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/klass.inline.hpp" 36 #include "oops/objArrayKlass.hpp" 37 #include "opto/addnode.hpp" 38 #include "opto/arraycopynode.hpp" 39 #include "opto/c2compiler.hpp" 40 #include "opto/callGenerator.hpp" 41 #include "opto/castnode.hpp" 42 #include "opto/cfgnode.hpp" 43 #include "opto/convertnode.hpp" 44 #include "opto/countbitsnode.hpp" 45 #include "opto/intrinsicnode.hpp" 46 #include "opto/idealKit.hpp" 47 #include "opto/mathexactnode.hpp" 48 #include "opto/movenode.hpp" 49 #include "opto/mulnode.hpp" 50 #include "opto/narrowptrnode.hpp" 51 #include "opto/opaquenode.hpp" 52 #include "opto/parse.hpp" 53 #include "opto/runtime.hpp" 54 #include "opto/rootnode.hpp" 55 #include "opto/subnode.hpp" 56 #include "opto/valuetypenode.hpp" 57 #include "prims/nativeLookup.hpp" 58 #include "prims/unsafe.hpp" 59 #include "runtime/objectMonitor.hpp" 60 #include "runtime/sharedRuntime.hpp" 61 #include "utilities/macros.hpp" 62 63 64 class LibraryIntrinsic : public InlineCallGenerator { 65 // Extend the set of intrinsics known to the runtime: 66 public: 67 private: 68 bool _is_virtual; 69 bool _does_virtual_dispatch; 70 int8_t _predicates_count; // Intrinsic is predicated by several conditions 71 int8_t _last_predicate; // Last generated predicate 72 vmIntrinsics::ID _intrinsic_id; 73 74 public: 75 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id) 76 : InlineCallGenerator(m), 77 _is_virtual(is_virtual), 78 _does_virtual_dispatch(does_virtual_dispatch), 79 _predicates_count((int8_t)predicates_count), 80 _last_predicate((int8_t)-1), 81 _intrinsic_id(id) 82 { 83 } 84 virtual bool is_intrinsic() const { return true; } 85 virtual bool is_virtual() const { return _is_virtual; } 86 virtual bool is_predicated() const { return _predicates_count > 0; } 87 virtual int predicates_count() const { return _predicates_count; } 88 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; } 89 virtual JVMState* generate(JVMState* jvms); 90 virtual Node* generate_predicate(JVMState* jvms, int predicate); 91 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; } 92 }; 93 94 95 // Local helper class for LibraryIntrinsic: 96 class LibraryCallKit : public GraphKit { 97 private: 98 LibraryIntrinsic* _intrinsic; // the library intrinsic being called 99 Node* _result; // the result node, if any 100 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted 101 102 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type); 103 104 public: 105 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic) 106 : GraphKit(jvms), 107 _intrinsic(intrinsic), 108 _result(NULL) 109 { 110 // Check if this is a root compile. In that case we don't have a caller. 111 if (!jvms->has_method()) { 112 _reexecute_sp = sp(); 113 } else { 114 // Find out how many arguments the interpreter needs when deoptimizing 115 // and save the stack pointer value so it can used by uncommon_trap. 116 // We find the argument count by looking at the declared signature. 117 bool ignored_will_link; 118 ciSignature* declared_signature = NULL; 119 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature); 120 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci())); 121 _reexecute_sp = sp() + nargs; // "push" arguments back on stack 122 } 123 } 124 125 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; } 126 127 ciMethod* caller() const { return jvms()->method(); } 128 int bci() const { return jvms()->bci(); } 129 LibraryIntrinsic* intrinsic() const { return _intrinsic; } 130 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); } 131 ciMethod* callee() const { return _intrinsic->method(); } 132 133 bool try_to_inline(int predicate); 134 Node* try_to_predicate(int predicate); 135 136 void push_result() { 137 // Push the result onto the stack. 138 Node* res = result(); 139 if (!stopped() && res != NULL) { 140 BasicType bt = res->bottom_type()->basic_type(); 141 if (C->inlining_incrementally() && res->is_ValueType()) { 142 // The caller expects an oop when incrementally inlining an intrinsic that returns an 143 // inline type. Make sure the call is re-executed if the allocation triggers a deoptimization. 