1958
1959 set_control(fast_path);
1960 set_result(math);
1961 }
1962
1963 template <typename OverflowOp>
1964 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1965 typedef typename OverflowOp::MathOp MathOp;
1966
1967 MathOp* mathOp = new MathOp(arg1, arg2);
1968 Node* operation = _gvn.transform( mathOp );
1969 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1970 inline_math_mathExact(operation, ofcheck);
1971 return true;
1972 }
1973
1974 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1975 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1976 }
1977
1978 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1979 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1980 }
1981
1982 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1983 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1984 }
1985
1986 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1987 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1988 }
1989
1990 bool LibraryCallKit::inline_math_negateExactI() {
1991 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1992 }
1993
1994 bool LibraryCallKit::inline_math_negateExactL() {
1995 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1996 }
1997
1998 bool LibraryCallKit::inline_math_multiplyExactI() {
1999 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2000 }
2001
2002 bool LibraryCallKit::inline_math_multiplyExactL() {
2003 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2004 }
2005
2006 bool LibraryCallKit::inline_math_multiplyHigh() {
2007 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2008 return true;
2009 }
2010
2011 Node*
2012 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2013 // These are the candidate return value:
2014 Node* xvalue = x0;
2015 Node* yvalue = y0;
2016
2017 if (xvalue == yvalue) {
2018 return xvalue;
2019 }
2020
2021 bool want_max = (id == vmIntrinsics::_max);
2022
2023 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2024 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2025 if (txvalue == NULL || tyvalue == NULL) return top();
2026 // This is not really necessary, but it is consistent with a
2027 // hypothetical MaxINode::Value method:
2028 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2029
2030 // %%% This folding logic should (ideally) be in a different place.
2031 // Some should be inside IfNode, and there to be a more reliable
2032 // transformation of ?: style patterns into cmoves. We also want
2033 // more powerful optimizations around cmove and min/max.
2034
2035 // Try to find a dominating comparison of these guys.
2036 // It can simplify the index computation for Arrays.copyOf
2037 // and similar uses of System.arraycopy.
2038 // First, compute the normalized version of CmpI(x, y).
2039 int cmp_op = Op_CmpI;
2040 Node* xkey = xvalue;
2041 Node* ykey = yvalue;
2042 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2043 if (ideal_cmpxy->is_Cmp()) {
2044 // E.g., if we have CmpI(length - offset, count),
2045 // it might idealize to CmpI(length, count + offset)
2046 cmp_op = ideal_cmpxy->Opcode();
2047 xkey = ideal_cmpxy->in(1);
2048 ykey = ideal_cmpxy->in(2);
2049 }
2050
2051 // Start by locating any relevant comparisons.
2052 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2053 Node* cmpxy = NULL;
2054 Node* cmpyx = NULL;
2055 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2056 Node* cmp = start_from->fast_out(k);
2057 if (cmp->outcnt() > 0 && // must have prior uses
2058 cmp->in(0) == NULL && // must be context-independent
2059 cmp->Opcode() == cmp_op) { // right kind of compare
2060 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2061 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2062 }
2063 }
2064
2065 const int NCMPS = 2;
2066 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2067 int cmpn;
2068 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2069 if (cmps[cmpn] != NULL) break; // find a result
2070 }
2071 if (cmpn < NCMPS) {
2072 // Look for a dominating test that tells us the min and max.
2073 int depth = 0; // Limit search depth for speed
2074 Node* dom = control();
2075 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2076 if (++depth >= 100) break;
2077 Node* ifproj = dom;
2078 if (!ifproj->is_Proj()) continue;
2079 Node* iff = ifproj->in(0);
2080 if (!iff->is_If()) continue;
2081 Node* bol = iff->in(1);
2082 if (!bol->is_Bool()) continue;
2083 Node* cmp = bol->in(1);
2084 if (cmp == NULL) continue;
2085 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2086 if (cmps[cmpn] == cmp) break;
2087 if (cmpn == NCMPS) continue;
2088 BoolTest::mask btest = bol->as_Bool()->_test._test;
2089 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2090 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2091 // At this point, we know that 'x btest y' is true.
2092 switch (btest) {
2093 case BoolTest::eq:
2094 // They are proven equal, so we can collapse the min/max.
2095 // Either value is the answer. Choose the simpler.
2096 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2097 return yvalue;
2098 return xvalue;
2099 case BoolTest::lt: // x < y
2100 case BoolTest::le: // x <= y
2101 return (want_max ? yvalue : xvalue);
2102 case BoolTest::gt: // x > y
2103 case BoolTest::ge: // x >= y
2104 return (want_max ? xvalue : yvalue);
2105 default:
2106 break;
2107 }
2108 }
2109 }
2110
2111 // We failed to find a dominating test.
2112 // Let's pick a test that might GVN with prior tests.
2113 Node* best_bol = NULL;
2114 BoolTest::mask best_btest = BoolTest::illegal;
2115 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2116 Node* cmp = cmps[cmpn];
2117 if (cmp == NULL) continue;
2118 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2119 Node* bol = cmp->fast_out(j);
2120 if (!bol->is_Bool()) continue;
2121 BoolTest::mask btest = bol->as_Bool()->_test._test;
2122 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2123 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2124 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2125 best_bol = bol->as_Bool();
2126 best_btest = btest;
2127 }
2128 }
2129 }
2130
2131 Node* answer_if_true = NULL;
2132 Node* answer_if_false = NULL;
2133 switch (best_btest) {
2134 default:
2135 if (cmpxy == NULL)
2136 cmpxy = ideal_cmpxy;
2137 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2138 // and fall through:
2139 case BoolTest::lt: // x < y
2140 case BoolTest::le: // x <= y
2141 answer_if_true = (want_max ? yvalue : xvalue);
2142 answer_if_false = (want_max ? xvalue : yvalue);
2143 break;
2144 case BoolTest::gt: // x > y
2145 case BoolTest::ge: // x >= y
2146 answer_if_true = (want_max ? xvalue : yvalue);
2147 answer_if_false = (want_max ? yvalue : xvalue);
2148 break;
2149 }
2150
2151 jint hi, lo;
2152 if (want_max) {
2153 // We can sharpen the minimum.
2154 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2155 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2156 } else {
2157 // We can sharpen the maximum.
2158 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2159 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2160 }
2161
2162 // Use a flow-free graph structure, to avoid creating excess control edges
2163 // which could hinder other optimizations.
2164 // Since Math.min/max is often used with arraycopy, we want
2165 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2166 Node* cmov = CMoveNode::make(NULL, best_bol,
2167 answer_if_false, answer_if_true,
2168 TypeInt::make(lo, hi, widen));
2169
2170 return _gvn.transform(cmov);
2171
2172 /*
2173 // This is not as desirable as it may seem, since Min and Max
2174 // nodes do not have a full set of optimizations.
2175 // And they would interfere, anyway, with 'if' optimizations
2176 // and with CMoveI canonical forms.
2177 switch (id) {
2178 case vmIntrinsics::_min:
2179 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2180 case vmIntrinsics::_max:
2181 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2182 default:
2183 ShouldNotReachHere();
2184 }
2185 */
2186 }
2187
2188 inline int
2189 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2190 const TypePtr* base_type = TypePtr::NULL_PTR;
2191 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2192 if (base_type == NULL) {
2193 // Unknown type.
2194 return Type::AnyPtr;
2195 } else if (base_type == TypePtr::NULL_PTR) {
2196 // Since this is a NULL+long form, we have to switch to a rawptr.
2197 base = _gvn.transform(new CastX2PNode(offset));
2198 offset = MakeConX(0);
2199 return Type::RawPtr;
2200 } else if (base_type->base() == Type::RawPtr) {
2201 return Type::RawPtr;
2202 } else if (base_type->isa_oopptr()) {
2203 // Base is never null => always a heap address.
2204 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2205 return Type::OopPtr;
2206 }
2207 // Offset is small => always a heap address.
2208 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2209 if (offset_type != NULL &&
2210 base_type->offset() == 0 && // (should always be?)
2211 offset_type->_lo >= 0 &&
2212 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2213 return Type::OopPtr;
2214 } else if (type == T_OBJECT) {
2215 // off heap access to an oop doesn't make any sense. Has to be on
2216 // heap.
2217 return Type::OopPtr;
2218 }
2219 // Otherwise, it might either be oop+off or NULL+addr.
2220 return Type::AnyPtr;
2221 } else {
2222 // No information:
2223 return Type::AnyPtr;
2224 }
2225 }
2226
2227 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2228 Node* uncasted_base = base;
2229 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2230 if (kind == Type::RawPtr) {
2231 return basic_plus_adr(top(), uncasted_base, offset);
2232 } else if (kind == Type::AnyPtr) {
2233 assert(base == uncasted_base, "unexpected base change");
2234 if (can_cast) {
2235 if (!_gvn.type(base)->speculative_maybe_null() &&
2236 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2237 // According to profiling, this access is always on
2238 // heap. Casting the base to not null and thus avoiding membars
2239 // around the access should allow better optimizations
2240 Node* null_ctl = top();
2241 base = null_check_oop(base, &null_ctl, true, true, true);
2242 assert(null_ctl->is_top(), "no null control here");
2243 return basic_plus_adr(base, offset);
2244 } else if (_gvn.type(base)->speculative_always_null() &&
2245 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2246 // According to profiling, this access is always off
2247 // heap.
2248 base = null_assert(base);
2249 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2250 offset = MakeConX(0);
2251 return basic_plus_adr(top(), raw_base, offset);
2252 }
2253 }
2254 // We don't know if it's an on heap or off heap access. Fall back
2255 // to raw memory access.
2256 base = access_resolve(base, decorators);
2257 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2258 return basic_plus_adr(top(), raw, offset);
2259 } else {
2260 assert(base == uncasted_base, "unexpected base change");
2261 // We know it's an on heap access so base can't be null
2262 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2263 base = must_be_not_null(base, true);
2264 }
2265 return basic_plus_adr(base, offset);
2266 }
2267 }
2268
2269 //--------------------------inline_number_methods-----------------------------
2270 // inline int Integer.numberOfLeadingZeros(int)
2271 // inline int Long.numberOfLeadingZeros(long)
2272 //
2273 // inline int Integer.numberOfTrailingZeros(int)
2274 // inline int Long.numberOfTrailingZeros(long)
2275 //
2276 // inline int Integer.bitCount(int)
2277 // inline int Long.bitCount(long)
2278 //
2279 // inline char Character.reverseBytes(char)
2280 // inline short Short.reverseBytes(short)
2281 // inline int Integer.reverseBytes(int)
2282 // inline long Long.reverseBytes(long)
2283 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2284 Node* arg = argument(0);
2285 Node* n = NULL;
2286 switch (id) {
2287 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2288 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2289 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2290 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2291 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2292 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2293 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2294 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2295 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2296 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2297 default: fatal_unexpected_iid(id); break;
2298 }
2299 set_result(_gvn.transform(n));
2300 return true;
2301 }
2302
2303 //----------------------------inline_unsafe_access----------------------------
2304
2305 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2306 // Attempt to infer a sharper value type from the offset and base type.
2307 ciKlass* sharpened_klass = NULL;
2308
2309 // See if it is an instance field, with an object type.
2310 if (alias_type->field() != NULL) {
2311 if (alias_type->field()->type()->is_klass()) {
2312 sharpened_klass = alias_type->field()->type()->as_klass();
2313 }
2314 }
2315
2316 // See if it is a narrow oop array.
2317 if (adr_type->isa_aryptr()) {
2318 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2319 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2320 if (elem_type != NULL) {
2321 sharpened_klass = elem_type->klass();
2322 }
2323 }
2324 }
2325
2326 // The sharpened class might be unloaded if there is no class loader
2327 // contraint in place.
2328 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2329 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2330
2331 #ifndef PRODUCT
2332 if (C->print_intrinsics() || C->print_inlining()) {
2333 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2334 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2335 }
2336 #endif
2337 // Sharpen the value type.
2338 return tjp;
2339 }
2340 return NULL;
2341 }
2342
2343 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2344 switch (kind) {
2345 case Relaxed:
2346 return MO_UNORDERED;
2347 case Opaque:
2348 return MO_RELAXED;
2349 case Acquire:
2350 return MO_ACQUIRE;
2351 case Release:
2352 return MO_RELEASE;
2353 case Volatile:
2354 return MO_SEQ_CST;
2355 default:
2356 ShouldNotReachHere();
2357 return 0;
2358 }
2359 }
2360
2361 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2362 if (callee()->is_static()) return false; // caller must have the capability!
2363 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2364 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2365 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2366 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2367
2368 if (type == T_OBJECT || type == T_ARRAY) {
2369 decorators |= ON_UNKNOWN_OOP_REF;
2370 }
2371
2372 if (unaligned) {
2373 decorators |= C2_UNALIGNED;
2374 }
2375
2376 #ifndef PRODUCT
2377 {
2378 ResourceMark rm;
2379 // Check the signatures.
2380 ciSignature* sig = callee()->signature();
2381 #ifdef ASSERT
2382 if (!is_store) {
2383 // Object getReference(Object base, int/long offset), etc.
2384 BasicType rtype = sig->return_type()->basic_type();
2385 assert(rtype == type || (rtype == T_OBJECT && type == T_VALUETYPE), "getter must return the expected value");
2386 assert(sig->count() == 2 || (type == T_VALUETYPE && sig->count() == 3), "oop getter has 2 or 3 arguments");
2387 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2388 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2389 } else {
2390 // void putReference(Object base, int/long offset, Object x), etc.
2391 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2392 assert(sig->count() == 3 || (type == T_VALUETYPE && sig->count() == 4), "oop putter has 3 arguments");
2393 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2394 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2395 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2396 assert(vtype == type || (type == T_VALUETYPE && vtype == T_OBJECT), "putter must accept the expected value");
2397 }
2398 #endif // ASSERT
2399 }
2400 #endif //PRODUCT
2401
2402 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2403
2404 Node* receiver = argument(0); // type: oop
2405
2406 // Build address expression.
2407 Node* adr;
2408 Node* heap_base_oop = top();
2409 Node* offset = top();
2410 Node* val;
2411
2412 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2413 Node* base = argument(1); // type: oop
2414 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2415 offset = argument(2); // type: long
2416 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2417 // to be plain byte offsets, which are also the same as those accepted
2418 // by oopDesc::field_addr.
2419 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2420 "fieldOffset must be byte-scaled");
2421
2422 ciValueKlass* value_klass = NULL;
2423 if (type == T_VALUETYPE) {
2424 Node* cls = null_check(argument(4));
2425 if (stopped()) {
2426 return true;
2427 }
2428 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2429 const TypeKlassPtr* kls_t = _gvn.type(kls)->isa_klassptr();
2430 if (!kls_t->klass_is_exact()) {
2431 return false;
2432 }
2433 ciKlass* klass = kls_t->klass();
2434 if (!klass->is_valuetype()) {
2435 return false;
2436 }
2437 value_klass = klass->as_value_klass();
2438 }
2439
2440 receiver = null_check(receiver);
2441 if (stopped()) {
2442 return true;
2443 }
2444
2445 if (base->is_ValueType()) {
2446 ValueTypeNode* vt = base->as_ValueType();
2447
2448 if (is_store) {
2449 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) {
2450 return false;
2451 }
2452 base = vt->get_oop();
2453 } else {
2454 if (offset->is_Con()) {
2455 long off = find_long_con(offset, 0);
2456 ciValueKlass* vk = _gvn.type(vt)->is_valuetype()->value_klass();
2457 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2458 return false;
2459 }
2460
2461 ciField* f = vk->get_non_flattened_field_by_offset((int)off);
2462
2463 if (f != NULL) {
2464 BasicType bt = f->layout_type();
2465 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2466 bt = T_OBJECT;
2467 }
2468 if (bt == type) {
2469 if (bt != T_VALUETYPE || f->type() == value_klass) {
2470 set_result(vt->field_value_by_offset((int)off, false));
2471 return true;
2472 }
2473 }
2474 }
2475 }
2476 vt = vt->allocate(this)->as_ValueType();
2477 base = vt->get_oop();
2478 }
2479 }
2480
2481 // 32-bit machines ignore the high half!
2482 offset = ConvL2X(offset);
2483 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2484
2485 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2486 heap_base_oop = base;
2487 } else if (type == T_OBJECT || (value_klass != NULL && value_klass->has_object_fields())) {
2488 return false; // off-heap oop accesses are not supported
2489 }
2490
2491 // Can base be NULL? Otherwise, always on-heap access.
2492 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2493
2494 if (!can_access_non_heap) {
2495 decorators |= IN_HEAP;
2496 }
2497
2498 val = is_store ? argument(4 + (type == T_VALUETYPE ? 1 : 0)) : NULL;
2499
2500 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2501
2502 // Try to categorize the address.
2503 Compile::AliasType* alias_type = C->alias_type(adr_type);
2504 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2505
2506 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2507 alias_type->adr_type() == TypeAryPtr::RANGE) {
2508 return false; // not supported
2509 }
2510
2511 bool mismatched = false;
2512 BasicType bt = T_ILLEGAL;
2513 ciField* field = NULL;
2514 if (adr_type->isa_instptr()) {
2515 const TypeInstPtr* instptr = adr_type->is_instptr();
2516 ciInstanceKlass* k = instptr->klass()->as_instance_klass();
2517 int off = instptr->offset();
2518 if (instptr->const_oop() != NULL &&
2519 instptr->klass() == ciEnv::current()->Class_klass() &&
2520 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) {
2521 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2522 field = k->get_field_by_offset(off, true);
2523 } else {
2524 field = k->get_non_flattened_field_by_offset(off);
2525 }
2526 if (field != NULL) {
2527 bt = field->layout_type();
2528 }
2529 assert(bt == alias_type->basic_type() || bt == T_VALUETYPE, "should match");
2530 if (field != NULL && bt == T_VALUETYPE && !field->is_flattened()) {
2531 bt = T_OBJECT;
2532 }
2533 } else {
2534 bt = alias_type->basic_type();
2535 }
2536
2537 if (bt != T_ILLEGAL) {
2538 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2539 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2540 // Alias type doesn't differentiate between byte[] and boolean[]).
2541 // Use address type to get the element type.
2542 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2543 }
2544 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2545 // accessing an array field with getReference is not a mismatch
2546 bt = T_OBJECT;
2547 }
2548 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2549 // Don't intrinsify mismatched object accesses
2550 return false;
2551 }
2552 mismatched = (bt != type);
2553 } else if (alias_type->adr_type()->isa_oopptr()) {
2554 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2555 }
2556
2557 if (type == T_VALUETYPE) {
2558 if (adr_type->isa_instptr()) {
2559 if (field == NULL || field->type() != value_klass) {
2560 mismatched = true;
2561 }
2562 } else if (adr_type->isa_aryptr()) {
2563 const Type* elem = adr_type->is_aryptr()->elem();
2564 if (!elem->isa_valuetype()) {
2565 mismatched = true;
2566 } else if (elem->is_valuetype()->value_klass() != value_klass) {
2567 mismatched = true;
2568 }
2569 }
2570 if (is_store) {
2571 const Type* val_t = _gvn.type(val);
2572 if (!val_t->isa_valuetype() ||
2573 val_t->is_valuetype()->value_klass() != value_klass) {
2574 return false;
2575 }
2576 }
2577 }
2578
2579 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2580
2581 if (mismatched) {
2582 decorators |= C2_MISMATCHED;
2583 }
2584
2585 // First guess at the value type.
2586 const Type *value_type = Type::get_const_basic_type(type);
2587
2588 // Figure out the memory ordering.
2589 decorators |= mo_decorator_for_access_kind(kind);
2590
2591 if (!is_store) {
2592 if (type == T_OBJECT) {
2593 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2594 if (tjp != NULL) {
2595 value_type = tjp;
2596 }
2597 } else if (type == T_VALUETYPE) {
2598 value_type = NULL;
2599 }
2600 }
2601
2602 // Heap pointers get a null-check from the interpreter,
2603 // as a courtesy. However, this is not guaranteed by Unsafe,
2604 // and it is not possible to fully distinguish unintended nulls
2605 // from intended ones in this API.
