1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/callGenerator.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/intrinsicnode.hpp"
45 #include "opto/idealKit.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/movenode.hpp"
48 #include "opto/mulnode.hpp"
49 #include "opto/narrowptrnode.hpp"
50 #include "opto/opaquenode.hpp"
51 #include "opto/parse.hpp"
52 #include "opto/runtime.hpp"
53 #include "opto/rootnode.hpp"
54 #include "opto/subnode.hpp"
55 #include "prims/nativeLookup.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "utilities/macros.hpp"
60
61
62 class LibraryIntrinsic : public InlineCallGenerator {
63 // Extend the set of intrinsics known to the runtime:
64 public:
65 private:
66 bool _is_virtual;
67 bool _does_virtual_dispatch;
68 int8_t _predicates_count; // Intrinsic is predicated by several conditions
69 int8_t _last_predicate; // Last generated predicate
70 vmIntrinsics::ID _intrinsic_id;
71
72 public:
73 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
74 : InlineCallGenerator(m),
75 _is_virtual(is_virtual),
76 _does_virtual_dispatch(does_virtual_dispatch),
77 _predicates_count((int8_t)predicates_count),
78 _last_predicate((int8_t)-1),
79 _intrinsic_id(id)
80 {
81 }
82 virtual bool is_intrinsic() const { return true; }
83 virtual bool is_virtual() const { return _is_virtual; }
84 virtual bool is_predicated() const { return _predicates_count > 0; }
85 virtual int predicates_count() const { return _predicates_count; }
86 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
87 virtual JVMState* generate(JVMState* jvms);
88 virtual Node* generate_predicate(JVMState* jvms, int predicate);
89 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
90 };
91
92
93 // Local helper class for LibraryIntrinsic:
94 class LibraryCallKit : public GraphKit {
95 private:
96 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
97 Node* _result; // the result node, if any
98 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
99
100 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
101
102 public:
103 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
104 : GraphKit(jvms),
105 _intrinsic(intrinsic),
106 _result(NULL)
107 {
108 // Check if this is a root compile. In that case we don't have a caller.
109 if (!jvms->has_method()) {
110 _reexecute_sp = sp();
111 } else {
112 // Find out how many arguments the interpreter needs when deoptimizing
113 // and save the stack pointer value so it can used by uncommon_trap.
114 // We find the argument count by looking at the declared signature.
115 bool ignored_will_link;
116 ciSignature* declared_signature = NULL;
117 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
118 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
119 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
120 }
121 }
122
123 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
124
125 ciMethod* caller() const { return jvms()->method(); }
126 int bci() const { return jvms()->bci(); }
127 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
128 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
129 ciMethod* callee() const { return _intrinsic->method(); }
130
131 bool try_to_inline(int predicate);
132 Node* try_to_predicate(int predicate);
133
134 void push_result() {
135 // Push the result onto the stack.
136 if (!stopped() && result() != NULL) {
137 BasicType bt = result()->bottom_type()->basic_type();
138 push_node(bt, result());
139 }
140 }
141
142 private:
143 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
144 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
145 }
146
147 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
148 void set_result(RegionNode* region, PhiNode* value);
149 Node* result() { return _result; }
150
151 virtual int reexecute_sp() { return _reexecute_sp; }
152
153 // Helper functions to inline natives
154 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
155 Node* generate_slow_guard(Node* test, RegionNode* region);
156 Node* generate_fair_guard(Node* test, RegionNode* region);
157 Node* generate_negative_guard(Node* index, RegionNode* region,
158 // resulting CastII of index:
159 Node* *pos_index = NULL);
160 Node* generate_limit_guard(Node* offset, Node* subseq_length,
161 Node* array_length,
162 RegionNode* region);
163 void generate_string_range_check(Node* array, Node* offset,
164 Node* length, bool char_count);
165 Node* generate_current_thread(Node* &tls_output);
166 Node* load_mirror_from_klass(Node* klass);
167 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path,
169 int offset);
170 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
171 RegionNode* region, int null_path) {
172 int offset = java_lang_Class::klass_offset_in_bytes();
173 return load_klass_from_mirror_common(mirror, never_see_null,
174 region, null_path,
175 offset);
176 }
177 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
178 RegionNode* region, int null_path) {
179 int offset = java_lang_Class::array_klass_offset_in_bytes();
180 return load_klass_from_mirror_common(mirror, never_see_null,
181 region, null_path,
182 offset);
183 }
184 Node* generate_access_flags_guard(Node* kls,
185 int modifier_mask, int modifier_bits,
186 RegionNode* region);
187 Node* generate_interface_guard(Node* kls, RegionNode* region);
188 Node* generate_array_guard(Node* kls, RegionNode* region) {
189 return generate_array_guard_common(kls, region, false, false);
190 }
191 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
192 return generate_array_guard_common(kls, region, false, true);
193 }
194 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
195 return generate_array_guard_common(kls, region, true, false);
196 }
197 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
198 return generate_array_guard_common(kls, region, true, true);
199 }
200 Node* generate_array_guard_common(Node* kls, RegionNode* region,
201 bool obj_array, bool not_array);
202 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
203 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
204 bool is_virtual = false, bool is_static = false);
205 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
206 return generate_method_call(method_id, false, true);
207 }
208 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
209 return generate_method_call(method_id, true, false);
210 }
211 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
212 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
213
214 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
215 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
218 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
219 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
220 bool inline_string_indexOfChar();
221 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
222 bool inline_string_toBytesU();
223 bool inline_string_getCharsU();
224 bool inline_string_copy(bool compress);
225 bool inline_string_char_access(bool is_store);
226 Node* round_double_node(Node* n);
227 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
228 bool inline_math_native(vmIntrinsics::ID id);
229 bool inline_math(vmIntrinsics::ID id);
230 bool inline_double_math(vmIntrinsics::ID id);
231 template <typename OverflowOp>
232 bool inline_math_overflow(Node* arg1, Node* arg2);
233 void inline_math_mathExact(Node* math, Node* test);
234 bool inline_math_addExactI(bool is_increment);
235 bool inline_math_addExactL(bool is_increment);
236 bool inline_math_multiplyExactI();
237 bool inline_math_multiplyExactL();
238 bool inline_math_multiplyHigh();
239 bool inline_math_negateExactI();
240 bool inline_math_negateExactL();
241 bool inline_math_subtractExactI(bool is_decrement);
242 bool inline_math_subtractExactL(bool is_decrement);
243 bool inline_min_max(vmIntrinsics::ID id);
244 bool inline_notify(vmIntrinsics::ID id);
245 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
246 // This returns Type::AnyPtr, RawPtr, or OopPtr.
247 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
248 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
249
250 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
251 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
252 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
253 static bool klass_needs_init_guard(Node* kls);
254 bool inline_unsafe_allocate();
255 bool inline_unsafe_newArray(bool uninitialized);
256 bool inline_unsafe_copyMemory();
257 bool inline_native_currentThread();
258
259 bool inline_native_time_funcs(address method, const char* funcName);
260 #ifdef JFR_HAVE_INTRINSICS
261 bool inline_native_classID();
262 bool inline_native_getEventWriter();
263 #endif
264 bool inline_native_isInterrupted();
265 bool inline_native_Class_query(vmIntrinsics::ID id);
266 bool inline_native_subtype_check();
267 bool inline_native_getLength();
268 bool inline_array_copyOf(bool is_copyOfRange);
269 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
270 bool inline_preconditions_checkIndex();
271 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
272 bool inline_native_clone(bool is_virtual);
273 bool inline_native_Reflection_getCallerClass();
274 // Helper function for inlining native object hash method
275 bool inline_native_hashcode(bool is_virtual, bool is_static);
276 bool inline_native_getClass();
277
278 // Helper functions for inlining arraycopy
279 bool inline_arraycopy();
280 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
281 RegionNode* slow_region);
282 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
283 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
284 uint new_idx);
285
286 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
287 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
288 bool inline_unsafe_fence(vmIntrinsics::ID id);
289 bool inline_onspinwait();
290 bool inline_fp_conversions(vmIntrinsics::ID id);
291 bool inline_number_methods(vmIntrinsics::ID id);
292 bool inline_reference_get();
293 bool inline_Class_cast();
294 bool inline_aescrypt_Block(vmIntrinsics::ID id);
295 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
296 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
297 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
298 Node* inline_counterMode_AESCrypt_predicate();
299 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
300 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
301 bool inline_ghash_processBlocks();
302 bool inline_base64_encodeBlock();
303 bool inline_sha_implCompress(vmIntrinsics::ID id);
304 bool inline_digestBase_implCompressMB(int predicate);
305 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
306 bool long_state, address stubAddr, const char *stubName,
307 Node* src_start, Node* ofs, Node* limit);
308 Node* get_state_from_sha_object(Node *sha_object);
309 Node* get_state_from_sha5_object(Node *sha_object);
310 Node* inline_digestBase_implCompressMB_predicate(int predicate);
311 bool inline_encodeISOArray();
312 bool inline_updateCRC32();
313 bool inline_updateBytesCRC32();
314 bool inline_updateByteBufferCRC32();
315 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
316 bool inline_updateBytesCRC32C();
317 bool inline_updateDirectByteBufferCRC32C();
318 bool inline_updateBytesAdler32();
319 bool inline_updateByteBufferAdler32();
320 bool inline_multiplyToLen();
321 bool inline_hasNegatives();
322 bool inline_squareToLen();
323 bool inline_mulAdd();
324 bool inline_montgomeryMultiply();
325 bool inline_montgomerySquare();
326 bool inline_vectorizedMismatch();
327 bool inline_fma(vmIntrinsics::ID id);
328 bool inline_character_compare(vmIntrinsics::ID id);
329 bool inline_fp_min_max(vmIntrinsics::ID id);
330
331 bool inline_profileBoolean();
332 bool inline_isCompileConstant();
333 void clear_upper_avx() {
334 #ifdef X86
335 if (UseAVX >= 2) {
336 C->set_clear_upper_avx(true);
337 }
338 #endif
339 }
340 };
341
342 //---------------------------make_vm_intrinsic----------------------------
343 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
344 vmIntrinsics::ID id = m->intrinsic_id();
345 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
346
347 if (!m->is_loaded()) {
348 // Do not attempt to inline unloaded methods.
349 return NULL;
350 }
351
352 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
353 bool is_available = false;
354
355 {
356 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
357 // the compiler must transition to '_thread_in_vm' state because both
358 // methods access VM-internal data.
359 VM_ENTRY_MARK;
360 methodHandle mh(THREAD, m->get_Method());
361 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
362 !C->directive()->is_intrinsic_disabled(mh) &&
363 !vmIntrinsics::is_disabled_by_flags(mh);
364
365 }
366
367 if (is_available) {
368 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
369 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
370 return new LibraryIntrinsic(m, is_virtual,
371 vmIntrinsics::predicates_needed(id),
372 vmIntrinsics::does_virtual_dispatch(id),
373 (vmIntrinsics::ID) id);
374 } else {
375 return NULL;
376 }
377 }
378
379 //----------------------register_library_intrinsics-----------------------
380 // Initialize this file's data structures, for each Compile instance.
381 void Compile::register_library_intrinsics() {
382 // Nothing to do here.
383 }
384
385 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
386 LibraryCallKit kit(jvms, this);
387 Compile* C = kit.C;
388 int nodes = C->unique();
389 #ifndef PRODUCT
390 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
391 char buf[1000];
392 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
393 tty->print_cr("Intrinsic %s", str);
394 }
395 #endif
396 ciMethod* callee = kit.callee();
397 const int bci = kit.bci();
398
399 // Try to inline the intrinsic.
400 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
401 kit.try_to_inline(_last_predicate)) {
402 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
403 : "(intrinsic)";
404 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
405 if (C->print_intrinsics() || C->print_inlining()) {
406 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
407 }
408 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
409 if (C->log()) {
410 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
411 vmIntrinsics::name_at(intrinsic_id()),
412 (is_virtual() ? " virtual='1'" : ""),
413 C->unique() - nodes);
414 }
415 // Push the result from the inlined method onto the stack.
416 kit.push_result();
417 C->print_inlining_update(this);
418 return kit.transfer_exceptions_into_jvms();
419 }
420
421 // The intrinsic bailed out
422 if (jvms->has_method()) {
423 // Not a root compile.
424 const char* msg;
425 if (callee->intrinsic_candidate()) {
426 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
427 } else {
428 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
429 : "failed to inline (intrinsic), method not annotated";
430 }
431 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
432 if (C->print_intrinsics() || C->print_inlining()) {
433 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
434 }
435 } else {
436 // Root compile
437 ResourceMark rm;
438 stringStream msg_stream;
439 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
440 vmIntrinsics::name_at(intrinsic_id()),
441 is_virtual() ? " (virtual)" : "", bci);
442 const char *msg = msg_stream.as_string();
443 log_debug(jit, inlining)("%s", msg);
444 if (C->print_intrinsics() || C->print_inlining()) {
445 tty->print("%s", msg);
446 }
447 }
448 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
449 C->print_inlining_update(this);
450 return NULL;
451 }
452
453 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
454 LibraryCallKit kit(jvms, this);
455 Compile* C = kit.C;
456 int nodes = C->unique();
457 _last_predicate = predicate;
458 #ifndef PRODUCT
459 assert(is_predicated() && predicate < predicates_count(), "sanity");
460 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
461 char buf[1000];
462 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
463 tty->print_cr("Predicate for intrinsic %s", str);
464 }
465 #endif
466 ciMethod* callee = kit.callee();
467 const int bci = kit.bci();
468
469 Node* slow_ctl = kit.try_to_predicate(predicate);
470 if (!kit.failing()) {
471 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
472 : "(intrinsic, predicate)";
473 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
474 if (C->print_intrinsics() || C->print_inlining()) {
475 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
476 }
477 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
478 if (C->log()) {
479 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
480 vmIntrinsics::name_at(intrinsic_id()),
481 (is_virtual() ? " virtual='1'" : ""),
482 C->unique() - nodes);
483 }
484 return slow_ctl; // Could be NULL if the check folds.
485 }
486
487 // The intrinsic bailed out
488 if (jvms->has_method()) {
489 // Not a root compile.
490 const char* msg = "failed to generate predicate for intrinsic";
491 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
492 if (C->print_intrinsics() || C->print_inlining()) {
493 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
494 }
495 } else {
496 // Root compile
497 ResourceMark rm;
498 stringStream msg_stream;
499 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
500 vmIntrinsics::name_at(intrinsic_id()),
501 is_virtual() ? " (virtual)" : "", bci);
502 const char *msg = msg_stream.as_string();
503 log_debug(jit, inlining)("%s", msg);
504 if (C->print_intrinsics() || C->print_inlining()) {
505 C->print_inlining_stream()->print("%s", msg);
506 }
507 }
508 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
509 return NULL;
510 }
511
512 bool LibraryCallKit::try_to_inline(int predicate) {
513 // Handle symbolic names for otherwise undistinguished boolean switches:
514 const bool is_store = true;
515 const bool is_compress = true;
516 const bool is_static = true;
517 const bool is_volatile = true;
518
519 if (!jvms()->has_method()) {
520 // Root JVMState has a null method.
521 assert(map()->memory()->Opcode() == Op_Parm, "");
522 // Insert the memory aliasing node
523 set_all_memory(reset_memory());
524 }
525 assert(merged_memory(), "");
526
527
528 switch (intrinsic_id()) {
529 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
530 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
531 case vmIntrinsics::_getClass: return inline_native_getClass();
532
533 case vmIntrinsics::_dsin:
534 case vmIntrinsics::_dcos:
535 case vmIntrinsics::_dtan:
536 case vmIntrinsics::_dabs:
537 case vmIntrinsics::_fabs:
538 case vmIntrinsics::_iabs:
539 case vmIntrinsics::_labs:
540 case vmIntrinsics::_datan2:
541 case vmIntrinsics::_dsqrt:
542 case vmIntrinsics::_dexp:
543 case vmIntrinsics::_dlog:
544 case vmIntrinsics::_dlog10:
545 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
546
547 case vmIntrinsics::_min:
548 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
549
550 case vmIntrinsics::_notify:
551 case vmIntrinsics::_notifyAll:
552 return inline_notify(intrinsic_id());
553
554 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
555 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
556 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
557 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
558 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
559 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
560 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
561 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
562 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
563 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
564 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
565 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
566 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
567
568 case vmIntrinsics::_arraycopy: return inline_arraycopy();
569
570 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
571 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
572 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
573 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
574
575 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
576 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
577 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
578 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
579 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
580 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
581 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
582
583 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
584 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
585
586 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
587 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
588 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
589 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
590
591 case vmIntrinsics::_compressStringC:
592 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
593 case vmIntrinsics::_inflateStringC:
594 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
595
596 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
597 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
598 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
599 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
600 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
601 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
602 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
603 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
604 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
605
606 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
607 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
608 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
609 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
610 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
611 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
612 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
613 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
614 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
615
616 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
617 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
618 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
619 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
620 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
621 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
622 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
623 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
624 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
625
626 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
627 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
628 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
629 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
630 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
631 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
632 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
633 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
634 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
635
636 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
637 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
638 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
639 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
640
641 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
642 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
643 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
644 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
645
646 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
647 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
648 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
649 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
650 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
651 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
652 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
653 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
654 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
655
656 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
657 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
658 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
659 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
660 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
661 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
662 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
663 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
664 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
665
666 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
667 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
668 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
669 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
670 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
671 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
672 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
673 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
674 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
675
676 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
677 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
678 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
679 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
680 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
681 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
682 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
683 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
684 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
685
686 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
687 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
688 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
689 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
690 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
691
692 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
693 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
694 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
695 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
696 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
697 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
698 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
699 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
700 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
701 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
702 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
703 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
704 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
705 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
706 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
707 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
708 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
709 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
710 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
711 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
712
713 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
714 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
715 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
716 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
717 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
718 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
719 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
720 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
721 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
722 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
723 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
724 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
725 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
726 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
727 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
728
729 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
730 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
731 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
732 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
733
734 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
735 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
736 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
737 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
738 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
739
740 case vmIntrinsics::_loadFence:
741 case vmIntrinsics::_storeFence:
742 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
743
744 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
745
746 case vmIntrinsics::_currentThread: return inline_native_currentThread();
747 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
748
749 #ifdef JFR_HAVE_INTRINSICS
750 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
751 case vmIntrinsics::_getClassId: return inline_native_classID();
752 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
753 #endif
754 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
755 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
756 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
757 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
758 case vmIntrinsics::_getLength: return inline_native_getLength();
759 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
760 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
761 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
762 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
763 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
764 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
765
766 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
767 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
768
769 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
770
771 case vmIntrinsics::_isInstance:
772 case vmIntrinsics::_getModifiers:
773 case vmIntrinsics::_isInterface:
774 case vmIntrinsics::_isArray:
775 case vmIntrinsics::_isPrimitive:
776 case vmIntrinsics::_getSuperclass:
777 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
778
779 case vmIntrinsics::_floatToRawIntBits:
780 case vmIntrinsics::_floatToIntBits:
781 case vmIntrinsics::_intBitsToFloat:
782 case vmIntrinsics::_doubleToRawLongBits:
783 case vmIntrinsics::_doubleToLongBits:
784 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
785
786 case vmIntrinsics::_numberOfLeadingZeros_i:
787 case vmIntrinsics::_numberOfLeadingZeros_l:
788 case vmIntrinsics::_numberOfTrailingZeros_i:
789 case vmIntrinsics::_numberOfTrailingZeros_l:
790 case vmIntrinsics::_bitCount_i:
791 case vmIntrinsics::_bitCount_l:
792 case vmIntrinsics::_reverseBytes_i:
793 case vmIntrinsics::_reverseBytes_l:
794 case vmIntrinsics::_reverseBytes_s:
795 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
796
797 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
798
799 case vmIntrinsics::_Reference_get: return inline_reference_get();
800
801 case vmIntrinsics::_Class_cast: return inline_Class_cast();
802
803 case vmIntrinsics::_aescrypt_encryptBlock:
804 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
805
806 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
807 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
809
810 case vmIntrinsics::_counterMode_AESCrypt:
811 return inline_counterMode_AESCrypt(intrinsic_id());
812
813 case vmIntrinsics::_sha_implCompress:
814 case vmIntrinsics::_sha2_implCompress:
815 case vmIntrinsics::_sha5_implCompress:
816 return inline_sha_implCompress(intrinsic_id());
817
818 case vmIntrinsics::_digestBase_implCompressMB:
819 return inline_digestBase_implCompressMB(predicate);
820
821 case vmIntrinsics::_multiplyToLen:
822 return inline_multiplyToLen();
823
824 case vmIntrinsics::_squareToLen:
825 return inline_squareToLen();
826
827 case vmIntrinsics::_mulAdd:
828 return inline_mulAdd();
829
830 case vmIntrinsics::_montgomeryMultiply:
831 return inline_montgomeryMultiply();
832 case vmIntrinsics::_montgomerySquare:
833 return inline_montgomerySquare();
834
835 case vmIntrinsics::_vectorizedMismatch:
836 return inline_vectorizedMismatch();
837
838 case vmIntrinsics::_ghash_processBlocks:
839 return inline_ghash_processBlocks();
840 case vmIntrinsics::_base64_encodeBlock:
841 return inline_base64_encodeBlock();
842
843 case vmIntrinsics::_encodeISOArray:
844 case vmIntrinsics::_encodeByteISOArray:
845 return inline_encodeISOArray();
846
847 case vmIntrinsics::_updateCRC32:
848 return inline_updateCRC32();
849 case vmIntrinsics::_updateBytesCRC32:
850 return inline_updateBytesCRC32();
851 case vmIntrinsics::_updateByteBufferCRC32:
852 return inline_updateByteBufferCRC32();
853
854 case vmIntrinsics::_updateBytesCRC32C:
855 return inline_updateBytesCRC32C();
856 case vmIntrinsics::_updateDirectByteBufferCRC32C:
857 return inline_updateDirectByteBufferCRC32C();
858
859 case vmIntrinsics::_updateBytesAdler32:
860 return inline_updateBytesAdler32();
861 case vmIntrinsics::_updateByteBufferAdler32:
862 return inline_updateByteBufferAdler32();
863
864 case vmIntrinsics::_profileBoolean:
865 return inline_profileBoolean();
866 case vmIntrinsics::_isCompileConstant:
867 return inline_isCompileConstant();
868
869 case vmIntrinsics::_hasNegatives:
870 return inline_hasNegatives();
871
872 case vmIntrinsics::_fmaD:
873 case vmIntrinsics::_fmaF:
874 return inline_fma(intrinsic_id());
875
876 case vmIntrinsics::_isDigit:
877 case vmIntrinsics::_isLowerCase:
878 case vmIntrinsics::_isUpperCase:
879 case vmIntrinsics::_isWhitespace:
880 return inline_character_compare(intrinsic_id());
881
882 case vmIntrinsics::_maxF:
883 case vmIntrinsics::_minF:
884 case vmIntrinsics::_maxD:
885 case vmIntrinsics::_minD:
886 return inline_fp_min_max(intrinsic_id());
887
888 default:
889 // If you get here, it may be that someone has added a new intrinsic
890 // to the list in vmSymbols.hpp without implementing it here.
891 #ifndef PRODUCT
892 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
893 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
894 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
895 }
896 #endif
897 return false;
898 }
899 }
900
901 Node* LibraryCallKit::try_to_predicate(int predicate) {
902 if (!jvms()->has_method()) {
903 // Root JVMState has a null method.
904 assert(map()->memory()->Opcode() == Op_Parm, "");
905 // Insert the memory aliasing node
906 set_all_memory(reset_memory());
907 }
908 assert(merged_memory(), "");
909
910 switch (intrinsic_id()) {
911 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
912 return inline_cipherBlockChaining_AESCrypt_predicate(false);
913 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
914 return inline_cipherBlockChaining_AESCrypt_predicate(true);
915 case vmIntrinsics::_counterMode_AESCrypt:
916 return inline_counterMode_AESCrypt_predicate();
917 case vmIntrinsics::_digestBase_implCompressMB:
918 return inline_digestBase_implCompressMB_predicate(predicate);
919
920 default:
921 // If you get here, it may be that someone has added a new intrinsic
922 // to the list in vmSymbols.hpp without implementing it here.
923 #ifndef PRODUCT
924 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
925 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
926 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
927 }
928 #endif
929 Node* slow_ctl = control();
930 set_control(top()); // No fast path instrinsic
931 return slow_ctl;
932 }
933 }
934
935 //------------------------------set_result-------------------------------
936 // Helper function for finishing intrinsics.
937 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
938 record_for_igvn(region);
939 set_control(_gvn.transform(region));
940 set_result( _gvn.transform(value));
941 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
942 }
943
944 //------------------------------generate_guard---------------------------
945 // Helper function for generating guarded fast-slow graph structures.
946 // The given 'test', if true, guards a slow path. If the test fails
947 // then a fast path can be taken. (We generally hope it fails.)
948 // In all cases, GraphKit::control() is updated to the fast path.
949 // The returned value represents the control for the slow path.
950 // The return value is never 'top'; it is either a valid control
951 // or NULL if it is obvious that the slow path can never be taken.
952 // Also, if region and the slow control are not NULL, the slow edge
953 // is appended to the region.
954 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
955 if (stopped()) {
956 // Already short circuited.
957 return NULL;
958 }
959
960 // Build an if node and its projections.
961 // If test is true we take the slow path, which we assume is uncommon.
962 if (_gvn.type(test) == TypeInt::ZERO) {
963 // The slow branch is never taken. No need to build this guard.
964 return NULL;
965 }
966
967 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
968
969 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
970 if (if_slow == top()) {
971 // The slow branch is never taken. No need to build this guard.