2606
2607 if (!is_store) {
2608 Node* p = NULL;
2609 // Try to constant fold a load from a constant field
2610
2611 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2612 // final or stable field
2613 p = make_constant_from_field(field, heap_base_oop);
2614 }
2615
2616 if (p == NULL) { // Could not constant fold the load
2617 if (type == T_VALUETYPE) {
2618 if (adr_type->isa_instptr() && !mismatched) {
2619 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2620 int offset = adr_type->is_instptr()->offset();
2621 p = ValueTypeNode::make_from_flattened(this, value_klass, base, base, holder, offset, decorators);
2622 } else {
2623 p = ValueTypeNode::make_from_flattened(this, value_klass, base, adr, NULL, 0, decorators);
2624 }
2625 } else {
2626 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2627 }
2628 // Normalize the value returned by getBoolean in the following cases
2629 if (type == T_BOOLEAN &&
2630 (mismatched ||
2631 heap_base_oop == top() || // - heap_base_oop is NULL or
2632 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2633 // and the unsafe access is made to large offset
2634 // (i.e., larger than the maximum offset necessary for any
2635 // field access)
2636 ) {
2637 IdealKit ideal = IdealKit(this);
2638 #define __ ideal.
2639 IdealVariable normalized_result(ideal);
2640 __ declarations_done();
2641 __ set(normalized_result, p);
2642 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2643 __ set(normalized_result, ideal.ConI(1));
2644 ideal.end_if();
2645 final_sync(ideal);
2646 p = __ value(normalized_result);
2647 #undef __
2648 }
2649 }
2650 if (type == T_ADDRESS) {
2651 p = gvn().transform(new CastP2XNode(NULL, p));
2652 p = ConvX2UL(p);
2653 }
2654 if (field != NULL && field->is_flattenable()&& !field->is_flattened()) {
2655 // Load a non-flattened but flattenable value type from memory
2656 if (value_type->value_klass()->is_scalarizable()) {
2657 p = ValueTypeNode::make_from_oop(this, p, value_type->value_klass());
2658 } else {
2659 p = null2default(p, value_type->value_klass());
2660 }
2661 }
2662 // The load node has the control of the preceding MemBarCPUOrder. All
2663 // following nodes will have the control of the MemBarCPUOrder inserted at
2664 // the end of this method. So, pushing the load onto the stack at a later
2665 // point is fine.
2666 set_result(p);
2667 } else {
2668 if (bt == T_ADDRESS) {
2669 // Repackage the long as a pointer.
2670 val = ConvL2X(val);
2671 val = gvn().transform(new CastX2PNode(val));
2672 }
2673 if (type == T_VALUETYPE) {
2674 if (adr_type->isa_instptr() && !mismatched) {
2675 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2676 int offset = adr_type->is_instptr()->offset();
2677 val->as_ValueType()->store_flattened(this, base, base, holder, offset, decorators);
2678 } else {
2679 val->as_ValueType()->store_flattened(this, base, adr, NULL, 0, decorators);
2680 }
2681 } else {
2682 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2683 }
2684 }
2685
2686 if (argument(1)->is_ValueType() && is_store) {
2687 Node* value = ValueTypeNode::make_from_oop(this, base, _gvn.type(base)->value_klass());
2688 value = value->as_ValueType()->make_larval(this, false);
2689 replace_in_map(argument(1), value);
2690 }
2691
2692 return true;
2693 }
2694
2695 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2696 Node* receiver = argument(0);
2697 Node* value = argument(1);
2698
2699 receiver = null_check(receiver);
2700 if (stopped()) {
2701 return true;
2702 }
2703
2704 if (!value->is_ValueType()) {
2705 return false;
2706 }
2707
2708 set_result(value->as_ValueType()->make_larval(this, true));
2709
2710 return true;
2711 }
2712 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2713 Node* receiver = argument(0);
2714 Node* buffer = argument(1);
2715
2716 receiver = null_check(receiver);
2717 if (stopped()) {
2718 return true;
2719 }
2720
2721 if (!buffer->is_ValueType()) {
2722 return false;
2723 }
2724
2725 ValueTypeNode* vt = buffer->as_ValueType();
2726 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) {
2727 return false;
2728 }
2729
2730 set_result(vt->finish_larval(this));
2731
2732 return true;
2733 }
2734
2735 //----------------------------inline_unsafe_load_store----------------------------
2736 // This method serves a couple of different customers (depending on LoadStoreKind):
2737 //
2738 // LS_cmp_swap:
2739 //
2740 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2741 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2742 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2743 //
2744 // LS_cmp_swap_weak:
2745 //
2746 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2747 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2748 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2749 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2750 //
2751 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2752 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2753 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2754 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2755 //
2756 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2757 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2758 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2759 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2760 //
2761 // LS_cmp_exchange:
2762 //
2763 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2764 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2765 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2766 //
2767 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2768 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2769 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2770 //
2771 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2772 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2773 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2774 //
2775 // LS_get_add:
2776 //
2777 // int getAndAddInt( Object o, long offset, int delta)
2778 // long getAndAddLong(Object o, long offset, long delta)
2779 //
2780 // LS_get_set:
2781 //
2782 // int getAndSet(Object o, long offset, int newValue)
2783 // long getAndSet(Object o, long offset, long newValue)
2784 // Object getAndSet(Object o, long offset, Object newValue)
2785 //
2786 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2787 // This basic scheme here is the same as inline_unsafe_access, but
2788 // differs in enough details that combining them would make the code
2789 // overly confusing. (This is a true fact! I originally combined
2790 // them, but even I was confused by it!) As much code/comments as
2791 // possible are retained from inline_unsafe_access though to make
2792 // the correspondences clearer. - dl
2793
2794 if (callee()->is_static()) return false; // caller must have the capability!
2795
2796 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2797 decorators |= mo_decorator_for_access_kind(access_kind);
2798
2799 #ifndef PRODUCT
2800 BasicType rtype;
2801 {
2802 ResourceMark rm;
2803 // Check the signatures.
2804 ciSignature* sig = callee()->signature();
2805 rtype = sig->return_type()->basic_type();
2806 switch(kind) {
2807 case LS_get_add:
2808 case LS_get_set: {
2809 // Check the signatures.
2810 #ifdef ASSERT
2811 assert(rtype == type, "get and set must return the expected type");
2812 assert(sig->count() == 3, "get and set has 3 arguments");
2813 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2814 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2815 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2816 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2817 #endif // ASSERT
2818 break;
2819 }
2820 case LS_cmp_swap:
2821 case LS_cmp_swap_weak: {
2822 // Check the signatures.
2823 #ifdef ASSERT
2824 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2825 assert(sig->count() == 4, "CAS has 4 arguments");
2826 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2827 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2828 #endif // ASSERT
2829 break;
2830 }
2831 case LS_cmp_exchange: {
2832 // Check the signatures.
2833 #ifdef ASSERT
2834 assert(rtype == type, "CAS must return the expected type");
2835 assert(sig->count() == 4, "CAS has 4 arguments");
2836 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2837 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2838 #endif // ASSERT
2839 break;
2840 }
2841 default:
2842 ShouldNotReachHere();
2843 }
2844 }
2845 #endif //PRODUCT
2846
2847 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2848
2849 // Get arguments:
2850 Node* receiver = NULL;
2851 Node* base = NULL;
2852 Node* offset = NULL;
2853 Node* oldval = NULL;
2854 Node* newval = NULL;
2855 switch(kind) {
2856 case LS_cmp_swap:
2857 case LS_cmp_swap_weak:
2858 case LS_cmp_exchange: {
2859 const bool two_slot_type = type2size[type] == 2;
2860 receiver = argument(0); // type: oop
2861 base = argument(1); // type: oop
2862 offset = argument(2); // type: long
2863 oldval = argument(4); // type: oop, int, or long
2864 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2865 break;
2866 }
2867 case LS_get_add:
2868 case LS_get_set: {
2869 receiver = argument(0); // type: oop
2870 base = argument(1); // type: oop
2871 offset = argument(2); // type: long
2872 oldval = NULL;
2873 newval = argument(4); // type: oop, int, or long
2874 break;
2875 }
2876 default:
2877 ShouldNotReachHere();
2878 }
2879
2880 // Build field offset expression.
2881 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2882 // to be plain byte offsets, which are also the same as those accepted
2883 // by oopDesc::field_addr.
2884 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2885 // 32-bit machines ignore the high half of long offsets
2886 offset = ConvL2X(offset);
2887 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2888 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2889
2890 Compile::AliasType* alias_type = C->alias_type(adr_type);
2891 BasicType bt = alias_type->basic_type();
2892 if (bt != T_ILLEGAL &&
2893 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2894 // Don't intrinsify mismatched object accesses.
2895 return false;
2896 }
2897
2898 // For CAS, unlike inline_unsafe_access, there seems no point in
2899 // trying to refine types. Just use the coarse types here.
2900 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2901 const Type *value_type = Type::get_const_basic_type(type);
2902
2903 switch (kind) {
2904 case LS_get_set:
2905 case LS_cmp_exchange: {
2906 if (type == T_OBJECT) {
2907 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2908 if (tjp != NULL) {
2909 value_type = tjp;
2910 }
2911 }
2912 break;
2913 }
2914 case LS_cmp_swap:
2915 case LS_cmp_swap_weak:
2916 case LS_get_add:
2917 break;
2918 default:
2919 ShouldNotReachHere();
2920 }
2921
2922 // Null check receiver.
2923 receiver = null_check(receiver);
2924 if (stopped()) {
2925 return true;
2926 }
2927
2928 int alias_idx = C->get_alias_index(adr_type);
2929
2930 if (type == T_OBJECT || type == T_ARRAY) {
2931 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2932
2933 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2934 // could be delayed during Parse (for example, in adjust_map_after_if()).
2935 // Execute transformation here to avoid barrier generation in such case.
2936 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2937 newval = _gvn.makecon(TypePtr::NULL_PTR);
2938
2939 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2940 // Refine the value to a null constant, when it is known to be null
2941 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2942 }
2943 }
2944
2945 Node* result = NULL;
2946 switch (kind) {
2947 case LS_cmp_exchange: {
2948 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2949 oldval, newval, value_type, type, decorators);
2950 break;
2951 }
2952 case LS_cmp_swap_weak:
2953 decorators |= C2_WEAK_CMPXCHG;
2954 case LS_cmp_swap: {
2955 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2956 oldval, newval, value_type, type, decorators);
2957 break;
2958 }
2959 case LS_get_set: {
2960 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2961 newval, value_type, type, decorators);
2962 break;
2963 }
2964 case LS_get_add: {
2965 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2966 newval, value_type, type, decorators);
2967 break;
2968 }
2969 default:
2970 ShouldNotReachHere();
2971 }
2972
2973 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2974 set_result(result);
2975 return true;
2976 }
2977
2978 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2979 // Regardless of form, don't allow previous ld/st to move down,
2980 // then issue acquire, release, or volatile mem_bar.
2981 insert_mem_bar(Op_MemBarCPUOrder);
2982 switch(id) {
2983 case vmIntrinsics::_loadFence:
2984 insert_mem_bar(Op_LoadFence);
2985 return true;
2986 case vmIntrinsics::_storeFence:
2987 insert_mem_bar(Op_StoreFence);
2988 return true;
2989 case vmIntrinsics::_fullFence:
2990 insert_mem_bar(Op_MemBarVolatile);
2991 return true;
2992 default:
2993 fatal_unexpected_iid(id);
2994 return false;
2995 }
2996 }
2997
2998 bool LibraryCallKit::inline_onspinwait() {
2999 insert_mem_bar(Op_OnSpinWait);
3000 return true;
3001 }
3002
3003 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3004 if (!kls->is_Con()) {
3005 return true;
3006 }
3007 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3008 if (klsptr == NULL) {
3009 return true;
3010 }
3011 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3012 // don't need a guard for a klass that is already initialized
3013 return !ik->is_initialized();
3014 }
3015
3016 //----------------------------inline_unsafe_allocate---------------------------
3017 // public native Object Unsafe.allocateInstance(Class<?> cls);
3018 bool LibraryCallKit::inline_unsafe_allocate() {
3019 if (callee()->is_static()) return false; // caller must have the capability!
3020
3021 null_check_receiver(); // null-check, then ignore
3022 Node* cls = null_check(argument(1));
3023 if (stopped()) return true;
3024
3025 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3026 kls = null_check(kls);
3027 if (stopped()) return true; // argument was like int.class
3028
3029 Node* test = NULL;
3030 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3031 // Note: The argument might still be an illegal value like
3032 // Serializable.class or Object[].class. The runtime will handle it.
3033 // But we must make an explicit check for initialization.
3034 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3035 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3036 // can generate code to load it as unsigned byte.
3037 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3038 Node* bits = intcon(InstanceKlass::fully_initialized);
3039 test = _gvn.transform(new SubINode(inst, bits));
3040 // The 'test' is non-zero if we need to take a slow path.
3041 }
3042
3043 Node* obj = new_instance(kls, test);
3044 set_result(obj);
3045 return true;
3046 }
3047
3048 //------------------------inline_native_time_funcs--------------
3049 // inline code for System.currentTimeMillis() and System.nanoTime()
3050 // these have the same type and signature
3051 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3052 const TypeFunc* tf = OptoRuntime::void_long_Type();
3053 const TypePtr* no_memory_effects = NULL;
3054 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3055 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3056 #ifdef ASSERT
3057 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3058 assert(value_top == top(), "second value must be top");
3059 #endif
3060 set_result(value);
3061 return true;
3062 }
3063
3064 #ifdef JFR_HAVE_INTRINSICS
3065
3066 /*
3067 * oop -> myklass
3068 * myklass->trace_id |= USED
3069 * return myklass->trace_id & ~0x3
3070 */
3071 bool LibraryCallKit::inline_native_classID() {
3072 Node* cls = null_check(argument(0), T_OBJECT);
3073 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3074 kls = null_check(kls, T_OBJECT);
3075
3076 ByteSize offset = KLASS_TRACE_ID_OFFSET;
3077 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3078 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3079
3080 Node* clsused = longcon(0x01l); // set the class bit
3081 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3082 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3083 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3084
3085 #ifdef TRACE_ID_META_BITS
3086 Node* mbits = longcon(~TRACE_ID_META_BITS);
3087 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
3088 #endif
3089 #ifdef TRACE_ID_SHIFT
3090 Node* cbits = intcon(TRACE_ID_SHIFT);
3091 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
3092 #endif
3093
3094 set_result(tvalue);
3095 return true;
3096
3097 }
3098
3099 bool LibraryCallKit::inline_native_getEventWriter() {
3100 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3101
3102 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3103 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3104
3105 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3106
3107 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3108 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3109
3110 IfNode* iff_jobj_null =
3111 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3112
3113 enum { _normal_path = 1,
3114 _null_path = 2,
3115 PATH_LIMIT };
3116
3117 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3118 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3119
3120 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3121 result_rgn->init_req(_null_path, jobj_is_null);
3122 result_val->init_req(_null_path, null());
3123
3124 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3125 set_control(jobj_is_not_null);
3126 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3127 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3128 result_rgn->init_req(_normal_path, control());
3129 result_val->init_req(_normal_path, res);
3130
3131 set_result(result_rgn, result_val);
3132
3133 return true;
3134 }
3135
3136 #endif // JFR_HAVE_INTRINSICS
3137
3138 //------------------------inline_native_currentThread------------------
3139 bool LibraryCallKit::inline_native_currentThread() {
3140 Node* junk = NULL;
3141 set_result(generate_current_thread(junk));
3142 return true;
3143 }
3144
3145 //------------------------inline_native_isInterrupted------------------
3146 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3147 bool LibraryCallKit::inline_native_isInterrupted() {
3148 // Add a fast path to t.isInterrupted(clear_int):
3149 // (t == Thread.current() &&
3150 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3151 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3152 // So, in the common case that the interrupt bit is false,
3153 // we avoid making a call into the VM. Even if the interrupt bit
3154 // is true, if the clear_int argument is false, we avoid the VM call.
3155 // However, if the receiver is not currentThread, we must call the VM,
3156 // because there must be some locking done around the operation.
3157
3158 // We only go to the fast case code if we pass two guards.
3159 // Paths which do not pass are accumulated in the slow_region.
3160
3161 enum {
3162 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3163 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3164 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3165 PATH_LIMIT
3166 };
3167
3168 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3169 // out of the function.
3170 insert_mem_bar(Op_MemBarCPUOrder);
3171
3172 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3173 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3174
3175 RegionNode* slow_region = new RegionNode(1);
3176 record_for_igvn(slow_region);
3177
3178 // (a) Receiving thread must be the current thread.
3179 Node* rec_thr = argument(0);
3180 Node* tls_ptr = NULL;
3181 Node* cur_thr = generate_current_thread(tls_ptr);
3182
3183 // Resolve oops to stable for CmpP below.
3184 cur_thr = access_resolve(cur_thr, 0);
3185 rec_thr = access_resolve(rec_thr, 0);
3186
3187 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3188 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3189
3190 generate_slow_guard(bol_thr, slow_region);
3191
3192 // (b) Interrupt bit on TLS must be false.
3193 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3194 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3195 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3196
3197 // Set the control input on the field _interrupted read to prevent it floating up.
3198 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3199 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3200 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3201
3202 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3203
3204 // First fast path: if (!TLS._interrupted) return false;
3205 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3206 result_rgn->init_req(no_int_result_path, false_bit);
3207 result_val->init_req(no_int_result_path, intcon(0));
3208
3209 // drop through to next case
3210 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3211
3212 #ifndef _WINDOWS
3213 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3214 Node* clr_arg = argument(1);
3215 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3216 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3217 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3218
3219 // Second fast path: ... else if (!clear_int) return true;
3220 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3221 result_rgn->init_req(no_clear_result_path, false_arg);
3222 result_val->init_req(no_clear_result_path, intcon(1));
3223
3224 // drop through to next case
3225 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3226 #else
3227 // To return true on Windows you must read the _interrupted field
3228 // and check the event state i.e. take the slow path.
3229 #endif // _WINDOWS
3230
3231 // (d) Otherwise, go to the slow path.
3232 slow_region->add_req(control());
3233 set_control( _gvn.transform(slow_region));
3234
3235 if (stopped()) {
3236 // There is no slow path.
3237 result_rgn->init_req(slow_result_path, top());
3238 result_val->init_req(slow_result_path, top());
3239 } else {
3240 // non-virtual because it is a private non-static
3241 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3242
3243 Node* slow_val = set_results_for_java_call(slow_call);
3244 // this->control() comes from set_results_for_java_call
3245
3246 Node* fast_io = slow_call->in(TypeFunc::I_O);
3247 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3248
3249 // These two phis are pre-filled with copies of of the fast IO and Memory
3250 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3251 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3252
3253 result_rgn->init_req(slow_result_path, control());
3254 result_io ->init_req(slow_result_path, i_o());
3255 result_mem->init_req(slow_result_path, reset_memory());
3256 result_val->init_req(slow_result_path, slow_val);
3257
3258 set_all_memory(_gvn.transform(result_mem));
3259 set_i_o( _gvn.transform(result_io));
3260 }
3261
3262 C->set_has_split_ifs(true); // Has chance for split-if optimization
3263 set_result(result_rgn, result_val);
3264 return true;
3265 }
3266
3267 //-----------------------load_klass_from_mirror_common-------------------------
3268 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3269 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3270 // and branch to the given path on the region.
3271 // If never_see_null, take an uncommon trap on null, so we can optimistically
3272 // compile for the non-null case.
3273 // If the region is NULL, force never_see_null = true.
3274 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3275 bool never_see_null,
3276 RegionNode* region,
3277 int null_path,
3278 int offset) {
3279 if (region == NULL) never_see_null = true;
3280 Node* p = basic_plus_adr(mirror, offset);
3281 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3282 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3283 Node* null_ctl = top();
3284 kls = null_check_oop(kls, &null_ctl, never_see_null);
3285 if (region != NULL) {
3286 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3287 region->init_req(null_path, null_ctl);
3288 } else {
3289 assert(null_ctl == top(), "no loose ends");
3290 }
3291 return kls;
3292 }
3293
3294 //--------------------(inline_native_Class_query helpers)---------------------
3295 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3296 // Fall through if (mods & mask) == bits, take the guard otherwise.