972 return NULL;
973 }
974
975 if (region != NULL)
976 region->add_req(if_slow);
977
978 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
979 set_control(if_fast);
980
981 return if_slow;
982 }
983
984 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
985 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
986 }
987 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
988 return generate_guard(test, region, PROB_FAIR);
989 }
990
991 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
992 Node* *pos_index) {
993 if (stopped())
994 return NULL; // already stopped
995 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
996 return NULL; // index is already adequately typed
997 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
998 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
999 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1000 if (is_neg != NULL && pos_index != NULL) {
1001 // Emulate effect of Parse::adjust_map_after_if.
1002 Node* ccast = new CastIINode(index, TypeInt::POS);
1003 ccast->set_req(0, control());
1004 (*pos_index) = _gvn.transform(ccast);
1005 }
1006 return is_neg;
1007 }
1008
1009 // Make sure that 'position' is a valid limit index, in [0..length].
1010 // There are two equivalent plans for checking this:
1011 // A. (offset + copyLength) unsigned<= arrayLength
1012 // B. offset <= (arrayLength - copyLength)
1013 // We require that all of the values above, except for the sum and
1014 // difference, are already known to be non-negative.
1015 // Plan A is robust in the face of overflow, if offset and copyLength
1016 // are both hugely positive.
1017 //
1018 // Plan B is less direct and intuitive, but it does not overflow at
1019 // all, since the difference of two non-negatives is always
1020 // representable. Whenever Java methods must perform the equivalent
1021 // check they generally use Plan B instead of Plan A.
1022 // For the moment we use Plan A.
1023 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1024 Node* subseq_length,
1025 Node* array_length,
1026 RegionNode* region) {
1027 if (stopped())
1028 return NULL; // already stopped
1029 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1030 if (zero_offset && subseq_length->eqv_uncast(array_length))
1031 return NULL; // common case of whole-array copy
1032 Node* last = subseq_length;
1033 if (!zero_offset) // last += offset
1034 last = _gvn.transform(new AddINode(last, offset));
1035 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1036 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1037 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1038 return is_over;
1039 }
1040
1041 // Emit range checks for the given String.value byte array
1042 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1043 if (stopped()) {
1044 return; // already stopped
1045 }
1046 RegionNode* bailout = new RegionNode(1);
1047 record_for_igvn(bailout);
1048 if (char_count) {
1049 // Convert char count to byte count
1050 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1051 }
1052
1053 // Offset and count must not be negative
1054 generate_negative_guard(offset, bailout);
1055 generate_negative_guard(count, bailout);
1056 // Offset + count must not exceed length of array
1057 generate_limit_guard(offset, count, load_array_length(array), bailout);
1058
1059 if (bailout->req() > 1) {
1060 PreserveJVMState pjvms(this);
1061 set_control(_gvn.transform(bailout));
1062 uncommon_trap(Deoptimization::Reason_intrinsic,
1063 Deoptimization::Action_maybe_recompile);
1064 }
1065 }
1066
1067 //--------------------------generate_current_thread--------------------
1068 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1069 ciKlass* thread_klass = env()->Thread_klass();
1070 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1071 Node* thread = _gvn.transform(new ThreadLocalNode());
1072 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1073 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1074 tls_output = thread;
1075 return threadObj;
1076 }
1077
1078
1079 //------------------------------make_string_method_node------------------------
1080 // Helper method for String intrinsic functions. This version is called with
1081 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1082 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1083 // containing the lengths of str1 and str2.
1084 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1085 Node* result = NULL;
1086 switch (opcode) {
1087 case Op_StrIndexOf:
1088 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1089 str1_start, cnt1, str2_start, cnt2, ae);
1090 break;
1091 case Op_StrComp:
1092 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1093 str1_start, cnt1, str2_start, cnt2, ae);
1094 break;
1095 case Op_StrEquals:
1096 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1097 // Use the constant length if there is one because optimized match rule may exist.
1098 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1099 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1100 break;
1101 default:
1102 ShouldNotReachHere();
1103 return NULL;
1104 }
1105
1106 // All these intrinsics have checks.
1107 C->set_has_split_ifs(true); // Has chance for split-if optimization
1108 clear_upper_avx();
1109
1110 return _gvn.transform(result);
1111 }
1112
1113 //------------------------------inline_string_compareTo------------------------
1114 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1115 Node* arg1 = argument(0);
1116 Node* arg2 = argument(1);
1117
1118 arg1 = must_be_not_null(arg1, true);
1119 arg2 = must_be_not_null(arg2, true);
1120
1121 arg1 = access_resolve(arg1, ACCESS_READ);
1122 arg2 = access_resolve(arg2, ACCESS_READ);
1123
1124 // Get start addr and length of first argument
1125 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1126 Node* arg1_cnt = load_array_length(arg1);
1127
1128 // Get start addr and length of second argument
1129 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1130 Node* arg2_cnt = load_array_length(arg2);
1131
1132 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1133 set_result(result);
1134 return true;
1135 }
1136
1137 //------------------------------inline_string_equals------------------------
1138 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1139 Node* arg1 = argument(0);
1140 Node* arg2 = argument(1);
1141
1142 // paths (plus control) merge
1143 RegionNode* region = new RegionNode(3);
1144 Node* phi = new PhiNode(region, TypeInt::BOOL);
1145
1146 if (!stopped()) {
1147
1148 arg1 = must_be_not_null(arg1, true);
1149 arg2 = must_be_not_null(arg2, true);
1150
1151 arg1 = access_resolve(arg1, ACCESS_READ);
1152 arg2 = access_resolve(arg2, ACCESS_READ);
1153
1154 // Get start addr and length of first argument
1155 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1156 Node* arg1_cnt = load_array_length(arg1);
1157
1158 // Get start addr and length of second argument
1159 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1160 Node* arg2_cnt = load_array_length(arg2);
1161
1162 // Check for arg1_cnt != arg2_cnt
1163 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1164 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1165 Node* if_ne = generate_slow_guard(bol, NULL);
1166 if (if_ne != NULL) {
1167 phi->init_req(2, intcon(0));
1168 region->init_req(2, if_ne);
1169 }
1170
1171 // Check for count == 0 is done by assembler code for StrEquals.
1172
1173 if (!stopped()) {
1174 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1175 phi->init_req(1, equals);
1176 region->init_req(1, control());
1177 }
1178 }
1179
1180 // post merge
1181 set_control(_gvn.transform(region));
1182 record_for_igvn(region);
1183
1184 set_result(_gvn.transform(phi));
1185 return true;
1186 }
1187
1188 //------------------------------inline_array_equals----------------------------
1189 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1190 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1191 Node* arg1 = argument(0);
1192 Node* arg2 = argument(1);
1193
1194 arg1 = access_resolve(arg1, ACCESS_READ);
1195 arg2 = access_resolve(arg2, ACCESS_READ);
1196
1197 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1198 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1199 clear_upper_avx();
1200
1201 return true;
1202 }
1203
1204 //------------------------------inline_hasNegatives------------------------------
1205 bool LibraryCallKit::inline_hasNegatives() {
1206 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1207 return false;
1208 }
1209
1210 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1211 // no receiver since it is static method
1212 Node* ba = argument(0);
1213 Node* offset = argument(1);
1214 Node* len = argument(2);
1215
1216 ba = must_be_not_null(ba, true);
1217
1218 // Range checks
1219 generate_string_range_check(ba, offset, len, false);
1220 if (stopped()) {
1221 return true;
1222 }
1223 ba = access_resolve(ba, ACCESS_READ);
1224 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1225 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1226 set_result(_gvn.transform(result));
1227 return true;
1228 }
1229
1230 bool LibraryCallKit::inline_preconditions_checkIndex() {
1231 Node* index = argument(0);
1232 Node* length = argument(1);
1233 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1234 return false;
1235 }
1236
1237 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1238 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1239
1240 {
1241 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1242 uncommon_trap(Deoptimization::Reason_intrinsic,
1243 Deoptimization::Action_make_not_entrant);
1244 }
1245
1246 if (stopped()) {
1247 return false;
1248 }
1249
1250 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1251 BoolTest::mask btest = BoolTest::lt;
1252 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1253 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1254 _gvn.set_type(rc, rc->Value(&_gvn));
1255 if (!rc_bool->is_Con()) {
1256 record_for_igvn(rc);
1257 }
1258 set_control(_gvn.transform(new IfTrueNode(rc)));
1259 {
1260 PreserveJVMState pjvms(this);
1261 set_control(_gvn.transform(new IfFalseNode(rc)));
1262 uncommon_trap(Deoptimization::Reason_range_check,
1263 Deoptimization::Action_make_not_entrant);
1264 }
1265
1266 if (stopped()) {
1267 return false;
1268 }
1269
1270 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1271 result->set_req(0, control());
1272 result = _gvn.transform(result);
1273 set_result(result);
1274 replace_in_map(index, result);
1275 clear_upper_avx();
1276 return true;
1277 }
1278
1279 //------------------------------inline_string_indexOf------------------------
1280 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1281 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1282 return false;
1283 }
1284 Node* src = argument(0);
1285 Node* tgt = argument(1);
1286
1287 // Make the merge point
1288 RegionNode* result_rgn = new RegionNode(4);
1289 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1290
1291 src = must_be_not_null(src, true);
1292 tgt = must_be_not_null(tgt, true);
1293
1294 src = access_resolve(src, ACCESS_READ);
1295 tgt = access_resolve(tgt, ACCESS_READ);
1296
1297 // Get start addr and length of source string
1298 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1299 Node* src_count = load_array_length(src);
1300
1301 // Get start addr and length of substring
1302 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1303 Node* tgt_count = load_array_length(tgt);
1304
1305 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1306 // Divide src size by 2 if String is UTF16 encoded
1307 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1308 }
1309 if (ae == StrIntrinsicNode::UU) {
1310 // Divide substring size by 2 if String is UTF16 encoded
1311 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1312 }
1313
1314 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1315 if (result != NULL) {
1316 result_phi->init_req(3, result);
1317 result_rgn->init_req(3, control());
1318 }
1319 set_control(_gvn.transform(result_rgn));
1320 record_for_igvn(result_rgn);
1321 set_result(_gvn.transform(result_phi));
1322
1323 return true;
1324 }
1325
1326 //-----------------------------inline_string_indexOf-----------------------
1327 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1328 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1329 return false;
1330 }
1331 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1332 return false;
1333 }
1334 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1335 Node* src = argument(0); // byte[]
1336 Node* src_count = argument(1); // char count
1337 Node* tgt = argument(2); // byte[]
1338 Node* tgt_count = argument(3); // char count
1339 Node* from_index = argument(4); // char index
1340
1341 src = must_be_not_null(src, true);
1342 tgt = must_be_not_null(tgt, true);
1343
1344 src = access_resolve(src, ACCESS_READ);
1345 tgt = access_resolve(tgt, ACCESS_READ);
1346
1347 // Multiply byte array index by 2 if String is UTF16 encoded
1348 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1349 src_count = _gvn.transform(new SubINode(src_count, from_index));
1350 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1351 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1352
1353 // Range checks
1354 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1355 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1356 if (stopped()) {
1357 return true;
1358 }
1359
1360 RegionNode* region = new RegionNode(5);
1361 Node* phi = new PhiNode(region, TypeInt::INT);
1362
1363 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1364 if (result != NULL) {
1365 // The result is index relative to from_index if substring was found, -1 otherwise.
1366 // Generate code which will fold into cmove.
1367 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1368 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1369
1370 Node* if_lt = generate_slow_guard(bol, NULL);
1371 if (if_lt != NULL) {
1372 // result == -1
1373 phi->init_req(3, result);
1374 region->init_req(3, if_lt);
1375 }
1376 if (!stopped()) {
1377 result = _gvn.transform(new AddINode(result, from_index));
1378 phi->init_req(4, result);
1379 region->init_req(4, control());
1380 }
1381 }
1382
1383 set_control(_gvn.transform(region));
1384 record_for_igvn(region);
1385 set_result(_gvn.transform(phi));
1386 clear_upper_avx();
1387
1388 return true;
1389 }
1390
1391 // Create StrIndexOfNode with fast path checks
1392 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1393 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1394 // Check for substr count > string count
1395 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1396 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1397 Node* if_gt = generate_slow_guard(bol, NULL);
1398 if (if_gt != NULL) {
1399 phi->init_req(1, intcon(-1));
1400 region->init_req(1, if_gt);
1401 }
1402 if (!stopped()) {
1403 // Check for substr count == 0
1404 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1405 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1406 Node* if_zero = generate_slow_guard(bol, NULL);
1407 if (if_zero != NULL) {
1408 phi->init_req(2, intcon(0));
1409 region->init_req(2, if_zero);
1410 }
1411 }
1412 if (!stopped()) {
1413 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1414 }
1415 return NULL;
1416 }
1417
1418 //-----------------------------inline_string_indexOfChar-----------------------
1419 bool LibraryCallKit::inline_string_indexOfChar() {
1420 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1421 return false;
1422 }
1423 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1424 return false;
1425 }
1426 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1427 Node* src = argument(0); // byte[]
1428 Node* tgt = argument(1); // tgt is int ch
1429 Node* from_index = argument(2);
1430 Node* max = argument(3);
1431
1432 src = must_be_not_null(src, true);
1433 src = access_resolve(src, ACCESS_READ);
1434
1435 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1436 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1437 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1438
1439 // Range checks
1440 generate_string_range_check(src, src_offset, src_count, true);
1441 if (stopped()) {
1442 return true;
1443 }
1444
1445 RegionNode* region = new RegionNode(3);
1446 Node* phi = new PhiNode(region, TypeInt::INT);
1447
1448 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1449 C->set_has_split_ifs(true); // Has chance for split-if optimization
1450 _gvn.transform(result);
1451
1452 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1453 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1454
1455 Node* if_lt = generate_slow_guard(bol, NULL);
1456 if (if_lt != NULL) {
1457 // result == -1
1458 phi->init_req(2, result);
1459 region->init_req(2, if_lt);
1460 }
1461 if (!stopped()) {
1462 result = _gvn.transform(new AddINode(result, from_index));
1463 phi->init_req(1, result);
1464 region->init_req(1, control());
1465 }
1466 set_control(_gvn.transform(region));
1467 record_for_igvn(region);
1468 set_result(_gvn.transform(phi));
1469
1470 return true;
1471 }
1472 //---------------------------inline_string_copy---------------------
1473 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1474 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1475 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1476 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1477 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1478 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1479 bool LibraryCallKit::inline_string_copy(bool compress) {
1480 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1481 return false;
1482 }
1483 int nargs = 5; // 2 oops, 3 ints
1484 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1485
1486 Node* src = argument(0);
1487 Node* src_offset = argument(1);
1488 Node* dst = argument(2);
1489 Node* dst_offset = argument(3);
1490 Node* length = argument(4);
1491
1492 // Check for allocation before we add nodes that would confuse
1493 // tightly_coupled_allocation()
1494 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1495
1496 // Figure out the size and type of the elements we will be copying.
1497 const Type* src_type = src->Value(&_gvn);
1498 const Type* dst_type = dst->Value(&_gvn);
1499 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1500 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1501 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1502 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1503 "Unsupported array types for inline_string_copy");
1504
1505 src = must_be_not_null(src, true);
1506 dst = must_be_not_null(dst, true);
1507
1508 // Convert char[] offsets to byte[] offsets
1509 bool convert_src = (compress && src_elem == T_BYTE);
1510 bool convert_dst = (!compress && dst_elem == T_BYTE);
1511 if (convert_src) {
1512 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1513 } else if (convert_dst) {
1514 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1515 }
1516
1517 // Range checks
1518 generate_string_range_check(src, src_offset, length, convert_src);
1519 generate_string_range_check(dst, dst_offset, length, convert_dst);
1520 if (stopped()) {
1521 return true;
1522 }
1523
1524 src = access_resolve(src, ACCESS_READ);
1525 dst = access_resolve(dst, ACCESS_WRITE);
1526
1527 Node* src_start = array_element_address(src, src_offset, src_elem);
1528 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1529 // 'src_start' points to src array + scaled offset
1530 // 'dst_start' points to dst array + scaled offset
1531 Node* count = NULL;
1532 if (compress) {
1533 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1534 } else {
1535 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1536 }
1537
1538 if (alloc != NULL) {
1539 if (alloc->maybe_set_complete(&_gvn)) {
1540 // "You break it, you buy it."
1541 InitializeNode* init = alloc->initialization();
1542 assert(init->is_complete(), "we just did this");
1543 init->set_complete_with_arraycopy();
1544 assert(dst->is_CheckCastPP(), "sanity");
1545 assert(dst->in(0)->in(0) == init, "dest pinned");
1546 }
1547 // Do not let stores that initialize this object be reordered with
1548 // a subsequent store that would make this object accessible by
1549 // other threads.
1550 // Record what AllocateNode this StoreStore protects so that
1551 // escape analysis can go from the MemBarStoreStoreNode to the
1552 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1553 // based on the escape status of the AllocateNode.
1554 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1555 }
1556 if (compress) {
1557 set_result(_gvn.transform(count));
1558 }
1559 clear_upper_avx();
1560
1561 return true;
1562 }
1563
1564 #ifdef _LP64
1565 #define XTOP ,top() /*additional argument*/
1566 #else //_LP64
1567 #define XTOP /*no additional argument*/
1568 #endif //_LP64
1569
1570 //------------------------inline_string_toBytesU--------------------------
1571 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1572 bool LibraryCallKit::inline_string_toBytesU() {
1573 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1574 return false;
1575 }
1576 // Get the arguments.
1577 Node* value = argument(0);
1578 Node* offset = argument(1);
1579 Node* length = argument(2);
1580
1581 Node* newcopy = NULL;
1582
1583 // Set the original stack and the reexecute bit for the interpreter to reexecute
1584 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1585 { PreserveReexecuteState preexecs(this);
1586 jvms()->set_should_reexecute(true);
1587
1588 // Check if a null path was taken unconditionally.
1589 value = null_check(value);
1590
1591 RegionNode* bailout = new RegionNode(1);
1592 record_for_igvn(bailout);
1593
1594 // Range checks
1595 generate_negative_guard(offset, bailout);
1596 generate_negative_guard(length, bailout);
1597 generate_limit_guard(offset, length, load_array_length(value), bailout);
1598 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1599 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1600
1601 if (bailout->req() > 1) {
1602 PreserveJVMState pjvms(this);
1603 set_control(_gvn.transform(bailout));
1604 uncommon_trap(Deoptimization::Reason_intrinsic,
1605 Deoptimization::Action_maybe_recompile);
1606 }
1607 if (stopped()) {
1608 return true;
1609 }
1610
1611 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1612 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1613 newcopy = new_array(klass_node, size, 0); // no arguments to push
1614 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1615
1616 // Calculate starting addresses.
1617 value = access_resolve(value, ACCESS_READ);
1618 Node* src_start = array_element_address(value, offset, T_CHAR);
1619 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1620
1621 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1622 const TypeInt* toffset = gvn().type(offset)->is_int();
1623 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1624
1625 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1626 const char* copyfunc_name = "arraycopy";
1627 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1628 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1629 OptoRuntime::fast_arraycopy_Type(),
1630 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1631 src_start, dst_start, ConvI2X(length) XTOP);
1632 // Do not let reads from the cloned object float above the arraycopy.
1633 if (alloc != NULL) {
1634 if (alloc->maybe_set_complete(&_gvn)) {
1635 // "You break it, you buy it."
1636 InitializeNode* init = alloc->initialization();
1637 assert(init->is_complete(), "we just did this");
1638 init->set_complete_with_arraycopy();
1639 assert(newcopy->is_CheckCastPP(), "sanity");
1640 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1641 }
1642 // Do not let stores that initialize this object be reordered with
1643 // a subsequent store that would make this object accessible by
1644 // other threads.
1645 // Record what AllocateNode this StoreStore protects so that
1646 // escape analysis can go from the MemBarStoreStoreNode to the
1647 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1648 // based on the escape status of the AllocateNode.
1649 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1650 } else {
1651 insert_mem_bar(Op_MemBarCPUOrder);
1652 }
1653 } // original reexecute is set back here
1654
1655 C->set_has_split_ifs(true); // Has chance for split-if optimization
1656 if (!stopped()) {
1657 set_result(newcopy);
1658 }
1659 clear_upper_avx();
1660
1661 return true;
1662 }
1663
1664 //------------------------inline_string_getCharsU--------------------------
1665 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1666 bool LibraryCallKit::inline_string_getCharsU() {
1667 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1668 return false;
1669 }
1670
1671 // Get the arguments.
1672 Node* src = argument(0);
1673 Node* src_begin = argument(1);
1674 Node* src_end = argument(2); // exclusive offset (i < src_end)
1675 Node* dst = argument(3);
1676 Node* dst_begin = argument(4);
1677
1678 // Check for allocation before we add nodes that would confuse
1679 // tightly_coupled_allocation()
1680 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1681
1682 // Check if a null path was taken unconditionally.
1683 src = null_check(src);
1684 dst = null_check(dst);
1685 if (stopped()) {
1686 return true;
1687 }
1688
1689 // Get length and convert char[] offset to byte[] offset
1690 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1691 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1692
1693 // Range checks
1694 generate_string_range_check(src, src_begin, length, true);
1695 generate_string_range_check(dst, dst_begin, length, false);
1696 if (stopped()) {
1697 return true;
1698 }
1699
1700 if (!stopped()) {
1701 src = access_resolve(src, ACCESS_READ);
1702 dst = access_resolve(dst, ACCESS_WRITE);
1703
1704 // Calculate starting addresses.
1705 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1706 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1707
1708 // Check if array addresses are aligned to HeapWordSize
1709 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1710 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1711 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1712 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1713
1714 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1715 const char* copyfunc_name = "arraycopy";
1716 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1717 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1718 OptoRuntime::fast_arraycopy_Type(),
1719 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1720 src_start, dst_start, ConvI2X(length) XTOP);
1721 // Do not let reads from the cloned object float above the arraycopy.
1722 if (alloc != NULL) {
1723 if (alloc->maybe_set_complete(&_gvn)) {
1724 // "You break it, you buy it."
1725 InitializeNode* init = alloc->initialization();
1726 assert(init->is_complete(), "we just did this");
1727 init->set_complete_with_arraycopy();
1728 assert(dst->is_CheckCastPP(), "sanity");
1729 assert(dst->in(0)->in(0) == init, "dest pinned");
1730 }
1731 // Do not let stores that initialize this object be reordered with
1732 // a subsequent store that would make this object accessible by
1733 // other threads.
1734 // Record what AllocateNode this StoreStore protects so that
1735 // escape analysis can go from the MemBarStoreStoreNode to the
1736 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1737 // based on the escape status of the AllocateNode.
1738 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1739 } else {
1740 insert_mem_bar(Op_MemBarCPUOrder);
1741 }
1742 }
1743
1744 C->set_has_split_ifs(true); // Has chance for split-if optimization
1745 return true;
1746 }
1747
1748 //----------------------inline_string_char_access----------------------------
1749 // Store/Load char to/from byte[] array.
1750 // static void StringUTF16.putChar(byte[] val, int index, int c)
1751 // static char StringUTF16.getChar(byte[] val, int index)
1752 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1753 Node* value = argument(0);
1754 Node* index = argument(1);
1755 Node* ch = is_store ? argument(2) : NULL;
1756
1757 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1758 // correctly requires matched array shapes.
1759 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1760 "sanity: byte[] and char[] bases agree");
1761 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1762 "sanity: byte[] and char[] scales agree");
1763
1764 // Bail when getChar over constants is requested: constant folding would
1765 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1766 // Java method would constant fold nicely instead.
1767 if (!is_store && value->is_Con() && index->is_Con()) {
1768 return false;
1769 }
1770
1771 value = must_be_not_null(value, true);
1772 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1773
1774 Node* adr = array_element_address(value, index, T_CHAR);
1775 if (adr->is_top()) {
1776 return false;
1777 }
1778 if (is_store) {
1779 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1780 } else {
1781 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1782 set_result(ch);
1783 }
1784 return true;
1785 }
1786
1787 //--------------------------round_double_node--------------------------------
1788 // Round a double node if necessary.
1789 Node* LibraryCallKit::round_double_node(Node* n) {
1790 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1791 n = _gvn.transform(new RoundDoubleNode(0, n));
1792 return n;
1793 }
1794
1795 //------------------------------inline_math-----------------------------------
1796 // public static double Math.abs(double)
1797 // public static double Math.sqrt(double)
1798 // public static double Math.log(double)
1799 // public static double Math.log10(double)
1800 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1801 Node* arg = round_double_node(argument(0));
1802 Node* n = NULL;
1803 switch (id) {
1804 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1805 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1806 default: fatal_unexpected_iid(id); break;
1807 }
1808 set_result(_gvn.transform(n));
1809 return true;
1810 }
1811
1812 //------------------------------inline_math-----------------------------------
1813 // public static float Math.abs(float)
1814 // public static int Math.abs(int)
1815 // public static long Math.abs(long)
1816 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1817 Node* arg = argument(0);
1818 Node* n = NULL;
1819 switch (id) {
1820 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1821 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1822 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1823 default: fatal_unexpected_iid(id); break;
1824 }
1825 set_result(_gvn.transform(n));
1826 return true;
1827 }
1828
1829 //------------------------------runtime_math-----------------------------
1830 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1831 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1832 "must be (DD)D or (D)D type");
1833
1834 // Inputs
1835 Node* a = round_double_node(argument(0));
1836 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1837
1838 const TypePtr* no_memory_effects = NULL;
1839 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1840 no_memory_effects,
1841 a, top(), b, b ? top() : NULL);
1842 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1843 #ifdef ASSERT
1844 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1845 assert(value_top == top(), "second value must be top");
1846 #endif
1847
1848 set_result(value);
1849 return true;
1850 }
1851
1852 //------------------------------inline_math_native-----------------------------
1853 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1854 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1855 switch (id) {
1856 // These intrinsics are not properly supported on all hardware
1857 case vmIntrinsics::_dsin:
1858 return StubRoutines::dsin() != NULL ?
1859 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1860 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1861 case vmIntrinsics::_dcos:
1862 return StubRoutines::dcos() != NULL ?
1863 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1864 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1865 case vmIntrinsics::_dtan:
1866 return StubRoutines::dtan() != NULL ?
1867 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1868 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1869 case vmIntrinsics::_dlog:
1870 return StubRoutines::dlog() != NULL ?
1871 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1872 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1873 case vmIntrinsics::_dlog10:
1874 return StubRoutines::dlog10() != NULL ?
1875 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1876 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1877
1878 // These intrinsics are supported on all hardware
1879 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1880 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1881 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1882 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1883 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1884
1885 case vmIntrinsics::_dexp:
1886 return StubRoutines::dexp() != NULL ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1889 case vmIntrinsics::_dpow: {
1890 Node* exp = round_double_node(argument(2));
1891 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1892 if (d != NULL && d->getd() == 2.0) {
1893 // Special case: pow(x, 2.0) => x * x
1894 Node* base = round_double_node(argument(0));
1895 set_result(_gvn.transform(new MulDNode(base, base)));
1896 return true;
1897 }
1898 return StubRoutines::dpow() != NULL ?
1899 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1900 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1901 }
1902 #undef FN_PTR
1903
1904 // These intrinsics are not yet correctly implemented
1905 case vmIntrinsics::_datan2:
1906 return false;
1907
1908 default:
1909 fatal_unexpected_iid(id);
1910 return false;
1911 }
1912 }
1913
1914 static bool is_simple_name(Node* n) {
1915 return (n->req() == 1 // constant
1916 || (n->is_Type() && n->as_Type()->type()->singleton())
1917 || n->is_Proj() // parameter or return value
1918 || n->is_Phi() // local of some sort
1919 );
1920 }
1921
1922 //----------------------------inline_notify-----------------------------------*
1923 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1924 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1925 address func;
1926 if (id == vmIntrinsics::_notify) {
1927 func = OptoRuntime::monitor_notify_Java();
1928 } else {
1929 func = OptoRuntime::monitor_notifyAll_Java();
1930 }
1931 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1932 make_slow_call_ex(call, env()->Throwable_klass(), false);
1933 return true;
1934 }
1935
1936
1937 //----------------------------inline_min_max-----------------------------------
1938 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1939 set_result(generate_min_max(id, argument(0), argument(1)));
1940 return true;
1941 }
1942
1943 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1944 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1945 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1946 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1947 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1948
1949 {
1950 PreserveJVMState pjvms(this);
1951 PreserveReexecuteState preexecs(this);
1952 jvms()->set_should_reexecute(true);
1953
1954 set_control(slow_path);
1955 set_i_o(i_o());
1956
1957 uncommon_trap(Deoptimization::Reason_intrinsic,
1958 Deoptimization::Action_none);
1959 }
1960
1961 set_control(fast_path);
1962 set_result(math);
1963 }
1964
1965 template <typename OverflowOp>
1966 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1967 typedef typename OverflowOp::MathOp MathOp;
1968
1969 MathOp* mathOp = new MathOp(arg1, arg2);
1970 Node* operation = _gvn.transform( mathOp );
1971 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1972 inline_math_mathExact(operation, ofcheck);
1973 return true;
1974 }
1975
1976 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1977 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1978 }
1979
1980 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1981 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1982 }
1983
1984 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1985 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1986 }
1987
1988 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1989 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1990 }
1991
1992 bool LibraryCallKit::inline_math_negateExactI() {
1993 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1994 }
1995
1996 bool LibraryCallKit::inline_math_negateExactL() {
1997 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1998 }
1999
2000 bool LibraryCallKit::inline_math_multiplyExactI() {
2001 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2002 }
2003
2004 bool LibraryCallKit::inline_math_multiplyExactL() {
2005 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2006 }
2007
2008 bool LibraryCallKit::inline_math_multiplyHigh() {
2009 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2010 return true;
2011 }
2012
2013 Node*
2014 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2015 // These are the candidate return value:
2016 Node* xvalue = x0;
2017 Node* yvalue = y0;
2018
2019 if (xvalue == yvalue) {
2020 return xvalue;
2021 }
2022
2023 bool want_max = (id == vmIntrinsics::_max);
2024
2025 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2026 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2027 if (txvalue == NULL || tyvalue == NULL) return top();
2028 // This is not really necessary, but it is consistent with a
2029 // hypothetical MaxINode::Value method:
2030 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2031
2032 // %%% This folding logic should (ideally) be in a different place.
2033 // Some should be inside IfNode, and there to be a more reliable
2034 // transformation of ?: style patterns into cmoves. We also want
2035 // more powerful optimizations around cmove and min/max.
2036
2037 // Try to find a dominating comparison of these guys.
2038 // It can simplify the index computation for Arrays.copyOf
2039 // and similar uses of System.arraycopy.
2040 // First, compute the normalized version of CmpI(x, y).
2041 int cmp_op = Op_CmpI;
2042 Node* xkey = xvalue;
2043 Node* ykey = yvalue;
2044 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2045 if (ideal_cmpxy->is_Cmp()) {
2046 // E.g., if we have CmpI(length - offset, count),
2047 // it might idealize to CmpI(length, count + offset)
2048 cmp_op = ideal_cmpxy->Opcode();
2049 xkey = ideal_cmpxy->in(1);
2050 ykey = ideal_cmpxy->in(2);
2051 }
2052
2053 // Start by locating any relevant comparisons.
2054 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2055 Node* cmpxy = NULL;
2056 Node* cmpyx = NULL;
2057 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2058 Node* cmp = start_from->fast_out(k);
2059 if (cmp->outcnt() > 0 && // must have prior uses
2060 cmp->in(0) == NULL && // must be context-independent
2061 cmp->Opcode() == cmp_op) { // right kind of compare
2062 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2063 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2064 }
2065 }
2066
2067 const int NCMPS = 2;
2068 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2069 int cmpn;
2070 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2071 if (cmps[cmpn] != NULL) break; // find a result
2072 }
2073 if (cmpn < NCMPS) {
2074 // Look for a dominating test that tells us the min and max.
2075 int depth = 0; // Limit search depth for speed
2076 Node* dom = control();
2077 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2078 if (++depth >= 100) break;
2079 Node* ifproj = dom;
2080 if (!ifproj->is_Proj()) continue;
2081 Node* iff = ifproj->in(0);
2082 if (!iff->is_If()) continue;
2083 Node* bol = iff->in(1);
2084 if (!bol->is_Bool()) continue;
2085 Node* cmp = bol->in(1);
2086 if (cmp == NULL) continue;
2087 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2088 if (cmps[cmpn] == cmp) break;
2089 if (cmpn == NCMPS) continue;
2090 BoolTest::mask btest = bol->as_Bool()->_test._test;
2091 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2092 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2093 // At this point, we know that 'x btest y' is true.
2094 switch (btest) {
2095 case BoolTest::eq:
2096 // They are proven equal, so we can collapse the min/max.
2097 // Either value is the answer. Choose the simpler.
2098 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2099 return yvalue;
2100 return xvalue;
2101 case BoolTest::lt: // x < y
2102 case BoolTest::le: // x <= y
2103 return (want_max ? yvalue : xvalue);
2104 case BoolTest::gt: // x > y
2105 case BoolTest::ge: // x >= y
2106 return (want_max ? xvalue : yvalue);
2107 default:
2108 break;
2109 }
2110 }
2111 }
2112
2113 // We failed to find a dominating test.
2114 // Let's pick a test that might GVN with prior tests.
2115 Node* best_bol = NULL;
2116 BoolTest::mask best_btest = BoolTest::illegal;
2117 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2118 Node* cmp = cmps[cmpn];
2119 if (cmp == NULL) continue;
2120 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2121 Node* bol = cmp->fast_out(j);
2122 if (!bol->is_Bool()) continue;
2123 BoolTest::mask btest = bol->as_Bool()->_test._test;
2124 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2125 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2126 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2127 best_bol = bol->as_Bool();
2128 best_btest = btest;
2129 }
2130 }
2131 }
2132
2133 Node* answer_if_true = NULL;
2134 Node* answer_if_false = NULL;
2135 switch (best_btest) {
2136 default:
2137 if (cmpxy == NULL)
2138 cmpxy = ideal_cmpxy;
2139 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2140 // and fall through:
2141 case BoolTest::lt: // x < y
2142 case BoolTest::le: // x <= y
2143 answer_if_true = (want_max ? yvalue : xvalue);
2144 answer_if_false = (want_max ? xvalue : yvalue);
2145 break;
2146 case BoolTest::gt: // x > y
2147 case BoolTest::ge: // x >= y
2148 answer_if_true = (want_max ? xvalue : yvalue);
2149 answer_if_false = (want_max ? yvalue : xvalue);
2150 break;
2151 }
2152
2153 jint hi, lo;
2154 if (want_max) {
2155 // We can sharpen the minimum.
2156 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2157 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2158 } else {
2159 // We can sharpen the maximum.
2160 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2161 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2162 }
2163
2164 // Use a flow-free graph structure, to avoid creating excess control edges
2165 // which could hinder other optimizations.
2166 // Since Math.min/max is often used with arraycopy, we want
2167 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2168 Node* cmov = CMoveNode::make(NULL, best_bol,
2169 answer_if_false, answer_if_true,
2170 TypeInt::make(lo, hi, widen));
2171
2172 return _gvn.transform(cmov);
2173
2174 /*
2175 // This is not as desirable as it may seem, since Min and Max
2176 // nodes do not have a full set of optimizations.
2177 // And they would interfere, anyway, with 'if' optimizations
2178 // and with CMoveI canonical forms.
2179 switch (id) {
2180 case vmIntrinsics::_min:
2181 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2182 case vmIntrinsics::_max:
2183 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2184 default:
2185 ShouldNotReachHere();
2186 }
2187 */
2188 }
2189
2190 inline int
2191 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2192 const TypePtr* base_type = TypePtr::NULL_PTR;
2193 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2194 if (base_type == NULL) {
2195 // Unknown type.
2196 return Type::AnyPtr;
2197 } else if (base_type == TypePtr::NULL_PTR) {
2198 // Since this is a NULL+long form, we have to switch to a rawptr.
2199 base = _gvn.transform(new CastX2PNode(offset));
2200 offset = MakeConX(0);
2201 return Type::RawPtr;
2202 } else if (base_type->base() == Type::RawPtr) {
2203 return Type::RawPtr;
2204 } else if (base_type->isa_oopptr()) {
2205 // Base is never null => always a heap address.
2206 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2207 return Type::OopPtr;
2208 }
2209 // Offset is small => always a heap address.
2210 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2211 if (offset_type != NULL &&
2212 base_type->offset() == 0 && // (should always be?)
2213 offset_type->_lo >= 0 &&
2214 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2215 return Type::OopPtr;
2216 } else if (type == T_OBJECT) {
2217 // off heap access to an oop doesn't make any sense. Has to be on
2218 // heap.
2219 return Type::OopPtr;
2220 }
2221 // Otherwise, it might either be oop+off or NULL+addr.
2222 return Type::AnyPtr;
2223 } else {
2224 // No information:
2225 return Type::AnyPtr;
2226 }
2227 }
2228
2229 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2230 Node* uncasted_base = base;
2231 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2232 if (kind == Type::RawPtr) {
2233 return basic_plus_adr(top(), uncasted_base, offset);
2234 } else if (kind == Type::AnyPtr) {
2235 assert(base == uncasted_base, "unexpected base change");
2236 if (can_cast) {
2237 if (!_gvn.type(base)->speculative_maybe_null() &&
2238 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2239 // According to profiling, this access is always on
2240 // heap. Casting the base to not null and thus avoiding membars
2241 // around the access should allow better optimizations
2242 Node* null_ctl = top();
2243 base = null_check_oop(base, &null_ctl, true, true, true);
2244 assert(null_ctl->is_top(), "no null control here");
2245 return basic_plus_adr(base, offset);
2246 } else if (_gvn.type(base)->speculative_always_null() &&
2247 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2248 // According to profiling, this access is always off
2249 // heap.
2250 base = null_assert(base);
2251 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2252 offset = MakeConX(0);
2253 return basic_plus_adr(top(), raw_base, offset);
2254 }
2255 }
2256 // We don't know if it's an on heap or off heap access. Fall back
2257 // to raw memory access.
2258 base = access_resolve(base, decorators);
2259 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2260 return basic_plus_adr(top(), raw, offset);
2261 } else {
2262 assert(base == uncasted_base, "unexpected base change");
2263 // We know it's an on heap access so base can't be null
2264 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2265 base = must_be_not_null(base, true);
2266 }
2267 return basic_plus_adr(base, offset);
2268 }
2269 }
2270
2271 //--------------------------inline_number_methods-----------------------------
2272 // inline int Integer.numberOfLeadingZeros(int)
2273 // inline int Long.numberOfLeadingZeros(long)
2274 //
2275 // inline int Integer.numberOfTrailingZeros(int)
2276 // inline int Long.numberOfTrailingZeros(long)
2277 //
2278 // inline int Integer.bitCount(int)
2279 // inline int Long.bitCount(long)
2280 //
2281 // inline char Character.reverseBytes(char)
2282 // inline short Short.reverseBytes(short)
2283 // inline int Integer.reverseBytes(int)
2284 // inline long Long.reverseBytes(long)
2285 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2286 Node* arg = argument(0);
2287 Node* n = NULL;
2288 switch (id) {
2289 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2290 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2291 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2292 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2293 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2294 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2295 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2296 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2297 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2298 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2299 default: fatal_unexpected_iid(id); break;
2300 }
2301 set_result(_gvn.transform(n));
2302 return true;
2303 }
2304
2305 //----------------------------inline_unsafe_access----------------------------
2306
2307 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2308 // Attempt to infer a sharper value type from the offset and base type.
2309 ciKlass* sharpened_klass = NULL;
2310
2311 // See if it is an instance field, with an object type.
2312 if (alias_type->field() != NULL) {
2313 if (alias_type->field()->type()->is_klass()) {
2314 sharpened_klass = alias_type->field()->type()->as_klass();
2315 }
2316 }
2317
2318 // See if it is a narrow oop array.
2319 if (adr_type->isa_aryptr()) {
2320 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2321 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2322 if (elem_type != NULL) {
2323 sharpened_klass = elem_type->klass();
2324 }
2325 }
2326 }
2327
2328 // The sharpened class might be unloaded if there is no class loader
2329 // contraint in place.
2330 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2331 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2332
2333 #ifndef PRODUCT
2334 if (C->print_intrinsics() || C->print_inlining()) {
2335 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2336 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2337 }
2338 #endif
2339 // Sharpen the value type.
2340 return tjp;
2341 }
2342 return NULL;
2343 }
2344
2345 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2346 switch (kind) {
2347 case Relaxed:
2348 return MO_UNORDERED;
2349 case Opaque:
2350 return MO_RELAXED;
2351 case Acquire:
2352 return MO_ACQUIRE;
2353 case Release:
2354 return MO_RELEASE;
2355 case Volatile:
2356 return MO_SEQ_CST;
2357 default:
2358 ShouldNotReachHere();
2359 return 0;
2360 }
2361 }
2362
2363 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2364 if (callee()->is_static()) return false; // caller must have the capability!
2365 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2366 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2367 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2368 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2369
2370 if (type == T_OBJECT || type == T_ARRAY) {
2371 decorators |= ON_UNKNOWN_OOP_REF;
2372 }
2373
2374 if (unaligned) {
2375 decorators |= C2_UNALIGNED;
2376 }
2377
2378 #ifndef PRODUCT
2379 {
2380 ResourceMark rm;
2381 // Check the signatures.
2382 ciSignature* sig = callee()->signature();
2383 #ifdef ASSERT
2384 if (!is_store) {
2385 // Object getReference(Object base, int/long offset), etc.
2386 BasicType rtype = sig->return_type()->basic_type();
2387 assert(rtype == type, "getter must return the expected value");
2388 assert(sig->count() == 2, "oop getter has 2 arguments");
2389 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2390 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2391 } else {
2392 // void putReference(Object base, int/long offset, Object x), etc.
2393 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2394 assert(sig->count() == 3, "oop putter has 3 arguments");
2395 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2396 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2397 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2398 assert(vtype == type, "putter must accept the expected value");
2399 }
2400 #endif // ASSERT
2401 }
2402 #endif //PRODUCT
2403
2404 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2405
2406 Node* receiver = argument(0); // type: oop
2407
2408 // Build address expression.
2409 Node* adr;
2410 Node* heap_base_oop = top();
2411 Node* offset = top();
2412 Node* val;
2413
2414 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2415 Node* base = argument(1); // type: oop
2416 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2417 offset = argument(2); // type: long
2418 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2419 // to be plain byte offsets, which are also the same as those accepted
2420 // by oopDesc::field_addr.
2421 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2422 "fieldOffset must be byte-scaled");
2423 // 32-bit machines ignore the high half!
2424 offset = ConvL2X(offset);
2425 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2426
2427 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2428 heap_base_oop = base;
2429 } else if (type == T_OBJECT) {
2430 return false; // off-heap oop accesses are not supported
2431 }
2432
2433 // Can base be NULL? Otherwise, always on-heap access.
2434 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2435
2436 if (!can_access_non_heap) {
2437 decorators |= IN_HEAP;
2438 }
2439
2440 val = is_store ? argument(4) : NULL;
2441
2442 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2443
2444 // Try to categorize the address.
2445 Compile::AliasType* alias_type = C->alias_type(adr_type);
2446 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2447
2448 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2449 alias_type->adr_type() == TypeAryPtr::RANGE) {
2450 return false; // not supported
2451 }
2452
2453 bool mismatched = false;
2454 BasicType bt = alias_type->basic_type();
2455 if (bt != T_ILLEGAL) {
2456 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2457 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2458 // Alias type doesn't differentiate between byte[] and boolean[]).
2459 // Use address type to get the element type.
2460 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2461 }
2462 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2463 // accessing an array field with getReference is not a mismatch
2464 bt = T_OBJECT;
2465 }
2466 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2467 // Don't intrinsify mismatched object accesses
2468 return false;
2469 }
2470 mismatched = (bt != type);
2471 } else if (alias_type->adr_type()->isa_oopptr()) {
2472 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2473 }
2474
2475 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2476
2477 if (mismatched) {
2478 decorators |= C2_MISMATCHED;
2479 }
2480
2481 // First guess at the value type.
2482 const Type *value_type = Type::get_const_basic_type(type);
2483
2484 // Figure out the memory ordering.
2485 decorators |= mo_decorator_for_access_kind(kind);
2486
2487 if (!is_store && type == T_OBJECT) {
2488 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2489 if (tjp != NULL) {
2490 value_type = tjp;
2491 }
2492 }
2493
2494 receiver = null_check(receiver);
2495 if (stopped()) {
2496 return true;
2497 }
2498 // Heap pointers get a null-check from the interpreter,
2499 // as a courtesy. However, this is not guaranteed by Unsafe,
2500 // and it is not possible to fully distinguish unintended nulls
2501 // from intended ones in this API.
2502
2503 if (!is_store) {
2504 Node* p = NULL;
2505 // Try to constant fold a load from a constant field
2506 ciField* field = alias_type->field();
2507 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2508 // final or stable field
2509 p = make_constant_from_field(field, heap_base_oop);
2510 }
2511
2512 if (p == NULL) { // Could not constant fold the load
2513 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2514 // Normalize the value returned by getBoolean in the following cases
2515 if (type == T_BOOLEAN &&
2516 (mismatched ||
2517 heap_base_oop == top() || // - heap_base_oop is NULL or
2518 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2519 // and the unsafe access is made to large offset
2520 // (i.e., larger than the maximum offset necessary for any
2521 // field access)
2522 ) {
2523 IdealKit ideal = IdealKit(this);
2524 #define __ ideal.
2525 IdealVariable normalized_result(ideal);
2526 __ declarations_done();
2527 __ set(normalized_result, p);
2528 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2529 __ set(normalized_result, ideal.ConI(1));
2530 ideal.end_if();
2531 final_sync(ideal);
2532 p = __ value(normalized_result);
2533 #undef __
2534 }
2535 }
2536 if (type == T_ADDRESS) {
2537 p = gvn().transform(new CastP2XNode(NULL, p));
2538 p = ConvX2UL(p);
2539 }
2540 // The load node has the control of the preceding MemBarCPUOrder. All
2541 // following nodes will have the control of the MemBarCPUOrder inserted at
2542 // the end of this method. So, pushing the load onto the stack at a later
2543 // point is fine.
2544 set_result(p);
2545 } else {
2546 if (bt == T_ADDRESS) {
2547 // Repackage the long as a pointer.
2548 val = ConvL2X(val);
2549 val = gvn().transform(new CastX2PNode(val));
2550 }
2551 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2552 }
2553
2554 return true;
2555 }
2556
2557 //----------------------------inline_unsafe_load_store----------------------------
2558 // This method serves a couple of different customers (depending on LoadStoreKind):
2559 //
2560 // LS_cmp_swap:
2561 //
2562 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2563 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2564 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2565 //
2566 // LS_cmp_swap_weak:
2567 //
2568 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2569 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2570 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2571 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2572 //
2573 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2574 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2575 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2576 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2577 //
2578 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2579 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2580 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2581 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2582 //
2583 // LS_cmp_exchange:
2584 //
2585 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2586 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2587 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2588 //
2589 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2590 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2591 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2592 //
2593 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2594 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2595 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2596 //
2597 // LS_get_add:
2598 //
2599 // int getAndAddInt( Object o, long offset, int delta)
2600 // long getAndAddLong(Object o, long offset, long delta)
2601 //
2602 // LS_get_set:
2603 //
2604 // int getAndSet(Object o, long offset, int newValue)
2605 // long getAndSet(Object o, long offset, long newValue)
2606 // Object getAndSet(Object o, long offset, Object newValue)
2607 //
2608 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2609 // This basic scheme here is the same as inline_unsafe_access, but
2610 // differs in enough details that combining them would make the code
2611 // overly confusing. (This is a true fact! I originally combined
2612 // them, but even I was confused by it!) As much code/comments as
2613 // possible are retained from inline_unsafe_access though to make
2614 // the correspondences clearer. - dl
2615
2616 if (callee()->is_static()) return false; // caller must have the capability!
2617
2618 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2619 decorators |= mo_decorator_for_access_kind(access_kind);
2620
2621 #ifndef PRODUCT
2622 BasicType rtype;
2623 {
2624 ResourceMark rm;
2625 // Check the signatures.
2626 ciSignature* sig = callee()->signature();
2627 rtype = sig->return_type()->basic_type();
2628 switch(kind) {
2629 case LS_get_add:
2630 case LS_get_set: {
2631 // Check the signatures.
2632 #ifdef ASSERT
2633 assert(rtype == type, "get and set must return the expected type");
2634 assert(sig->count() == 3, "get and set has 3 arguments");
2635 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2636 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2637 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2638 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2639 #endif // ASSERT
2640 break;
2641 }
2642 case LS_cmp_swap:
2643 case LS_cmp_swap_weak: {
2644 // Check the signatures.
2645 #ifdef ASSERT
2646 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2647 assert(sig->count() == 4, "CAS has 4 arguments");
2648 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2649 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2650 #endif // ASSERT
2651 break;
2652 }
2653 case LS_cmp_exchange: {
2654 // Check the signatures.
2655 #ifdef ASSERT
2656 assert(rtype == type, "CAS must return the expected type");
2657 assert(sig->count() == 4, "CAS has 4 arguments");
2658 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2659 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2660 #endif // ASSERT
2661 break;
2662 }
2663 default:
2664 ShouldNotReachHere();
2665 }
2666 }
2667 #endif //PRODUCT
2668
2669 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2670
2671 // Get arguments:
2672 Node* receiver = NULL;
2673 Node* base = NULL;
2674 Node* offset = NULL;
2675 Node* oldval = NULL;
2676 Node* newval = NULL;
2677 switch(kind) {
2678 case LS_cmp_swap:
2679 case LS_cmp_swap_weak:
2680 case LS_cmp_exchange: {
2681 const bool two_slot_type = type2size[type] == 2;
2682 receiver = argument(0); // type: oop
2683 base = argument(1); // type: oop
2684 offset = argument(2); // type: long
2685 oldval = argument(4); // type: oop, int, or long
2686 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2687 break;
2688 }
2689 case LS_get_add:
2690 case LS_get_set: {
2691 receiver = argument(0); // type: oop
2692 base = argument(1); // type: oop
2693 offset = argument(2); // type: long
2694 oldval = NULL;
2695 newval = argument(4); // type: oop, int, or long
2696 break;
2697 }
2698 default:
2699 ShouldNotReachHere();
2700 }
2701
2702 // Build field offset expression.
2703 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2704 // to be plain byte offsets, which are also the same as those accepted
2705 // by oopDesc::field_addr.
2706 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2707 // 32-bit machines ignore the high half of long offsets
2708 offset = ConvL2X(offset);
2709 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2710 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2711
2712 Compile::AliasType* alias_type = C->alias_type(adr_type);
2713 BasicType bt = alias_type->basic_type();
2714 if (bt != T_ILLEGAL &&
2715 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2716 // Don't intrinsify mismatched object accesses.
2717 return false;
2718 }
2719
2720 // For CAS, unlike inline_unsafe_access, there seems no point in
2721 // trying to refine types. Just use the coarse types here.