3297 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3298 // Branch around if the given klass has the given modifier bit set.
3299 // Like generate_guard, adds a new path onto the region.
3300 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3301 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3302 Node* mask = intcon(modifier_mask);
3303 Node* bits = intcon(modifier_bits);
3304 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3305 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3306 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3307 return generate_fair_guard(bol, region);
3308 }
3309 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3310 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3311 }
3312
3313 Node* LibraryCallKit::generate_value_guard(Node* kls, RegionNode* region) {
3314 return generate_access_flags_guard(kls, JVM_ACC_VALUE, 0, region);
3315 }
3316
3317 //-------------------------inline_native_Class_query-------------------
3318 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3319 const Type* return_type = TypeInt::BOOL;
3320 Node* prim_return_value = top(); // what happens if it's a primitive class?
3321 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3322 bool expect_prim = false; // most of these guys expect to work on refs
3323
3324 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3325
3326 Node* mirror = argument(0);
3327 Node* obj = top();
3328
3329 switch (id) {
3330 case vmIntrinsics::_isInstance:
3331 // nothing is an instance of a primitive type
3332 prim_return_value = intcon(0);
3333 obj = argument(1);
3334 break;
3335 case vmIntrinsics::_getModifiers:
3336 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3337 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3338 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3339 break;
3340 case vmIntrinsics::_isInterface:
3341 prim_return_value = intcon(0);
3342 break;
3343 case vmIntrinsics::_isArray:
3344 prim_return_value = intcon(0);
3345 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3346 break;
3347 case vmIntrinsics::_isPrimitive:
3348 prim_return_value = intcon(1);
3349 expect_prim = true; // obviously
3350 break;
3351 case vmIntrinsics::_getSuperclass:
3352 prim_return_value = null();
3353 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3354 break;
3355 case vmIntrinsics::_getClassAccessFlags:
3356 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3357 return_type = TypeInt::INT; // not bool! 6297094
3358 break;
3359 default:
3360 fatal_unexpected_iid(id);
3361 break;
3362 }
3363
3364 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3365 if (mirror_con == NULL) return false; // cannot happen?
3366
3367 #ifndef PRODUCT
3368 if (C->print_intrinsics() || C->print_inlining()) {
3369 ciType* k = mirror_con->java_mirror_type();
3370 if (k) {
3371 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3372 k->print_name();
3373 tty->cr();
3374 }
3375 }
3376 #endif
3377
3378 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3379 RegionNode* region = new RegionNode(PATH_LIMIT);
3380 record_for_igvn(region);
3381 PhiNode* phi = new PhiNode(region, return_type);
3382
3383 // The mirror will never be null of Reflection.getClassAccessFlags, however
3384 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3385 // if it is. See bug 4774291.
3386
3387 // For Reflection.getClassAccessFlags(), the null check occurs in
3388 // the wrong place; see inline_unsafe_access(), above, for a similar
3389 // situation.
3390 mirror = null_check(mirror);
3391 // If mirror or obj is dead, only null-path is taken.
3392 if (stopped()) return true;
3393
3394 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3395
3396 // Now load the mirror's klass metaobject, and null-check it.
3397 // Side-effects region with the control path if the klass is null.
3398 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3399 // If kls is null, we have a primitive mirror.
3400 phi->init_req(_prim_path, prim_return_value);
3401 if (stopped()) { set_result(region, phi); return true; }
3402 bool safe_for_replace = (region->in(_prim_path) == top());
3403
3404 Node* p; // handy temp
3405 Node* null_ctl;
3406
3407 // Now that we have the non-null klass, we can perform the real query.
3408 // For constant classes, the query will constant-fold in LoadNode::Value.
3409 Node* query_value = top();
3410 switch (id) {
3411 case vmIntrinsics::_isInstance:
3412 // nothing is an instance of a primitive type
3413 query_value = gen_instanceof(obj, kls, safe_for_replace);
3414 break;
3415
3416 case vmIntrinsics::_getModifiers:
3417 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3418 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3419 break;
3420
3421 case vmIntrinsics::_isInterface:
3422 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3423 if (generate_interface_guard(kls, region) != NULL)
3424 // A guard was added. If the guard is taken, it was an interface.
3425 phi->add_req(intcon(1));
3426 // If we fall through, it's a plain class.
3427 query_value = intcon(0);
3428 break;
3429
3430 case vmIntrinsics::_isArray:
3431 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3432 if (generate_array_guard(kls, region) != NULL)
3433 // A guard was added. If the guard is taken, it was an array.
3434 phi->add_req(intcon(1));
3435 // If we fall through, it's a plain class.
3436 query_value = intcon(0);
3437 break;
3438
3439 case vmIntrinsics::_isPrimitive:
3440 query_value = intcon(0); // "normal" path produces false
3441 break;
3442
3443 case vmIntrinsics::_getSuperclass:
3444 // The rules here are somewhat unfortunate, but we can still do better
3445 // with random logic than with a JNI call.
3446 // Interfaces store null or Object as _super, but must report null.
3447 // Arrays store an intermediate super as _super, but must report Object.
3448 // Other types can report the actual _super.
3449 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3450 if (generate_interface_guard(kls, region) != NULL)
3451 // A guard was added. If the guard is taken, it was an interface.
3452 phi->add_req(null());
3453 if (generate_array_guard(kls, region) != NULL)
3454 // A guard was added. If the guard is taken, it was an array.
3455 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3456 // If we fall through, it's a plain class. Get its _super.
3457 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3458 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3459 null_ctl = top();
3460 kls = null_check_oop(kls, &null_ctl);
3461 if (null_ctl != top()) {
3462 // If the guard is taken, Object.superClass is null (both klass and mirror).
3463 region->add_req(null_ctl);
3464 phi ->add_req(null());
3465 }
3466 if (!stopped()) {
3467 query_value = load_mirror_from_klass(kls);
3468 }
3469 break;
3470
3471 case vmIntrinsics::_getClassAccessFlags:
3472 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3473 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3474 break;
3475
3476 default:
3477 fatal_unexpected_iid(id);
3478 break;
3479 }
3480
3481 // Fall-through is the normal case of a query to a real class.
3482 phi->init_req(1, query_value);
3483 region->init_req(1, control());
3484
3485 C->set_has_split_ifs(true); // Has chance for split-if optimization
3486 set_result(region, phi);
3487 return true;
3488 }
3489
3490 //-------------------------inline_Class_cast-------------------
3491 bool LibraryCallKit::inline_Class_cast() {
3492 Node* mirror = argument(0); // Class
3493 Node* obj = argument(1);
3494 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3495 if (mirror_con == NULL) {
3496 return false; // dead path (mirror->is_top()).
3497 }
3498 if (obj == NULL || obj->is_top()) {
3499 return false; // dead path
3500 }
3501
3502 ciKlass* obj_klass = NULL;
3503 if (obj->is_ValueType()) {
3504 const TypeValueType* tvt = _gvn.type(obj)->is_valuetype();
3505 obj_klass = tvt->value_klass();
3506 } else {
3507 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3508 if (tp != NULL) {
3509 obj_klass = tp->klass();
3510 }
3511 }
3512
3513 // First, see if Class.cast() can be folded statically.
3514 // java_mirror_type() returns non-null for compile-time Class constants.
3515 ciType* tm = mirror_con->java_mirror_type();
3516 if (tm != NULL && tm->is_klass() &&
3517 obj_klass != NULL) {
3518 if (!obj_klass->is_loaded()) {
3519 // Don't use intrinsic when class is not loaded.
3520 return false;
3521 } else {
3522 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass);
3523 if (static_res == Compile::SSC_always_true) {
3524 // isInstance() is true - fold the code.
3525 set_result(obj);
3526 return true;
3527 } else if (static_res == Compile::SSC_always_false) {
3528 // Don't use intrinsic, have to throw ClassCastException.
3529 // If the reference is null, the non-intrinsic bytecode will
3530 // be optimized appropriately.
3531 return false;
3532 }
3533 }
3534 }
3535
3536 // Bailout intrinsic and do normal inlining if exception path is frequent.
3537 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3538 return false;
3539 }
3540
3541 // Generate dynamic checks.
3542 // Class.cast() is java implementation of _checkcast bytecode.
3543 // Do checkcast (Parse::do_checkcast()) optimizations here.
3544
3545 mirror = null_check(mirror);
3546 // If mirror is dead, only null-path is taken.
3547 if (stopped()) {
3548 return true;
3549 }
3550
3551 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3552 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3553 RegionNode* region = new RegionNode(PATH_LIMIT);
3554 record_for_igvn(region);
3555
3556 // Now load the mirror's klass metaobject, and null-check it.
3557 // If kls is null, we have a primitive mirror and
3558 // nothing is an instance of a primitive type.
3559 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3560
3561 Node* res = top();
3562 if (!stopped()) {
3563 Node* bad_type_ctrl = top();
3564 // Do checkcast optimizations.
3565 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3566 region->init_req(_bad_type_path, bad_type_ctrl);
3567 }
3568 if (region->in(_prim_path) != top() ||
3569 region->in(_bad_type_path) != top()) {
3570 // Let Interpreter throw ClassCastException.
3571 PreserveJVMState pjvms(this);
3572 set_control(_gvn.transform(region));
3573 uncommon_trap(Deoptimization::Reason_intrinsic,
3574 Deoptimization::Action_maybe_recompile);
3575 }
3576 if (!stopped()) {
3577 set_result(res);
3578 }
3579 return true;
3580 }
3581
3582
3583 //--------------------------inline_native_subtype_check------------------------
3584 // This intrinsic takes the JNI calls out of the heart of
3585 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3586 bool LibraryCallKit::inline_native_subtype_check() {
3587 // Pull both arguments off the stack.
3588 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3589 args[0] = argument(0);
3590 args[1] = argument(1);
3591 Node* klasses[2]; // corresponding Klasses: superk, subk
3592 klasses[0] = klasses[1] = top();
3593
3594 enum {
3595 // A full decision tree on {superc is prim, subc is prim}:
3596 _prim_0_path = 1, // {P,N} => false
3597 // {P,P} & superc!=subc => false
3598 _prim_same_path, // {P,P} & superc==subc => true
3599 _prim_1_path, // {N,P} => false
3600 _ref_subtype_path, // {N,N} & subtype check wins => true
3601 _both_ref_path, // {N,N} & subtype check loses => false
3602 PATH_LIMIT
3603 };
3604
3605 RegionNode* region = new RegionNode(PATH_LIMIT);
3606 Node* phi = new PhiNode(region, TypeInt::BOOL);
3607 record_for_igvn(region);
3608
3609 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3610 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3611 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3612
3613 // First null-check both mirrors and load each mirror's klass metaobject.
3614 int which_arg;
3615 for (which_arg = 0; which_arg <= 1; which_arg++) {
3616 Node* arg = args[which_arg];
3617 arg = null_check(arg);
3618 if (stopped()) break;
3619 args[which_arg] = arg;
3620
3621 Node* p = basic_plus_adr(arg, class_klass_offset);
3622 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3623 klasses[which_arg] = _gvn.transform(kls);
3624 }
3625
3626 // Resolve oops to stable for CmpP below.
3627 args[0] = access_resolve(args[0], 0);
3628 args[1] = access_resolve(args[1], 0);
3629
3630 // Having loaded both klasses, test each for null.
3631 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3632 for (which_arg = 0; which_arg <= 1; which_arg++) {
3633 Node* kls = klasses[which_arg];
3634 Node* null_ctl = top();
3635 kls = null_check_oop(kls, &null_ctl, never_see_null);
3636 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3637 region->init_req(prim_path, null_ctl);
3638 if (stopped()) break;
3639 klasses[which_arg] = kls;
3640 }
3641
3642 if (!stopped()) {
3643 // now we have two reference types, in klasses[0..1]
3644 Node* subk = klasses[1]; // the argument to isAssignableFrom
3645 Node* superk = klasses[0]; // the receiver
3646 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3647 // now we have a successful reference subtype check
3648 region->set_req(_ref_subtype_path, control());
3649 }
3650
3651 // If both operands are primitive (both klasses null), then
3652 // we must return true when they are identical primitives.
3653 // It is convenient to test this after the first null klass check.
3654 set_control(region->in(_prim_0_path)); // go back to first null check
3655 if (!stopped()) {
3656 // Since superc is primitive, make a guard for the superc==subc case.
3657 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3658 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3659 generate_guard(bol_eq, region, PROB_FAIR);
3660 if (region->req() == PATH_LIMIT+1) {
3661 // A guard was added. If the added guard is taken, superc==subc.
3662 region->swap_edges(PATH_LIMIT, _prim_same_path);
3663 region->del_req(PATH_LIMIT);
3664 }
3665 region->set_req(_prim_0_path, control()); // Not equal after all.
3666 }
3667
3668 // these are the only paths that produce 'true':
3669 phi->set_req(_prim_same_path, intcon(1));
3670 phi->set_req(_ref_subtype_path, intcon(1));
3671
3672 // pull together the cases:
3673 assert(region->req() == PATH_LIMIT, "sane region");
3674 for (uint i = 1; i < region->req(); i++) {
3675 Node* ctl = region->in(i);
3676 if (ctl == NULL || ctl == top()) {
3677 region->set_req(i, top());
3678 phi ->set_req(i, top());
3679 } else if (phi->in(i) == NULL) {
3680 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3681 }
3682 }
3683
3684 set_control(_gvn.transform(region));
3685 set_result(_gvn.transform(phi));
3686 return true;
3687 }
3688
3689 //---------------------generate_array_guard_common------------------------
3690 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
3691
3692 if (stopped()) {
3693 return NULL;
3694 }
3695
3696 // Like generate_guard, adds a new path onto the region.
3697 jint layout_con = 0;
3698 Node* layout_val = get_layout_helper(kls, layout_con);
3699 if (layout_val == NULL) {
3700 bool query = 0;
3701 switch(kind) {
3702 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
3703 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
3704 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
3705 case ValueArray: query = Klass::layout_helper_is_valueArray(layout_con); break;
3706 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
3707 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
3708 default:
3709 ShouldNotReachHere();
3710 }
3711 if (!query) {
3712 return NULL; // never a branch
3713 } else { // always a branch
3714 Node* always_branch = control();
3715 if (region != NULL)
3716 region->add_req(always_branch);
3717 set_control(top());
3718 return always_branch;
3719 }
3720 }
3721 unsigned int value = 0;
3722 BoolTest::mask btest = BoolTest::illegal;
3723 switch(kind) {
3724 case ObjectArray:
3725 case NonObjectArray: {
3726 value = Klass::_lh_array_tag_obj_value;
3727 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3728 btest = kind == ObjectArray ? BoolTest::eq : BoolTest::ne;
3729 break;
3730 }
3731 case TypeArray: {
3732 value = Klass::_lh_array_tag_type_value;
3733 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3734 btest = BoolTest::eq;
3735 break;
3736 }
3737 case ValueArray: {
3738 value = Klass::_lh_array_tag_vt_value;
3739 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3740 btest = BoolTest::eq;
3741 break;
3742 }
3743 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
3744 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
3745 default:
3746 ShouldNotReachHere();
3747 }
3748 // Now test the correct condition.
3749 jint nval = (jint)value;
3750 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3751 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3752 return generate_fair_guard(bol, region);
3753 }
3754
3755
3756 //-----------------------inline_native_newArray--------------------------
3757 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
3758 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3759 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3760 Node* mirror;
3761 Node* count_val;
3762 if (uninitialized) {
3763 mirror = argument(1);
3764 count_val = argument(2);
3765 } else {
3766 mirror = argument(0);
3767 count_val = argument(1);
3768 }
3769
3770 mirror = null_check(mirror);
3771 // If mirror or obj is dead, only null-path is taken.
3772 if (stopped()) return true;
3773
3774 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3775 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3776 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3777 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3778 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3779
3780 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3781 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3782 result_reg, _slow_path);
3783 Node* normal_ctl = control();
3784 Node* no_array_ctl = result_reg->in(_slow_path);
3785
3786 // Generate code for the slow case. We make a call to newArray().
3787 set_control(no_array_ctl);
3788 if (!stopped()) {
3789 // Either the input type is void.class, or else the
3790 // array klass has not yet been cached. Either the
3791 // ensuing call will throw an exception, or else it
3792 // will cache the array klass for next time.
3793 PreserveJVMState pjvms(this);
3794 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3795 Node* slow_result = set_results_for_java_call(slow_call);
3796 // this->control() comes from set_results_for_java_call
3797 result_reg->set_req(_slow_path, control());
3798 result_val->set_req(_slow_path, slow_result);
3799 result_io ->set_req(_slow_path, i_o());
3800 result_mem->set_req(_slow_path, reset_memory());
3801 }
3802
3803 set_control(normal_ctl);
3804 if (!stopped()) {
3805 // Normal case: The array type has been cached in the java.lang.Class.
3806 // The following call works fine even if the array type is polymorphic.
3807 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3808 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3809 result_reg->init_req(_normal_path, control());
3810 result_val->init_req(_normal_path, obj);
3811 result_io ->init_req(_normal_path, i_o());
3812 result_mem->init_req(_normal_path, reset_memory());
3813
3814 if (uninitialized) {
3815 // Mark the allocation so that zeroing is skipped
3816 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3817 alloc->maybe_set_complete(&_gvn);
3818 }
3819 }
3820
3821 // Return the combined state.
3822 set_i_o( _gvn.transform(result_io) );
3823 set_all_memory( _gvn.transform(result_mem));
3824
3825 C->set_has_split_ifs(true); // Has chance for split-if optimization
3826 set_result(result_reg, result_val);
3827 return true;
3828 }
3829
3830 //----------------------inline_native_getLength--------------------------
3831 // public static native int java.lang.reflect.Array.getLength(Object array);
3832 bool LibraryCallKit::inline_native_getLength() {
3833 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3834
3835 Node* array = null_check(argument(0));
3836 // If array is dead, only null-path is taken.
3837 if (stopped()) return true;
3838
3839 // Deoptimize if it is a non-array.
3840 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3841
3842 if (non_array != NULL) {
3843 PreserveJVMState pjvms(this);
3844 set_control(non_array);
3845 uncommon_trap(Deoptimization::Reason_intrinsic,
3846 Deoptimization::Action_maybe_recompile);
3847 }
3848
3849 // If control is dead, only non-array-path is taken.
3850 if (stopped()) return true;
3851
3852 // The works fine even if the array type is polymorphic.
3853 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3854 Node* result = load_array_length(array);
3855
3856 C->set_has_split_ifs(true); // Has chance for split-if optimization
3857 set_result(result);
3858 return true;
3859 }
3860
3861 //------------------------inline_array_copyOf----------------------------
3862 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3863 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3864 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3865 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3866
3867 // Get the arguments.
3868 Node* original = argument(0);
3869 Node* start = is_copyOfRange? argument(1): intcon(0);
3870 Node* end = is_copyOfRange? argument(2): argument(1);
3871 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3872
3873 const TypeAryPtr* original_t = _gvn.type(original)->isa_aryptr();
3874 const TypeInstPtr* mirror_t = _gvn.type(array_type_mirror)->isa_instptr();
3875 if (EnableValhalla && ValueArrayFlatten &&
3876 (original_t == NULL || mirror_t == NULL ||
3877 (mirror_t->java_mirror_type() == NULL &&
3878 (original_t->elem()->isa_valuetype() ||
3879 (original_t->elem()->make_oopptr() != NULL &&
3880 original_t->elem()->make_oopptr()->can_be_value_type()))))) {
3881 // We need to know statically if the copy is to a flattened array
3882 // or not but can't tell.
3883 return false;
3884 }
3885
3886 Node* newcopy = NULL;
3887
3888 // Set the original stack and the reexecute bit for the interpreter to reexecute
3889 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3890 { PreserveReexecuteState preexecs(this);
3891 jvms()->set_should_reexecute(true);
3892
3893 array_type_mirror = null_check(array_type_mirror);
3894 original = null_check(original);
3895
3896 // Check if a null path was taken unconditionally.
3897 if (stopped()) return true;
3898
3899 Node* orig_length = load_array_length(original);
3900
3901 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3902 klass_node = null_check(klass_node);
3903
3904 RegionNode* bailout = new RegionNode(1);
3905 record_for_igvn(bailout);
3906
3907 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3908 // Bail out if that is so.