2722 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2723 const Type *value_type = Type::get_const_basic_type(type);
2724
2725 switch (kind) {
2726 case LS_get_set:
2727 case LS_cmp_exchange: {
2728 if (type == T_OBJECT) {
2729 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2730 if (tjp != NULL) {
2731 value_type = tjp;
2732 }
2733 }
2734 break;
2735 }
2736 case LS_cmp_swap:
2737 case LS_cmp_swap_weak:
2738 case LS_get_add:
2739 break;
2740 default:
2741 ShouldNotReachHere();
2742 }
2743
2744 // Null check receiver.
2745 receiver = null_check(receiver);
2746 if (stopped()) {
2747 return true;
2748 }
2749
2750 int alias_idx = C->get_alias_index(adr_type);
2751
2752 if (type == T_OBJECT || type == T_ARRAY) {
2753 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2754
2755 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2756 // could be delayed during Parse (for example, in adjust_map_after_if()).
2757 // Execute transformation here to avoid barrier generation in such case.
2758 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2759 newval = _gvn.makecon(TypePtr::NULL_PTR);
2760
2761 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2762 // Refine the value to a null constant, when it is known to be null
2763 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2764 }
2765 }
2766
2767 Node* result = NULL;
2768 switch (kind) {
2769 case LS_cmp_exchange: {
2770 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2771 oldval, newval, value_type, type, decorators);
2772 break;
2773 }
2774 case LS_cmp_swap_weak:
2775 decorators |= C2_WEAK_CMPXCHG;
2776 case LS_cmp_swap: {
2777 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2778 oldval, newval, value_type, type, decorators);
2779 break;
2780 }
2781 case LS_get_set: {
2782 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2783 newval, value_type, type, decorators);
2784 break;
2785 }
2786 case LS_get_add: {
2787 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2788 newval, value_type, type, decorators);
2789 break;
2790 }
2791 default:
2792 ShouldNotReachHere();
2793 }
2794
2795 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2796 set_result(result);
2797 return true;
2798 }
2799
2800 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2801 // Regardless of form, don't allow previous ld/st to move down,
2802 // then issue acquire, release, or volatile mem_bar.
2803 insert_mem_bar(Op_MemBarCPUOrder);
2804 switch(id) {
2805 case vmIntrinsics::_loadFence:
2806 insert_mem_bar(Op_LoadFence);
2807 return true;
2808 case vmIntrinsics::_storeFence:
2809 insert_mem_bar(Op_StoreFence);
2810 return true;
2811 case vmIntrinsics::_fullFence:
2812 insert_mem_bar(Op_MemBarVolatile);
2813 return true;
2814 default:
2815 fatal_unexpected_iid(id);
2816 return false;
2817 }
2818 }
2819
2820 bool LibraryCallKit::inline_onspinwait() {
2821 insert_mem_bar(Op_OnSpinWait);
2822 return true;
2823 }
2824
2825 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2826 if (!kls->is_Con()) {
2827 return true;
2828 }
2829 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2830 if (klsptr == NULL) {
2831 return true;
2832 }
2833 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2834 // don't need a guard for a klass that is already initialized
2835 return !ik->is_initialized();
2836 }
2837
2838 //----------------------------inline_unsafe_allocate---------------------------
2839 // public native Object Unsafe.allocateInstance(Class<?> cls);
2840 bool LibraryCallKit::inline_unsafe_allocate() {
2841 if (callee()->is_static()) return false; // caller must have the capability!
2842
2843 null_check_receiver(); // null-check, then ignore
2844 Node* cls = null_check(argument(1));
2845 if (stopped()) return true;
2846
2847 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2848 kls = null_check(kls);
2849 if (stopped()) return true; // argument was like int.class
2850
2851 Node* test = NULL;
2852 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2853 // Note: The argument might still be an illegal value like
2854 // Serializable.class or Object[].class. The runtime will handle it.
2855 // But we must make an explicit check for initialization.
2856 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2857 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2858 // can generate code to load it as unsigned byte.
2859 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2860 Node* bits = intcon(InstanceKlass::fully_initialized);
2861 test = _gvn.transform(new SubINode(inst, bits));
2862 // The 'test' is non-zero if we need to take a slow path.
2863 }
2864
2865 Node* obj = new_instance(kls, test);
2866 set_result(obj);
2867 return true;
2868 }
2869
2870 //------------------------inline_native_time_funcs--------------
2871 // inline code for System.currentTimeMillis() and System.nanoTime()
2872 // these have the same type and signature
2873 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2874 const TypeFunc* tf = OptoRuntime::void_long_Type();
2875 const TypePtr* no_memory_effects = NULL;
2876 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2877 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2878 #ifdef ASSERT
2879 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2880 assert(value_top == top(), "second value must be top");
2881 #endif
2882 set_result(value);
2883 return true;
2884 }
2885
2886 #ifdef JFR_HAVE_INTRINSICS
2887
2888 /*
2889 * oop -> myklass
2890 * myklass->trace_id |= USED
2891 * return myklass->trace_id & ~0x3
2892 */
2893 bool LibraryCallKit::inline_native_classID() {
2894 Node* cls = null_check(argument(0), T_OBJECT);
2895 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2896 kls = null_check(kls, T_OBJECT);
2897
2898 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2899 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2900 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2901
2902 Node* clsused = longcon(0x01l); // set the class bit
2903 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2904 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2905 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2906
2907 #ifdef TRACE_ID_META_BITS
2908 Node* mbits = longcon(~TRACE_ID_META_BITS);
2909 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2910 #endif
2911 #ifdef TRACE_ID_SHIFT
2912 Node* cbits = intcon(TRACE_ID_SHIFT);
2913 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2914 #endif
2915
2916 set_result(tvalue);
2917 return true;
2918
2919 }
2920
2921 bool LibraryCallKit::inline_native_getEventWriter() {
2922 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2923
2924 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2925 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2926
2927 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2928
2929 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2930 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2931
2932 IfNode* iff_jobj_null =
2933 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2934
2935 enum { _normal_path = 1,
2936 _null_path = 2,
2937 PATH_LIMIT };
2938
2939 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2940 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
2941
2942 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2943 result_rgn->init_req(_null_path, jobj_is_null);
2944 result_val->init_req(_null_path, null());
2945
2946 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2947 set_control(jobj_is_not_null);
2948 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
2949 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
2950 result_rgn->init_req(_normal_path, control());
2951 result_val->init_req(_normal_path, res);
2952
2953 set_result(result_rgn, result_val);
2954
2955 return true;
2956 }
2957
2958 #endif // JFR_HAVE_INTRINSICS
2959
2960 //------------------------inline_native_currentThread------------------
2961 bool LibraryCallKit::inline_native_currentThread() {
2962 Node* junk = NULL;
2963 set_result(generate_current_thread(junk));
2964 return true;
2965 }
2966
2967 //------------------------inline_native_isInterrupted------------------
2968 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
2969 bool LibraryCallKit::inline_native_isInterrupted() {
2970 // Add a fast path to t.isInterrupted(clear_int):
2971 // (t == Thread.current() &&
2972 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
2973 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
2974 // So, in the common case that the interrupt bit is false,
2975 // we avoid making a call into the VM. Even if the interrupt bit
2976 // is true, if the clear_int argument is false, we avoid the VM call.
2977 // However, if the receiver is not currentThread, we must call the VM,
2978 // because there must be some locking done around the operation.
2979
2980 // We only go to the fast case code if we pass two guards.
2981 // Paths which do not pass are accumulated in the slow_region.
2982
2983 enum {
2984 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
2985 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
2986 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
2987 PATH_LIMIT
2988 };
2989
2990 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
2991 // out of the function.
2992 insert_mem_bar(Op_MemBarCPUOrder);
2993
2994 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2995 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
2996
2997 RegionNode* slow_region = new RegionNode(1);
2998 record_for_igvn(slow_region);
2999
3000 // (a) Receiving thread must be the current thread.
3001 Node* rec_thr = argument(0);
3002 Node* tls_ptr = NULL;
3003 Node* cur_thr = generate_current_thread(tls_ptr);
3004
3005 // Resolve oops to stable for CmpP below.
3006 cur_thr = access_resolve(cur_thr, 0);
3007 rec_thr = access_resolve(rec_thr, 0);
3008
3009 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3010 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3011
3012 generate_slow_guard(bol_thr, slow_region);
3013
3014 // (b) Interrupt bit on TLS must be false.
3015 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3016 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3017 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3018
3019 // Set the control input on the field _interrupted read to prevent it floating up.
3020 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3021 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3022 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3023
3024 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3025
3026 // First fast path: if (!TLS._interrupted) return false;
3027 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3028 result_rgn->init_req(no_int_result_path, false_bit);
3029 result_val->init_req(no_int_result_path, intcon(0));
3030
3031 // drop through to next case
3032 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3033
3034 #ifndef _WINDOWS
3035 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3036 Node* clr_arg = argument(1);
3037 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3038 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3039 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3040
3041 // Second fast path: ... else if (!clear_int) return true;
3042 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3043 result_rgn->init_req(no_clear_result_path, false_arg);
3044 result_val->init_req(no_clear_result_path, intcon(1));
3045
3046 // drop through to next case
3047 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3048 #else
3049 // To return true on Windows you must read the _interrupted field
3050 // and check the event state i.e. take the slow path.
3051 #endif // _WINDOWS
3052
3053 // (d) Otherwise, go to the slow path.
3054 slow_region->add_req(control());
3055 set_control( _gvn.transform(slow_region));
3056
3057 if (stopped()) {
3058 // There is no slow path.
3059 result_rgn->init_req(slow_result_path, top());
3060 result_val->init_req(slow_result_path, top());
3061 } else {
3062 // non-virtual because it is a private non-static
3063 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3064
3065 Node* slow_val = set_results_for_java_call(slow_call);
3066 // this->control() comes from set_results_for_java_call
3067
3068 Node* fast_io = slow_call->in(TypeFunc::I_O);
3069 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3070
3071 // These two phis are pre-filled with copies of of the fast IO and Memory
3072 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3073 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3074
3075 result_rgn->init_req(slow_result_path, control());
3076 result_io ->init_req(slow_result_path, i_o());
3077 result_mem->init_req(slow_result_path, reset_memory());
3078 result_val->init_req(slow_result_path, slow_val);
3079
3080 set_all_memory(_gvn.transform(result_mem));
3081 set_i_o( _gvn.transform(result_io));
3082 }
3083
3084 C->set_has_split_ifs(true); // Has chance for split-if optimization
3085 set_result(result_rgn, result_val);
3086 return true;
3087 }
3088
3089 //---------------------------load_mirror_from_klass----------------------------
3090 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3091 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3092 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3093 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3094 // mirror = ((OopHandle)mirror)->resolve();
3095 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3096 }
3097
3098 //-----------------------load_klass_from_mirror_common-------------------------
3099 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3100 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3101 // and branch to the given path on the region.
3102 // If never_see_null, take an uncommon trap on null, so we can optimistically
3103 // compile for the non-null case.
3104 // If the region is NULL, force never_see_null = true.
3105 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3106 bool never_see_null,
3107 RegionNode* region,
3108 int null_path,
3109 int offset) {
3110 if (region == NULL) never_see_null = true;
3111 Node* p = basic_plus_adr(mirror, offset);
3112 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3113 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3114 Node* null_ctl = top();
3115 kls = null_check_oop(kls, &null_ctl, never_see_null);
3116 if (region != NULL) {
3117 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3118 region->init_req(null_path, null_ctl);
3119 } else {
3120 assert(null_ctl == top(), "no loose ends");
3121 }
3122 return kls;
3123 }
3124
3125 //--------------------(inline_native_Class_query helpers)---------------------
3126 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3127 // Fall through if (mods & mask) == bits, take the guard otherwise.
3128 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3129 // Branch around if the given klass has the given modifier bit set.
3130 // Like generate_guard, adds a new path onto the region.
3131 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3132 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3133 Node* mask = intcon(modifier_mask);
3134 Node* bits = intcon(modifier_bits);
3135 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3136 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3137 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3138 return generate_fair_guard(bol, region);
3139 }
3140 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3141 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3142 }
3143
3144 //-------------------------inline_native_Class_query-------------------
3145 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3146 const Type* return_type = TypeInt::BOOL;
3147 Node* prim_return_value = top(); // what happens if it's a primitive class?
3148 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3149 bool expect_prim = false; // most of these guys expect to work on refs
3150
3151 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3152
3153 Node* mirror = argument(0);
3154 Node* obj = top();
3155
3156 switch (id) {
3157 case vmIntrinsics::_isInstance:
3158 // nothing is an instance of a primitive type
3159 prim_return_value = intcon(0);
3160 obj = argument(1);
3161 break;
3162 case vmIntrinsics::_getModifiers:
3163 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3164 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3165 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3166 break;
3167 case vmIntrinsics::_isInterface:
3168 prim_return_value = intcon(0);
3169 break;
3170 case vmIntrinsics::_isArray:
3171 prim_return_value = intcon(0);
3172 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3173 break;
3174 case vmIntrinsics::_isPrimitive:
3175 prim_return_value = intcon(1);
3176 expect_prim = true; // obviously
3177 break;
3178 case vmIntrinsics::_getSuperclass:
3179 prim_return_value = null();
3180 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3181 break;
3182 case vmIntrinsics::_getClassAccessFlags:
3183 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3184 return_type = TypeInt::INT; // not bool! 6297094
3185 break;
3186 default:
3187 fatal_unexpected_iid(id);
3188 break;
3189 }
3190
3191 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3192 if (mirror_con == NULL) return false; // cannot happen?
3193
3194 #ifndef PRODUCT
3195 if (C->print_intrinsics() || C->print_inlining()) {
3196 ciType* k = mirror_con->java_mirror_type();
3197 if (k) {
3198 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3199 k->print_name();
3200 tty->cr();
3201 }
3202 }
3203 #endif
3204
3205 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3206 RegionNode* region = new RegionNode(PATH_LIMIT);
3207 record_for_igvn(region);
3208 PhiNode* phi = new PhiNode(region, return_type);
3209
3210 // The mirror will never be null of Reflection.getClassAccessFlags, however
3211 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3212 // if it is. See bug 4774291.
3213
3214 // For Reflection.getClassAccessFlags(), the null check occurs in
3215 // the wrong place; see inline_unsafe_access(), above, for a similar
3216 // situation.
3217 mirror = null_check(mirror);
3218 // If mirror or obj is dead, only null-path is taken.
3219 if (stopped()) return true;
3220
3221 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3222
3223 // Now load the mirror's klass metaobject, and null-check it.
3224 // Side-effects region with the control path if the klass is null.
3225 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3226 // If kls is null, we have a primitive mirror.
3227 phi->init_req(_prim_path, prim_return_value);
3228 if (stopped()) { set_result(region, phi); return true; }
3229 bool safe_for_replace = (region->in(_prim_path) == top());
3230
3231 Node* p; // handy temp
3232 Node* null_ctl;
3233
3234 // Now that we have the non-null klass, we can perform the real query.
3235 // For constant classes, the query will constant-fold in LoadNode::Value.
3236 Node* query_value = top();
3237 switch (id) {
3238 case vmIntrinsics::_isInstance:
3239 // nothing is an instance of a primitive type
3240 query_value = gen_instanceof(obj, kls, safe_for_replace);
3241 break;
3242
3243 case vmIntrinsics::_getModifiers:
3244 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3245 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3246 break;
3247
3248 case vmIntrinsics::_isInterface:
3249 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3250 if (generate_interface_guard(kls, region) != NULL)
3251 // A guard was added. If the guard is taken, it was an interface.
3252 phi->add_req(intcon(1));
3253 // If we fall through, it's a plain class.
3254 query_value = intcon(0);
3255 break;
3256
3257 case vmIntrinsics::_isArray:
3258 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3259 if (generate_array_guard(kls, region) != NULL)
3260 // A guard was added. If the guard is taken, it was an array.
3261 phi->add_req(intcon(1));
3262 // If we fall through, it's a plain class.
3263 query_value = intcon(0);
3264 break;
3265
3266 case vmIntrinsics::_isPrimitive:
3267 query_value = intcon(0); // "normal" path produces false
3268 break;
3269
3270 case vmIntrinsics::_getSuperclass:
3271 // The rules here are somewhat unfortunate, but we can still do better
3272 // with random logic than with a JNI call.
3273 // Interfaces store null or Object as _super, but must report null.
3274 // Arrays store an intermediate super as _super, but must report Object.
3275 // Other types can report the actual _super.
3276 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3277 if (generate_interface_guard(kls, region) != NULL)
3278 // A guard was added. If the guard is taken, it was an interface.
3279 phi->add_req(null());
3280 if (generate_array_guard(kls, region) != NULL)
3281 // A guard was added. If the guard is taken, it was an array.
3282 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3283 // If we fall through, it's a plain class. Get its _super.
3284 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3285 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3286 null_ctl = top();
3287 kls = null_check_oop(kls, &null_ctl);
3288 if (null_ctl != top()) {
3289 // If the guard is taken, Object.superClass is null (both klass and mirror).
3290 region->add_req(null_ctl);
3291 phi ->add_req(null());
3292 }
3293 if (!stopped()) {
3294 query_value = load_mirror_from_klass(kls);
3295 }
3296 break;
3297
3298 case vmIntrinsics::_getClassAccessFlags:
3299 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3300 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3301 break;
3302
3303 default:
3304 fatal_unexpected_iid(id);
3305 break;
3306 }
3307
3308 // Fall-through is the normal case of a query to a real class.
3309 phi->init_req(1, query_value);
3310 region->init_req(1, control());
3311
3312 C->set_has_split_ifs(true); // Has chance for split-if optimization
3313 set_result(region, phi);
3314 return true;
3315 }
3316
3317 //-------------------------inline_Class_cast-------------------
3318 bool LibraryCallKit::inline_Class_cast() {
3319 Node* mirror = argument(0); // Class
3320 Node* obj = argument(1);
3321 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3322 if (mirror_con == NULL) {
3323 return false; // dead path (mirror->is_top()).
3324 }
3325 if (obj == NULL || obj->is_top()) {
3326 return false; // dead path
3327 }
3328 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3329
3330 // First, see if Class.cast() can be folded statically.
3331 // java_mirror_type() returns non-null for compile-time Class constants.
3332 ciType* tm = mirror_con->java_mirror_type();
3333 if (tm != NULL && tm->is_klass() &&
3334 tp != NULL && tp->klass() != NULL) {
3335 if (!tp->klass()->is_loaded()) {
3336 // Don't use intrinsic when class is not loaded.
3337 return false;
3338 } else {
3339 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3340 if (static_res == Compile::SSC_always_true) {
3341 // isInstance() is true - fold the code.
3342 set_result(obj);
3343 return true;
3344 } else if (static_res == Compile::SSC_always_false) {
3345 // Don't use intrinsic, have to throw ClassCastException.
3346 // If the reference is null, the non-intrinsic bytecode will
3347 // be optimized appropriately.
3348 return false;
3349 }
3350 }
3351 }
3352
3353 // Bailout intrinsic and do normal inlining if exception path is frequent.
3354 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3355 return false;
3356 }
3357
3358 // Generate dynamic checks.
3359 // Class.cast() is java implementation of _checkcast bytecode.
3360 // Do checkcast (Parse::do_checkcast()) optimizations here.
3361
3362 mirror = null_check(mirror);
3363 // If mirror is dead, only null-path is taken.
3364 if (stopped()) {
3365 return true;
3366 }
3367
3368 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3369 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3370 RegionNode* region = new RegionNode(PATH_LIMIT);
3371 record_for_igvn(region);
3372
3373 // Now load the mirror's klass metaobject, and null-check it.
3374 // If kls is null, we have a primitive mirror and
3375 // nothing is an instance of a primitive type.
3376 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3377
3378 Node* res = top();
3379 if (!stopped()) {
3380 Node* bad_type_ctrl = top();
3381 // Do checkcast optimizations.
3382 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3383 region->init_req(_bad_type_path, bad_type_ctrl);
3384 }
3385 if (region->in(_prim_path) != top() ||
3386 region->in(_bad_type_path) != top()) {
3387 // Let Interpreter throw ClassCastException.
3388 PreserveJVMState pjvms(this);
3389 set_control(_gvn.transform(region));
3390 uncommon_trap(Deoptimization::Reason_intrinsic,
3391 Deoptimization::Action_maybe_recompile);
3392 }
3393 if (!stopped()) {
3394 set_result(res);
3395 }
3396 return true;
3397 }
3398
3399
3400 //--------------------------inline_native_subtype_check------------------------
3401 // This intrinsic takes the JNI calls out of the heart of
3402 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3403 bool LibraryCallKit::inline_native_subtype_check() {
3404 // Pull both arguments off the stack.
3405 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3406 args[0] = argument(0);
3407 args[1] = argument(1);
3408 Node* klasses[2]; // corresponding Klasses: superk, subk
3409 klasses[0] = klasses[1] = top();
3410
3411 enum {
3412 // A full decision tree on {superc is prim, subc is prim}:
3413 _prim_0_path = 1, // {P,N} => false
3414 // {P,P} & superc!=subc => false
3415 _prim_same_path, // {P,P} & superc==subc => true
3416 _prim_1_path, // {N,P} => false
3417 _ref_subtype_path, // {N,N} & subtype check wins => true
3418 _both_ref_path, // {N,N} & subtype check loses => false
3419 PATH_LIMIT
3420 };
3421
3422 RegionNode* region = new RegionNode(PATH_LIMIT);
3423 Node* phi = new PhiNode(region, TypeInt::BOOL);
3424 record_for_igvn(region);
3425
3426 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3427 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3428 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3429
3430 // First null-check both mirrors and load each mirror's klass metaobject.
3431 int which_arg;
3432 for (which_arg = 0; which_arg <= 1; which_arg++) {
3433 Node* arg = args[which_arg];
3434 arg = null_check(arg);
3435 if (stopped()) break;
3436 args[which_arg] = arg;
3437
3438 Node* p = basic_plus_adr(arg, class_klass_offset);
3439 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3440 klasses[which_arg] = _gvn.transform(kls);
3441 }
3442
3443 // Resolve oops to stable for CmpP below.
3444 args[0] = access_resolve(args[0], 0);
3445 args[1] = access_resolve(args[1], 0);
3446
3447 // Having loaded both klasses, test each for null.
3448 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3449 for (which_arg = 0; which_arg <= 1; which_arg++) {
3450 Node* kls = klasses[which_arg];
3451 Node* null_ctl = top();
3452 kls = null_check_oop(kls, &null_ctl, never_see_null);
3453 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3454 region->init_req(prim_path, null_ctl);
3455 if (stopped()) break;
3456 klasses[which_arg] = kls;
3457 }
3458
3459 if (!stopped()) {
3460 // now we have two reference types, in klasses[0..1]
3461 Node* subk = klasses[1]; // the argument to isAssignableFrom
3462 Node* superk = klasses[0]; // the receiver
3463 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3464 // now we have a successful reference subtype check
3465 region->set_req(_ref_subtype_path, control());
3466 }
3467
3468 // If both operands are primitive (both klasses null), then
3469 // we must return true when they are identical primitives.
3470 // It is convenient to test this after the first null klass check.
3471 set_control(region->in(_prim_0_path)); // go back to first null check
3472 if (!stopped()) {
3473 // Since superc is primitive, make a guard for the superc==subc case.
3474 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3475 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3476 generate_guard(bol_eq, region, PROB_FAIR);
3477 if (region->req() == PATH_LIMIT+1) {
3478 // A guard was added. If the added guard is taken, superc==subc.
3479 region->swap_edges(PATH_LIMIT, _prim_same_path);
3480 region->del_req(PATH_LIMIT);
3481 }
3482 region->set_req(_prim_0_path, control()); // Not equal after all.
3483 }
3484
3485 // these are the only paths that produce 'true':
3486 phi->set_req(_prim_same_path, intcon(1));
3487 phi->set_req(_ref_subtype_path, intcon(1));
3488
3489 // pull together the cases:
3490 assert(region->req() == PATH_LIMIT, "sane region");
3491 for (uint i = 1; i < region->req(); i++) {
3492 Node* ctl = region->in(i);
3493 if (ctl == NULL || ctl == top()) {
3494 region->set_req(i, top());
3495 phi ->set_req(i, top());
3496 } else if (phi->in(i) == NULL) {
3497 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3498 }
3499 }
3500
3501 set_control(_gvn.transform(region));
3502 set_result(_gvn.transform(phi));
3503 return true;
3504 }
3505
3506 //---------------------generate_array_guard_common------------------------
3507 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3508 bool obj_array, bool not_array) {
3509
3510 if (stopped()) {
3511 return NULL;
3512 }
3513
3514 // If obj_array/non_array==false/false:
3515 // Branch around if the given klass is in fact an array (either obj or prim).
3516 // If obj_array/non_array==false/true:
3517 // Branch around if the given klass is not an array klass of any kind.
3518 // If obj_array/non_array==true/true:
3519 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3520 // If obj_array/non_array==true/false:
3521 // Branch around if the kls is an oop array (Object[] or subtype)
3522 //
3523 // Like generate_guard, adds a new path onto the region.
3524 jint layout_con = 0;
3525 Node* layout_val = get_layout_helper(kls, layout_con);
3526 if (layout_val == NULL) {
3527 bool query = (obj_array
3528 ? Klass::layout_helper_is_objArray(layout_con)
3529 : Klass::layout_helper_is_array(layout_con));
3530 if (query == not_array) {
3531 return NULL; // never a branch
3532 } else { // always a branch
3533 Node* always_branch = control();
3534 if (region != NULL)
3535 region->add_req(always_branch);
3536 set_control(top());
3537 return always_branch;
3538 }
3539 }
3540 // Now test the correct condition.