3909 // Value type array may have object field that would require a
3910 // write barrier. Conservatively, go to slow path.
3911 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
3912 Node* not_objArray = !bs->array_copy_requires_gc_barriers(false, T_OBJECT, false, BarrierSetC2::Parsing) ?
3913 generate_typeArray_guard(klass_node, bailout) : generate_non_objArray_guard(klass_node, bailout);
3914 if (not_objArray != NULL) {
3915 // Improve the klass node's type from the new optimistic assumption:
3916 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3917 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
3918 Node* cast = new CastPPNode(klass_node, akls);
3919 cast->init_req(0, control());
3920 klass_node = _gvn.transform(cast);
3921 }
3922
3923 Node* original_kls = load_object_klass(original);
3924 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3925 // loads/stores but it is legal only if we're sure the
3926 // Arrays.copyOf would succeed. So we need all input arguments
3927 // to the copyOf to be validated, including that the copy to the
3928 // new array won't trigger an ArrayStoreException. That subtype
3929 // check can be optimized if we know something on the type of
3930 // the input array from type speculation.
3931 if (_gvn.type(klass_node)->singleton() && !stopped()) {
3932 ciKlass* subk = _gvn.type(original_kls)->is_klassptr()->klass();
3933 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3934
3935 int test = C->static_subtype_check(superk, subk);
3936 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3937 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3938 if (t_original->speculative_type() != NULL) {
3939 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3940 original_kls = load_object_klass(original);
3941 }
3942 }
3943 }
3944
3945 if (EnableValhalla) {
3946 // Either both or neither new array klass and original array
3947 // klass must be flattened
3948 Node* flattened_klass = generate_valueArray_guard(klass_node, NULL);
3949 generate_valueArray_guard(original_kls, bailout);
3950 if (flattened_klass != NULL) {
3951 RegionNode* r = new RegionNode(2);
3952 record_for_igvn(r);
3953 r->init_req(1, control());
3954 set_control(flattened_klass);
3955 generate_valueArray_guard(original_kls, r);
3956 bailout->add_req(control());
3957 set_control(_gvn.transform(r));
3958 }
3959 }
3960
3961 // Bail out if either start or end is negative.
3962 generate_negative_guard(start, bailout, &start);
3963 generate_negative_guard(end, bailout, &end);
3964
3965 Node* length = end;
3966 if (_gvn.type(start) != TypeInt::ZERO) {
3967 length = _gvn.transform(new SubINode(end, start));
3968 }
3969
3970 // Bail out if length is negative.
3971 // Without this the new_array would throw
3972 // NegativeArraySizeException but IllegalArgumentException is what
3973 // should be thrown
3974 generate_negative_guard(length, bailout, &length);
3975
3976 if (bailout->req() > 1) {
3977 PreserveJVMState pjvms(this);
3978 set_control(_gvn.transform(bailout));
3979 uncommon_trap(Deoptimization::Reason_intrinsic,
3980 Deoptimization::Action_maybe_recompile);
3981 }
3982
3983 if (!stopped()) {
3984 // How many elements will we copy from the original?
3985 // The answer is MinI(orig_length - start, length).
3986 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3987 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3988
3989 original = access_resolve(original, ACCESS_READ);
3990
3991 // Generate a direct call to the right arraycopy function(s).
3992 // We know the copy is disjoint but we might not know if the
3993 // oop stores need checking.
3994 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3995 // This will fail a store-check if x contains any non-nulls.
3996
3997 bool validated = false;
3998 // Reason_class_check rather than Reason_intrinsic because we
3999 // want to intrinsify even if this traps.
4000 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4001 Node* not_subtype_ctrl = gen_subtype_check(original_kls,
4002 klass_node);
4003
4004 if (not_subtype_ctrl != top()) {
4005 PreserveJVMState pjvms(this);
4006 set_control(not_subtype_ctrl);
4007 uncommon_trap(Deoptimization::Reason_class_check,
4008 Deoptimization::Action_make_not_entrant);
4009 assert(stopped(), "Should be stopped");
4010 }
4011 validated = true;
4012 }
4013
4014 if (!stopped()) {
4015 newcopy = new_array(klass_node, length, 0); // no arguments to push
4016
4017 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4018 original_kls, klass_node);
4019 if (!is_copyOfRange) {
4020 ac->set_copyof(validated);
4021 } else {
4022 ac->set_copyofrange(validated);
4023 }
4024 Node* n = _gvn.transform(ac);
4025 if (n == ac) {
4026 ac->connect_outputs(this);
4027 } else {
4028 assert(validated, "shouldn't transform if all arguments not validated");
4029 set_all_memory(n);
4030 }
4031 }
4032 }
4033 } // original reexecute is set back here
4034
4035 C->set_has_split_ifs(true); // Has chance for split-if optimization
4036 if (!stopped()) {
4037 set_result(newcopy);
4038 }
4039 return true;
4040 }
4041
4042
4043 //----------------------generate_virtual_guard---------------------------
4044 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4045 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4046 RegionNode* slow_region) {
4047 ciMethod* method = callee();
4048 int vtable_index = method->vtable_index();
4049 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4050 "bad index %d", vtable_index);
4051 // Get the Method* out of the appropriate vtable entry.
4052 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4053 vtable_index*vtableEntry::size_in_bytes() +
4054 vtableEntry::method_offset_in_bytes();
4055 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4056 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4057
4058 // Compare the target method with the expected method (e.g., Object.hashCode).
4059 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4060
4061 Node* native_call = makecon(native_call_addr);
4062 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4063 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4064
4065 return generate_slow_guard(test_native, slow_region);
4066 }
4067
4068 //-----------------------generate_method_call----------------------------
4069 // Use generate_method_call to make a slow-call to the real
4070 // method if the fast path fails. An alternative would be to
4071 // use a stub like OptoRuntime::slow_arraycopy_Java.
4072 // This only works for expanding the current library call,
4073 // not another intrinsic. (E.g., don't use this for making an
4074 // arraycopy call inside of the copyOf intrinsic.)
4075 CallJavaNode*
4076 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4077 // When compiling the intrinsic method itself, do not use this technique.
4078 guarantee(callee() != C->method(), "cannot make slow-call to self");
4079
4080 ciMethod* method = callee();
4081 // ensure the JVMS we have will be correct for this call
4082 guarantee(method_id == method->intrinsic_id(), "must match");
4083
4084 const TypeFunc* tf = TypeFunc::make(method);
4085 CallJavaNode* slow_call;
4086 if (is_static) {
4087 assert(!is_virtual, "");
4088 slow_call = new CallStaticJavaNode(C, tf,
4089 SharedRuntime::get_resolve_static_call_stub(),
4090 method, bci());
4091 } else if (is_virtual) {
4092 null_check_receiver();
4093 int vtable_index = Method::invalid_vtable_index;
4094 if (UseInlineCaches) {
4095 // Suppress the vtable call
4096 } else {
4097 // hashCode and clone are not a miranda methods,
4098 // so the vtable index is fixed.
4099 // No need to use the linkResolver to get it.
4100 vtable_index = method->vtable_index();
4101 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4102 "bad index %d", vtable_index);
4103 }
4104 slow_call = new CallDynamicJavaNode(tf,
4105 SharedRuntime::get_resolve_virtual_call_stub(),
4106 method, vtable_index, bci());
4107 } else { // neither virtual nor static: opt_virtual
4108 null_check_receiver();
4109 slow_call = new CallStaticJavaNode(C, tf,
4110 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4111 method, bci());
4112 slow_call->set_optimized_virtual(true);
4113 }
4114 set_arguments_for_java_call(slow_call);
4115 set_edges_for_java_call(slow_call);
4116 return slow_call;
4117 }
4118
4119
4120 /**
4121 * Build special case code for calls to hashCode on an object. This call may
4122 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4123 * slightly different code.
4124 */
4125 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4126 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4127 assert(!(is_virtual && is_static), "either virtual, special, or static");
4128
4129 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4130
4131 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4132 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4133 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4134 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4135 Node* obj = argument(0);
4136
4137 if (obj->is_ValueType() || gvn().type(obj)->is_valuetypeptr()) {
4138 return false;
4139 }
4140
4141 if (!is_static) {
4142 // Check for hashing null object
4143 obj = null_check_receiver();
4144 if (stopped()) return true; // unconditionally null
4145 result_reg->init_req(_null_path, top());
4146 result_val->init_req(_null_path, top());
4147 } else {
4148 // Do a null check, and return zero if null.
4149 // System.identityHashCode(null) == 0
4150 Node* null_ctl = top();
4151 obj = null_check_oop(obj, &null_ctl);
4152 result_reg->init_req(_null_path, null_ctl);
4153 result_val->init_req(_null_path, _gvn.intcon(0));
4154 }
4155
4156 // Unconditionally null? Then return right away.
4157 if (stopped()) {
4158 set_control( result_reg->in(_null_path));
4159 if (!stopped())
4160 set_result(result_val->in(_null_path));
4161 return true;
4162 }
4163
4164 // We only go to the fast case code if we pass a number of guards. The
4165 // paths which do not pass are accumulated in the slow_region.
4166 RegionNode* slow_region = new RegionNode(1);
4167 record_for_igvn(slow_region);
4168
4169 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4170 assert(!obj_type->isa_valuetype() || !obj_type->is_valuetypeptr(), "no value type here");
4171 if (is_static && obj_type->can_be_value_type()) {
4172 Node* obj_klass = load_object_klass(obj);
4173 generate_value_guard(obj_klass, slow_region);
4174 }
4175
4176 // If this is a virtual call, we generate a funny guard. We pull out
4177 // the vtable entry corresponding to hashCode() from the target object.
4178 // If the target method which we are calling happens to be the native
4179 // Object hashCode() method, we pass the guard. We do not need this
4180 // guard for non-virtual calls -- the caller is known to be the native
4181 // Object hashCode().
4182 if (is_virtual) {
4183 // After null check, get the object's klass.
4184 Node* obj_klass = load_object_klass(obj);
4185 generate_virtual_guard(obj_klass, slow_region);
4186 }
4187
4188 // Get the header out of the object, use LoadMarkNode when available
4189 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4190 // The control of the load must be NULL. Otherwise, the load can move before
4191 // the null check after castPP removal.
4192 Node* no_ctrl = NULL;
4193 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4194
4195 // Test the header to see if it is unlocked.
4196 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4197 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4198 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4199 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4200 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4201
4202 generate_slow_guard(test_unlocked, slow_region);
4203
4204 // Get the hash value and check to see that it has been properly assigned.
4205 // We depend on hash_mask being at most 32 bits and avoid the use of
4206 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4207 // vm: see markOop.hpp.
4208 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4209 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4210 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4211 // This hack lets the hash bits live anywhere in the mark object now, as long
4212 // as the shift drops the relevant bits into the low 32 bits. Note that
4213 // Java spec says that HashCode is an int so there's no point in capturing
4214 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4215 hshifted_header = ConvX2I(hshifted_header);
4216 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4217
4218 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4219 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4220 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4221
4222 generate_slow_guard(test_assigned, slow_region);
4223
4224 Node* init_mem = reset_memory();
4225 // fill in the rest of the null path:
4226 result_io ->init_req(_null_path, i_o());
4227 result_mem->init_req(_null_path, init_mem);
4228
4229 result_val->init_req(_fast_path, hash_val);
4230 result_reg->init_req(_fast_path, control());
4231 result_io ->init_req(_fast_path, i_o());
4232 result_mem->init_req(_fast_path, init_mem);
4233
4234 // Generate code for the slow case. We make a call to hashCode().
4235 set_control(_gvn.transform(slow_region));
4236 if (!stopped()) {
4237 // No need for PreserveJVMState, because we're using up the present state.
4238 set_all_memory(init_mem);
4239 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4240 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4241 Node* slow_result = set_results_for_java_call(slow_call);
4242 // this->control() comes from set_results_for_java_call
4243 result_reg->init_req(_slow_path, control());
4244 result_val->init_req(_slow_path, slow_result);
4245 result_io ->set_req(_slow_path, i_o());
4246 result_mem ->set_req(_slow_path, reset_memory());
4247 }
4248
4249 // Return the combined state.
4250 set_i_o( _gvn.transform(result_io) );
4251 set_all_memory( _gvn.transform(result_mem));
4252
4253 set_result(result_reg, result_val);
4254 return true;
4255 }
4256
4257 //---------------------------inline_native_getClass----------------------------
4258 // public final native Class<?> java.lang.Object.getClass();
4259 //
4260 // Build special case code for calls to getClass on an object.
4261 bool LibraryCallKit::inline_native_getClass() {
4262 Node* obj = argument(0);
4263 if (obj->is_ValueType()) {
4264 ciKlass* vk = _gvn.type(obj)->is_valuetype()->value_klass();
4265 set_result(makecon(TypeInstPtr::make(vk->java_mirror())));
4266 return true;
4267 }
4268 obj = null_check_receiver();
4269 if (stopped()) return true;
4270 set_result(load_mirror_from_klass(load_object_klass(obj)));
4271 return true;
4272 }
4273
4274 //-----------------inline_native_Reflection_getCallerClass---------------------
4275 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4276 //
4277 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4278 //
4279 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4280 // in that it must skip particular security frames and checks for
4281 // caller sensitive methods.
4282 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4283 #ifndef PRODUCT
4284 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4285 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4286 }
4287 #endif
4288
4289 if (!jvms()->has_method()) {
4290 #ifndef PRODUCT
4291 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4292 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4293 }
4294 #endif
4295 return false;
4296 }
4297
4298 // Walk back up the JVM state to find the caller at the required
4299 // depth.
4300 JVMState* caller_jvms = jvms();
4301
4302 // Cf. JVM_GetCallerClass
4303 // NOTE: Start the loop at depth 1 because the current JVM state does
4304 // not include the Reflection.getCallerClass() frame.
4305 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4306 ciMethod* m = caller_jvms->method();
4307 switch (n) {
4308 case 0:
4309 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4310 break;
4311 case 1:
4312 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4313 if (!m->caller_sensitive()) {
4314 #ifndef PRODUCT
4315 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4316 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4317 }
4318 #endif
4319 return false; // bail-out; let JVM_GetCallerClass do the work
4320 }
4321 break;
4322 default:
4323 if (!m->is_ignored_by_security_stack_walk()) {
4324 // We have reached the desired frame; return the holder class.
4325 // Acquire method holder as java.lang.Class and push as constant.
4326 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4327 ciInstance* caller_mirror = caller_klass->java_mirror();
4328 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4329
4330 #ifndef PRODUCT
4331 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4332 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());
4333 tty->print_cr(" JVM state at this point:");
4334 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4335 ciMethod* m = jvms()->of_depth(i)->method();
4336 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4337 }
4338 }
4339 #endif
4340 return true;
4341 }
4342 break;
4343 }
4344 }
4345
4346 #ifndef PRODUCT
4347 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4348 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4349 tty->print_cr(" JVM state at this point:");
4350 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4351 ciMethod* m = jvms()->of_depth(i)->method();
4352 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4353 }
4354 }
4355 #endif
4356
4357 return false; // bail-out; let JVM_GetCallerClass do the work
4358 }
4359
4360 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4361 Node* arg = argument(0);
4362 Node* result = NULL;
4363
4364 switch (id) {
4365 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4366 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4367 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4368 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4369
4370 case vmIntrinsics::_doubleToLongBits: {
4371 // two paths (plus control) merge in a wood
4372 RegionNode *r = new RegionNode(3);
4373 Node *phi = new PhiNode(r, TypeLong::LONG);
4374
4375 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4376 // Build the boolean node
4377 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4378
4379 // Branch either way.
4380 // NaN case is less traveled, which makes all the difference.
4381 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4382 Node *opt_isnan = _gvn.transform(ifisnan);
4383 assert( opt_isnan->is_If(), "Expect an IfNode");
4384 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4385 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4386
4387 set_control(iftrue);
4388
4389 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4390 Node *slow_result = longcon(nan_bits); // return NaN
4391 phi->init_req(1, _gvn.transform( slow_result ));
4392 r->init_req(1, iftrue);
4393
4394 // Else fall through
4395 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4396 set_control(iffalse);
4397
4398 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4399 r->init_req(2, iffalse);
4400
4401 // Post merge
4402 set_control(_gvn.transform(r));
4403 record_for_igvn(r);
4404
4405 C->set_has_split_ifs(true); // Has chance for split-if optimization
4406 result = phi;
4407 assert(result->bottom_type()->isa_long(), "must be");
4408 break;
4409 }
4410
4411 case vmIntrinsics::_floatToIntBits: {
4412 // two paths (plus control) merge in a wood
4413 RegionNode *r = new RegionNode(3);
4414 Node *phi = new PhiNode(r, TypeInt::INT);
4415
4416 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4417 // Build the boolean node
4418 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4419
4420 // Branch either way.
4421 // NaN case is less traveled, which makes all the difference.
4422 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4423 Node *opt_isnan = _gvn.transform(ifisnan);
4424 assert( opt_isnan->is_If(), "Expect an IfNode");
4425 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4426 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4427
4428 set_control(iftrue);
4429
4430 static const jint nan_bits = 0x7fc00000;
4431 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4432 phi->init_req(1, _gvn.transform( slow_result ));
4433 r->init_req(1, iftrue);
4434
4435 // Else fall through
4436 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4437 set_control(iffalse);
4438
4439 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4440 r->init_req(2, iffalse);
4441
4442 // Post merge
4443 set_control(_gvn.transform(r));
4444 record_for_igvn(r);
4445
4446 C->set_has_split_ifs(true); // Has chance for split-if optimization
4447 result = phi;
4448 assert(result->bottom_type()->isa_int(), "must be");
4449 break;
4450 }
4451
4452 default:
4453 fatal_unexpected_iid(id);
4454 break;
4455 }
4456 set_result(_gvn.transform(result));
4457 return true;
4458 }
4459
4460 //----------------------inline_unsafe_copyMemory-------------------------
4461 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4462 bool LibraryCallKit::inline_unsafe_copyMemory() {
4463 if (callee()->is_static()) return false; // caller must have the capability!
4464 null_check_receiver(); // null-check receiver
4465 if (stopped()) return true;
4466
4467 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4468
4469 Node* src_ptr = argument(1); // type: oop
4470 Node* src_off = ConvL2X(argument(2)); // type: long
4471 Node* dst_ptr = argument(4); // type: oop
4472 Node* dst_off = ConvL2X(argument(5)); // type: long
4473 Node* size = ConvL2X(argument(7)); // type: long
4474
4475 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4476 "fieldOffset must be byte-scaled");
4477
4478 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4479 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4480 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4481 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4482
4483 // Conservatively insert a memory barrier on all memory slices.
4484 // Do not let writes of the copy source or destination float below the copy.
4485 insert_mem_bar(Op_MemBarCPUOrder);
4486
4487 // Call it. Note that the length argument is not scaled.
4488 make_runtime_call(RC_LEAF|RC_NO_FP,
4489 OptoRuntime::fast_arraycopy_Type(),
4490 StubRoutines::unsafe_arraycopy(),
4491 "unsafe_arraycopy",
4492 TypeRawPtr::BOTTOM,
4493 src, dst, size XTOP);
4494
4495 // Do not let reads of the copy destination float above the copy.
4496 insert_mem_bar(Op_MemBarCPUOrder);
4497
4498 return true;
4499 }
4500
4501 //------------------------clone_coping-----------------------------------
4502 // Helper function for inline_native_clone.
4503 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4504 assert(obj_size != NULL, "");
4505 Node* raw_obj = alloc_obj->in(1);
4506 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4507
4508 AllocateNode* alloc = NULL;
4509 if (ReduceBulkZeroing) {
4510 // We will be completely responsible for initializing this object -
4511 // mark Initialize node as complete.
4512 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4513 // The object was just allocated - there should be no any stores!
4514 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4515 // Mark as complete_with_arraycopy so that on AllocateNode
4516 // expansion, we know this AllocateNode is initialized by an array
4517 // copy and a StoreStore barrier exists after the array copy.
4518 alloc->initialization()->set_complete_with_arraycopy();
4519 }
4520
4521 // Copy the fastest available way.
4522 // TODO: generate fields copies for small objects instead.
4523 Node* size = _gvn.transform(obj_size);
4524
4525 // Exclude the header but include array length to copy by 8 bytes words.
4526 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4527 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4528 instanceOopDesc::base_offset_in_bytes();
4529 // base_off:
4530 // 8 - 32-bit VM
4531 // 12 - 64-bit VM, compressed klass
4532 // 16 - 64-bit VM, normal klass
4533 if (base_off % BytesPerLong != 0) {
4534 assert(UseCompressedClassPointers, "");
4535 if (is_array) {
4536 // Exclude length to copy by 8 bytes words.
4537 base_off += sizeof(int);
4538 } else {
4539 // Include klass to copy by 8 bytes words.
4540 base_off = instanceOopDesc::klass_offset_in_bytes();
4541 }
4542 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4543 }
4544 Node* src_base = basic_plus_adr(obj, base_off);
4545 Node* dst_base = basic_plus_adr(alloc_obj, base_off);
4546
4547 // Compute the length also, if needed:
4548 Node* countx = size;
4549 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4550 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong)));
4551
4552 access_clone(src_base, dst_base, countx, is_array);
4553
4554 // Do not let reads from the cloned object float above the arraycopy.
4555 if (alloc != NULL) {
4556 // Do not let stores that initialize this object be reordered with
4557 // a subsequent store that would make this object accessible by
4558 // other threads.
4559 // Record what AllocateNode this StoreStore protects so that
4560 // escape analysis can go from the MemBarStoreStoreNode to the
4561 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4562 // based on the escape status of the AllocateNode.
4563 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4564 } else {
4565 insert_mem_bar(Op_MemBarCPUOrder);
4566 }
4567 }
4568
4569 //------------------------inline_native_clone----------------------------
4570 // protected native Object java.lang.Object.clone();
4571 //
4572 // Here are the simple edge cases:
4573 // null receiver => normal trap
4574 // virtual and clone was overridden => slow path to out-of-line clone
4575 // not cloneable or finalizer => slow path to out-of-line Object.clone
4576 //
4577 // The general case has two steps, allocation and copying.
4578 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4579 //
4580 // Copying also has two cases, oop arrays and everything else.
4581 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4582 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4583 //
4584 // These steps fold up nicely if and when the cloned object's klass
4585 // can be sharply typed as an object array, a type array, or an instance.
4586 //
4587 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4588 PhiNode* result_val;
4589
4590 // Set the reexecute bit for the interpreter to reexecute
4591 // the bytecode that invokes Object.clone if deoptimization happens.
4592 { PreserveReexecuteState preexecs(this);
4593 jvms()->set_should_reexecute(true);
4594
4595 Node* obj = argument(0);
4596 if (obj->is_ValueType()) {
4597 return false;
4598 }
4599
4600 obj = null_check_receiver();
4601 if (stopped()) return true;
4602
4603 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4604
4605 // If we are going to clone an instance, we need its exact type to
4606 // know the number and types of fields to convert the clone to
4607 // loads/stores. Maybe a speculative type can help us.
4608 if (!obj_type->klass_is_exact() &&
4609 obj_type->speculative_type() != NULL &&
4610 obj_type->speculative_type()->is_instance_klass() &&
4611 !obj_type->speculative_type()->is_valuetype()) {
4612 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4613 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4614 !spec_ik->has_injected_fields()) {
4615 ciKlass* k = obj_type->klass();
4616 if (!k->is_instance_klass() ||
4617 k->as_instance_klass()->is_interface() ||
4618 k->as_instance_klass()->has_subklass()) {
4619 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4620 }
4621 }
4622 }
4623
4624 Node* obj_klass = load_object_klass(obj);
4625 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4626 const TypeOopPtr* toop = ((tklass != NULL)
4627 ? tklass->as_instance_type()
4628 : TypeInstPtr::NOTNULL);
4629
4630 // Conservatively insert a memory barrier on all memory slices.
4631 // Do not let writes into the original float below the clone.
4632 insert_mem_bar(Op_MemBarCPUOrder);
4633
4634 // paths into result_reg:
4635 enum {
4636 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4637 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4638 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4639 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4640 PATH_LIMIT
4641 };
4642 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4643 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4644 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4645 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4646 record_for_igvn(result_reg);
4647
4648 // We only go to the fast case code if we pass a number of guards.
4649 // The paths which do not pass are accumulated in the slow_region.
4650 RegionNode* slow_region = new RegionNode(1);
4651 record_for_igvn(slow_region);
4652
4653 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4654 if (array_ctl != NULL) {
4655 // It's an array.
4656 PreserveJVMState pjvms(this);
4657 set_control(array_ctl);
4658
4659 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4660 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4661 // Value type array may have object field that would require a
4662 // write barrier. Conservatively, go to slow path.
4663 generate_valueArray_guard(obj_klass, slow_region);
4664 }
4665
4666 if (!stopped()) {
4667 Node* obj_length = load_array_length(obj);
4668 Node* obj_size = NULL;
4669 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4670
4671 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4672 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4673 // If it is an oop array, it requires very special treatment,
4674 // because gc barriers are required when accessing the array.
4675 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4676 if (is_obja != NULL) {
4677 PreserveJVMState pjvms2(this);
4678 set_control(is_obja);
4679 // Generate a direct call to the right arraycopy function(s).
4680 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4681 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4682 ac->set_cloneoop();
4683 Node* n = _gvn.transform(ac);
4684 assert(n == ac, "cannot disappear");
4685 ac->connect_outputs(this);
4686
4687 result_reg->init_req(_objArray_path, control());
4688 result_val->init_req(_objArray_path, alloc_obj);
4689 result_i_o ->set_req(_objArray_path, i_o());
4690 result_mem ->set_req(_objArray_path, reset_memory());
4691 }
4692 }
4693
4694 // Otherwise, there are no barriers to worry about.
4695 // (We can dispense with card marks if we know the allocation
4696 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4697 // causes the non-eden paths to take compensating steps to
4698 // simulate a fresh allocation, so that no further
4699 // card marks are required in compiled code to initialize
4700 // the object.)
4701
4702 if (!stopped()) {
4703 copy_to_clone(obj, alloc_obj, obj_size, true);
4704
4705 // Present the results of the copy.
4706 result_reg->init_req(_array_path, control());
4707 result_val->init_req(_array_path, alloc_obj);
4708 result_i_o ->set_req(_array_path, i_o());
4709 result_mem ->set_req(_array_path, reset_memory());
4710 }
4711 }
4712 }
4713
4714 if (!stopped()) {
4715 // It's an instance (we did array above). Make the slow-path tests.
4716 // If this is a virtual call, we generate a funny guard. We grab
4717 // the vtable entry corresponding to clone() from the target object.
4718 // If the target method which we are calling happens to be the
4719 // Object clone() method, we pass the guard. We do not need this
4720 // guard for non-virtual calls; the caller is known to be the native
4721 // Object clone().
4722 if (is_virtual) {
4723 generate_virtual_guard(obj_klass, slow_region);
4724 }
4725
4726 // The object must be easily cloneable and must not have a finalizer.
4727 // Both of these conditions may be checked in a single test.
4728 // We could optimize the test further, but we don't care.
4729 generate_access_flags_guard(obj_klass,
4730 // Test both conditions:
4731 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4732 // Must be cloneable but not finalizer:
4733 JVM_ACC_IS_CLONEABLE_FAST,
4734 slow_region);
4735 }
4736
4737 if (!stopped()) {
4738 // It's an instance, and it passed the slow-path tests.
4739 PreserveJVMState pjvms(this);
4740 Node* obj_size = NULL;
4741 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4742 // is reexecuted if deoptimization occurs and there could be problems when merging
4743 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4744 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4745
4746 copy_to_clone(obj, alloc_obj, obj_size, false);
4747
4748 // Present the results of the slow call.
4749 result_reg->init_req(_instance_path, control());
4750 result_val->init_req(_instance_path, alloc_obj);
4751 result_i_o ->set_req(_instance_path, i_o());
4752 result_mem ->set_req(_instance_path, reset_memory());
4753 }
4754
4755 // Generate code for the slow case. We make a call to clone().
4756 set_control(_gvn.transform(slow_region));
4757 if (!stopped()) {
4758 PreserveJVMState pjvms(this);
4759 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4760 // We need to deoptimize on exception (see comment above)
4761 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4762 // this->control() comes from set_results_for_java_call
4763 result_reg->init_req(_slow_path, control());
4764 result_val->init_req(_slow_path, slow_result);
4765 result_i_o ->set_req(_slow_path, i_o());
4766 result_mem ->set_req(_slow_path, reset_memory());
4767 }
4768
4769 // Return the combined state.
4770 set_control( _gvn.transform(result_reg));
4771 set_i_o( _gvn.transform(result_i_o));
4772 set_all_memory( _gvn.transform(result_mem));
4773 } // original reexecute is set back here
4774
4775 set_result(_gvn.transform(result_val));
4776 return true;
4777 }
4778
4779 // If we have a tighly coupled allocation, the arraycopy may take care
4780 // of the array initialization. If one of the guards we insert between
4781 // the allocation and the arraycopy causes a deoptimization, an
4782 // unitialized array will escape the compiled method. To prevent that
4783 // we set the JVM state for uncommon traps between the allocation and
4784 // the arraycopy to the state before the allocation so, in case of
4785 // deoptimization, we'll reexecute the allocation and the
4786 // initialization.
4787 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4788 if (alloc != NULL) {
4789 ciMethod* trap_method = alloc->jvms()->method();
4790 int trap_bci = alloc->jvms()->bci();
4791
4792 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4793 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4794 // Make sure there's no store between the allocation and the
4795 // arraycopy otherwise visible side effects could be rexecuted
4796 // in case of deoptimization and cause incorrect execution.
4797 bool no_interfering_store = true;
4798 Node* mem = alloc->in(TypeFunc::Memory);
4799 if (mem->is_MergeMem()) {
4800 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4801 Node* n = mms.memory();
4802 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4803 assert(n->is_Store() || n->Opcode() == Op_ShenandoahWBMemProj, "what else?");
4804 no_interfering_store = false;
4805 break;
4806 }
4807 }
4808 } else {
4809 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4810 Node* n = mms.memory();
4811 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4812 assert(n->is_Store() || n->Opcode() == Op_ShenandoahWBMemProj, "what else?");
4813 no_interfering_store = false;
4814 break;
4815 }
4816 }
4817 }
4818
4819 if (no_interfering_store) {
4820 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4821 uint size = alloc->req();
4822 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4823 old_jvms->set_map(sfpt);
4824 for (uint i = 0; i < size; i++) {
4825 sfpt->init_req(i, alloc->in(i));
4826 }
4827 // re-push array length for deoptimization
4828 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4829 old_jvms->set_sp(old_jvms->sp()+1);
4830 old_jvms->set_monoff(old_jvms->monoff()+1);
4831 old_jvms->set_scloff(old_jvms->scloff()+1);
4832 old_jvms->set_endoff(old_jvms->endoff()+1);
4833 old_jvms->set_should_reexecute(true);
4834
4835 sfpt->set_i_o(map()->i_o());
4836 sfpt->set_memory(map()->memory());
4837 sfpt->set_control(map()->control());
4838
4839 JVMState* saved_jvms = jvms();
4840 saved_reexecute_sp = _reexecute_sp;
4841
4842 set_jvms(sfpt->jvms());
4843 _reexecute_sp = jvms()->sp();
4844
4845 return saved_jvms;
4846 }
4847 }
4848 }
4849 return NULL;
4850 }
4851
4852 // In case of a deoptimization, we restart execution at the
4853 // allocation, allocating a new array. We would leave an uninitialized
4854 // array in the heap that GCs wouldn't expect. Move the allocation
4855 // after the traps so we don't allocate the array if we
4856 // deoptimize. This is possible because tightly_coupled_allocation()
4857 // guarantees there's no observer of the allocated array at this point
4858 // and the control flow is simple enough.
4859 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4860 int saved_reexecute_sp, uint new_idx) {
4861 if (saved_jvms != NULL && !stopped()) {
4862 assert(alloc != NULL, "only with a tightly coupled allocation");
4863 // restore JVM state to the state at the arraycopy
4864 saved_jvms->map()->set_control(map()->control());
4865 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4866 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4867 // If we've improved the types of some nodes (null check) while
4868 // emitting the guards, propagate them to the current state
4869 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4870 set_jvms(saved_jvms);
4871 _reexecute_sp = saved_reexecute_sp;
4872
4873 // Remove the allocation from above the guards
4874 CallProjections* callprojs = alloc->extract_projections(true);
4875 InitializeNode* init = alloc->initialization();
4876 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4877 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4878 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4879 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4880
4881 // move the allocation here (after the guards)
4882 _gvn.hash_delete(alloc);
4883 alloc->set_req(TypeFunc::Control, control());
4884 alloc->set_req(TypeFunc::I_O, i_o());
4885 Node *mem = reset_memory();
4886 set_all_memory(mem);
4887 alloc->set_req(TypeFunc::Memory, mem);
4888 set_control(init->proj_out_or_null(TypeFunc::Control));
4889 set_i_o(callprojs->fallthrough_ioproj);
4890
4891 // Update memory as done in GraphKit::set_output_for_allocation()
4892 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4893 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4894 if (ary_type->isa_aryptr() && length_type != NULL) {
4895 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4896 }
4897 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4898 int elemidx = C->get_alias_index(telemref);
4899 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4900 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4901
4902 Node* allocx = _gvn.transform(alloc);
4903 assert(allocx == alloc, "where has the allocation gone?");
4904 assert(dest->is_CheckCastPP(), "not an allocation result?");
4905
4906 _gvn.hash_delete(dest);
4907 dest->set_req(0, control());
4908 Node* destx = _gvn.transform(dest);
4909 assert(destx == dest, "where has the allocation result gone?");
4910 }
4911 }
4912
4913
4914 //------------------------------inline_arraycopy-----------------------
4915 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4916 // Object dest, int destPos,
4917 // int length);
4918 bool LibraryCallKit::inline_arraycopy() {
4919 // Get the arguments.
4920 Node* src = argument(0); // type: oop
4921 Node* src_offset = argument(1); // type: int
4922 Node* dest = argument(2); // type: oop
4923 Node* dest_offset = argument(3); // type: int
4924 Node* length = argument(4); // type: int
4925
4926 uint new_idx = C->unique();
4927
4928 // Check for allocation before we add nodes that would confuse
4929 // tightly_coupled_allocation()
4930 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4931
4932 int saved_reexecute_sp = -1;
4933 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4934 // See arraycopy_restore_alloc_state() comment
4935 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4936 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4937 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4938 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4939
4940 // The following tests must be performed
4941 // (1) src and dest are arrays.
4942 // (2) src and dest arrays must have elements of the same BasicType
4943 // (3) src and dest must not be null.
4944 // (4) src_offset must not be negative.
4945 // (5) dest_offset must not be negative.
4946 // (6) length must not be negative.
4947 // (7) src_offset + length must not exceed length of src.
4948 // (8) dest_offset + length must not exceed length of dest.
4949 // (9) each element of an oop array must be assignable
4950
4951 // (3) src and dest must not be null.
4952 // always do this here because we need the JVM state for uncommon traps
4953 Node* null_ctl = top();
4954 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4955 assert(null_ctl->is_top(), "no null control here");
4956 dest = null_check(dest, T_ARRAY);
4957
4958 if (!can_emit_guards) {
4959 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4960 // guards but the arraycopy node could still take advantage of a
4961 // tightly allocated allocation. tightly_coupled_allocation() is
4962 // called again to make sure it takes the null check above into
4963 // account: the null check is mandatory and if it caused an
4964 // uncommon trap to be emitted then the allocation can't be
4965 // considered tightly coupled in this context.
4966 alloc = tightly_coupled_allocation(dest, NULL);
4967 }
4968
4969 bool validated = false;
4970
4971 const Type* src_type = _gvn.type(src);
4972 const Type* dest_type = _gvn.type(dest);
4973 const TypeAryPtr* top_src = src_type->isa_aryptr();
4974 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4975
4976 // Do we have the type of src?
4977 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4978 // Do we have the type of dest?
4979 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4980 // Is the type for src from speculation?
4981 bool src_spec = false;
4982 // Is the type for dest from speculation?
4983 bool dest_spec = false;
4984
4985 if ((!has_src || !has_dest) && can_emit_guards) {
4986 // We don't have sufficient type information, let's see if
4987 // speculative types can help. We need to have types for both src
4988 // and dest so that it pays off.
4989
4990 // Do we already have or could we have type information for src
4991 bool could_have_src = has_src;
4992 // Do we already have or could we have type information for dest
4993 bool could_have_dest = has_dest;
4994
4995 ciKlass* src_k = NULL;
4996 if (!has_src) {
4997 src_k = src_type->speculative_type_not_null();
4998 if (src_k != NULL && src_k->is_array_klass()) {
4999 could_have_src = true;
5000 }
5001 }
5002
5003 ciKlass* dest_k = NULL;
5004 if (!has_dest) {
5005 dest_k = dest_type->speculative_type_not_null();
5006 if (dest_k != NULL && dest_k->is_array_klass()) {
5007 could_have_dest = true;
5008 }
5009 }
5010
5011 if (could_have_src && could_have_dest) {
5012 // This is going to pay off so emit the required guards
5013 if (!has_src) {
5014 src = maybe_cast_profiled_obj(src, src_k, true);
5015 src_type = _gvn.type(src);
5016 top_src = src_type->isa_aryptr();
5017 has_src = (top_src != NULL && top_src->klass() != NULL);
5018 src_spec = true;
5019 }
5020 if (!has_dest) {
5021 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5022 dest_type = _gvn.type(dest);
5023 top_dest = dest_type->isa_aryptr();
5024 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5025 dest_spec = true;
5026 }
5027 }
5028 }
5029
5030 if (has_src && has_dest && can_emit_guards) {
5031 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5032 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5033 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5034 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5035
5036 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5037 // If both arrays are object arrays then having the exact types
5038 // for both will remove the need for a subtype check at runtime
5039 // before the call and may make it possible to pick a faster copy
5040 // routine (without a subtype check on every element)
5041 // Do we have the exact type of src?
5042 bool could_have_src = src_spec;
5043 // Do we have the exact type of dest?
5044 bool could_have_dest = dest_spec;
5045 ciKlass* src_k = top_src->klass();
5046 ciKlass* dest_k = top_dest->klass();
5047 if (!src_spec) {
5048 src_k = src_type->speculative_type_not_null();
5049 if (src_k != NULL && src_k->is_array_klass()) {
5050 could_have_src = true;
5051 }
5052 }
5053 if (!dest_spec) {
5054 dest_k = dest_type->speculative_type_not_null();
5055 if (dest_k != NULL && dest_k->is_array_klass()) {
5056 could_have_dest = true;
5057 }
5058 }
5059 if (could_have_src && could_have_dest) {
5060 // If we can have both exact types, emit the missing guards
5061 if (could_have_src && !src_spec) {
5062 src = maybe_cast_profiled_obj(src, src_k, true);
5063 }
5064 if (could_have_dest && !dest_spec) {
5065 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5066 }
5067 }
5068 }
5069 }
5070
5071 ciMethod* trap_method = method();
5072 int trap_bci = bci();
5073 if (saved_jvms != NULL) {
5074 trap_method = alloc->jvms()->method();
5075 trap_bci = alloc->jvms()->bci();
5076 }
5077
5078 bool negative_length_guard_generated = false;
5079
5080 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5081 can_emit_guards &&
5082 !src->is_top() && !dest->is_top()) {
5083 // validate arguments: enables transformation the ArrayCopyNode
5084 validated = true;
5085
5086 RegionNode* slow_region = new RegionNode(1);
5087 record_for_igvn(slow_region);
5088
5089 // (1) src and dest are arrays.
5090 generate_non_array_guard(load_object_klass(src), slow_region);
5091 generate_non_array_guard(load_object_klass(dest), slow_region);
5092
5093 // (2) src and dest arrays must have elements of the same BasicType
5094 // done at macro expansion or at Ideal transformation time
5095
5096 // (4) src_offset must not be negative.
5097 generate_negative_guard(src_offset, slow_region);
5098
5099 // (5) dest_offset must not be negative.