3541 jint nval = (obj_array
3542 ? (jint)(Klass::_lh_array_tag_type_value
3543 << Klass::_lh_array_tag_shift)
3544 : Klass::_lh_neutral_value);
3545 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3546 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3547 // invert the test if we are looking for a non-array
3548 if (not_array) btest = BoolTest(btest).negate();
3549 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3550 return generate_fair_guard(bol, region);
3551 }
3552
3553
3554 //-----------------------inline_native_newArray--------------------------
3555 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3556 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3557 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3558 Node* mirror;
3559 Node* count_val;
3560 if (uninitialized) {
3561 mirror = argument(1);
3562 count_val = argument(2);
3563 } else {
3564 mirror = argument(0);
3565 count_val = argument(1);
3566 }
3567
3568 mirror = null_check(mirror);
3569 // If mirror or obj is dead, only null-path is taken.
3570 if (stopped()) return true;
3571
3572 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3573 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3574 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3575 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3576 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3577
3578 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3579 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3580 result_reg, _slow_path);
3581 Node* normal_ctl = control();
3582 Node* no_array_ctl = result_reg->in(_slow_path);
3583
3584 // Generate code for the slow case. We make a call to newArray().
3585 set_control(no_array_ctl);
3586 if (!stopped()) {
3587 // Either the input type is void.class, or else the
3588 // array klass has not yet been cached. Either the
3589 // ensuing call will throw an exception, or else it
3590 // will cache the array klass for next time.
3591 PreserveJVMState pjvms(this);
3592 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3593 Node* slow_result = set_results_for_java_call(slow_call);
3594 // this->control() comes from set_results_for_java_call
3595 result_reg->set_req(_slow_path, control());
3596 result_val->set_req(_slow_path, slow_result);
3597 result_io ->set_req(_slow_path, i_o());
3598 result_mem->set_req(_slow_path, reset_memory());
3599 }
3600
3601 set_control(normal_ctl);
3602 if (!stopped()) {
3603 // Normal case: The array type has been cached in the java.lang.Class.
3604 // The following call works fine even if the array type is polymorphic.
3605 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3606 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3607 result_reg->init_req(_normal_path, control());
3608 result_val->init_req(_normal_path, obj);
3609 result_io ->init_req(_normal_path, i_o());
3610 result_mem->init_req(_normal_path, reset_memory());
3611
3612 if (uninitialized) {
3613 // Mark the allocation so that zeroing is skipped
3614 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3615 alloc->maybe_set_complete(&_gvn);
3616 }
3617 }
3618
3619 // Return the combined state.
3620 set_i_o( _gvn.transform(result_io) );
3621 set_all_memory( _gvn.transform(result_mem));
3622
3623 C->set_has_split_ifs(true); // Has chance for split-if optimization
3624 set_result(result_reg, result_val);
3625 return true;
3626 }
3627
3628 //----------------------inline_native_getLength--------------------------
3629 // public static native int java.lang.reflect.Array.getLength(Object array);
3630 bool LibraryCallKit::inline_native_getLength() {
3631 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3632
3633 Node* array = null_check(argument(0));
3634 // If array is dead, only null-path is taken.
3635 if (stopped()) return true;
3636
3637 // Deoptimize if it is a non-array.
3638 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3639
3640 if (non_array != NULL) {
3641 PreserveJVMState pjvms(this);
3642 set_control(non_array);
3643 uncommon_trap(Deoptimization::Reason_intrinsic,
3644 Deoptimization::Action_maybe_recompile);
3645 }
3646
3647 // If control is dead, only non-array-path is taken.
3648 if (stopped()) return true;
3649
3650 // The works fine even if the array type is polymorphic.
3651 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3652 Node* result = load_array_length(array);
3653
3654 C->set_has_split_ifs(true); // Has chance for split-if optimization
3655 set_result(result);
3656 return true;
3657 }
3658
3659 //------------------------inline_array_copyOf----------------------------
3660 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3661 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3662 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3663 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3664
3665 // Get the arguments.
3666 Node* original = argument(0);
3667 Node* start = is_copyOfRange? argument(1): intcon(0);
3668 Node* end = is_copyOfRange? argument(2): argument(1);
3669 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3670
3671 Node* newcopy = NULL;
3672
3673 // Set the original stack and the reexecute bit for the interpreter to reexecute
3674 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3675 { PreserveReexecuteState preexecs(this);
3676 jvms()->set_should_reexecute(true);
3677
3678 array_type_mirror = null_check(array_type_mirror);
3679 original = null_check(original);
3680
3681 // Check if a null path was taken unconditionally.
3682 if (stopped()) return true;
3683
3684 Node* orig_length = load_array_length(original);
3685
3686 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3687 klass_node = null_check(klass_node);
3688
3689 RegionNode* bailout = new RegionNode(1);
3690 record_for_igvn(bailout);
3691
3692 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3693 // Bail out if that is so.
3694 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3695 if (not_objArray != NULL) {
3696 // Improve the klass node's type from the new optimistic assumption:
3697 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3698 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3699 Node* cast = new CastPPNode(klass_node, akls);
3700 cast->init_req(0, control());
3701 klass_node = _gvn.transform(cast);
3702 }
3703
3704 // Bail out if either start or end is negative.
3705 generate_negative_guard(start, bailout, &start);
3706 generate_negative_guard(end, bailout, &end);
3707
3708 Node* length = end;
3709 if (_gvn.type(start) != TypeInt::ZERO) {
3710 length = _gvn.transform(new SubINode(end, start));
3711 }
3712
3713 // Bail out if length is negative.
3714 // Without this the new_array would throw
3715 // NegativeArraySizeException but IllegalArgumentException is what
3716 // should be thrown
3717 generate_negative_guard(length, bailout, &length);
3718
3719 if (bailout->req() > 1) {
3720 PreserveJVMState pjvms(this);
3721 set_control(_gvn.transform(bailout));
3722 uncommon_trap(Deoptimization::Reason_intrinsic,
3723 Deoptimization::Action_maybe_recompile);
3724 }
3725
3726 if (!stopped()) {
3727 // How many elements will we copy from the original?
3728 // The answer is MinI(orig_length - start, length).
3729 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3730 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3731
3732 original = access_resolve(original, ACCESS_READ);
3733
3734 // Generate a direct call to the right arraycopy function(s).
3735 // We know the copy is disjoint but we might not know if the
3736 // oop stores need checking.
3737 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3738 // This will fail a store-check if x contains any non-nulls.
3739
3740 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3741 // loads/stores but it is legal only if we're sure the
3742 // Arrays.copyOf would succeed. So we need all input arguments
3743 // to the copyOf to be validated, including that the copy to the
3744 // new array won't trigger an ArrayStoreException. That subtype
3745 // check can be optimized if we know something on the type of
3746 // the input array from type speculation.
3747 if (_gvn.type(klass_node)->singleton()) {
3748 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3749 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3750
3751 int test = C->static_subtype_check(superk, subk);
3752 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3753 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3754 if (t_original->speculative_type() != NULL) {
3755 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3756 }
3757 }
3758 }
3759
3760 bool validated = false;
3761 // Reason_class_check rather than Reason_intrinsic because we
3762 // want to intrinsify even if this traps.
3763 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3764 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3765 klass_node);
3766
3767 if (not_subtype_ctrl != top()) {
3768 PreserveJVMState pjvms(this);
3769 set_control(not_subtype_ctrl);
3770 uncommon_trap(Deoptimization::Reason_class_check,
3771 Deoptimization::Action_make_not_entrant);
3772 assert(stopped(), "Should be stopped");
3773 }
3774 validated = true;
3775 }
3776
3777 if (!stopped()) {
3778 newcopy = new_array(klass_node, length, 0); // no arguments to push
3779
3780 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3781 load_object_klass(original), klass_node);
3782 if (!is_copyOfRange) {
3783 ac->set_copyof(validated);
3784 } else {
3785 ac->set_copyofrange(validated);
3786 }
3787 Node* n = _gvn.transform(ac);
3788 if (n == ac) {
3789 ac->connect_outputs(this);
3790 } else {
3791 assert(validated, "shouldn't transform if all arguments not validated");
3792 set_all_memory(n);
3793 }
3794 }
3795 }
3796 } // original reexecute is set back here
3797
3798 C->set_has_split_ifs(true); // Has chance for split-if optimization
3799 if (!stopped()) {
3800 set_result(newcopy);
3801 }
3802 return true;
3803 }
3804
3805
3806 //----------------------generate_virtual_guard---------------------------
3807 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3808 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3809 RegionNode* slow_region) {
3810 ciMethod* method = callee();
3811 int vtable_index = method->vtable_index();
3812 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3813 "bad index %d", vtable_index);
3814 // Get the Method* out of the appropriate vtable entry.
3815 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3816 vtable_index*vtableEntry::size_in_bytes() +
3817 vtableEntry::method_offset_in_bytes();
3818 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3819 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3820
3821 // Compare the target method with the expected method (e.g., Object.hashCode).
3822 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3823
3824 Node* native_call = makecon(native_call_addr);
3825 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3826 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3827
3828 return generate_slow_guard(test_native, slow_region);
3829 }
3830
3831 //-----------------------generate_method_call----------------------------
3832 // Use generate_method_call to make a slow-call to the real
3833 // method if the fast path fails. An alternative would be to
3834 // use a stub like OptoRuntime::slow_arraycopy_Java.
3835 // This only works for expanding the current library call,
3836 // not another intrinsic. (E.g., don't use this for making an
3837 // arraycopy call inside of the copyOf intrinsic.)
3838 CallJavaNode*
3839 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3840 // When compiling the intrinsic method itself, do not use this technique.
3841 guarantee(callee() != C->method(), "cannot make slow-call to self");
3842
3843 ciMethod* method = callee();
3844 // ensure the JVMS we have will be correct for this call
3845 guarantee(method_id == method->intrinsic_id(), "must match");
3846
3847 const TypeFunc* tf = TypeFunc::make(method);
3848 CallJavaNode* slow_call;
3849 if (is_static) {
3850 assert(!is_virtual, "");
3851 slow_call = new CallStaticJavaNode(C, tf,
3852 SharedRuntime::get_resolve_static_call_stub(),
3853 method, bci());
3854 } else if (is_virtual) {
3855 null_check_receiver();
3856 int vtable_index = Method::invalid_vtable_index;
3857 if (UseInlineCaches) {
3858 // Suppress the vtable call
3859 } else {
3860 // hashCode and clone are not a miranda methods,
3861 // so the vtable index is fixed.
3862 // No need to use the linkResolver to get it.
3863 vtable_index = method->vtable_index();
3864 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3865 "bad index %d", vtable_index);
3866 }
3867 slow_call = new CallDynamicJavaNode(tf,
3868 SharedRuntime::get_resolve_virtual_call_stub(),
3869 method, vtable_index, bci());
3870 } else { // neither virtual nor static: opt_virtual
3871 null_check_receiver();
3872 slow_call = new CallStaticJavaNode(C, tf,
3873 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3874 method, bci());
3875 slow_call->set_optimized_virtual(true);
3876 }
3877 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3878 // To be able to issue a direct call (optimized virtual or virtual)
3879 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3880 // about the method being invoked should be attached to the call site to
3881 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3882 slow_call->set_override_symbolic_info(true);
3883 }
3884 set_arguments_for_java_call(slow_call);
3885 set_edges_for_java_call(slow_call);
3886 return slow_call;
3887 }
3888
3889
3890 /**
3891 * Build special case code for calls to hashCode on an object. This call may
3892 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3893 * slightly different code.
3894 */
3895 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3896 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3897 assert(!(is_virtual && is_static), "either virtual, special, or static");
3898
3899 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3900
3901 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3902 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3903 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3904 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3905 Node* obj = NULL;
3906 if (!is_static) {
3907 // Check for hashing null object
3908 obj = null_check_receiver();
3909 if (stopped()) return true; // unconditionally null
3910 result_reg->init_req(_null_path, top());
3911 result_val->init_req(_null_path, top());
3912 } else {
3913 // Do a null check, and return zero if null.
3914 // System.identityHashCode(null) == 0
3915 obj = argument(0);
3916 Node* null_ctl = top();
3917 obj = null_check_oop(obj, &null_ctl);
3918 result_reg->init_req(_null_path, null_ctl);
3919 result_val->init_req(_null_path, _gvn.intcon(0));
3920 }
3921
3922 // Unconditionally null? Then return right away.
3923 if (stopped()) {
3924 set_control( result_reg->in(_null_path));
3925 if (!stopped())
3926 set_result(result_val->in(_null_path));
3927 return true;
3928 }
3929
3930 // We only go to the fast case code if we pass a number of guards. The
3931 // paths which do not pass are accumulated in the slow_region.
3932 RegionNode* slow_region = new RegionNode(1);
3933 record_for_igvn(slow_region);
3934
3935 // If this is a virtual call, we generate a funny guard. We pull out
3936 // the vtable entry corresponding to hashCode() from the target object.
3937 // If the target method which we are calling happens to be the native
3938 // Object hashCode() method, we pass the guard. We do not need this
3939 // guard for non-virtual calls -- the caller is known to be the native
3940 // Object hashCode().
3941 if (is_virtual) {
3942 // After null check, get the object's klass.
3943 Node* obj_klass = load_object_klass(obj);
3944 generate_virtual_guard(obj_klass, slow_region);
3945 }
3946
3947 // Get the header out of the object, use LoadMarkNode when available
3948 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3949 // The control of the load must be NULL. Otherwise, the load can move before
3950 // the null check after castPP removal.
3951 Node* no_ctrl = NULL;
3952 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3953
3954 // Test the header to see if it is unlocked.
3955 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
3956 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3957 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
3958 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3959 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3960
3961 generate_slow_guard(test_unlocked, slow_region);
3962
3963 // Get the hash value and check to see that it has been properly assigned.
3964 // We depend on hash_mask being at most 32 bits and avoid the use of
3965 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3966 // vm: see markOop.hpp.
3967 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
3968 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
3969 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3970 // This hack lets the hash bits live anywhere in the mark object now, as long
3971 // as the shift drops the relevant bits into the low 32 bits. Note that
3972 // Java spec says that HashCode is an int so there's no point in capturing
3973 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3974 hshifted_header = ConvX2I(hshifted_header);
3975 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3976
3977 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
3978 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3979 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3980
3981 generate_slow_guard(test_assigned, slow_region);
3982
3983 Node* init_mem = reset_memory();
3984 // fill in the rest of the null path:
3985 result_io ->init_req(_null_path, i_o());
3986 result_mem->init_req(_null_path, init_mem);
3987
3988 result_val->init_req(_fast_path, hash_val);
3989 result_reg->init_req(_fast_path, control());
3990 result_io ->init_req(_fast_path, i_o());
3991 result_mem->init_req(_fast_path, init_mem);
3992
3993 // Generate code for the slow case. We make a call to hashCode().
3994 set_control(_gvn.transform(slow_region));
3995 if (!stopped()) {
3996 // No need for PreserveJVMState, because we're using up the present state.
3997 set_all_memory(init_mem);
3998 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3999 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4000 Node* slow_result = set_results_for_java_call(slow_call);
4001 // this->control() comes from set_results_for_java_call
4002 result_reg->init_req(_slow_path, control());
4003 result_val->init_req(_slow_path, slow_result);
4004 result_io ->set_req(_slow_path, i_o());
4005 result_mem ->set_req(_slow_path, reset_memory());
4006 }
4007
4008 // Return the combined state.
4009 set_i_o( _gvn.transform(result_io) );
4010 set_all_memory( _gvn.transform(result_mem));
4011
4012 set_result(result_reg, result_val);
4013 return true;
4014 }
4015
4016 //---------------------------inline_native_getClass----------------------------
4017 // public final native Class<?> java.lang.Object.getClass();
4018 //
4019 // Build special case code for calls to getClass on an object.
4020 bool LibraryCallKit::inline_native_getClass() {
4021 Node* obj = null_check_receiver();
4022 if (stopped()) return true;
4023 set_result(load_mirror_from_klass(load_object_klass(obj)));
4024 return true;
4025 }
4026
4027 //-----------------inline_native_Reflection_getCallerClass---------------------
4028 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4029 //
4030 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4031 //
4032 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4033 // in that it must skip particular security frames and checks for
4034 // caller sensitive methods.
4035 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4036 #ifndef PRODUCT
4037 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4038 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4039 }
4040 #endif
4041
4042 if (!jvms()->has_method()) {
4043 #ifndef PRODUCT
4044 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4045 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4046 }
4047 #endif
4048 return false;
4049 }
4050
4051 // Walk back up the JVM state to find the caller at the required
4052 // depth.
4053 JVMState* caller_jvms = jvms();
4054
4055 // Cf. JVM_GetCallerClass
4056 // NOTE: Start the loop at depth 1 because the current JVM state does
4057 // not include the Reflection.getCallerClass() frame.
4058 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4059 ciMethod* m = caller_jvms->method();
4060 switch (n) {
4061 case 0:
4062 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4063 break;
4064 case 1:
4065 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4066 if (!m->caller_sensitive()) {
4067 #ifndef PRODUCT
4068 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4069 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4070 }
4071 #endif
4072 return false; // bail-out; let JVM_GetCallerClass do the work
4073 }
4074 break;
4075 default:
4076 if (!m->is_ignored_by_security_stack_walk()) {
4077 // We have reached the desired frame; return the holder class.
4078 // Acquire method holder as java.lang.Class and push as constant.
4079 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4080 ciInstance* caller_mirror = caller_klass->java_mirror();
4081 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4082
4083 #ifndef PRODUCT
4084 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4085 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());
4086 tty->print_cr(" JVM state at this point:");
4087 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4088 ciMethod* m = jvms()->of_depth(i)->method();
4089 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4090 }
4091 }
4092 #endif
4093 return true;
4094 }
4095 break;
4096 }
4097 }
4098
4099 #ifndef PRODUCT
4100 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4101 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4102 tty->print_cr(" JVM state at this point:");
4103 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4104 ciMethod* m = jvms()->of_depth(i)->method();
4105 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4106 }
4107 }
4108 #endif
4109
4110 return false; // bail-out; let JVM_GetCallerClass do the work
4111 }
4112
4113 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4114 Node* arg = argument(0);
4115 Node* result = NULL;
4116
4117 switch (id) {
4118 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4119 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4120 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4121 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4122
4123 case vmIntrinsics::_doubleToLongBits: {
4124 // two paths (plus control) merge in a wood
4125 RegionNode *r = new RegionNode(3);
4126 Node *phi = new PhiNode(r, TypeLong::LONG);
4127
4128 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4129 // Build the boolean node
4130 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4131
4132 // Branch either way.
4133 // NaN case is less traveled, which makes all the difference.
4134 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4135 Node *opt_isnan = _gvn.transform(ifisnan);
4136 assert( opt_isnan->is_If(), "Expect an IfNode");
4137 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4138 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4139
4140 set_control(iftrue);
4141
4142 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4143 Node *slow_result = longcon(nan_bits); // return NaN
4144 phi->init_req(1, _gvn.transform( slow_result ));
4145 r->init_req(1, iftrue);
4146
4147 // Else fall through
4148 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4149 set_control(iffalse);
4150
4151 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4152 r->init_req(2, iffalse);
4153
4154 // Post merge
4155 set_control(_gvn.transform(r));
4156 record_for_igvn(r);
4157
4158 C->set_has_split_ifs(true); // Has chance for split-if optimization
4159 result = phi;
4160 assert(result->bottom_type()->isa_long(), "must be");
4161 break;
4162 }
4163
4164 case vmIntrinsics::_floatToIntBits: {
4165 // two paths (plus control) merge in a wood
4166 RegionNode *r = new RegionNode(3);
4167 Node *phi = new PhiNode(r, TypeInt::INT);
4168
4169 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4170 // Build the boolean node
4171 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4172
4173 // Branch either way.
4174 // NaN case is less traveled, which makes all the difference.
4175 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4176 Node *opt_isnan = _gvn.transform(ifisnan);
4177 assert( opt_isnan->is_If(), "Expect an IfNode");
4178 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4179 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4180
4181 set_control(iftrue);
4182
4183 static const jint nan_bits = 0x7fc00000;
4184 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4185 phi->init_req(1, _gvn.transform( slow_result ));
4186 r->init_req(1, iftrue);
4187
4188 // Else fall through
4189 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4190 set_control(iffalse);
4191
4192 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4193 r->init_req(2, iffalse);
4194
4195 // Post merge
4196 set_control(_gvn.transform(r));
4197 record_for_igvn(r);
4198
4199 C->set_has_split_ifs(true); // Has chance for split-if optimization
4200 result = phi;
4201 assert(result->bottom_type()->isa_int(), "must be");
4202 break;
4203 }
4204
4205 default:
4206 fatal_unexpected_iid(id);
4207 break;
4208 }
4209 set_result(_gvn.transform(result));
4210 return true;
4211 }
4212
4213 //----------------------inline_unsafe_copyMemory-------------------------
4214 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4215 bool LibraryCallKit::inline_unsafe_copyMemory() {
4216 if (callee()->is_static()) return false; // caller must have the capability!
4217 null_check_receiver(); // null-check receiver
4218 if (stopped()) return true;
4219
4220 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4221
4222 Node* src_ptr = argument(1); // type: oop
4223 Node* src_off = ConvL2X(argument(2)); // type: long
4224 Node* dst_ptr = argument(4); // type: oop
4225 Node* dst_off = ConvL2X(argument(5)); // type: long
4226 Node* size = ConvL2X(argument(7)); // type: long
4227
4228 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4229 "fieldOffset must be byte-scaled");
4230
4231 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4232 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4233 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4234 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4235
4236 // Conservatively insert a memory barrier on all memory slices.
4237 // Do not let writes of the copy source or destination float below the copy.
4238 insert_mem_bar(Op_MemBarCPUOrder);
4239
4240 // Call it. Note that the length argument is not scaled.
4241 make_runtime_call(RC_LEAF|RC_NO_FP,
4242 OptoRuntime::fast_arraycopy_Type(),
4243 StubRoutines::unsafe_arraycopy(),
4244 "unsafe_arraycopy",
4245 TypeRawPtr::BOTTOM,
4246 src, dst, size XTOP);
4247
4248 // Do not let reads of the copy destination float above the copy.
4249 insert_mem_bar(Op_MemBarCPUOrder);
4250
4251 return true;
4252 }
4253
4254 //------------------------clone_coping-----------------------------------
4255 // Helper function for inline_native_clone.
4256 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4257 assert(obj_size != NULL, "");
4258 Node* raw_obj = alloc_obj->in(1);
4259 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4260
4261 AllocateNode* alloc = NULL;
4262 if (ReduceBulkZeroing) {
4263 // We will be completely responsible for initializing this object -
4264 // mark Initialize node as complete.
4265 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4266 // The object was just allocated - there should be no any stores!
4267 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4268 // Mark as complete_with_arraycopy so that on AllocateNode
4269 // expansion, we know this AllocateNode is initialized by an array
4270 // copy and a StoreStore barrier exists after the array copy.
4271 alloc->initialization()->set_complete_with_arraycopy();
4272 }
4273
4274 // Copy the fastest available way.
4275 // TODO: generate fields copies for small objects instead.
4276 Node* size = _gvn.transform(obj_size);
4277
4278 access_clone(obj, alloc_obj, size, is_array);
4279
4280 // Do not let reads from the cloned object float above the arraycopy.
4281 if (alloc != NULL) {
4282 // Do not let stores that initialize this object be reordered with
4283 // a subsequent store that would make this object accessible by
4284 // other threads.
4285 // Record what AllocateNode this StoreStore protects so that
4286 // escape analysis can go from the MemBarStoreStoreNode to the
4287 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4288 // based on the escape status of the AllocateNode.
4289 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4290 } else {
4291 insert_mem_bar(Op_MemBarCPUOrder);
4292 }
4293 }
4294
4295 //------------------------inline_native_clone----------------------------
4296 // protected native Object java.lang.Object.clone();
4297 //
4298 // Here are the simple edge cases:
4299 // null receiver => normal trap
4300 // virtual and clone was overridden => slow path to out-of-line clone
4301 // not cloneable or finalizer => slow path to out-of-line Object.clone
4302 //
4303 // The general case has two steps, allocation and copying.
4304 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4305 //
4306 // Copying also has two cases, oop arrays and everything else.
4307 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4308 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4309 //
4310 // These steps fold up nicely if and when the cloned object's klass
4311 // can be sharply typed as an object array, a type array, or an instance.
4312 //
4313 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4314 PhiNode* result_val;
4315
4316 // Set the reexecute bit for the interpreter to reexecute
4317 // the bytecode that invokes Object.clone if deoptimization happens.
4318 { PreserveReexecuteState preexecs(this);
4319 jvms()->set_should_reexecute(true);
4320
4321 Node* obj = null_check_receiver();
4322 if (stopped()) return true;
4323
4324 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4325
4326 // If we are going to clone an instance, we need its exact type to
4327 // know the number and types of fields to convert the clone to
4328 // loads/stores. Maybe a speculative type can help us.
4329 if (!obj_type->klass_is_exact() &&
4330 obj_type->speculative_type() != NULL &&
4331 obj_type->speculative_type()->is_instance_klass()) {
4332 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4333 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4334 !spec_ik->has_injected_fields()) {
4335 ciKlass* k = obj_type->klass();
4336 if (!k->is_instance_klass() ||
4337 k->as_instance_klass()->is_interface() ||
4338 k->as_instance_klass()->has_subklass()) {
4339 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4340 }
4341 }
4342 }
4343
4344 Node* obj_klass = load_object_klass(obj);
4345 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4346 const TypeOopPtr* toop = ((tklass != NULL)
4347 ? tklass->as_instance_type()
4348 : TypeInstPtr::NOTNULL);
4349
4350 // Conservatively insert a memory barrier on all memory slices.
4351 // Do not let writes into the original float below the clone.