5100 generate_negative_guard(dest_offset, slow_region);
5101
5102 // (7) src_offset + length must not exceed length of src.
5103 generate_limit_guard(src_offset, length,
5104 load_array_length(src),
5105 slow_region);
5106
5107 // (8) dest_offset + length must not exceed length of dest.
5108 generate_limit_guard(dest_offset, length,
5109 load_array_length(dest),
5110 slow_region);
5111
5112 // (6) length must not be negative.
5113 // This is also checked in generate_arraycopy() during macro expansion, but
5114 // we also have to check it here for the case where the ArrayCopyNode will
5115 // be eliminated by Escape Analysis.
5116 if (EliminateAllocations) {
5117 generate_negative_guard(length, slow_region);
5118 negative_length_guard_generated = true;
5119 }
5120
5121 // (9) each element of an oop array must be assignable
5122 Node* src_klass = load_object_klass(src);
5123 Node* dest_klass = load_object_klass(dest);
5124 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5125
5126 if (not_subtype_ctrl != top()) {
5127 PreserveJVMState pjvms(this);
5128 set_control(not_subtype_ctrl);
5129 uncommon_trap(Deoptimization::Reason_intrinsic,
5130 Deoptimization::Action_make_not_entrant);
5131 assert(stopped(), "Should be stopped");
5132 }
5133
5134 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5135 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
5136 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5137
5138 src_type = _gvn.type(src);
5139 top_src = src_type->isa_aryptr();
5140
5141 if (top_dest != NULL &&
5142 top_dest->elem()->make_oopptr() != NULL &&
5143 top_dest->elem()->make_oopptr()->can_be_value_type()) {
5144 generate_valueArray_guard(load_object_klass(dest), slow_region);
5145 }
5146
5147 if (top_src != NULL &&
5148 top_src->elem()->make_oopptr() != NULL &&
5149 top_src->elem()->make_oopptr()->can_be_value_type()) {
5150 generate_valueArray_guard(load_object_klass(src), slow_region);
5151 }
5152
5153 {
5154 PreserveJVMState pjvms(this);
5155 set_control(_gvn.transform(slow_region));
5156 uncommon_trap(Deoptimization::Reason_intrinsic,
5157 Deoptimization::Action_make_not_entrant);
5158 assert(stopped(), "Should be stopped");
5159 }
5160 }
5161
5162 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5163
5164 if (stopped()) {
5165 return true;
5166 }
5167
5168 Node* new_src = access_resolve(src, ACCESS_READ);
5169 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
5170
5171 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5172 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5173 // so the compiler has a chance to eliminate them: during macro expansion,
5174 // we have to set their control (CastPP nodes are eliminated).
5175 load_object_klass(src), load_object_klass(dest),
5176 load_array_length(src), load_array_length(dest));
5177
5178 ac->set_arraycopy(validated);
5179
5180 Node* n = _gvn.transform(ac);
5181 if (n == ac) {
5182 ac->connect_outputs(this);
5183 } else {
5184 assert(validated, "shouldn't transform if all arguments not validated");
5185 set_all_memory(n);
5186 }
5187 clear_upper_avx();
5188
5189
5190 return true;
5191 }
5192
5193
5194 // Helper function which determines if an arraycopy immediately follows
5195 // an allocation, with no intervening tests or other escapes for the object.
5196 AllocateArrayNode*
5197 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5198 RegionNode* slow_region) {
5199 if (stopped()) return NULL; // no fast path
5200 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5201
5202 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5203 if (alloc == NULL) return NULL;
5204
5205 Node* rawmem = memory(Compile::AliasIdxRaw);
5206 // Is the allocation's memory state untouched?
5207 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5208 // Bail out if there have been raw-memory effects since the allocation.
5209 // (Example: There might have been a call or safepoint.)
5210 return NULL;
5211 }
5212 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5213 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5214 return NULL;
5215 }
5216
5217 // There must be no unexpected observers of this allocation.
5218 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5219 Node* obs = ptr->fast_out(i);
5220 if (obs != this->map()) {
5221 return NULL;
5222 }
5223 }
5224
5225 // This arraycopy must unconditionally follow the allocation of the ptr.
5226 Node* alloc_ctl = ptr->in(0);
5227 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5228
5229 Node* ctl = control();
5230 while (ctl != alloc_ctl) {
5231 // There may be guards which feed into the slow_region.
5232 // Any other control flow means that we might not get a chance
5233 // to finish initializing the allocated object.
5234 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5235 IfNode* iff = ctl->in(0)->as_If();
5236 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5237 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5238 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5239 ctl = iff->in(0); // This test feeds the known slow_region.
5240 continue;
5241 }
5242 // One more try: Various low-level checks bottom out in
5243 // uncommon traps. If the debug-info of the trap omits
5244 // any reference to the allocation, as we've already
5245 // observed, then there can be no objection to the trap.
5246 bool found_trap = false;
5247 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5248 Node* obs = not_ctl->fast_out(j);
5249 if (obs->in(0) == not_ctl && obs->is_Call() &&
5250 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5251 found_trap = true; break;
5252 }
5253 }
5254 if (found_trap) {
5255 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5256 continue;
5257 }
5258 }
5259 return NULL;
5260 }
5261
5262 // If we get this far, we have an allocation which immediately
5263 // precedes the arraycopy, and we can take over zeroing the new object.
5264 // The arraycopy will finish the initialization, and provide
5265 // a new control state to which we will anchor the destination pointer.
5266
5267 return alloc;
5268 }
5269
5270 //-------------inline_encodeISOArray-----------------------------------
5271 // encode char[] to byte[] in ISO_8859_1
5272 bool LibraryCallKit::inline_encodeISOArray() {
5273 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5274 // no receiver since it is static method
5275 Node *src = argument(0);
5276 Node *src_offset = argument(1);
5277 Node *dst = argument(2);
5278 Node *dst_offset = argument(3);
5279 Node *length = argument(4);
5280
5281 src = must_be_not_null(src, true);
5282 dst = must_be_not_null(dst, true);
5283
5284 src = access_resolve(src, ACCESS_READ);
5285 dst = access_resolve(dst, ACCESS_WRITE);
5286
5287 const Type* src_type = src->Value(&_gvn);
5288 const Type* dst_type = dst->Value(&_gvn);
5289 const TypeAryPtr* top_src = src_type->isa_aryptr();
5290 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5291 if (top_src == NULL || top_src->klass() == NULL ||
5292 top_dest == NULL || top_dest->klass() == NULL) {
5293 // failed array check
5294 return false;
5295 }
5296
5297 // Figure out the size and type of the elements we will be copying.
5298 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5299 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5300 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5301 return false;
5302 }
5303
5304 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5305 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5306 // 'src_start' points to src array + scaled offset
5307 // 'dst_start' points to dst array + scaled offset
5308
5309 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5310 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5311 enc = _gvn.transform(enc);
5312 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5313 set_memory(res_mem, mtype);
5314 set_result(enc);
5315 clear_upper_avx();
5316
5317 return true;
5318 }
5319
5320 //-------------inline_multiplyToLen-----------------------------------
5321 bool LibraryCallKit::inline_multiplyToLen() {
5322 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5323
5324 address stubAddr = StubRoutines::multiplyToLen();
5325 if (stubAddr == NULL) {
5326 return false; // Intrinsic's stub is not implemented on this platform
5327 }
5328 const char* stubName = "multiplyToLen";
5329
5330 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5331
5332 // no receiver because it is a static method
5333 Node* x = argument(0);
5334 Node* xlen = argument(1);
5335 Node* y = argument(2);
5336 Node* ylen = argument(3);
5337 Node* z = argument(4);
5338
5339 x = must_be_not_null(x, true);
5340 y = must_be_not_null(y, true);
5341
5342 x = access_resolve(x, ACCESS_READ);
5343 y = access_resolve(y, ACCESS_READ);
5344 z = access_resolve(z, ACCESS_WRITE);
5345
5346 const Type* x_type = x->Value(&_gvn);
5347 const Type* y_type = y->Value(&_gvn);
5348 const TypeAryPtr* top_x = x_type->isa_aryptr();
5349 const TypeAryPtr* top_y = y_type->isa_aryptr();
5350 if (top_x == NULL || top_x->klass() == NULL ||
5351 top_y == NULL || top_y->klass() == NULL) {
5352 // failed array check
5353 return false;
5354 }
5355
5356 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5357 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5358 if (x_elem != T_INT || y_elem != T_INT) {
5359 return false;
5360 }
5361
5362 // Set the original stack and the reexecute bit for the interpreter to reexecute
5363 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5364 // on the return from z array allocation in runtime.
5365 { PreserveReexecuteState preexecs(this);
5366 jvms()->set_should_reexecute(true);
5367
5368 Node* x_start = array_element_address(x, intcon(0), x_elem);
5369 Node* y_start = array_element_address(y, intcon(0), y_elem);
5370 // 'x_start' points to x array + scaled xlen
5371 // 'y_start' points to y array + scaled ylen
5372
5373 // Allocate the result array
5374 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5375 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5376 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5377
5378 IdealKit ideal(this);
5379
5380 #define __ ideal.
5381 Node* one = __ ConI(1);
5382 Node* zero = __ ConI(0);
5383 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5384 __ set(need_alloc, zero);
5385 __ set(z_alloc, z);
5386 __ if_then(z, BoolTest::eq, null()); {
5387 __ increment (need_alloc, one);
5388 } __ else_(); {
5389 // Update graphKit memory and control from IdealKit.
5390 sync_kit(ideal);
5391 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5392 cast->init_req(0, control());
5393 _gvn.set_type(cast, cast->bottom_type());
5394 C->record_for_igvn(cast);
5395
5396 Node* zlen_arg = load_array_length(cast);
5397 // Update IdealKit memory and control from graphKit.
5398 __ sync_kit(this);
5399 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5400 __ increment (need_alloc, one);
5401 } __ end_if();
5402 } __ end_if();
5403
5404 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5405 // Update graphKit memory and control from IdealKit.
5406 sync_kit(ideal);
5407 Node * narr = new_array(klass_node, zlen, 1);
5408 // Update IdealKit memory and control from graphKit.
5409 __ sync_kit(this);
5410 __ set(z_alloc, narr);
5411 } __ end_if();
5412
5413 sync_kit(ideal);
5414 z = __ value(z_alloc);
5415 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5416 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5417 // Final sync IdealKit and GraphKit.
5418 final_sync(ideal);
5419 #undef __
5420
5421 Node* z_start = array_element_address(z, intcon(0), T_INT);
5422
5423 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5424 OptoRuntime::multiplyToLen_Type(),
5425 stubAddr, stubName, TypePtr::BOTTOM,
5426 x_start, xlen, y_start, ylen, z_start, zlen);
5427 } // original reexecute is set back here
5428
5429 C->set_has_split_ifs(true); // Has chance for split-if optimization
5430 set_result(z);
5431 return true;
5432 }
5433
5434 //-------------inline_squareToLen------------------------------------
5435 bool LibraryCallKit::inline_squareToLen() {
5436 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5437
5438 address stubAddr = StubRoutines::squareToLen();
5439 if (stubAddr == NULL) {
5440 return false; // Intrinsic's stub is not implemented on this platform
5441 }
5442 const char* stubName = "squareToLen";
5443
5444 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5445
5446 Node* x = argument(0);
5447 Node* len = argument(1);
5448 Node* z = argument(2);
5449 Node* zlen = argument(3);
5450
5451 x = must_be_not_null(x, true);
5452 z = must_be_not_null(z, true);
5453
5454 x = access_resolve(x, ACCESS_READ);
5455 z = access_resolve(z, ACCESS_WRITE);
5456
5457 const Type* x_type = x->Value(&_gvn);
5458 const Type* z_type = z->Value(&_gvn);
5459 const TypeAryPtr* top_x = x_type->isa_aryptr();
5460 const TypeAryPtr* top_z = z_type->isa_aryptr();
5461 if (top_x == NULL || top_x->klass() == NULL ||
5462 top_z == NULL || top_z->klass() == NULL) {
5463 // failed array check
5464 return false;
5465 }
5466
5467 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5468 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5469 if (x_elem != T_INT || z_elem != T_INT) {
5470 return false;
5471 }
5472
5473
5474 Node* x_start = array_element_address(x, intcon(0), x_elem);
5475 Node* z_start = array_element_address(z, intcon(0), z_elem);
5476
5477 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5478 OptoRuntime::squareToLen_Type(),
5479 stubAddr, stubName, TypePtr::BOTTOM,
5480 x_start, len, z_start, zlen);
5481
5482 set_result(z);
5483 return true;
5484 }
5485
5486 //-------------inline_mulAdd------------------------------------------
5487 bool LibraryCallKit::inline_mulAdd() {
5488 assert(UseMulAddIntrinsic, "not implemented on this platform");
5489
5490 address stubAddr = StubRoutines::mulAdd();
5491 if (stubAddr == NULL) {
5492 return false; // Intrinsic's stub is not implemented on this platform
5493 }
5494 const char* stubName = "mulAdd";
5495
5496 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5497
5498 Node* out = argument(0);
5499 Node* in = argument(1);
5500 Node* offset = argument(2);
5501 Node* len = argument(3);
5502 Node* k = argument(4);
5503
5504 out = must_be_not_null(out, true);
5505
5506 in = access_resolve(in, ACCESS_READ);
5507 out = access_resolve(out, ACCESS_WRITE);
5508
5509 const Type* out_type = out->Value(&_gvn);
5510 const Type* in_type = in->Value(&_gvn);
5511 const TypeAryPtr* top_out = out_type->isa_aryptr();
5512 const TypeAryPtr* top_in = in_type->isa_aryptr();
5513 if (top_out == NULL || top_out->klass() == NULL ||
5514 top_in == NULL || top_in->klass() == NULL) {
5515 // failed array check
5516 return false;
5517 }
5518
5519 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5520 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5521 if (out_elem != T_INT || in_elem != T_INT) {
5522 return false;
5523 }
5524
5525 Node* outlen = load_array_length(out);
5526 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5527 Node* out_start = array_element_address(out, intcon(0), out_elem);
5528 Node* in_start = array_element_address(in, intcon(0), in_elem);
5529
5530 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5531 OptoRuntime::mulAdd_Type(),
5532 stubAddr, stubName, TypePtr::BOTTOM,
5533 out_start,in_start, new_offset, len, k);
5534 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5535 set_result(result);
5536 return true;
5537 }
5538
5539 //-------------inline_montgomeryMultiply-----------------------------------
5540 bool LibraryCallKit::inline_montgomeryMultiply() {
5541 address stubAddr = StubRoutines::montgomeryMultiply();
5542 if (stubAddr == NULL) {
5543 return false; // Intrinsic's stub is not implemented on this platform
5544 }
5545
5546 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5547 const char* stubName = "montgomery_multiply";
5548
5549 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5550
5551 Node* a = argument(0);
5552 Node* b = argument(1);
5553 Node* n = argument(2);
5554 Node* len = argument(3);
5555 Node* inv = argument(4);
5556 Node* m = argument(6);
5557
5558 a = access_resolve(a, ACCESS_READ);
5559 b = access_resolve(b, ACCESS_READ);
5560 n = access_resolve(n, ACCESS_READ);
5561 m = access_resolve(m, ACCESS_WRITE);
5562
5563 const Type* a_type = a->Value(&_gvn);
5564 const TypeAryPtr* top_a = a_type->isa_aryptr();
5565 const Type* b_type = b->Value(&_gvn);
5566 const TypeAryPtr* top_b = b_type->isa_aryptr();
5567 const Type* n_type = a->Value(&_gvn);
5568 const TypeAryPtr* top_n = n_type->isa_aryptr();
5569 const Type* m_type = a->Value(&_gvn);
5570 const TypeAryPtr* top_m = m_type->isa_aryptr();
5571 if (top_a == NULL || top_a->klass() == NULL ||
5572 top_b == NULL || top_b->klass() == NULL ||
5573 top_n == NULL || top_n->klass() == NULL ||
5574 top_m == NULL || top_m->klass() == NULL) {
5575 // failed array check
5576 return false;
5577 }
5578
5579 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5580 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5581 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5582 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5583 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5584 return false;
5585 }
5586
5587 // Make the call
5588 {
5589 Node* a_start = array_element_address(a, intcon(0), a_elem);
5590 Node* b_start = array_element_address(b, intcon(0), b_elem);
5591 Node* n_start = array_element_address(n, intcon(0), n_elem);
5592 Node* m_start = array_element_address(m, intcon(0), m_elem);
5593
5594 Node* call = make_runtime_call(RC_LEAF,
5595 OptoRuntime::montgomeryMultiply_Type(),
5596 stubAddr, stubName, TypePtr::BOTTOM,
5597 a_start, b_start, n_start, len, inv, top(),
5598 m_start);
5599 set_result(m);
5600 }
5601
5602 return true;
5603 }
5604
5605 bool LibraryCallKit::inline_montgomerySquare() {
5606 address stubAddr = StubRoutines::montgomerySquare();
5607 if (stubAddr == NULL) {
5608 return false; // Intrinsic's stub is not implemented on this platform
5609 }
5610
5611 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5612 const char* stubName = "montgomery_square";
5613
5614 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5615
5616 Node* a = argument(0);
5617 Node* n = argument(1);
5618 Node* len = argument(2);
5619 Node* inv = argument(3);
5620 Node* m = argument(5);
5621
5622 a = access_resolve(a, ACCESS_READ);
5623 n = access_resolve(n, ACCESS_READ);
5624 m = access_resolve(m, ACCESS_WRITE);
5625
5626 const Type* a_type = a->Value(&_gvn);
5627 const TypeAryPtr* top_a = a_type->isa_aryptr();
5628 const Type* n_type = a->Value(&_gvn);
5629 const TypeAryPtr* top_n = n_type->isa_aryptr();
5630 const Type* m_type = a->Value(&_gvn);
5631 const TypeAryPtr* top_m = m_type->isa_aryptr();
5632 if (top_a == NULL || top_a->klass() == NULL ||
5633 top_n == NULL || top_n->klass() == NULL ||
5634 top_m == NULL || top_m->klass() == NULL) {
5635 // failed array check
5636 return false;
5637 }
5638
5639 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5640 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5641 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5642 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5643 return false;
5644 }
5645
5646 // Make the call
5647 {
5648 Node* a_start = array_element_address(a, intcon(0), a_elem);
5649 Node* n_start = array_element_address(n, intcon(0), n_elem);
5650 Node* m_start = array_element_address(m, intcon(0), m_elem);
5651
5652 Node* call = make_runtime_call(RC_LEAF,
5653 OptoRuntime::montgomerySquare_Type(),
5654 stubAddr, stubName, TypePtr::BOTTOM,
5655 a_start, n_start, len, inv, top(),
5656 m_start);
5657 set_result(m);
5658 }
5659
5660 return true;
5661 }
5662
5663 //-------------inline_vectorizedMismatch------------------------------
5664 bool LibraryCallKit::inline_vectorizedMismatch() {
5665 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5666
5667 address stubAddr = StubRoutines::vectorizedMismatch();
5668 if (stubAddr == NULL) {
5669 return false; // Intrinsic's stub is not implemented on this platform
5670 }
5671 const char* stubName = "vectorizedMismatch";
5672 int size_l = callee()->signature()->size();
5673 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5674
5675 Node* obja = argument(0);
5676 Node* aoffset = argument(1);
5677 Node* objb = argument(3);
5678 Node* boffset = argument(4);
5679 Node* length = argument(6);
5680 Node* scale = argument(7);
5681
5682 const Type* a_type = obja->Value(&_gvn);
5683 const Type* b_type = objb->Value(&_gvn);
5684 const TypeAryPtr* top_a = a_type->isa_aryptr();
5685 const TypeAryPtr* top_b = b_type->isa_aryptr();
5686 if (top_a == NULL || top_a->klass() == NULL ||
5687 top_b == NULL || top_b->klass() == NULL) {
5688 // failed array check
5689 return false;
5690 }
5691
5692 Node* call;
5693 jvms()->set_should_reexecute(true);
5694
5695 obja = access_resolve(obja, ACCESS_READ);
5696 objb = access_resolve(objb, ACCESS_READ);
5697 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5698 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5699
5700 call = make_runtime_call(RC_LEAF,
5701 OptoRuntime::vectorizedMismatch_Type(),
5702 stubAddr, stubName, TypePtr::BOTTOM,
5703 obja_adr, objb_adr, length, scale);
5704
5705 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5706 set_result(result);
5707 return true;
5708 }
5709
5710 /**
5711 * Calculate CRC32 for byte.
5712 * int java.util.zip.CRC32.update(int crc, int b)
5713 */
5714 bool LibraryCallKit::inline_updateCRC32() {
5715 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5716 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5717 // no receiver since it is static method
5718 Node* crc = argument(0); // type: int
5719 Node* b = argument(1); // type: int
5720
5721 /*
5722 * int c = ~ crc;
5723 * b = timesXtoThe32[(b ^ c) & 0xFF];
5724 * b = b ^ (c >>> 8);
5725 * crc = ~b;
5726 */
5727
5728 Node* M1 = intcon(-1);
5729 crc = _gvn.transform(new XorINode(crc, M1));
5730 Node* result = _gvn.transform(new XorINode(crc, b));
5731 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5732
5733 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5734 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5735 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5736 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5737
5738 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5739 result = _gvn.transform(new XorINode(crc, result));
5740 result = _gvn.transform(new XorINode(result, M1));
5741 set_result(result);
5742 return true;
5743 }
5744
5745 /**
5746 * Calculate CRC32 for byte[] array.
5747 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5748 */
5749 bool LibraryCallKit::inline_updateBytesCRC32() {
5750 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5751 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5752 // no receiver since it is static method
5753 Node* crc = argument(0); // type: int
5754 Node* src = argument(1); // type: oop
5755 Node* offset = argument(2); // type: int
5756 Node* length = argument(3); // type: int
5757
5758 const Type* src_type = src->Value(&_gvn);
5759 const TypeAryPtr* top_src = src_type->isa_aryptr();
5760 if (top_src == NULL || top_src->klass() == NULL) {
5761 // failed array check
5762 return false;
5763 }
5764
5765 // Figure out the size and type of the elements we will be copying.
5766 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5767 if (src_elem != T_BYTE) {
5768 return false;
5769 }
5770
5771 // 'src_start' points to src array + scaled offset
5772 src = must_be_not_null(src, true);
5773 src = access_resolve(src, ACCESS_READ);
5774 Node* src_start = array_element_address(src, offset, src_elem);
5775
5776 // We assume that range check is done by caller.
5777 // TODO: generate range check (offset+length < src.length) in debug VM.
5778
5779 // Call the stub.
5780 address stubAddr = StubRoutines::updateBytesCRC32();
5781 const char *stubName = "updateBytesCRC32";
5782
5783 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5784 stubAddr, stubName, TypePtr::BOTTOM,
5785 crc, src_start, length);
5786 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5787 set_result(result);
5788 return true;
5789 }
5790
5791 /**
5792 * Calculate CRC32 for ByteBuffer.
5793 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5794 */
5795 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5796 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5797 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5798 // no receiver since it is static method
5799 Node* crc = argument(0); // type: int
5800 Node* src = argument(1); // type: long
5801 Node* offset = argument(3); // type: int
5802 Node* length = argument(4); // type: int
5803
5804 src = ConvL2X(src); // adjust Java long to machine word
5805 Node* base = _gvn.transform(new CastX2PNode(src));
5806 offset = ConvI2X(offset);
5807
5808 // 'src_start' points to src array + scaled offset
5809 Node* src_start = basic_plus_adr(top(), base, offset);
5810
5811 // Call the stub.
5812 address stubAddr = StubRoutines::updateBytesCRC32();
5813 const char *stubName = "updateBytesCRC32";
5814
5815 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5816 stubAddr, stubName, TypePtr::BOTTOM,
5817 crc, src_start, length);
5818 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5819 set_result(result);
5820 return true;
5821 }
5822
5823 //------------------------------get_table_from_crc32c_class-----------------------
5824 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5825 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5826 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5827
5828 return table;
5829 }
5830
5831 //------------------------------inline_updateBytesCRC32C-----------------------
5832 //
5833 // Calculate CRC32C for byte[] array.
5834 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5835 //
5836 bool LibraryCallKit::inline_updateBytesCRC32C() {
5837 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5838 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5839 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5840 // no receiver since it is a static method
5841 Node* crc = argument(0); // type: int
5842 Node* src = argument(1); // type: oop
5843 Node* offset = argument(2); // type: int
5844 Node* end = argument(3); // type: int
5845
5846 Node* length = _gvn.transform(new SubINode(end, offset));
5847
5848 const Type* src_type = src->Value(&_gvn);
5849 const TypeAryPtr* top_src = src_type->isa_aryptr();
5850 if (top_src == NULL || top_src->klass() == NULL) {
5851 // failed array check
5852 return false;
5853 }
5854
5855 // Figure out the size and type of the elements we will be copying.
5856 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5857 if (src_elem != T_BYTE) {
5858 return false;
5859 }
5860
5861 // 'src_start' points to src array + scaled offset
5862 src = must_be_not_null(src, true);
5863 src = access_resolve(src, ACCESS_READ);
5864 Node* src_start = array_element_address(src, offset, src_elem);
5865
5866 // static final int[] byteTable in class CRC32C
5867 Node* table = get_table_from_crc32c_class(callee()->holder());
5868 table = must_be_not_null(table, true);
5869 table = access_resolve(table, ACCESS_READ);
5870 Node* table_start = array_element_address(table, intcon(0), T_INT);
5871
5872 // We assume that range check is done by caller.
5873 // TODO: generate range check (offset+length < src.length) in debug VM.
5874
5875 // Call the stub.
5876 address stubAddr = StubRoutines::updateBytesCRC32C();
5877 const char *stubName = "updateBytesCRC32C";
5878
5879 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5880 stubAddr, stubName, TypePtr::BOTTOM,
5881 crc, src_start, length, table_start);
5882 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5883 set_result(result);
5884 return true;
5885 }
5886
5887 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5888 //
5889 // Calculate CRC32C for DirectByteBuffer.
5890 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5891 //
5892 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5893 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5894 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5895 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5896 // no receiver since it is a static method
5897 Node* crc = argument(0); // type: int
5898 Node* src = argument(1); // type: long
5899 Node* offset = argument(3); // type: int
5900 Node* end = argument(4); // type: int
5901
5902 Node* length = _gvn.transform(new SubINode(end, offset));
5903
5904 src = ConvL2X(src); // adjust Java long to machine word
5905 Node* base = _gvn.transform(new CastX2PNode(src));
5906 offset = ConvI2X(offset);
5907
5908 // 'src_start' points to src array + scaled offset
5909 Node* src_start = basic_plus_adr(top(), base, offset);
5910
5911 // static final int[] byteTable in class CRC32C
5912 Node* table = get_table_from_crc32c_class(callee()->holder());
5913 table = must_be_not_null(table, true);
5914 table = access_resolve(table, ACCESS_READ);
5915 Node* table_start = array_element_address(table, intcon(0), T_INT);
5916
5917 // Call the stub.
5918 address stubAddr = StubRoutines::updateBytesCRC32C();
5919 const char *stubName = "updateBytesCRC32C";
5920
5921 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5922 stubAddr, stubName, TypePtr::BOTTOM,
5923 crc, src_start, length, table_start);
5924 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5925 set_result(result);
5926 return true;
5927 }
5928
5929 //------------------------------inline_updateBytesAdler32----------------------
5930 //
5931 // Calculate Adler32 checksum for byte[] array.
5932 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5933 //
5934 bool LibraryCallKit::inline_updateBytesAdler32() {
5935 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5936 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5937 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5938 // no receiver since it is static method
5939 Node* crc = argument(0); // type: int
5940 Node* src = argument(1); // type: oop
5941 Node* offset = argument(2); // type: int
5942 Node* length = argument(3); // type: int
5943
5944 const Type* src_type = src->Value(&_gvn);
5945 const TypeAryPtr* top_src = src_type->isa_aryptr();
5946 if (top_src == NULL || top_src->klass() == NULL) {
5947 // failed array check
5948 return false;
5949 }
5950
5951 // Figure out the size and type of the elements we will be copying.
5952 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5953 if (src_elem != T_BYTE) {
5954 return false;
5955 }
5956
5957 // 'src_start' points to src array + scaled offset
5958 src = access_resolve(src, ACCESS_READ);
5959 Node* src_start = array_element_address(src, offset, src_elem);
5960
5961 // We assume that range check is done by caller.
5962 // TODO: generate range check (offset+length < src.length) in debug VM.
5963
5964 // Call the stub.
5965 address stubAddr = StubRoutines::updateBytesAdler32();
5966 const char *stubName = "updateBytesAdler32";
5967
5968 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5969 stubAddr, stubName, TypePtr::BOTTOM,
5970 crc, src_start, length);
5971 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5972 set_result(result);
5973 return true;
5974 }
5975
5976 //------------------------------inline_updateByteBufferAdler32---------------
5977 //
5978 // Calculate Adler32 checksum for DirectByteBuffer.
5979 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5980 //
5981 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5982 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5983 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5984 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5985 // no receiver since it is static method
5986 Node* crc = argument(0); // type: int
5987 Node* src = argument(1); // type: long
5988 Node* offset = argument(3); // type: int
5989 Node* length = argument(4); // type: int
5990
5991 src = ConvL2X(src); // adjust Java long to machine word
5992 Node* base = _gvn.transform(new CastX2PNode(src));
5993 offset = ConvI2X(offset);
5994
5995 // 'src_start' points to src array + scaled offset
5996 Node* src_start = basic_plus_adr(top(), base, offset);
5997
5998 // Call the stub.
5999 address stubAddr = StubRoutines::updateBytesAdler32();
6000 const char *stubName = "updateBytesAdler32";
6001
6002 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6003 stubAddr, stubName, TypePtr::BOTTOM,
6004 crc, src_start, length);
6005
6006 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6007 set_result(result);
6008 return true;
6009 }
6010
6011 //----------------------------inline_reference_get----------------------------
6012 // public T java.lang.ref.Reference.get();
6013 bool LibraryCallKit::inline_reference_get() {
6014 const int referent_offset = java_lang_ref_Reference::referent_offset;
6015 guarantee(referent_offset > 0, "should have already been set");
6016
6017 // Get the argument:
6018 Node* reference_obj = null_check_receiver();
6019 if (stopped()) return true;
6020
6021 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
6022 assert(tinst != NULL, "obj is null");
6023 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6024 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
6025 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
6026 ciSymbol::make("Ljava/lang/Object;"),
6027 false);
6028 assert (field != NULL, "undefined field");
6029
6030 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
6031 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6032
6033 ciInstanceKlass* klass = env()->Object_klass();
6034 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
6035
6036 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6037 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
6038 // Add memory barrier to prevent commoning reads from this field
6039 // across safepoint since GC can change its value.
6040 insert_mem_bar(Op_MemBarCPUOrder);
6041
6042 set_result(result);
6043 return true;
6044 }
6045
6046
6047 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6048 bool is_exact=true, bool is_static=false,
6049 ciInstanceKlass * fromKls=NULL) {
6050 if (fromKls == NULL) {
6051 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6052 assert(tinst != NULL, "obj is null");
6053 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6054 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6055 fromKls = tinst->klass()->as_instance_klass();
6056 } else {
6057 assert(is_static, "only for static field access");
6058 }
6059 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6060 ciSymbol::make(fieldTypeString),
6061 is_static);
6062
6063 assert (field != NULL, "undefined field");
6064 if (field == NULL) return (Node *) NULL;
6065
6066 if (is_static) {
6067 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6068 fromObj = makecon(tip);
6069 }
6070
6071 // Next code copied from Parse::do_get_xxx():
6072
6073 // Compute address and memory type.
6074 int offset = field->offset_in_bytes();
6075 bool is_vol = field->is_volatile();
6076 ciType* field_klass = field->type();
6077 assert(field_klass->is_loaded(), "should be loaded");
6078 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6079 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6080 BasicType bt = field->layout_type();
6081
6082 // Build the resultant type of the load
6083 const Type *type;
6084 if (bt == T_OBJECT) {
6085 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6086 } else {
6087 type = Type::get_const_basic_type(bt);
6088 }
6089
6090 DecoratorSet decorators = IN_HEAP;
6091
6092 if (is_vol) {
6093 decorators |= MO_SEQ_CST;
6094 }
6095
6096 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6097 }
6098
6099 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6100 bool is_exact = true, bool is_static = false,
6101 ciInstanceKlass * fromKls = NULL) {
6102 if (fromKls == NULL) {
6103 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6104 assert(tinst != NULL, "obj is null");
6105 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6106 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6107 fromKls = tinst->klass()->as_instance_klass();
6108 }
6109 else {
6110 assert(is_static, "only for static field access");
6111 }
6112 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6113 ciSymbol::make(fieldTypeString),
6114 is_static);
6115
6116 assert(field != NULL, "undefined field");
6117 assert(!field->is_volatile(), "not defined for volatile fields");
6118
6119 if (is_static) {
6120 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6121 fromObj = makecon(tip);
6122 }
6123
6124 // Next code copied from Parse::do_get_xxx():
6125
6126 // Compute address and memory type.
6127 int offset = field->offset_in_bytes();
6128 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6129
6130 return adr;
6131 }
6132
6133 //------------------------------inline_aescrypt_Block-----------------------
6134 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6135 address stubAddr = NULL;
6136 const char *stubName;
6137 assert(UseAES, "need AES instruction support");
6138
6139 switch(id) {
6140 case vmIntrinsics::_aescrypt_encryptBlock:
6141 stubAddr = StubRoutines::aescrypt_encryptBlock();
6142 stubName = "aescrypt_encryptBlock";
6143 break;
6144 case vmIntrinsics::_aescrypt_decryptBlock:
6145 stubAddr = StubRoutines::aescrypt_decryptBlock();
6146 stubName = "aescrypt_decryptBlock";
6147 break;
6148 default:
6149 break;
6150 }
6151 if (stubAddr == NULL) return false;
6152
6153 Node* aescrypt_object = argument(0);
6154 Node* src = argument(1);
6155 Node* src_offset = argument(2);
6156 Node* dest = argument(3);
6157 Node* dest_offset = argument(4);
6158
6159 src = must_be_not_null(src, true);
6160 dest = must_be_not_null(dest, true);
6161
6162 src = access_resolve(src, ACCESS_READ);
6163 dest = access_resolve(dest, ACCESS_WRITE);
6164
6165 // (1) src and dest are arrays.
6166 const Type* src_type = src->Value(&_gvn);
6167 const Type* dest_type = dest->Value(&_gvn);
6168 const TypeAryPtr* top_src = src_type->isa_aryptr();
6169 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6170 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6171
6172 // for the quick and dirty code we will skip all the checks.
6173 // we are just trying to get the call to be generated.
6174 Node* src_start = src;
6175 Node* dest_start = dest;
6176 if (src_offset != NULL || dest_offset != NULL) {
6177 assert(src_offset != NULL && dest_offset != NULL, "");
6178 src_start = array_element_address(src, src_offset, T_BYTE);
6179 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6180 }
6181
6182 // now need to get the start of its expanded key array
6183 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6184 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6185 if (k_start == NULL) return false;
6186
6187 if (Matcher::pass_original_key_for_aes()) {
6188 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6189 // compatibility issues between Java key expansion and SPARC crypto instructions
6190 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6191 if (original_k_start == NULL) return false;
6192
6193 // Call the stub.
6194 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6195 stubAddr, stubName, TypePtr::BOTTOM,
6196 src_start, dest_start, k_start, original_k_start);
6197 } else {
6198 // Call the stub.
6199 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6200 stubAddr, stubName, TypePtr::BOTTOM,
6201 src_start, dest_start, k_start);
6202 }
6203
6204 return true;
6205 }
6206
6207 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6208 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6209 address stubAddr = NULL;
6210 const char *stubName = NULL;
6211
6212 assert(UseAES, "need AES instruction support");
6213
6214 switch(id) {
6215 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6216 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6217 stubName = "cipherBlockChaining_encryptAESCrypt";
6218 break;
6219 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6220 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6221 stubName = "cipherBlockChaining_decryptAESCrypt";
6222 break;
6223 default:
6224 break;
6225 }
6226 if (stubAddr == NULL) return false;
6227
6228 Node* cipherBlockChaining_object = argument(0);
6229 Node* src = argument(1);
6230 Node* src_offset = argument(2);
6231 Node* len = argument(3);
6232 Node* dest = argument(4);
6233 Node* dest_offset = argument(5);
6234
6235 src = must_be_not_null(src, false);
6236 dest = must_be_not_null(dest, false);
6237
6238 src = access_resolve(src, ACCESS_READ);
6239 dest = access_resolve(dest, ACCESS_WRITE);
6240
6241 // (1) src and dest are arrays.
6242 const Type* src_type = src->Value(&_gvn);
6243 const Type* dest_type = dest->Value(&_gvn);
6244 const TypeAryPtr* top_src = src_type->isa_aryptr();
6245 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6246 assert (top_src != NULL && top_src->klass() != NULL
6247 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6248
6249 // checks are the responsibility of the caller
6250 Node* src_start = src;
6251 Node* dest_start = dest;
6252 if (src_offset != NULL || dest_offset != NULL) {
6253 assert(src_offset != NULL && dest_offset != NULL, "");
6254 src_start = array_element_address(src, src_offset, T_BYTE);
6255 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6256 }
6257
6258 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6259 // (because of the predicated logic executed earlier).
6260 // so we cast it here safely.
6261 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6262
6263 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6264 if (embeddedCipherObj == NULL) return false;
6265
6266 // cast it to what we know it will be at runtime
6267 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6268 assert(tinst != NULL, "CBC obj is null");
6269 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6270 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6271 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6272
6273 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6274 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6275 const TypeOopPtr* xtype = aklass->as_instance_type();
6276 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6277 aescrypt_object = _gvn.transform(aescrypt_object);
6278
6279 // we need to get the start of the aescrypt_object's expanded key array
6280 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6281 if (k_start == NULL) return false;
6282
6283 // similarly, get the start address of the r vector
6284 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6285 if (objRvec == NULL) return false;
6286 objRvec = access_resolve(objRvec, ACCESS_WRITE);
6287 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6288
6289 Node* cbcCrypt;
6290 if (Matcher::pass_original_key_for_aes()) {
6291 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6292 // compatibility issues between Java key expansion and SPARC crypto instructions
6293 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6294 if (original_k_start == NULL) return false;
6295
6296 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6297 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6298 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6299 stubAddr, stubName, TypePtr::BOTTOM,
6300 src_start, dest_start, k_start, r_start, len, original_k_start);
6301 } else {
6302 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6303 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6304 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6305 stubAddr, stubName, TypePtr::BOTTOM,
6306 src_start, dest_start, k_start, r_start, len);
6307 }
6308
6309 // return cipher length (int)
6310 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6311 set_result(retvalue);
6312 return true;
6313 }
6314
6315 //------------------------------inline_counterMode_AESCrypt-----------------------
6316 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6317 assert(UseAES, "need AES instruction support");
6318 if (!UseAESCTRIntrinsics) return false;
6319
6320 address stubAddr = NULL;
6321 const char *stubName = NULL;
6322 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6323 stubAddr = StubRoutines::counterMode_AESCrypt();
6324 stubName = "counterMode_AESCrypt";
6325 }
6326 if (stubAddr == NULL) return false;
6327
6328 Node* counterMode_object = argument(0);
6329 Node* src = argument(1);
6330 Node* src_offset = argument(2);
6331 Node* len = argument(3);
6332 Node* dest = argument(4);
6333 Node* dest_offset = argument(5);
6334
6335 src = access_resolve(src, ACCESS_READ);
6336 dest = access_resolve(dest, ACCESS_WRITE);
6337 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6338
6339 // (1) src and dest are arrays.
6340 const Type* src_type = src->Value(&_gvn);
6341 const Type* dest_type = dest->Value(&_gvn);
6342 const TypeAryPtr* top_src = src_type->isa_aryptr();
6343 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6344 assert(top_src != NULL && top_src->klass() != NULL &&
6345 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6346
6347 // checks are the responsibility of the caller
6348 Node* src_start = src;
6349 Node* dest_start = dest;
6350 if (src_offset != NULL || dest_offset != NULL) {
6351 assert(src_offset != NULL && dest_offset != NULL, "");
6352 src_start = array_element_address(src, src_offset, T_BYTE);
6353 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6354 }
6355
6356 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6357 // (because of the predicated logic executed earlier).