4352 insert_mem_bar(Op_MemBarCPUOrder);
4353
4354 // paths into result_reg:
4355 enum {
4356 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4357 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4358 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4359 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4360 PATH_LIMIT
4361 };
4362 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4363 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4364 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4365 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4366 record_for_igvn(result_reg);
4367
4368 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4369 if (array_ctl != NULL) {
4370 // It's an array.
4371 PreserveJVMState pjvms(this);
4372 set_control(array_ctl);
4373 Node* obj_length = load_array_length(obj);
4374 Node* obj_size = NULL;
4375 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4376
4377 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4378 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4379 // If it is an oop array, it requires very special treatment,
4380 // because gc barriers are required when accessing the array.
4381 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4382 if (is_obja != NULL) {
4383 PreserveJVMState pjvms2(this);
4384 set_control(is_obja);
4385 obj = access_resolve(obj, ACCESS_READ);
4386 // Generate a direct call to the right arraycopy function(s).
4387 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4388 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4389 ac->set_cloneoop();
4390 Node* n = _gvn.transform(ac);
4391 assert(n == ac, "cannot disappear");
4392 ac->connect_outputs(this);
4393
4394 result_reg->init_req(_objArray_path, control());
4395 result_val->init_req(_objArray_path, alloc_obj);
4396 result_i_o ->set_req(_objArray_path, i_o());
4397 result_mem ->set_req(_objArray_path, reset_memory());
4398 }
4399 }
4400 // Otherwise, there are no barriers to worry about.
4401 // (We can dispense with card marks if we know the allocation
4402 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4403 // causes the non-eden paths to take compensating steps to
4404 // simulate a fresh allocation, so that no further
4405 // card marks are required in compiled code to initialize
4406 // the object.)
4407
4408 if (!stopped()) {
4409 copy_to_clone(obj, alloc_obj, obj_size, true);
4410
4411 // Present the results of the copy.
4412 result_reg->init_req(_array_path, control());
4413 result_val->init_req(_array_path, alloc_obj);
4414 result_i_o ->set_req(_array_path, i_o());
4415 result_mem ->set_req(_array_path, reset_memory());
4416 }
4417 }
4418
4419 // We only go to the instance fast case code if we pass a number of guards.
4420 // The paths which do not pass are accumulated in the slow_region.
4421 RegionNode* slow_region = new RegionNode(1);
4422 record_for_igvn(slow_region);
4423 if (!stopped()) {
4424 // It's an instance (we did array above). Make the slow-path tests.
4425 // If this is a virtual call, we generate a funny guard. We grab
4426 // the vtable entry corresponding to clone() from the target object.
4427 // If the target method which we are calling happens to be the
4428 // Object clone() method, we pass the guard. We do not need this
4429 // guard for non-virtual calls; the caller is known to be the native
4430 // Object clone().
4431 if (is_virtual) {
4432 generate_virtual_guard(obj_klass, slow_region);
4433 }
4434
4435 // The object must be easily cloneable and must not have a finalizer.
4436 // Both of these conditions may be checked in a single test.
4437 // We could optimize the test further, but we don't care.
4438 generate_access_flags_guard(obj_klass,
4439 // Test both conditions:
4440 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4441 // Must be cloneable but not finalizer:
4442 JVM_ACC_IS_CLONEABLE_FAST,
4443 slow_region);
4444 }
4445
4446 if (!stopped()) {
4447 // It's an instance, and it passed the slow-path tests.
4448 PreserveJVMState pjvms(this);
4449 Node* obj_size = NULL;
4450 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4451 // is reexecuted if deoptimization occurs and there could be problems when merging
4452 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4453 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4454
4455 copy_to_clone(obj, alloc_obj, obj_size, false);
4456
4457 // Present the results of the slow call.
4458 result_reg->init_req(_instance_path, control());
4459 result_val->init_req(_instance_path, alloc_obj);
4460 result_i_o ->set_req(_instance_path, i_o());
4461 result_mem ->set_req(_instance_path, reset_memory());
4462 }
4463
4464 // Generate code for the slow case. We make a call to clone().
4465 set_control(_gvn.transform(slow_region));
4466 if (!stopped()) {
4467 PreserveJVMState pjvms(this);
4468 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4469 // We need to deoptimize on exception (see comment above)
4470 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4471 // this->control() comes from set_results_for_java_call
4472 result_reg->init_req(_slow_path, control());
4473 result_val->init_req(_slow_path, slow_result);
4474 result_i_o ->set_req(_slow_path, i_o());
4475 result_mem ->set_req(_slow_path, reset_memory());
4476 }
4477
4478 // Return the combined state.
4479 set_control( _gvn.transform(result_reg));
4480 set_i_o( _gvn.transform(result_i_o));
4481 set_all_memory( _gvn.transform(result_mem));
4482 } // original reexecute is set back here
4483
4484 set_result(_gvn.transform(result_val));
4485 return true;
4486 }
4487
4488 // If we have a tightly coupled allocation, the arraycopy may take care
4489 // of the array initialization. If one of the guards we insert between
4490 // the allocation and the arraycopy causes a deoptimization, an
4491 // unitialized array will escape the compiled method. To prevent that
4492 // we set the JVM state for uncommon traps between the allocation and
4493 // the arraycopy to the state before the allocation so, in case of
4494 // deoptimization, we'll reexecute the allocation and the
4495 // initialization.
4496 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4497 if (alloc != NULL) {
4498 ciMethod* trap_method = alloc->jvms()->method();
4499 int trap_bci = alloc->jvms()->bci();
4500
4501 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4502 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4503 // Make sure there's no store between the allocation and the
4504 // arraycopy otherwise visible side effects could be rexecuted
4505 // in case of deoptimization and cause incorrect execution.
4506 bool no_interfering_store = true;
4507 Node* mem = alloc->in(TypeFunc::Memory);
4508 if (mem->is_MergeMem()) {
4509 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4510 Node* n = mms.memory();
4511 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4512 assert(n->is_Store(), "what else?");
4513 no_interfering_store = false;
4514 break;
4515 }
4516 }
4517 } else {
4518 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4519 Node* n = mms.memory();
4520 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4521 assert(n->is_Store(), "what else?");
4522 no_interfering_store = false;
4523 break;
4524 }
4525 }
4526 }
4527
4528 if (no_interfering_store) {
4529 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4530 uint size = alloc->req();
4531 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4532 old_jvms->set_map(sfpt);
4533 for (uint i = 0; i < size; i++) {
4534 sfpt->init_req(i, alloc->in(i));
4535 }
4536 // re-push array length for deoptimization
4537 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4538 old_jvms->set_sp(old_jvms->sp()+1);
4539 old_jvms->set_monoff(old_jvms->monoff()+1);
4540 old_jvms->set_scloff(old_jvms->scloff()+1);
4541 old_jvms->set_endoff(old_jvms->endoff()+1);
4542 old_jvms->set_should_reexecute(true);
4543
4544 sfpt->set_i_o(map()->i_o());
4545 sfpt->set_memory(map()->memory());
4546 sfpt->set_control(map()->control());
4547
4548 JVMState* saved_jvms = jvms();
4549 saved_reexecute_sp = _reexecute_sp;
4550
4551 set_jvms(sfpt->jvms());
4552 _reexecute_sp = jvms()->sp();
4553
4554 return saved_jvms;
4555 }
4556 }
4557 }
4558 return NULL;
4559 }
4560
4561 // In case of a deoptimization, we restart execution at the
4562 // allocation, allocating a new array. We would leave an uninitialized
4563 // array in the heap that GCs wouldn't expect. Move the allocation
4564 // after the traps so we don't allocate the array if we
4565 // deoptimize. This is possible because tightly_coupled_allocation()
4566 // guarantees there's no observer of the allocated array at this point
4567 // and the control flow is simple enough.
4568 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4569 int saved_reexecute_sp, uint new_idx) {
4570 if (saved_jvms != NULL && !stopped()) {
4571 assert(alloc != NULL, "only with a tightly coupled allocation");
4572 // restore JVM state to the state at the arraycopy
4573 saved_jvms->map()->set_control(map()->control());
4574 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4575 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4576 // If we've improved the types of some nodes (null check) while
4577 // emitting the guards, propagate them to the current state
4578 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4579 set_jvms(saved_jvms);
4580 _reexecute_sp = saved_reexecute_sp;
4581
4582 // Remove the allocation from above the guards
4583 CallProjections callprojs;
4584 alloc->extract_projections(&callprojs, true);
4585 InitializeNode* init = alloc->initialization();
4586 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4587 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4588 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4589 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4590
4591 // move the allocation here (after the guards)
4592 _gvn.hash_delete(alloc);
4593 alloc->set_req(TypeFunc::Control, control());
4594 alloc->set_req(TypeFunc::I_O, i_o());
4595 Node *mem = reset_memory();
4596 set_all_memory(mem);
4597 alloc->set_req(TypeFunc::Memory, mem);
4598 set_control(init->proj_out_or_null(TypeFunc::Control));
4599 set_i_o(callprojs.fallthrough_ioproj);
4600
4601 // Update memory as done in GraphKit::set_output_for_allocation()
4602 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4603 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4604 if (ary_type->isa_aryptr() && length_type != NULL) {
4605 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4606 }
4607 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4608 int elemidx = C->get_alias_index(telemref);
4609 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4610 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4611
4612 Node* allocx = _gvn.transform(alloc);
4613 assert(allocx == alloc, "where has the allocation gone?");
4614 assert(dest->is_CheckCastPP(), "not an allocation result?");
4615
4616 _gvn.hash_delete(dest);
4617 dest->set_req(0, control());
4618 Node* destx = _gvn.transform(dest);
4619 assert(destx == dest, "where has the allocation result gone?");
4620 }
4621 }
4622
4623
4624 //------------------------------inline_arraycopy-----------------------
4625 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4626 // Object dest, int destPos,
4627 // int length);
4628 bool LibraryCallKit::inline_arraycopy() {
4629 // Get the arguments.
4630 Node* src = argument(0); // type: oop
4631 Node* src_offset = argument(1); // type: int
4632 Node* dest = argument(2); // type: oop
4633 Node* dest_offset = argument(3); // type: int
4634 Node* length = argument(4); // type: int
4635
4636 uint new_idx = C->unique();
4637
4638 // Check for allocation before we add nodes that would confuse
4639 // tightly_coupled_allocation()
4640 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4641
4642 int saved_reexecute_sp = -1;
4643 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4644 // See arraycopy_restore_alloc_state() comment
4645 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4646 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4647 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4648 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4649
4650 // The following tests must be performed
4651 // (1) src and dest are arrays.
4652 // (2) src and dest arrays must have elements of the same BasicType
4653 // (3) src and dest must not be null.
4654 // (4) src_offset must not be negative.
4655 // (5) dest_offset must not be negative.
4656 // (6) length must not be negative.
4657 // (7) src_offset + length must not exceed length of src.
4658 // (8) dest_offset + length must not exceed length of dest.
4659 // (9) each element of an oop array must be assignable
4660
4661 // (3) src and dest must not be null.
4662 // always do this here because we need the JVM state for uncommon traps
4663 Node* null_ctl = top();
4664 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4665 assert(null_ctl->is_top(), "no null control here");
4666 dest = null_check(dest, T_ARRAY);
4667
4668 if (!can_emit_guards) {
4669 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4670 // guards but the arraycopy node could still take advantage of a
4671 // tightly allocated allocation. tightly_coupled_allocation() is
4672 // called again to make sure it takes the null check above into
4673 // account: the null check is mandatory and if it caused an
4674 // uncommon trap to be emitted then the allocation can't be
4675 // considered tightly coupled in this context.
4676 alloc = tightly_coupled_allocation(dest, NULL);
4677 }
4678
4679 bool validated = false;
4680
4681 const Type* src_type = _gvn.type(src);
4682 const Type* dest_type = _gvn.type(dest);
4683 const TypeAryPtr* top_src = src_type->isa_aryptr();
4684 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4685
4686 // Do we have the type of src?
4687 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4688 // Do we have the type of dest?
4689 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4690 // Is the type for src from speculation?
4691 bool src_spec = false;
4692 // Is the type for dest from speculation?
4693 bool dest_spec = false;
4694
4695 if ((!has_src || !has_dest) && can_emit_guards) {
4696 // We don't have sufficient type information, let's see if
4697 // speculative types can help. We need to have types for both src
4698 // and dest so that it pays off.
4699
4700 // Do we already have or could we have type information for src
4701 bool could_have_src = has_src;
4702 // Do we already have or could we have type information for dest
4703 bool could_have_dest = has_dest;
4704
4705 ciKlass* src_k = NULL;
4706 if (!has_src) {
4707 src_k = src_type->speculative_type_not_null();
4708 if (src_k != NULL && src_k->is_array_klass()) {
4709 could_have_src = true;
4710 }
4711 }
4712
4713 ciKlass* dest_k = NULL;
4714 if (!has_dest) {
4715 dest_k = dest_type->speculative_type_not_null();
4716 if (dest_k != NULL && dest_k->is_array_klass()) {
4717 could_have_dest = true;
4718 }
4719 }
4720
4721 if (could_have_src && could_have_dest) {
4722 // This is going to pay off so emit the required guards
4723 if (!has_src) {
4724 src = maybe_cast_profiled_obj(src, src_k, true);
4725 src_type = _gvn.type(src);
4726 top_src = src_type->isa_aryptr();
4727 has_src = (top_src != NULL && top_src->klass() != NULL);
4728 src_spec = true;
4729 }
4730 if (!has_dest) {
4731 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4732 dest_type = _gvn.type(dest);
4733 top_dest = dest_type->isa_aryptr();
4734 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4735 dest_spec = true;
4736 }
4737 }
4738 }
4739
4740 if (has_src && has_dest && can_emit_guards) {
4741 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4742 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4743 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
4744 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
4745
4746 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4747 // If both arrays are object arrays then having the exact types
4748 // for both will remove the need for a subtype check at runtime
4749 // before the call and may make it possible to pick a faster copy
4750 // routine (without a subtype check on every element)
4751 // Do we have the exact type of src?
4752 bool could_have_src = src_spec;
4753 // Do we have the exact type of dest?
4754 bool could_have_dest = dest_spec;
4755 ciKlass* src_k = top_src->klass();
4756 ciKlass* dest_k = top_dest->klass();
4757 if (!src_spec) {
4758 src_k = src_type->speculative_type_not_null();
4759 if (src_k != NULL && src_k->is_array_klass()) {
4760 could_have_src = true;
4761 }
4762 }
4763 if (!dest_spec) {
4764 dest_k = dest_type->speculative_type_not_null();
4765 if (dest_k != NULL && dest_k->is_array_klass()) {
4766 could_have_dest = true;
4767 }
4768 }
4769 if (could_have_src && could_have_dest) {
4770 // If we can have both exact types, emit the missing guards
4771 if (could_have_src && !src_spec) {
4772 src = maybe_cast_profiled_obj(src, src_k, true);
4773 }
4774 if (could_have_dest && !dest_spec) {
4775 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4776 }
4777 }
4778 }
4779 }
4780
4781 ciMethod* trap_method = method();
4782 int trap_bci = bci();
4783 if (saved_jvms != NULL) {
4784 trap_method = alloc->jvms()->method();
4785 trap_bci = alloc->jvms()->bci();
4786 }
4787
4788 bool negative_length_guard_generated = false;
4789
4790 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4791 can_emit_guards &&
4792 !src->is_top() && !dest->is_top()) {
4793 // validate arguments: enables transformation the ArrayCopyNode
4794 validated = true;
4795
4796 RegionNode* slow_region = new RegionNode(1);
4797 record_for_igvn(slow_region);
4798
4799 // (1) src and dest are arrays.
4800 generate_non_array_guard(load_object_klass(src), slow_region);
4801 generate_non_array_guard(load_object_klass(dest), slow_region);
4802
4803 // (2) src and dest arrays must have elements of the same BasicType
4804 // done at macro expansion or at Ideal transformation time
4805
4806 // (4) src_offset must not be negative.
4807 generate_negative_guard(src_offset, slow_region);
4808
4809 // (5) dest_offset must not be negative.
4810 generate_negative_guard(dest_offset, slow_region);
4811
4812 // (7) src_offset + length must not exceed length of src.
4813 generate_limit_guard(src_offset, length,
4814 load_array_length(src),
4815 slow_region);
4816
4817 // (8) dest_offset + length must not exceed length of dest.
4818 generate_limit_guard(dest_offset, length,
4819 load_array_length(dest),
4820 slow_region);
4821
4822 // (6) length must not be negative.
4823 // This is also checked in generate_arraycopy() during macro expansion, but
4824 // we also have to check it here for the case where the ArrayCopyNode will
4825 // be eliminated by Escape Analysis.
4826 if (EliminateAllocations) {
4827 generate_negative_guard(length, slow_region);
4828 negative_length_guard_generated = true;
4829 }
4830
4831 // (9) each element of an oop array must be assignable
4832 Node* src_klass = load_object_klass(src);
4833 Node* dest_klass = load_object_klass(dest);
4834 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4835
4836 if (not_subtype_ctrl != top()) {
4837 PreserveJVMState pjvms(this);
4838 set_control(not_subtype_ctrl);
4839 uncommon_trap(Deoptimization::Reason_intrinsic,
4840 Deoptimization::Action_make_not_entrant);
4841 assert(stopped(), "Should be stopped");
4842 }
4843 {
4844 PreserveJVMState pjvms(this);
4845 set_control(_gvn.transform(slow_region));
4846 uncommon_trap(Deoptimization::Reason_intrinsic,
4847 Deoptimization::Action_make_not_entrant);
4848 assert(stopped(), "Should be stopped");
4849 }
4850
4851 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4852 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4853 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4854 }
4855
4856 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4857
4858 if (stopped()) {
4859 return true;
4860 }
4861
4862 Node* new_src = access_resolve(src, ACCESS_READ);
4863 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
4864
4865 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4866 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4867 // so the compiler has a chance to eliminate them: during macro expansion,
4868 // we have to set their control (CastPP nodes are eliminated).
4869 load_object_klass(src), load_object_klass(dest),
4870 load_array_length(src), load_array_length(dest));
4871
4872 ac->set_arraycopy(validated);
4873
4874 Node* n = _gvn.transform(ac);
4875 if (n == ac) {
4876 ac->connect_outputs(this);
4877 } else {
4878 assert(validated, "shouldn't transform if all arguments not validated");
4879 set_all_memory(n);
4880 }
4881 clear_upper_avx();
4882
4883
4884 return true;
4885 }
4886
4887
4888 // Helper function which determines if an arraycopy immediately follows
4889 // an allocation, with no intervening tests or other escapes for the object.
4890 AllocateArrayNode*
4891 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4892 RegionNode* slow_region) {
4893 if (stopped()) return NULL; // no fast path
4894 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4895
4896 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4897 if (alloc == NULL) return NULL;
4898
4899 Node* rawmem = memory(Compile::AliasIdxRaw);
4900 // Is the allocation's memory state untouched?
4901 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4902 // Bail out if there have been raw-memory effects since the allocation.
4903 // (Example: There might have been a call or safepoint.)
4904 return NULL;
4905 }
4906 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4907 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4908 return NULL;
4909 }
4910
4911 // There must be no unexpected observers of this allocation.
4912 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4913 Node* obs = ptr->fast_out(i);
4914 if (obs != this->map()) {
4915 return NULL;
4916 }
4917 }
4918
4919 // This arraycopy must unconditionally follow the allocation of the ptr.
4920 Node* alloc_ctl = ptr->in(0);
4921 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4922
4923 Node* ctl = control();
4924 while (ctl != alloc_ctl) {
4925 // There may be guards which feed into the slow_region.
4926 // Any other control flow means that we might not get a chance
4927 // to finish initializing the allocated object.
4928 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4929 IfNode* iff = ctl->in(0)->as_If();
4930 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4931 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4932 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4933 ctl = iff->in(0); // This test feeds the known slow_region.
4934 continue;
4935 }
4936 // One more try: Various low-level checks bottom out in
4937 // uncommon traps. If the debug-info of the trap omits
4938 // any reference to the allocation, as we've already
4939 // observed, then there can be no objection to the trap.
4940 bool found_trap = false;
4941 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4942 Node* obs = not_ctl->fast_out(j);
4943 if (obs->in(0) == not_ctl && obs->is_Call() &&
4944 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4945 found_trap = true; break;
4946 }
4947 }
4948 if (found_trap) {
4949 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4950 continue;
4951 }
4952 }
4953 return NULL;
4954 }
4955
4956 // If we get this far, we have an allocation which immediately
4957 // precedes the arraycopy, and we can take over zeroing the new object.
4958 // The arraycopy will finish the initialization, and provide
4959 // a new control state to which we will anchor the destination pointer.
4960
4961 return alloc;
4962 }
4963
4964 //-------------inline_encodeISOArray-----------------------------------
4965 // encode char[] to byte[] in ISO_8859_1
4966 bool LibraryCallKit::inline_encodeISOArray() {
4967 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4968 // no receiver since it is static method
4969 Node *src = argument(0);
4970 Node *src_offset = argument(1);
4971 Node *dst = argument(2);
4972 Node *dst_offset = argument(3);
4973 Node *length = argument(4);
4974
4975 src = must_be_not_null(src, true);
4976 dst = must_be_not_null(dst, true);
4977
4978 src = access_resolve(src, ACCESS_READ);
4979 dst = access_resolve(dst, ACCESS_WRITE);
4980
4981 const Type* src_type = src->Value(&_gvn);
4982 const Type* dst_type = dst->Value(&_gvn);
4983 const TypeAryPtr* top_src = src_type->isa_aryptr();
4984 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4985 if (top_src == NULL || top_src->klass() == NULL ||
4986 top_dest == NULL || top_dest->klass() == NULL) {
4987 // failed array check
4988 return false;
4989 }
4990
4991 // Figure out the size and type of the elements we will be copying.
4992 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4993 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4994 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4995 return false;
4996 }
4997
4998 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4999 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5000 // 'src_start' points to src array + scaled offset
5001 // 'dst_start' points to dst array + scaled offset
5002
5003 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5004 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5005 enc = _gvn.transform(enc);
5006 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5007 set_memory(res_mem, mtype);
5008 set_result(enc);
5009 clear_upper_avx();
5010
5011 return true;
5012 }
5013
5014 //-------------inline_multiplyToLen-----------------------------------
5015 bool LibraryCallKit::inline_multiplyToLen() {
5016 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5017
5018 address stubAddr = StubRoutines::multiplyToLen();
5019 if (stubAddr == NULL) {
5020 return false; // Intrinsic's stub is not implemented on this platform
5021 }
5022 const char* stubName = "multiplyToLen";
5023
5024 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5025
5026 // no receiver because it is a static method
5027 Node* x = argument(0);
5028 Node* xlen = argument(1);
5029 Node* y = argument(2);
5030 Node* ylen = argument(3);
5031 Node* z = argument(4);
5032
5033 x = must_be_not_null(x, true);
5034 y = must_be_not_null(y, true);
5035
5036 x = access_resolve(x, ACCESS_READ);
5037 y = access_resolve(y, ACCESS_READ);
5038 z = access_resolve(z, ACCESS_WRITE);
5039
5040 const Type* x_type = x->Value(&_gvn);
5041 const Type* y_type = y->Value(&_gvn);
5042 const TypeAryPtr* top_x = x_type->isa_aryptr();
5043 const TypeAryPtr* top_y = y_type->isa_aryptr();
5044 if (top_x == NULL || top_x->klass() == NULL ||
5045 top_y == NULL || top_y->klass() == NULL) {
5046 // failed array check
5047 return false;
5048 }
5049
5050 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5051 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5052 if (x_elem != T_INT || y_elem != T_INT) {
5053 return false;
5054 }
5055
5056 // Set the original stack and the reexecute bit for the interpreter to reexecute
5057 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5058 // on the return from z array allocation in runtime.
5059 { PreserveReexecuteState preexecs(this);
5060 jvms()->set_should_reexecute(true);
5061
5062 Node* x_start = array_element_address(x, intcon(0), x_elem);
5063 Node* y_start = array_element_address(y, intcon(0), y_elem);
5064 // 'x_start' points to x array + scaled xlen
5065 // 'y_start' points to y array + scaled ylen
5066
5067 // Allocate the result array
5068 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5069 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5070 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5071
5072 IdealKit ideal(this);
5073
5074 #define __ ideal.
5075 Node* one = __ ConI(1);
5076 Node* zero = __ ConI(0);
5077 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5078 __ set(need_alloc, zero);
5079 __ set(z_alloc, z);
5080 __ if_then(z, BoolTest::eq, null()); {
5081 __ increment (need_alloc, one);
5082 } __ else_(); {
5083 // Update graphKit memory and control from IdealKit.
5084 sync_kit(ideal);
5085 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5086 cast->init_req(0, control());
5087 _gvn.set_type(cast, cast->bottom_type());
5088 C->record_for_igvn(cast);
5089
5090 Node* zlen_arg = load_array_length(cast);
5091 // Update IdealKit memory and control from graphKit.
5092 __ sync_kit(this);
5093 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5094 __ increment (need_alloc, one);
5095 } __ end_if();
5096 } __ end_if();
5097
5098 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5099 // Update graphKit memory and control from IdealKit.
5100 sync_kit(ideal);
5101 Node * narr = new_array(klass_node, zlen, 1);
5102 // Update IdealKit memory and control from graphKit.
5103 __ sync_kit(this);
5104 __ set(z_alloc, narr);
5105 } __ end_if();
5106
5107 sync_kit(ideal);
5108 z = __ value(z_alloc);
5109 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5110 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5111 // Final sync IdealKit and GraphKit.
5112 final_sync(ideal);
5113 #undef __
5114
5115 Node* z_start = array_element_address(z, intcon(0), T_INT);
5116
5117 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5118 OptoRuntime::multiplyToLen_Type(),
5119 stubAddr, stubName, TypePtr::BOTTOM,
5120 x_start, xlen, y_start, ylen, z_start, zlen);
5121 } // original reexecute is set back here
5122
5123 C->set_has_split_ifs(true); // Has chance for split-if optimization
5124 set_result(z);
5125 return true;
5126 }
5127
5128 //-------------inline_squareToLen------------------------------------
5129 bool LibraryCallKit::inline_squareToLen() {
5130 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5131
5132 address stubAddr = StubRoutines::squareToLen();
5133 if (stubAddr == NULL) {
5134 return false; // Intrinsic's stub is not implemented on this platform
5135 }
5136 const char* stubName = "squareToLen";
5137
5138 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5139
5140 Node* x = argument(0);
5141 Node* len = argument(1);
5142 Node* z = argument(2);
5143 Node* zlen = argument(3);
5144
5145 x = must_be_not_null(x, true);
5146 z = must_be_not_null(z, true);
5147
5148 x = access_resolve(x, ACCESS_READ);
5149 z = access_resolve(z, ACCESS_WRITE);
5150
5151 const Type* x_type = x->Value(&_gvn);
5152 const Type* z_type = z->Value(&_gvn);
5153 const TypeAryPtr* top_x = x_type->isa_aryptr();
5154 const TypeAryPtr* top_z = z_type->isa_aryptr();
5155 if (top_x == NULL || top_x->klass() == NULL ||
5156 top_z == NULL || top_z->klass() == NULL) {
5157 // failed array check
5158 return false;
5159 }
5160
5161 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5162 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5163 if (x_elem != T_INT || z_elem != T_INT) {
5164 return false;
5165 }
5166
5167
5168 Node* x_start = array_element_address(x, intcon(0), x_elem);
5169 Node* z_start = array_element_address(z, intcon(0), z_elem);
5170
5171 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5172 OptoRuntime::squareToLen_Type(),
5173 stubAddr, stubName, TypePtr::BOTTOM,
5174 x_start, len, z_start, zlen);
5175
5176 set_result(z);
5177 return true;
5178 }
5179
5180 //-------------inline_mulAdd------------------------------------------
5181 bool LibraryCallKit::inline_mulAdd() {
5182 assert(UseMulAddIntrinsic, "not implemented on this platform");
5183
5184 address stubAddr = StubRoutines::mulAdd();
5185 if (stubAddr == NULL) {
5186 return false; // Intrinsic's stub is not implemented on this platform
5187 }
5188 const char* stubName = "mulAdd";
5189
5190 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5191
5192 Node* out = argument(0);
5193 Node* in = argument(1);
5194 Node* offset = argument(2);
5195 Node* len = argument(3);
5196 Node* k = argument(4);
5197
5198 out = must_be_not_null(out, true);
5199
5200 in = access_resolve(in, ACCESS_READ);
5201 out = access_resolve(out, ACCESS_WRITE);
5202
5203 const Type* out_type = out->Value(&_gvn);
5204 const Type* in_type = in->Value(&_gvn);
5205 const TypeAryPtr* top_out = out_type->isa_aryptr();
5206 const TypeAryPtr* top_in = in_type->isa_aryptr();
5207 if (top_out == NULL || top_out->klass() == NULL ||
5208 top_in == NULL || top_in->klass() == NULL) {
5209 // failed array check
5210 return false;
5211 }
5212
5213 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5214 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5215 if (out_elem != T_INT || in_elem != T_INT) {
5216 return false;
5217 }
5218
5219 Node* outlen = load_array_length(out);
5220 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5221 Node* out_start = array_element_address(out, intcon(0), out_elem);
5222 Node* in_start = array_element_address(in, intcon(0), in_elem);
5223
5224 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5225 OptoRuntime::mulAdd_Type(),
5226 stubAddr, stubName, TypePtr::BOTTOM,
5227 out_start,in_start, new_offset, len, k);
5228 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5229 set_result(result);
5230 return true;
5231 }
5232
5233 //-------------inline_montgomeryMultiply-----------------------------------
5234 bool LibraryCallKit::inline_montgomeryMultiply() {
5235 address stubAddr = StubRoutines::montgomeryMultiply();
5236 if (stubAddr == NULL) {
5237 return false; // Intrinsic's stub is not implemented on this platform
5238 }
5239
5240 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5241 const char* stubName = "montgomery_multiply";
5242
5243 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5244
5245 Node* a = argument(0);
5246 Node* b = argument(1);
5247 Node* n = argument(2);
5248 Node* len = argument(3);
5249 Node* inv = argument(4);
5250 Node* m = argument(6);
5251
5252 a = access_resolve(a, ACCESS_READ);
5253 b = access_resolve(b, ACCESS_READ);
5254 n = access_resolve(n, ACCESS_READ);
5255 m = access_resolve(m, ACCESS_WRITE);
5256
5257 const Type* a_type = a->Value(&_gvn);
5258 const TypeAryPtr* top_a = a_type->isa_aryptr();
5259 const Type* b_type = b->Value(&_gvn);
5260 const TypeAryPtr* top_b = b_type->isa_aryptr();
5261 const Type* n_type = a->Value(&_gvn);
5262 const TypeAryPtr* top_n = n_type->isa_aryptr();
5263 const Type* m_type = a->Value(&_gvn);
5264 const TypeAryPtr* top_m = m_type->isa_aryptr();
5265 if (top_a == NULL || top_a->klass() == NULL ||
5266 top_b == NULL || top_b->klass() == NULL ||
5267 top_n == NULL || top_n->klass() == NULL ||
5268 top_m == NULL || top_m->klass() == NULL) {
5269 // failed array check
5270 return false;
5271 }
5272
5273 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5274 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5275 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5276 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5277 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5278 return false;
5279 }
5280
5281 // Make the call
5282 {
5283 Node* a_start = array_element_address(a, intcon(0), a_elem);
5284 Node* b_start = array_element_address(b, intcon(0), b_elem);
5285 Node* n_start = array_element_address(n, intcon(0), n_elem);
5286 Node* m_start = array_element_address(m, intcon(0), m_elem);
5287
5288 Node* call = make_runtime_call(RC_LEAF,
5289 OptoRuntime::montgomeryMultiply_Type(),
5290 stubAddr, stubName, TypePtr::BOTTOM,
5291 a_start, b_start, n_start, len, inv, top(),
5292 m_start);
5293 set_result(m);
5294 }
5295
5296 return true;
5297 }
5298
5299 bool LibraryCallKit::inline_montgomerySquare() {
5300 address stubAddr = StubRoutines::montgomerySquare();
5301 if (stubAddr == NULL) {
5302 return false; // Intrinsic's stub is not implemented on this platform
5303 }
5304
5305 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5306 const char* stubName = "montgomery_square";
5307
5308 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5309
5310 Node* a = argument(0);
5311 Node* n = argument(1);
5312 Node* len = argument(2);
5313 Node* inv = argument(3);
5314 Node* m = argument(5);
5315
5316 a = access_resolve(a, ACCESS_READ);
5317 n = access_resolve(n, ACCESS_READ);
5318 m = access_resolve(m, ACCESS_WRITE);
5319
5320 const Type* a_type = a->Value(&_gvn);
5321 const TypeAryPtr* top_a = a_type->isa_aryptr();
5322 const Type* n_type = a->Value(&_gvn);
5323 const TypeAryPtr* top_n = n_type->isa_aryptr();
5324 const Type* m_type = a->Value(&_gvn);
5325 const TypeAryPtr* top_m = m_type->isa_aryptr();
5326 if (top_a == NULL || top_a->klass() == NULL ||
5327 top_n == NULL || top_n->klass() == NULL ||
5328 top_m == NULL || top_m->klass() == NULL) {
5329 // failed array check
5330 return false;
5331 }
5332
5333 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5334 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5335 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5336 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5337 return false;
5338 }
5339
5340 // Make the call
5341 {
5342 Node* a_start = array_element_address(a, intcon(0), a_elem);
5343 Node* n_start = array_element_address(n, intcon(0), n_elem);
5344 Node* m_start = array_element_address(m, intcon(0), m_elem);
5345
5346 Node* call = make_runtime_call(RC_LEAF,
5347 OptoRuntime::montgomerySquare_Type(),
5348 stubAddr, stubName, TypePtr::BOTTOM,
5349 a_start, n_start, len, inv, top(),
5350 m_start);
5351 set_result(m);
5352 }
5353
5354 return true;
5355 }
5356
5357 //-------------inline_vectorizedMismatch------------------------------
5358 bool LibraryCallKit::inline_vectorizedMismatch() {
5359 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5360
5361 address stubAddr = StubRoutines::vectorizedMismatch();
5362 if (stubAddr == NULL) {
5363 return false; // Intrinsic's stub is not implemented on this platform
5364 }
5365 const char* stubName = "vectorizedMismatch";
5366 int size_l = callee()->signature()->size();
5367 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5368
5369 Node* obja = argument(0);
5370 Node* aoffset = argument(1);
5371 Node* objb = argument(3);
5372 Node* boffset = argument(4);
5373 Node* length = argument(6);
5374 Node* scale = argument(7);
5375
5376 const Type* a_type = obja->Value(&_gvn);
5377 const Type* b_type = objb->Value(&_gvn);
5378 const TypeAryPtr* top_a = a_type->isa_aryptr();
5379 const TypeAryPtr* top_b = b_type->isa_aryptr();
5380 if (top_a == NULL || top_a->klass() == NULL ||
5381 top_b == NULL || top_b->klass() == NULL) {
5382 // failed array check
5383 return false;
5384 }
5385
5386 Node* call;
5387 jvms()->set_should_reexecute(true);
5388
5389 obja = access_resolve(obja, ACCESS_READ);
5390 objb = access_resolve(objb, ACCESS_READ);
5391 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5392 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5393
5394 call = make_runtime_call(RC_LEAF,
5395 OptoRuntime::vectorizedMismatch_Type(),
5396 stubAddr, stubName, TypePtr::BOTTOM,
5397 obja_adr, objb_adr, length, scale);
5398
5399 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5400 set_result(result);
5401 return true;
5402 }
5403
5404 /**
5405 * Calculate CRC32 for byte.
5406 * int java.util.zip.CRC32.update(int crc, int b)
5407 */
5408 bool LibraryCallKit::inline_updateCRC32() {
5409 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5410 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5411 // no receiver since it is static method
5412 Node* crc = argument(0); // type: int
5413 Node* b = argument(1); // type: int
5414
5415 /*
5416 * int c = ~ crc;
5417 * b = timesXtoThe32[(b ^ c) & 0xFF];
5418 * b = b ^ (c >>> 8);
5419 * crc = ~b;
5420 */
5421
5422 Node* M1 = intcon(-1);
5423 crc = _gvn.transform(new XorINode(crc, M1));
5424 Node* result = _gvn.transform(new XorINode(crc, b));
5425 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5426
5427 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5428 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5429 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5430 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5431
5432 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5433 result = _gvn.transform(new XorINode(crc, result));
5434 result = _gvn.transform(new XorINode(result, M1));
5435 set_result(result);
5436 return true;
5437 }
5438
5439 /**
5440 * Calculate CRC32 for byte[] array.
5441 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5442 */
5443 bool LibraryCallKit::inline_updateBytesCRC32() {
5444 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5445 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5446 // no receiver since it is static method
5447 Node* crc = argument(0); // type: int
5448 Node* src = argument(1); // type: oop
5449 Node* offset = argument(2); // type: int
5450 Node* length = argument(3); // type: int
5451
5452 const Type* src_type = src->Value(&_gvn);
5453 const TypeAryPtr* top_src = src_type->isa_aryptr();
5454 if (top_src == NULL || top_src->klass() == NULL) {
5455 // failed array check
5456 return false;
5457 }
5458
5459 // Figure out the size and type of the elements we will be copying.
5460 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5461 if (src_elem != T_BYTE) {
5462 return false;
5463 }
5464
5465 // 'src_start' points to src array + scaled offset
5466 src = must_be_not_null(src, true);
5467 src = access_resolve(src, ACCESS_READ);
5468 Node* src_start = array_element_address(src, offset, src_elem);
5469
5470 // We assume that range check is done by caller.
5471 // TODO: generate range check (offset+length < src.length) in debug VM.
5472
5473 // Call the stub.
5474 address stubAddr = StubRoutines::updateBytesCRC32();
5475 const char *stubName = "updateBytesCRC32";
5476
5477 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5478 stubAddr, stubName, TypePtr::BOTTOM,
5479 crc, src_start, length);
5480 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5481 set_result(result);
5482 return true;
5483 }
5484
5485 /**
5486 * Calculate CRC32 for ByteBuffer.
5487 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5488 */
5489 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5490 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5491 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5492 // no receiver since it is static method
5493 Node* crc = argument(0); // type: int
5494 Node* src = argument(1); // type: long
5495 Node* offset = argument(3); // type: int
5496 Node* length = argument(4); // type: int
5497
5498 src = ConvL2X(src); // adjust Java long to machine word
5499 Node* base = _gvn.transform(new CastX2PNode(src));
5500 offset = ConvI2X(offset);
5501
5502 // 'src_start' points to src array + scaled offset
5503 Node* src_start = basic_plus_adr(top(), base, offset);
5504
5505 // Call the stub.
5506 address stubAddr = StubRoutines::updateBytesCRC32();
5507 const char *stubName = "updateBytesCRC32";
5508
5509 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5510 stubAddr, stubName, TypePtr::BOTTOM,
5511 crc, src_start, length);
5512 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5513 set_result(result);
5514 return true;
5515 }
5516
5517 //------------------------------get_table_from_crc32c_class-----------------------
5518 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5519 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5520 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5521
5522 return table;
5523 }
5524
5525 //------------------------------inline_updateBytesCRC32C-----------------------
5526 //
5527 // Calculate CRC32C for byte[] array.
5528 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5529 //
5530 bool LibraryCallKit::inline_updateBytesCRC32C() {
5531 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5532 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5533 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5534 // no receiver since it is a static method
5535 Node* crc = argument(0); // type: int
5536 Node* src = argument(1); // type: oop
5537 Node* offset = argument(2); // type: int
5538 Node* end = argument(3); // type: int
5539
5540 Node* length = _gvn.transform(new SubINode(end, offset));
5541
5542 const Type* src_type = src->Value(&_gvn);
5543 const TypeAryPtr* top_src = src_type->isa_aryptr();
5544 if (top_src == NULL || top_src->klass() == NULL) {
5545 // failed array check
5546 return false;
5547 }
5548
5549 // Figure out the size and type of the elements we will be copying.
5550 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5551 if (src_elem != T_BYTE) {
5552 return false;
5553 }
5554
5555 // 'src_start' points to src array + scaled offset
5556 src = must_be_not_null(src, true);
5557 src = access_resolve(src, ACCESS_READ);
5558 Node* src_start = array_element_address(src, offset, src_elem);
5559
5560 // static final int[] byteTable in class CRC32C
5561 Node* table = get_table_from_crc32c_class(callee()->holder());
5562 table = must_be_not_null(table, true);
5563 table = access_resolve(table, ACCESS_READ);
5564 Node* table_start = array_element_address(table, intcon(0), T_INT);
5565
5566 // We assume that range check is done by caller.
5567 // TODO: generate range check (offset+length < src.length) in debug VM.
5568
5569 // Call the stub.
5570 address stubAddr = StubRoutines::updateBytesCRC32C();
5571 const char *stubName = "updateBytesCRC32C";
5572
5573 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5574 stubAddr, stubName, TypePtr::BOTTOM,
5575 crc, src_start, length, table_start);
5576 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5577 set_result(result);
5578 return true;
5579 }
5580
5581 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5582 //
5583 // Calculate CRC32C for DirectByteBuffer.
5584 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5585 //
5586 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5587 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5588 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5589 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5590 // no receiver since it is a static method
5591 Node* crc = argument(0); // type: int
5592 Node* src = argument(1); // type: long
5593 Node* offset = argument(3); // type: int
5594 Node* end = argument(4); // type: int
5595
5596 Node* length = _gvn.transform(new SubINode(end, offset));
5597
5598 src = ConvL2X(src); // adjust Java long to machine word
5599 Node* base = _gvn.transform(new CastX2PNode(src));
5600 offset = ConvI2X(offset);
5601
5602 // 'src_start' points to src array + scaled offset
5603 Node* src_start = basic_plus_adr(top(), base, offset);
5604
5605 // static final int[] byteTable in class CRC32C
5606 Node* table = get_table_from_crc32c_class(callee()->holder());
5607 table = must_be_not_null(table, true);
5608 table = access_resolve(table, ACCESS_READ);
5609 Node* table_start = array_element_address(table, intcon(0), T_INT);
5610
5611 // Call the stub.
5612 address stubAddr = StubRoutines::updateBytesCRC32C();
5613 const char *stubName = "updateBytesCRC32C";
5614
5615 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5616 stubAddr, stubName, TypePtr::BOTTOM,
5617 crc, src_start, length, table_start);
5618 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5619 set_result(result);
5620 return true;
5621 }
5622
5623 //------------------------------inline_updateBytesAdler32----------------------
5624 //
5625 // Calculate Adler32 checksum for byte[] array.
5626 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5627 //
5628 bool LibraryCallKit::inline_updateBytesAdler32() {
5629 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5630 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5631 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5632 // no receiver since it is static method
5633 Node* crc = argument(0); // type: int
5634 Node* src = argument(1); // type: oop
5635 Node* offset = argument(2); // type: int
5636 Node* length = argument(3); // type: int
5637
5638 const Type* src_type = src->Value(&_gvn);
5639 const TypeAryPtr* top_src = src_type->isa_aryptr();
5640 if (top_src == NULL || top_src->klass() == NULL) {
5641 // failed array check
5642 return false;
5643 }
5644
5645 // Figure out the size and type of the elements we will be copying.
5646 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5647 if (src_elem != T_BYTE) {
5648 return false;
5649 }
5650
5651 // 'src_start' points to src array + scaled offset
5652 src = access_resolve(src, ACCESS_READ);
5653 Node* src_start = array_element_address(src, offset, src_elem);
5654
5655 // We assume that range check is done by caller.
5656 // TODO: generate range check (offset+length < src.length) in debug VM.
5657
5658 // Call the stub.
5659 address stubAddr = StubRoutines::updateBytesAdler32();
5660 const char *stubName = "updateBytesAdler32";
5661
5662 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5663 stubAddr, stubName, TypePtr::BOTTOM,
5664 crc, src_start, length);
5665 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5666 set_result(result);
5667 return true;
5668 }
5669
5670 //------------------------------inline_updateByteBufferAdler32---------------
5671 //
5672 // Calculate Adler32 checksum for DirectByteBuffer.
5673 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5674 //
5675 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5676 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5677 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5678 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5679 // no receiver since it is static method
5680 Node* crc = argument(0); // type: int
5681 Node* src = argument(1); // type: long
5682 Node* offset = argument(3); // type: int
5683 Node* length = argument(4); // type: int
5684
5685 src = ConvL2X(src); // adjust Java long to machine word
5686 Node* base = _gvn.transform(new CastX2PNode(src));
5687 offset = ConvI2X(offset);
5688
5689 // 'src_start' points to src array + scaled offset
5690 Node* src_start = basic_plus_adr(top(), base, offset);
5691
5692 // Call the stub.
5693 address stubAddr = StubRoutines::updateBytesAdler32();
5694 const char *stubName = "updateBytesAdler32";
5695
5696 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5697 stubAddr, stubName, TypePtr::BOTTOM,
5698 crc, src_start, length);
5699
5700 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5701 set_result(result);
5702 return true;
5703 }
5704
5705 //----------------------------inline_reference_get----------------------------
5706 // public T java.lang.ref.Reference.get();
5707 bool LibraryCallKit::inline_reference_get() {
5708 const int referent_offset = java_lang_ref_Reference::referent_offset;
5709 guarantee(referent_offset > 0, "should have already been set");
5710
5711 // Get the argument:
5712 Node* reference_obj = null_check_receiver();
5713 if (stopped()) return true;
5714
5715 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5716 assert(tinst != NULL, "obj is null");
5717 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5718 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5719 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5720 ciSymbol::make("Ljava/lang/Object;"),
5721 false);
5722 assert (field != NULL, "undefined field");
5723
5724 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5725 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5726
5727 ciInstanceKlass* klass = env()->Object_klass();
5728 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5729
5730 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5731 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5732 // Add memory barrier to prevent commoning reads from this field
5733 // across safepoint since GC can change its value.
5734 insert_mem_bar(Op_MemBarCPUOrder);
5735
5736 set_result(result);
5737 return true;
5738 }
5739
5740
5741 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5742 bool is_exact=true, bool is_static=false,
5743 ciInstanceKlass * fromKls=NULL) {
5744 if (fromKls == NULL) {
5745 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5746 assert(tinst != NULL, "obj is null");
5747 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5748 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5749 fromKls = tinst->klass()->as_instance_klass();
5750 } else {
5751 assert(is_static, "only for static field access");
5752 }
5753 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5754 ciSymbol::make(fieldTypeString),
5755 is_static);
5756
5757 assert (field != NULL, "undefined field");
5758 if (field == NULL) return (Node *) NULL;
5759
5760 if (is_static) {
5761 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5762 fromObj = makecon(tip);
5763 }
5764
5765 // Next code copied from Parse::do_get_xxx():
5766
5767 // Compute address and memory type.
5768 int offset = field->offset_in_bytes();
5769 bool is_vol = field->is_volatile();
5770 ciType* field_klass = field->type();
5771 assert(field_klass->is_loaded(), "should be loaded");
5772 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5773 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5774 BasicType bt = field->layout_type();
5775
5776 // Build the resultant type of the load
5777 const Type *type;
5778 if (bt == T_OBJECT) {
5779 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5780 } else {
5781 type = Type::get_const_basic_type(bt);
5782 }
5783
5784 DecoratorSet decorators = IN_HEAP;
5785
5786 if (is_vol) {
5787 decorators |= MO_SEQ_CST;
5788 }
5789
5790 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5791 }
5792
5793 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5794 bool is_exact = true, bool is_static = false,
5795 ciInstanceKlass * fromKls = NULL) {
5796 if (fromKls == NULL) {
5797 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5798 assert(tinst != NULL, "obj is null");
5799 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5800 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5801 fromKls = tinst->klass()->as_instance_klass();
5802 }
5803 else {
5804 assert(is_static, "only for static field access");
5805 }
5806 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5807 ciSymbol::make(fieldTypeString),
5808 is_static);
5809
5810 assert(field != NULL, "undefined field");
5811 assert(!field->is_volatile(), "not defined for volatile fields");
5812
5813 if (is_static) {
5814 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5815 fromObj = makecon(tip);
5816 }
5817
5818 // Next code copied from Parse::do_get_xxx():
5819
5820 // Compute address and memory type.
5821 int offset = field->offset_in_bytes();
5822 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5823
5824 return adr;
5825 }
5826
5827 //------------------------------inline_aescrypt_Block-----------------------
5828 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5829 address stubAddr = NULL;
5830 const char *stubName;
5831 assert(UseAES, "need AES instruction support");
5832
5833 switch(id) {
5834 case vmIntrinsics::_aescrypt_encryptBlock:
5835 stubAddr = StubRoutines::aescrypt_encryptBlock();
5836 stubName = "aescrypt_encryptBlock";
5837 break;
5838 case vmIntrinsics::_aescrypt_decryptBlock:
5839 stubAddr = StubRoutines::aescrypt_decryptBlock();
5840 stubName = "aescrypt_decryptBlock";
5841 break;
5842 default:
5843 break;
5844 }
5845 if (stubAddr == NULL) return false;
5846
5847 Node* aescrypt_object = argument(0);
5848 Node* src = argument(1);
5849 Node* src_offset = argument(2);
5850 Node* dest = argument(3);
5851 Node* dest_offset = argument(4);
5852
5853 src = must_be_not_null(src, true);
5854 dest = must_be_not_null(dest, true);
5855
5856 src = access_resolve(src, ACCESS_READ);
5857 dest = access_resolve(dest, ACCESS_WRITE);
5858
5859 // (1) src and dest are arrays.
5860 const Type* src_type = src->Value(&_gvn);
5861 const Type* dest_type = dest->Value(&_gvn);
5862 const TypeAryPtr* top_src = src_type->isa_aryptr();
5863 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5864 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5865
5866 // for the quick and dirty code we will skip all the checks.
5867 // we are just trying to get the call to be generated.
5868 Node* src_start = src;
5869 Node* dest_start = dest;
5870 if (src_offset != NULL || dest_offset != NULL) {
5871 assert(src_offset != NULL && dest_offset != NULL, "");
5872 src_start = array_element_address(src, src_offset, T_BYTE);
5873 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5874 }
5875
5876 // now need to get the start of its expanded key array
5877 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5878 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5879 if (k_start == NULL) return false;
5880
5881 if (Matcher::pass_original_key_for_aes()) {
5882 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5883 // compatibility issues between Java key expansion and SPARC crypto instructions
5884 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5885 if (original_k_start == NULL) return false;
5886
5887 // Call the stub.
5888 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5889 stubAddr, stubName, TypePtr::BOTTOM,
5890 src_start, dest_start, k_start, original_k_start);
5891 } else {
5892 // Call the stub.