6358 // so we cast it here safely.
6359 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6360 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6361 if (embeddedCipherObj == NULL) return false;
6362 // cast it to what we know it will be at runtime
6363 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6364 assert(tinst != NULL, "CTR obj is null");
6365 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6366 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6367 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6368 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6369 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6370 const TypeOopPtr* xtype = aklass->as_instance_type();
6371 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6372 aescrypt_object = _gvn.transform(aescrypt_object);
6373 // we need to get the start of the aescrypt_object's expanded key array
6374 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6375 if (k_start == NULL) return false;
6376 // similarly, get the start address of the r vector
6377 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6378 if (obj_counter == NULL) return false;
6379 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6380 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6381
6382 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6383 if (saved_encCounter == NULL) return false;
6384 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6385 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6386 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6387
6388 Node* ctrCrypt;
6389 if (Matcher::pass_original_key_for_aes()) {
6390 // no SPARC version for AES/CTR intrinsics now.
6391 return false;
6392 }
6393 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6394 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6395 OptoRuntime::counterMode_aescrypt_Type(),
6396 stubAddr, stubName, TypePtr::BOTTOM,
6397 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6398
6399 // return cipher length (int)
6400 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6401 set_result(retvalue);
6402 return true;
6403 }
6404
6405 //------------------------------get_key_start_from_aescrypt_object-----------------------
6406 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6407 #if defined(PPC64) || defined(S390)
6408 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6409 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6410 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6411 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6412 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6413 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6414 if (objSessionK == NULL) {
6415 return (Node *) NULL;
6416 }
6417 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6418 #else
6419 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6420 #endif // PPC64
6421 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6422 if (objAESCryptKey == NULL) return (Node *) NULL;
6423
6424 // now have the array, need to get the start address of the K array
6425 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6426 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6427 return k_start;
6428 }
6429
6430 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6431 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6432 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6433 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6434 if (objAESCryptKey == NULL) return (Node *) NULL;
6435
6436 // now have the array, need to get the start address of the lastKey array
6437 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6438 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6439 return original_k_start;
6440 }
6441
6442 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6443 // Return node representing slow path of predicate check.
6444 // the pseudo code we want to emulate with this predicate is:
6445 // for encryption:
6446 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6447 // for decryption:
6448 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6449 // note cipher==plain is more conservative than the original java code but that's OK
6450 //
6451 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6452 // The receiver was checked for NULL already.
6453 Node* objCBC = argument(0);
6454
6455 Node* src = argument(1);
6456 Node* dest = argument(4);
6457
6458 // Load embeddedCipher field of CipherBlockChaining object.
6459 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6460
6461 // get AESCrypt klass for instanceOf check
6462 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6463 // will have same classloader as CipherBlockChaining object
6464 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6465 assert(tinst != NULL, "CBCobj is null");
6466 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6467
6468 // we want to do an instanceof comparison against the AESCrypt class
6469 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6470 if (!klass_AESCrypt->is_loaded()) {
6471 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6472 Node* ctrl = control();
6473 set_control(top()); // no regular fast path
6474 return ctrl;
6475 }
6476
6477 src = must_be_not_null(src, true);
6478 dest = must_be_not_null(dest, true);
6479
6480 // Resolve oops to stable for CmpP below.
6481 src = access_resolve(src, 0);
6482 dest = access_resolve(dest, 0);
6483
6484 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6485
6486 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6487 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6488 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6489
6490 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6491
6492 // for encryption, we are done
6493 if (!decrypting)
6494 return instof_false; // even if it is NULL
6495
6496 // for decryption, we need to add a further check to avoid
6497 // taking the intrinsic path when cipher and plain are the same
6498 // see the original java code for why.
6499 RegionNode* region = new RegionNode(3);
6500 region->init_req(1, instof_false);
6501
6502 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6503 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6504 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6505 region->init_req(2, src_dest_conjoint);
6506
6507 record_for_igvn(region);
6508 return _gvn.transform(region);
6509 }
6510
6511 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6512 // Return node representing slow path of predicate check.
6513 // the pseudo code we want to emulate with this predicate is:
6514 // for encryption:
6515 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6516 // for decryption:
6517 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6518 // note cipher==plain is more conservative than the original java code but that's OK
6519 //
6520
6521 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6522 // The receiver was checked for NULL already.
6523 Node* objCTR = argument(0);
6524
6525 // Load embeddedCipher field of CipherBlockChaining object.
6526 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6527
6528 // get AESCrypt klass for instanceOf check
6529 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6530 // will have same classloader as CipherBlockChaining object
6531 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6532 assert(tinst != NULL, "CTRobj is null");
6533 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6534
6535 // we want to do an instanceof comparison against the AESCrypt class
6536 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6537 if (!klass_AESCrypt->is_loaded()) {
6538 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6539 Node* ctrl = control();
6540 set_control(top()); // no regular fast path
6541 return ctrl;
6542 }
6543
6544 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6545 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6546 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6547 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6548 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6549
6550 return instof_false; // even if it is NULL
6551 }
6552
6553 //------------------------------inline_ghash_processBlocks
6554 bool LibraryCallKit::inline_ghash_processBlocks() {
6555 address stubAddr;
6556 const char *stubName;
6557 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6558
6559 stubAddr = StubRoutines::ghash_processBlocks();
6560 stubName = "ghash_processBlocks";
6561
6562 Node* data = argument(0);
6563 Node* offset = argument(1);
6564 Node* len = argument(2);
6565 Node* state = argument(3);
6566 Node* subkeyH = argument(4);
6567
6568 state = must_be_not_null(state, true);
6569 subkeyH = must_be_not_null(subkeyH, true);
6570 data = must_be_not_null(data, true);
6571
6572 state = access_resolve(state, ACCESS_WRITE);
6573 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6574 data = access_resolve(data, ACCESS_READ);
6575
6576 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6577 assert(state_start, "state is NULL");
6578 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6579 assert(subkeyH_start, "subkeyH is NULL");
6580 Node* data_start = array_element_address(data, offset, T_BYTE);
6581 assert(data_start, "data is NULL");
6582
6583 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6584 OptoRuntime::ghash_processBlocks_Type(),
6585 stubAddr, stubName, TypePtr::BOTTOM,
6586 state_start, subkeyH_start, data_start, len);
6587 return true;
6588 }
6589
6590 bool LibraryCallKit::inline_base64_encodeBlock() {
6591 address stubAddr;
6592 const char *stubName;
6593 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6594 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6595 stubAddr = StubRoutines::base64_encodeBlock();
6596 stubName = "encodeBlock";
6597
6598 if (!stubAddr) return false;
6599 Node* base64obj = argument(0);
6600 Node* src = argument(1);
6601 Node* offset = argument(2);
6602 Node* len = argument(3);
6603 Node* dest = argument(4);
6604 Node* dp = argument(5);
6605 Node* isURL = argument(6);
6606
6607 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6608 assert(src_start, "source array is NULL");
6609 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6610 assert(dest_start, "destination array is NULL");
6611
6612 Node* base64 = make_runtime_call(RC_LEAF,
6613 OptoRuntime::base64_encodeBlock_Type(),
6614 stubAddr, stubName, TypePtr::BOTTOM,
6615 src_start, offset, len, dest_start, dp, isURL);
6616 return true;
6617 }
6618
6619 //------------------------------inline_sha_implCompress-----------------------
6620 //
6621 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6622 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6623 //
6624 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6625 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6626 //
6627 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6628 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6629 //
6630 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6631 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6632
6633 Node* sha_obj = argument(0);
6634 Node* src = argument(1); // type oop
6635 Node* ofs = argument(2); // type int
6636
6637 const Type* src_type = src->Value(&_gvn);
6638 const TypeAryPtr* top_src = src_type->isa_aryptr();
6639 if (top_src == NULL || top_src->klass() == NULL) {
6640 // failed array check
6641 return false;
6642 }
6643 // Figure out the size and type of the elements we will be copying.
6644 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6645 if (src_elem != T_BYTE) {
6646 return false;
6647 }
6648 // 'src_start' points to src array + offset
6649 src = must_be_not_null(src, true);
6650 src = access_resolve(src, ACCESS_READ);
6651 Node* src_start = array_element_address(src, ofs, src_elem);
6652 Node* state = NULL;
6653 address stubAddr;
6654 const char *stubName;
6655
6656 switch(id) {
6657 case vmIntrinsics::_sha_implCompress:
6658 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6659 state = get_state_from_sha_object(sha_obj);
6660 stubAddr = StubRoutines::sha1_implCompress();
6661 stubName = "sha1_implCompress";
6662 break;
6663 case vmIntrinsics::_sha2_implCompress:
6664 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6665 state = get_state_from_sha_object(sha_obj);
6666 stubAddr = StubRoutines::sha256_implCompress();
6667 stubName = "sha256_implCompress";
6668 break;
6669 case vmIntrinsics::_sha5_implCompress:
6670 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6671 state = get_state_from_sha5_object(sha_obj);
6672 stubAddr = StubRoutines::sha512_implCompress();
6673 stubName = "sha512_implCompress";
6674 break;
6675 default:
6676 fatal_unexpected_iid(id);
6677 return false;
6678 }
6679 if (state == NULL) return false;
6680
6681 // Call the stub.
6682 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6683 stubAddr, stubName, TypePtr::BOTTOM,
6684 src_start, state);
6685
6686 return true;
6687 }
6688
6689 //------------------------------inline_digestBase_implCompressMB-----------------------
6690 //
6691 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6692 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6693 //
6694 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6695 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6696 "need SHA1/SHA256/SHA512 instruction support");
6697 assert((uint)predicate < 3, "sanity");
6698 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6699
6700 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6701 Node* src = argument(1); // byte[] array
6702 Node* ofs = argument(2); // type int
6703 Node* limit = argument(3); // type int
6704
6705 const Type* src_type = src->Value(&_gvn);
6706 const TypeAryPtr* top_src = src_type->isa_aryptr();
6707 if (top_src == NULL || top_src->klass() == NULL) {
6708 // failed array check
6709 return false;
6710 }
6711 // Figure out the size and type of the elements we will be copying.
6712 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6713 if (src_elem != T_BYTE) {
6714 return false;
6715 }
6716 // 'src_start' points to src array + offset
6717 src = must_be_not_null(src, false);
6718 src = access_resolve(src, ACCESS_READ);
6719 Node* src_start = array_element_address(src, ofs, src_elem);
6720
6721 const char* klass_SHA_name = NULL;
6722 const char* stub_name = NULL;
6723 address stub_addr = NULL;
6724 bool long_state = false;
6725
6726 switch (predicate) {
6727 case 0:
6728 if (UseSHA1Intrinsics) {
6729 klass_SHA_name = "sun/security/provider/SHA";
6730 stub_name = "sha1_implCompressMB";
6731 stub_addr = StubRoutines::sha1_implCompressMB();
6732 }
6733 break;
6734 case 1:
6735 if (UseSHA256Intrinsics) {
6736 klass_SHA_name = "sun/security/provider/SHA2";
6737 stub_name = "sha256_implCompressMB";
6738 stub_addr = StubRoutines::sha256_implCompressMB();
6739 }
6740 break;
6741 case 2:
6742 if (UseSHA512Intrinsics) {
6743 klass_SHA_name = "sun/security/provider/SHA5";
6744 stub_name = "sha512_implCompressMB";
6745 stub_addr = StubRoutines::sha512_implCompressMB();
6746 long_state = true;
6747 }
6748 break;
6749 default:
6750 fatal("unknown SHA intrinsic predicate: %d", predicate);
6751 }
6752 if (klass_SHA_name != NULL) {
6753 // get DigestBase klass to lookup for SHA klass
6754 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6755 assert(tinst != NULL, "digestBase_obj is not instance???");
6756 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6757
6758 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6759 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6760 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6761 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6762 }
6763 return false;
6764 }
6765 //------------------------------inline_sha_implCompressMB-----------------------
6766 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6767 bool long_state, address stubAddr, const char *stubName,
6768 Node* src_start, Node* ofs, Node* limit) {
6769 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6770 const TypeOopPtr* xtype = aklass->as_instance_type();
6771 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6772 sha_obj = _gvn.transform(sha_obj);
6773
6774 Node* state;
6775 if (long_state) {
6776 state = get_state_from_sha5_object(sha_obj);
6777 } else {
6778 state = get_state_from_sha_object(sha_obj);
6779 }
6780 if (state == NULL) return false;
6781
6782 // Call the stub.
6783 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6784 OptoRuntime::digestBase_implCompressMB_Type(),
6785 stubAddr, stubName, TypePtr::BOTTOM,
6786 src_start, state, ofs, limit);
6787 // return ofs (int)
6788 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6789 set_result(result);
6790
6791 return true;
6792 }
6793
6794 //------------------------------get_state_from_sha_object-----------------------
6795 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6796 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6797 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6798 if (sha_state == NULL) return (Node *) NULL;
6799
6800 // now have the array, need to get the start address of the state array
6801 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6802 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6803 return state;
6804 }
6805
6806 //------------------------------get_state_from_sha5_object-----------------------
6807 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6808 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6809 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6810 if (sha_state == NULL) return (Node *) NULL;
6811
6812 // now have the array, need to get the start address of the state array
6813 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6814 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6815 return state;
6816 }
6817
6818 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6819 // Return node representing slow path of predicate check.
6820 // the pseudo code we want to emulate with this predicate is:
6821 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6822 //
6823 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6824 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6825 "need SHA1/SHA256/SHA512 instruction support");
6826 assert((uint)predicate < 3, "sanity");
6827
6828 // The receiver was checked for NULL already.
6829 Node* digestBaseObj = argument(0);
6830
6831 // get DigestBase klass for instanceOf check
6832 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6833 assert(tinst != NULL, "digestBaseObj is null");
6834 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6835
6836 const char* klass_SHA_name = NULL;
6837 switch (predicate) {
6838 case 0:
6839 if (UseSHA1Intrinsics) {
6840 // we want to do an instanceof comparison against the SHA class
6841 klass_SHA_name = "sun/security/provider/SHA";
6842 }
6843 break;
6844 case 1:
6845 if (UseSHA256Intrinsics) {
6846 // we want to do an instanceof comparison against the SHA2 class
6847 klass_SHA_name = "sun/security/provider/SHA2";
6848 }
6849 break;
6850 case 2:
6851 if (UseSHA512Intrinsics) {
6852 // we want to do an instanceof comparison against the SHA5 class
6853 klass_SHA_name = "sun/security/provider/SHA5";
6854 }
6855 break;
6856 default:
6857 fatal("unknown SHA intrinsic predicate: %d", predicate);
6858 }
6859
6860 ciKlass* klass_SHA = NULL;
6861 if (klass_SHA_name != NULL) {
6862 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6863 }
6864 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6865 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6866 Node* ctrl = control();
6867 set_control(top()); // no intrinsic path
6868 return ctrl;
6869 }
6870 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6871
6872 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6873 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6874 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6875 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6876
6877 return instof_false; // even if it is NULL
6878 }
6879
6880 //-------------inline_fma-----------------------------------
6881 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6882 Node *a = NULL;
6883 Node *b = NULL;
6884 Node *c = NULL;
6885 Node* result = NULL;
6886 switch (id) {
6887 case vmIntrinsics::_fmaD:
6888 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6889 // no receiver since it is static method
6890 a = round_double_node(argument(0));
6891 b = round_double_node(argument(2));
6892 c = round_double_node(argument(4));
6893 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6894 break;
6895 case vmIntrinsics::_fmaF:
6896 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6897 a = argument(0);
6898 b = argument(1);
6899 c = argument(2);
6900 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6901 break;
6902 default:
6903 fatal_unexpected_iid(id); break;
6904 }
6905 set_result(result);
6906 return true;
6907 }
6908
6909 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6910 // argument(0) is receiver
6911 Node* codePoint = argument(1);
6912 Node* n = NULL;
6913
6914 switch (id) {
6915 case vmIntrinsics::_isDigit :
6916 n = new DigitNode(control(), codePoint);
6917 break;
6918 case vmIntrinsics::_isLowerCase :
6919 n = new LowerCaseNode(control(), codePoint);
6920 break;
6921 case vmIntrinsics::_isUpperCase :
6922 n = new UpperCaseNode(control(), codePoint);
6923 break;
6924 case vmIntrinsics::_isWhitespace :
6925 n = new WhitespaceNode(control(), codePoint);
6926 break;
6927 default:
6928 fatal_unexpected_iid(id);
6929 }
6930
6931 set_result(_gvn.transform(n));
6932 return true;
6933 }
6934
6935 //------------------------------inline_fp_min_max------------------------------
6936 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6937 Node *a = NULL;
6938 Node *b = NULL;
6939 Node *n = NULL;
6940 switch (id) {
6941 case vmIntrinsics::_maxF:
6942 case vmIntrinsics::_minF:
6943 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6944 a = argument(0);
6945 b = argument(1);
6946 break;
6947 case vmIntrinsics::_maxD:
6948 case vmIntrinsics::_minD:
6949 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6950 a = round_double_node(argument(0));
6951 b = round_double_node(argument(2));
6952 break;
6953 default:
6954 fatal_unexpected_iid(id);
6955 break;
6956 }
6957 if (a->is_Con() || b->is_Con()) {
6958 return false;
6959 }
6960 switch (id) {
6961 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6962 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6963 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6964 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6965 default: fatal_unexpected_iid(id); break;
6966 }
6967 set_result(_gvn.transform(n));
6968 return true;
6969 }
6970
6971 bool LibraryCallKit::inline_profileBoolean() {
6972 Node* counts = argument(1);
6973 const TypeAryPtr* ary = NULL;
6974 ciArray* aobj = NULL;
6975 if (counts->is_Con()
6976 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6977 && (aobj = ary->const_oop()->as_array()) != NULL
6978 && (aobj->length() == 2)) {
6979 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6980 jint false_cnt = aobj->element_value(0).as_int();
6981 jint true_cnt = aobj->element_value(1).as_int();
6982
6983 if (C->log() != NULL) {
6984 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6985 false_cnt, true_cnt);
6986 }
6987
6988 if (false_cnt + true_cnt == 0) {
6989 // According to profile, never executed.
6990 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6991 Deoptimization::Action_reinterpret);
6992 return true;
6993 }
6994
6995 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6996 // is a number of each value occurrences.
6997 Node* result = argument(0);
6998 if (false_cnt == 0 || true_cnt == 0) {
6999 // According to profile, one value has been never seen.
7000 int expected_val = (false_cnt == 0) ? 1 : 0;
7001
7002 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7003 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7004
7005 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7006 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7007 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7008
7009 { // Slow path: uncommon trap for never seen value and then reexecute
7010 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7011 // the value has been seen at least once.
7012 PreserveJVMState pjvms(this);
7013 PreserveReexecuteState preexecs(this);
7014 jvms()->set_should_reexecute(true);
7015
7016 set_control(slow_path);
7017 set_i_o(i_o());
7018
7019 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7020 Deoptimization::Action_reinterpret);
7021 }
7022 // The guard for never seen value enables sharpening of the result and
7023 // returning a constant. It allows to eliminate branches on the same value
7024 // later on.
7025 set_control(fast_path);
7026 result = intcon(expected_val);
7027 }
7028 // Stop profiling.
7029 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7030 // By replacing method body with profile data (represented as ProfileBooleanNode
7031 // on IR level) we effectively disable profiling.
7032 // It enables full speed execution once optimized code is generated.
7033 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7034 C->record_for_igvn(profile);
7035 set_result(profile);
7036 return true;
7037 } else {
7038 // Continue profiling.
7039 // Profile data isn't available at the moment. So, execute method's bytecode version.
7040 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7041 // is compiled and counters aren't available since corresponding MethodHandle
7042 // isn't a compile-time constant.
7043 return false;
7044 }
7045 }
7046
7047 bool LibraryCallKit::inline_isCompileConstant() {
7048 Node* n = argument(0);
7049 set_result(n->is_Con() ? intcon(1) : intcon(0));
7050 return true;
7051 }