5893 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5894 stubAddr, stubName, TypePtr::BOTTOM,
5895 src_start, dest_start, k_start);
5896 }
5897
5898 return true;
5899 }
5900
5901 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5902 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5903 address stubAddr = NULL;
5904 const char *stubName = NULL;
5905
5906 assert(UseAES, "need AES instruction support");
5907
5908 switch(id) {
5909 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5910 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5911 stubName = "cipherBlockChaining_encryptAESCrypt";
5912 break;
5913 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5914 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5915 stubName = "cipherBlockChaining_decryptAESCrypt";
5916 break;
5917 default:
5918 break;
5919 }
5920 if (stubAddr == NULL) return false;
5921
5922 Node* cipherBlockChaining_object = argument(0);
5923 Node* src = argument(1);
5924 Node* src_offset = argument(2);
5925 Node* len = argument(3);
5926 Node* dest = argument(4);
5927 Node* dest_offset = argument(5);
5928
5929 src = must_be_not_null(src, false);
5930 dest = must_be_not_null(dest, false);
5931
5932 src = access_resolve(src, ACCESS_READ);
5933 dest = access_resolve(dest, ACCESS_WRITE);
5934
5935 // (1) src and dest are arrays.
5936 const Type* src_type = src->Value(&_gvn);
5937 const Type* dest_type = dest->Value(&_gvn);
5938 const TypeAryPtr* top_src = src_type->isa_aryptr();
5939 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5940 assert (top_src != NULL && top_src->klass() != NULL
5941 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5942
5943 // checks are the responsibility of the caller
5944 Node* src_start = src;
5945 Node* dest_start = dest;
5946 if (src_offset != NULL || dest_offset != NULL) {
5947 assert(src_offset != NULL && dest_offset != NULL, "");
5948 src_start = array_element_address(src, src_offset, T_BYTE);
5949 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5950 }
5951
5952 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5953 // (because of the predicated logic executed earlier).
5954 // so we cast it here safely.
5955 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5956
5957 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5958 if (embeddedCipherObj == NULL) return false;
5959
5960 // cast it to what we know it will be at runtime
5961 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5962 assert(tinst != NULL, "CBC obj is null");
5963 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5964 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5965 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5966
5967 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5968 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5969 const TypeOopPtr* xtype = aklass->as_instance_type();
5970 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5971 aescrypt_object = _gvn.transform(aescrypt_object);
5972
5973 // we need to get the start of the aescrypt_object's expanded key array
5974 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5975 if (k_start == NULL) return false;
5976
5977 // similarly, get the start address of the r vector
5978 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5979 if (objRvec == NULL) return false;
5980 objRvec = access_resolve(objRvec, ACCESS_WRITE);
5981 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5982
5983 Node* cbcCrypt;
5984 if (Matcher::pass_original_key_for_aes()) {
5985 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5986 // compatibility issues between Java key expansion and SPARC crypto instructions
5987 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5988 if (original_k_start == NULL) return false;
5989
5990 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5991 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5992 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5993 stubAddr, stubName, TypePtr::BOTTOM,
5994 src_start, dest_start, k_start, r_start, len, original_k_start);
5995 } else {
5996 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5997 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5998 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5999 stubAddr, stubName, TypePtr::BOTTOM,
6000 src_start, dest_start, k_start, r_start, len);
6001 }
6002
6003 // return cipher length (int)
6004 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6005 set_result(retvalue);
6006 return true;
6007 }
6008
6009 //------------------------------inline_counterMode_AESCrypt-----------------------
6010 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6011 assert(UseAES, "need AES instruction support");
6012 if (!UseAESCTRIntrinsics) return false;
6013
6014 address stubAddr = NULL;
6015 const char *stubName = NULL;
6016 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6017 stubAddr = StubRoutines::counterMode_AESCrypt();
6018 stubName = "counterMode_AESCrypt";
6019 }
6020 if (stubAddr == NULL) return false;
6021
6022 Node* counterMode_object = argument(0);
6023 Node* src = argument(1);
6024 Node* src_offset = argument(2);
6025 Node* len = argument(3);
6026 Node* dest = argument(4);
6027 Node* dest_offset = argument(5);
6028
6029 src = access_resolve(src, ACCESS_READ);
6030 dest = access_resolve(dest, ACCESS_WRITE);
6031 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6032
6033 // (1) src and dest are arrays.
6034 const Type* src_type = src->Value(&_gvn);
6035 const Type* dest_type = dest->Value(&_gvn);
6036 const TypeAryPtr* top_src = src_type->isa_aryptr();
6037 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6038 assert(top_src != NULL && top_src->klass() != NULL &&
6039 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6040
6041 // checks are the responsibility of the caller
6042 Node* src_start = src;
6043 Node* dest_start = dest;
6044 if (src_offset != NULL || dest_offset != NULL) {
6045 assert(src_offset != NULL && dest_offset != NULL, "");
6046 src_start = array_element_address(src, src_offset, T_BYTE);
6047 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6048 }
6049
6050 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6051 // (because of the predicated logic executed earlier).
6052 // so we cast it here safely.
6053 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6054 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6055 if (embeddedCipherObj == NULL) return false;
6056 // cast it to what we know it will be at runtime
6057 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6058 assert(tinst != NULL, "CTR obj is null");
6059 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6060 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6061 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6062 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6063 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6064 const TypeOopPtr* xtype = aklass->as_instance_type();
6065 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6066 aescrypt_object = _gvn.transform(aescrypt_object);
6067 // we need to get the start of the aescrypt_object's expanded key array
6068 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6069 if (k_start == NULL) return false;
6070 // similarly, get the start address of the r vector
6071 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6072 if (obj_counter == NULL) return false;
6073 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6074 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6075
6076 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6077 if (saved_encCounter == NULL) return false;
6078 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6079 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6080 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6081
6082 Node* ctrCrypt;
6083 if (Matcher::pass_original_key_for_aes()) {
6084 // no SPARC version for AES/CTR intrinsics now.
6085 return false;
6086 }
6087 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6088 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6089 OptoRuntime::counterMode_aescrypt_Type(),
6090 stubAddr, stubName, TypePtr::BOTTOM,
6091 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6092
6093 // return cipher length (int)
6094 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6095 set_result(retvalue);
6096 return true;
6097 }
6098
6099 //------------------------------get_key_start_from_aescrypt_object-----------------------
6100 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6101 #if defined(PPC64) || defined(S390)
6102 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6103 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6104 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6105 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6106 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6107 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6108 if (objSessionK == NULL) {
6109 return (Node *) NULL;
6110 }
6111 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6112 #else
6113 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6114 #endif // PPC64
6115 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6116 if (objAESCryptKey == NULL) return (Node *) NULL;
6117
6118 // now have the array, need to get the start address of the K array
6119 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6120 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6121 return k_start;
6122 }
6123
6124 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6125 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6126 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6127 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6128 if (objAESCryptKey == NULL) return (Node *) NULL;
6129
6130 // now have the array, need to get the start address of the lastKey array
6131 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6132 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6133 return original_k_start;
6134 }
6135
6136 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6137 // Return node representing slow path of predicate check.
6138 // the pseudo code we want to emulate with this predicate is:
6139 // for encryption:
6140 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6141 // for decryption:
6142 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6143 // note cipher==plain is more conservative than the original java code but that's OK
6144 //
6145 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6146 // The receiver was checked for NULL already.
6147 Node* objCBC = argument(0);
6148
6149 Node* src = argument(1);
6150 Node* dest = argument(4);
6151
6152 // Load embeddedCipher field of CipherBlockChaining object.
6153 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6154
6155 // get AESCrypt klass for instanceOf check
6156 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6157 // will have same classloader as CipherBlockChaining object
6158 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6159 assert(tinst != NULL, "CBCobj is null");
6160 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6161
6162 // we want to do an instanceof comparison against the AESCrypt class
6163 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6164 if (!klass_AESCrypt->is_loaded()) {
6165 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6166 Node* ctrl = control();
6167 set_control(top()); // no regular fast path
6168 return ctrl;
6169 }
6170
6171 src = must_be_not_null(src, true);
6172 dest = must_be_not_null(dest, true);
6173
6174 // Resolve oops to stable for CmpP below.
6175 src = access_resolve(src, 0);
6176 dest = access_resolve(dest, 0);
6177
6178 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6179
6180 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6181 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6182 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6183
6184 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6185
6186 // for encryption, we are done
6187 if (!decrypting)
6188 return instof_false; // even if it is NULL
6189
6190 // for decryption, we need to add a further check to avoid
6191 // taking the intrinsic path when cipher and plain are the same
6192 // see the original java code for why.
6193 RegionNode* region = new RegionNode(3);
6194 region->init_req(1, instof_false);
6195
6196 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6197 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6198 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6199 region->init_req(2, src_dest_conjoint);
6200
6201 record_for_igvn(region);
6202 return _gvn.transform(region);
6203 }
6204
6205 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6206 // Return node representing slow path of predicate check.
6207 // the pseudo code we want to emulate with this predicate is:
6208 // for encryption:
6209 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6210 // for decryption:
6211 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6212 // note cipher==plain is more conservative than the original java code but that's OK
6213 //
6214
6215 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6216 // The receiver was checked for NULL already.
6217 Node* objCTR = argument(0);
6218
6219 // Load embeddedCipher field of CipherBlockChaining object.
6220 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6221
6222 // get AESCrypt klass for instanceOf check
6223 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6224 // will have same classloader as CipherBlockChaining object
6225 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6226 assert(tinst != NULL, "CTRobj is null");
6227 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6228
6229 // we want to do an instanceof comparison against the AESCrypt class
6230 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6231 if (!klass_AESCrypt->is_loaded()) {
6232 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6233 Node* ctrl = control();
6234 set_control(top()); // no regular fast path
6235 return ctrl;
6236 }
6237
6238 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6239 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6240 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6241 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6242 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6243
6244 return instof_false; // even if it is NULL
6245 }
6246
6247 //------------------------------inline_ghash_processBlocks
6248 bool LibraryCallKit::inline_ghash_processBlocks() {
6249 address stubAddr;
6250 const char *stubName;
6251 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6252
6253 stubAddr = StubRoutines::ghash_processBlocks();
6254 stubName = "ghash_processBlocks";
6255
6256 Node* data = argument(0);
6257 Node* offset = argument(1);
6258 Node* len = argument(2);
6259 Node* state = argument(3);
6260 Node* subkeyH = argument(4);
6261
6262 state = must_be_not_null(state, true);
6263 subkeyH = must_be_not_null(subkeyH, true);
6264 data = must_be_not_null(data, true);
6265
6266 state = access_resolve(state, ACCESS_WRITE);
6267 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6268 data = access_resolve(data, ACCESS_READ);
6269
6270 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6271 assert(state_start, "state is NULL");
6272 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6273 assert(subkeyH_start, "subkeyH is NULL");
6274 Node* data_start = array_element_address(data, offset, T_BYTE);
6275 assert(data_start, "data is NULL");
6276
6277 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6278 OptoRuntime::ghash_processBlocks_Type(),
6279 stubAddr, stubName, TypePtr::BOTTOM,
6280 state_start, subkeyH_start, data_start, len);
6281 return true;
6282 }
6283
6284 bool LibraryCallKit::inline_base64_encodeBlock() {
6285 address stubAddr;
6286 const char *stubName;
6287 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6288 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6289 stubAddr = StubRoutines::base64_encodeBlock();
6290 stubName = "encodeBlock";
6291
6292 if (!stubAddr) return false;
6293 Node* base64obj = argument(0);
6294 Node* src = argument(1);
6295 Node* offset = argument(2);
6296 Node* len = argument(3);
6297 Node* dest = argument(4);
6298 Node* dp = argument(5);
6299 Node* isURL = argument(6);
6300
6301 src = must_be_not_null(src, true);
6302 src = access_resolve(src, ACCESS_READ);
6303 dest = must_be_not_null(dest, true);
6304 dest = access_resolve(dest, ACCESS_WRITE);
6305
6306 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6307 assert(src_start, "source array is NULL");
6308 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6309 assert(dest_start, "destination array is NULL");
6310
6311 Node* base64 = make_runtime_call(RC_LEAF,
6312 OptoRuntime::base64_encodeBlock_Type(),
6313 stubAddr, stubName, TypePtr::BOTTOM,
6314 src_start, offset, len, dest_start, dp, isURL);
6315 return true;
6316 }
6317
6318 //------------------------------inline_sha_implCompress-----------------------
6319 //
6320 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6321 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6322 //
6323 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6324 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6325 //
6326 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6327 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6328 //
6329 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6330 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6331
6332 Node* sha_obj = argument(0);
6333 Node* src = argument(1); // type oop
6334 Node* ofs = argument(2); // type int
6335
6336 const Type* src_type = src->Value(&_gvn);
6337 const TypeAryPtr* top_src = src_type->isa_aryptr();
6338 if (top_src == NULL || top_src->klass() == NULL) {
6339 // failed array check
6340 return false;
6341 }
6342 // Figure out the size and type of the elements we will be copying.
6343 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6344 if (src_elem != T_BYTE) {
6345 return false;
6346 }
6347 // 'src_start' points to src array + offset
6348 src = must_be_not_null(src, true);
6349 src = access_resolve(src, ACCESS_READ);
6350 Node* src_start = array_element_address(src, ofs, src_elem);
6351 Node* state = NULL;
6352 address stubAddr;
6353 const char *stubName;
6354
6355 switch(id) {
6356 case vmIntrinsics::_sha_implCompress:
6357 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6358 state = get_state_from_sha_object(sha_obj);
6359 stubAddr = StubRoutines::sha1_implCompress();
6360 stubName = "sha1_implCompress";
6361 break;
6362 case vmIntrinsics::_sha2_implCompress:
6363 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6364 state = get_state_from_sha_object(sha_obj);
6365 stubAddr = StubRoutines::sha256_implCompress();
6366 stubName = "sha256_implCompress";
6367 break;
6368 case vmIntrinsics::_sha5_implCompress:
6369 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6370 state = get_state_from_sha5_object(sha_obj);
6371 stubAddr = StubRoutines::sha512_implCompress();
6372 stubName = "sha512_implCompress";
6373 break;
6374 default:
6375 fatal_unexpected_iid(id);
6376 return false;
6377 }
6378 if (state == NULL) return false;
6379
6380 assert(stubAddr != NULL, "Stub is generated");
6381 if (stubAddr == NULL) return false;
6382
6383 // Call the stub.
6384 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6385 stubAddr, stubName, TypePtr::BOTTOM,
6386 src_start, state);
6387
6388 return true;
6389 }
6390
6391 //------------------------------inline_digestBase_implCompressMB-----------------------
6392 //
6393 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6394 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6395 //
6396 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6397 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6398 "need SHA1/SHA256/SHA512 instruction support");
6399 assert((uint)predicate < 3, "sanity");
6400 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6401
6402 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6403 Node* src = argument(1); // byte[] array
6404 Node* ofs = argument(2); // type int
6405 Node* limit = argument(3); // type int
6406
6407 const Type* src_type = src->Value(&_gvn);
6408 const TypeAryPtr* top_src = src_type->isa_aryptr();
6409 if (top_src == NULL || top_src->klass() == NULL) {
6410 // failed array check
6411 return false;
6412 }
6413 // Figure out the size and type of the elements we will be copying.
6414 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6415 if (src_elem != T_BYTE) {
6416 return false;
6417 }
6418 // 'src_start' points to src array + offset
6419 src = must_be_not_null(src, false);
6420 src = access_resolve(src, ACCESS_READ);
6421 Node* src_start = array_element_address(src, ofs, src_elem);
6422
6423 const char* klass_SHA_name = NULL;
6424 const char* stub_name = NULL;
6425 address stub_addr = NULL;
6426 bool long_state = false;
6427
6428 switch (predicate) {
6429 case 0:
6430 if (UseSHA1Intrinsics) {
6431 klass_SHA_name = "sun/security/provider/SHA";
6432 stub_name = "sha1_implCompressMB";
6433 stub_addr = StubRoutines::sha1_implCompressMB();
6434 }
6435 break;
6436 case 1:
6437 if (UseSHA256Intrinsics) {
6438 klass_SHA_name = "sun/security/provider/SHA2";
6439 stub_name = "sha256_implCompressMB";
6440 stub_addr = StubRoutines::sha256_implCompressMB();
6441 }
6442 break;
6443 case 2:
6444 if (UseSHA512Intrinsics) {
6445 klass_SHA_name = "sun/security/provider/SHA5";
6446 stub_name = "sha512_implCompressMB";
6447 stub_addr = StubRoutines::sha512_implCompressMB();
6448 long_state = true;
6449 }
6450 break;
6451 default:
6452 fatal("unknown SHA intrinsic predicate: %d", predicate);
6453 }
6454 if (klass_SHA_name != NULL) {
6455 assert(stub_addr != NULL, "Stub is generated");
6456 if (stub_addr == NULL) return false;
6457
6458 // get DigestBase klass to lookup for SHA klass
6459 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6460 assert(tinst != NULL, "digestBase_obj is not instance???");
6461 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6462
6463 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6464 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6465 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6466 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6467 }
6468 return false;
6469 }
6470 //------------------------------inline_sha_implCompressMB-----------------------
6471 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6472 bool long_state, address stubAddr, const char *stubName,
6473 Node* src_start, Node* ofs, Node* limit) {
6474 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6475 const TypeOopPtr* xtype = aklass->as_instance_type();
6476 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6477 sha_obj = _gvn.transform(sha_obj);
6478
6479 Node* state;
6480 if (long_state) {
6481 state = get_state_from_sha5_object(sha_obj);
6482 } else {
6483 state = get_state_from_sha_object(sha_obj);
6484 }
6485 if (state == NULL) return false;
6486
6487 // Call the stub.
6488 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6489 OptoRuntime::digestBase_implCompressMB_Type(),
6490 stubAddr, stubName, TypePtr::BOTTOM,
6491 src_start, state, ofs, limit);
6492 // return ofs (int)
6493 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6494 set_result(result);
6495
6496 return true;
6497 }
6498
6499 //------------------------------get_state_from_sha_object-----------------------
6500 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6501 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6502 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6503 if (sha_state == NULL) return (Node *) NULL;
6504
6505 // now have the array, need to get the start address of the state array
6506 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6507 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6508 return state;
6509 }
6510
6511 //------------------------------get_state_from_sha5_object-----------------------
6512 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6513 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6514 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6515 if (sha_state == NULL) return (Node *) NULL;
6516
6517 // now have the array, need to get the start address of the state array
6518 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6519 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6520 return state;
6521 }
6522
6523 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6524 // Return node representing slow path of predicate check.
6525 // the pseudo code we want to emulate with this predicate is:
6526 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6527 //
6528 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6529 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6530 "need SHA1/SHA256/SHA512 instruction support");
6531 assert((uint)predicate < 3, "sanity");
6532
6533 // The receiver was checked for NULL already.
6534 Node* digestBaseObj = argument(0);
6535
6536 // get DigestBase klass for instanceOf check
6537 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6538 assert(tinst != NULL, "digestBaseObj is null");
6539 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6540
6541 const char* klass_SHA_name = NULL;
6542 switch (predicate) {
6543 case 0:
6544 if (UseSHA1Intrinsics) {
6545 // we want to do an instanceof comparison against the SHA class
6546 klass_SHA_name = "sun/security/provider/SHA";
6547 }
6548 break;
6549 case 1:
6550 if (UseSHA256Intrinsics) {
6551 // we want to do an instanceof comparison against the SHA2 class
6552 klass_SHA_name = "sun/security/provider/SHA2";
6553 }
6554 break;
6555 case 2:
6556 if (UseSHA512Intrinsics) {
6557 // we want to do an instanceof comparison against the SHA5 class
6558 klass_SHA_name = "sun/security/provider/SHA5";
6559 }
6560 break;
6561 default:
6562 fatal("unknown SHA intrinsic predicate: %d", predicate);
6563 }
6564
6565 ciKlass* klass_SHA = NULL;
6566 if (klass_SHA_name != NULL) {
6567 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6568 }
6569 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6570 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6571 Node* ctrl = control();
6572 set_control(top()); // no intrinsic path
6573 return ctrl;
6574 }
6575 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6576
6577 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6578 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6579 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6580 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6581
6582 return instof_false; // even if it is NULL
6583 }
6584
6585 //-------------inline_fma-----------------------------------
6586 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6587 Node *a = NULL;
6588 Node *b = NULL;
6589 Node *c = NULL;
6590 Node* result = NULL;
6591 switch (id) {
6592 case vmIntrinsics::_fmaD:
6593 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6594 // no receiver since it is static method
6595 a = round_double_node(argument(0));
6596 b = round_double_node(argument(2));
6597 c = round_double_node(argument(4));
6598 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6599 break;
6600 case vmIntrinsics::_fmaF:
6601 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6602 a = argument(0);
6603 b = argument(1);
6604 c = argument(2);
6605 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6606 break;
6607 default:
6608 fatal_unexpected_iid(id); break;
6609 }
6610 set_result(result);
6611 return true;
6612 }
6613
6614 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6615 // argument(0) is receiver
6616 Node* codePoint = argument(1);
6617 Node* n = NULL;
6618
6619 switch (id) {
6620 case vmIntrinsics::_isDigit :
6621 n = new DigitNode(control(), codePoint);
6622 break;
6623 case vmIntrinsics::_isLowerCase :
6624 n = new LowerCaseNode(control(), codePoint);
6625 break;
6626 case vmIntrinsics::_isUpperCase :
6627 n = new UpperCaseNode(control(), codePoint);
6628 break;
6629 case vmIntrinsics::_isWhitespace :
6630 n = new WhitespaceNode(control(), codePoint);
6631 break;
6632 default:
6633 fatal_unexpected_iid(id);
6634 }
6635
6636 set_result(_gvn.transform(n));
6637 return true;
6638 }
6639
6640 //------------------------------inline_fp_min_max------------------------------
6641 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6642 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6643
6644 // The intrinsic should be used only when the API branches aren't predictable,
6645 // the last one performing the most important comparison. The following heuristic
6646 // uses the branch statistics to eventually bail out if necessary.
6647
6648 ciMethodData *md = callee()->method_data();
6649
6650 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6651 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6652
6653 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6654 // Bail out if the call-site didn't contribute enough to the statistics.
6655 return false;
6656 }
6657
6658 uint taken = 0, not_taken = 0;
6659
6660 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6661 if (p->is_BranchData()) {
6662 taken = ((ciBranchData*)p)->taken();
6663 not_taken = ((ciBranchData*)p)->not_taken();
6664 }
6665 }
6666
6667 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6668 balance = balance < 0 ? -balance : balance;
6669 if ( balance > 0.2 ) {
6670 // Bail out if the most important branch is predictable enough.
6671 return false;
6672 }
6673 }
6674 */
6675
6676 Node *a = NULL;
6677 Node *b = NULL;
6678 Node *n = NULL;
6679 switch (id) {
6680 case vmIntrinsics::_maxF:
6681 case vmIntrinsics::_minF:
6682 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6683 a = argument(0);
6684 b = argument(1);
6685 break;
6686 case vmIntrinsics::_maxD:
6687 case vmIntrinsics::_minD:
6688 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6689 a = round_double_node(argument(0));
6690 b = round_double_node(argument(2));
6691 break;
6692 default:
6693 fatal_unexpected_iid(id);
6694 break;
6695 }
6696 if (a->is_Con() || b->is_Con()) {
6697 return false;
6698 }
6699 switch (id) {
6700 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6701 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6702 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6703 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6704 default: fatal_unexpected_iid(id); break;
6705 }
6706 set_result(_gvn.transform(n));
6707 return true;
6708 }
6709
6710 bool LibraryCallKit::inline_profileBoolean() {
6711 Node* counts = argument(1);
6712 const TypeAryPtr* ary = NULL;
6713 ciArray* aobj = NULL;
6714 if (counts->is_Con()
6715 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6716 && (aobj = ary->const_oop()->as_array()) != NULL
6717 && (aobj->length() == 2)) {
6718 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6719 jint false_cnt = aobj->element_value(0).as_int();
6720 jint true_cnt = aobj->element_value(1).as_int();
6721
6722 if (C->log() != NULL) {
6723 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6724 false_cnt, true_cnt);
6725 }
6726
6727 if (false_cnt + true_cnt == 0) {
6728 // According to profile, never executed.
6729 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6730 Deoptimization::Action_reinterpret);
6731 return true;
6732 }
6733
6734 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6735 // is a number of each value occurrences.
6736 Node* result = argument(0);
6737 if (false_cnt == 0 || true_cnt == 0) {
6738 // According to profile, one value has been never seen.
6739 int expected_val = (false_cnt == 0) ? 1 : 0;
6740
6741 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6742 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6743
6744 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6745 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6746 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6747
6748 { // Slow path: uncommon trap for never seen value and then reexecute
6749 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6750 // the value has been seen at least once.
6751 PreserveJVMState pjvms(this);
6752 PreserveReexecuteState preexecs(this);
6753 jvms()->set_should_reexecute(true);
6754
6755 set_control(slow_path);
6756 set_i_o(i_o());
6757
6758 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6759 Deoptimization::Action_reinterpret);
6760 }
6761 // The guard for never seen value enables sharpening of the result and
6762 // returning a constant. It allows to eliminate branches on the same value
6763 // later on.
6764 set_control(fast_path);
6765 result = intcon(expected_val);
6766 }
6767 // Stop profiling.
6768 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6769 // By replacing method body with profile data (represented as ProfileBooleanNode
6770 // on IR level) we effectively disable profiling.
6771 // It enables full speed execution once optimized code is generated.
6772 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6773 C->record_for_igvn(profile);
6774 set_result(profile);
6775 return true;
6776 } else {
6777 // Continue profiling.
6778 // Profile data isn't available at the moment. So, execute method's bytecode version.
6779 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6780 // is compiled and counters aren't available since corresponding MethodHandle
6781 // isn't a compile-time constant.
6782 return false;
6783 }
6784 }
6785
6786 bool LibraryCallKit::inline_isCompileConstant() {
6787 Node* n = argument(0);
6788 set_result(n->is_Con() ? intcon(1) : intcon(0));
6789 return true;
6790 }