144 PreserveReexecuteState preexecs(this); 145 jvms()->set_should_reexecute(true); 146 res = ValueTypePtrNode::make_from_value_type(this, res->as_ValueType()); 147 } 148 push_node(bt, res); 149 } 150 } 151 152 private: 153 void fatal_unexpected_iid(vmIntrinsics::ID iid) { 154 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid)); 155 } 156 157 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; } 158 void set_result(RegionNode* region, PhiNode* value); 159 Node* result() { return _result; } 160 161 virtual int reexecute_sp() { return _reexecute_sp; } 162 163 // Helper functions to inline natives 164 Node* generate_guard(Node* test, RegionNode* region, float true_prob); 165 Node* generate_slow_guard(Node* test, RegionNode* region); 166 Node* generate_fair_guard(Node* test, RegionNode* region); 167 Node* generate_negative_guard(Node* index, RegionNode* region, 168 // resulting CastII of index: 169 Node* *pos_index = NULL); 170 Node* generate_limit_guard(Node* offset, Node* subseq_length, 171 Node* array_length, 172 RegionNode* region); 173 void generate_string_range_check(Node* array, Node* offset, 174 Node* length, bool char_count); 175 Node* generate_current_thread(Node* &tls_output); 176 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null, 177 RegionNode* region, int null_path, 178 int offset); 179 Node* load_klass_from_mirror(Node* mirror, bool never_see_null, 180 RegionNode* region, int null_path) { 181 int offset = java_lang_Class::klass_offset_in_bytes(); 182 return load_klass_from_mirror_common(mirror, never_see_null, 183 region, null_path, 184 offset); 185 } 186 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null, 187 RegionNode* region, int null_path) { 188 int offset = java_lang_Class::array_klass_offset_in_bytes(); 189 return load_klass_from_mirror_common(mirror, never_see_null, 190 region, null_path, 191 offset); 192 } 193 Node* generate_access_flags_guard(Node* kls, 194 int modifier_mask, int modifier_bits, 195 RegionNode* region); 196 Node* generate_interface_guard(Node* kls, RegionNode* region); 197 198 enum ArrayKind { 199 AnyArray, 200 NonArray, 201 ObjectArray, 202 NonObjectArray, 203 TypeArray, 204 ValueArray 205 }; 206 207 Node* generate_array_guard(Node* kls, RegionNode* region) { 208 return generate_array_guard_common(kls, region, AnyArray); 209 } 210 Node* generate_non_array_guard(Node* kls, RegionNode* region) { 211 return generate_array_guard_common(kls, region, NonArray); 212 } 213 Node* generate_objArray_guard(Node* kls, RegionNode* region) { 214 return generate_array_guard_common(kls, region, ObjectArray); 215 } 216 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) { 217 return generate_array_guard_common(kls, region, NonObjectArray); 218 } 219 Node* generate_typeArray_guard(Node* kls, RegionNode* region) { 220 return generate_array_guard_common(kls, region, TypeArray); 221 } 222 Node* generate_valueArray_guard(Node* kls, RegionNode* region) { 223 assert(ValueArrayFlatten, "can never be flattened"); 224 return generate_array_guard_common(kls, region, ValueArray); 225 } 226 Node* generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind); 227 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region); 228 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id, 229 bool is_virtual = false, bool is_static = false); 230 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) { 231 return generate_method_call(method_id, false, true); 232 } 233 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) { 234 return generate_method_call(method_id, true, false); 235 } 236 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 237 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 238 239 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae); 240 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae); 241 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae); 242 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae); 243 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 244 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae); 245 bool inline_string_indexOfChar(); 246 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae); 247 bool inline_string_toBytesU(); 248 bool inline_string_getCharsU(); 249 bool inline_string_copy(bool compress); 250 bool inline_string_char_access(bool is_store); 251 Node* round_double_node(Node* n); 252 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName); 253 bool inline_math_native(vmIntrinsics::ID id); 254 bool inline_math(vmIntrinsics::ID id); 255 bool inline_double_math(vmIntrinsics::ID id); 256 template <typename OverflowOp> 257 bool inline_math_overflow(Node* arg1, Node* arg2); 258 void inline_math_mathExact(Node* math, Node* test); 259 bool inline_math_addExactI(bool is_increment); 260 bool inline_math_addExactL(bool is_increment); 261 bool inline_math_multiplyExactI(); 262 bool inline_math_multiplyExactL(); 263 bool inline_math_multiplyHigh(); 264 bool inline_math_negateExactI(); 265 bool inline_math_negateExactL(); 266 bool inline_math_subtractExactI(bool is_decrement); 267 bool inline_math_subtractExactL(bool is_decrement); 268 bool inline_min_max(vmIntrinsics::ID id); 269 bool inline_notify(vmIntrinsics::ID id); 270 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y); 271 // This returns Type::AnyPtr, RawPtr, or OopPtr. 272 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type); 273 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false); 274 275 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind; 276 DecoratorSet mo_decorator_for_access_kind(AccessKind kind); 277 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned); 278 static bool klass_needs_init_guard(Node* kls); 279 bool inline_unsafe_allocate(); 280 bool inline_unsafe_newArray(bool uninitialized); 281 bool inline_unsafe_writeback0(); 282 bool inline_unsafe_writebackSync0(bool is_pre); 283 bool inline_unsafe_copyMemory(); 284 bool inline_unsafe_make_private_buffer(); 285 bool inline_unsafe_finish_private_buffer(); 286 bool inline_native_currentThread(); 287 288 bool inline_native_time_funcs(address method, const char* funcName); 289 #ifdef JFR_HAVE_INTRINSICS 290 bool inline_native_classID(); 291 bool inline_native_getEventWriter(); 292 #endif 293 bool inline_native_Class_query(vmIntrinsics::ID id); 294 bool inline_value_Class_conversion(vmIntrinsics::ID id); 295 bool inline_native_subtype_check(); 296 bool inline_native_getLength(); 297 bool inline_array_copyOf(bool is_copyOfRange); 298 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae); 299 bool inline_preconditions_checkIndex(); 300 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array); 301 bool inline_native_clone(bool is_virtual); 302 bool inline_native_Reflection_getCallerClass(); 303 // Helper function for inlining native object hash method 304 bool inline_native_hashcode(bool is_virtual, bool is_static); 305 bool inline_native_getClass(); 306 307 // Helper functions for inlining arraycopy 308 bool inline_arraycopy(); 309 AllocateArrayNode* tightly_coupled_allocation(Node* ptr, 310 RegionNode* slow_region); 311 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp); 312 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp, 313 uint new_idx); 314 315 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind; 316 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind); 317 bool inline_unsafe_fence(vmIntrinsics::ID id); 318 bool inline_onspinwait(); 319 bool inline_fp_conversions(vmIntrinsics::ID id); 320 bool inline_number_methods(vmIntrinsics::ID id); 321 bool inline_reference_get(); 322 bool inline_Class_cast(); 323 bool inline_aescrypt_Block(vmIntrinsics::ID id); 324 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id); 325 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id); 326 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id); 327 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting); 328 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting); 329 Node* inline_counterMode_AESCrypt_predicate(); 330 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object); 331 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object); 332 bool inline_ghash_processBlocks(); 333 bool inline_base64_encodeBlock(); 334 bool inline_sha_implCompress(vmIntrinsics::ID id); 335 bool inline_digestBase_implCompressMB(int predicate); 336 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA, 337 bool long_state, address stubAddr, const char *stubName, 338 Node* src_start, Node* ofs, Node* limit); 339 Node* get_state_from_sha_object(Node *sha_object); 340 Node* get_state_from_sha5_object(Node *sha_object); 341 Node* inline_digestBase_implCompressMB_predicate(int predicate); 342 bool inline_encodeISOArray(); 343 bool inline_updateCRC32(); 344 bool inline_updateBytesCRC32(); 345 bool inline_updateByteBufferCRC32(); 346 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class); 347 bool inline_updateBytesCRC32C(); 348 bool inline_updateDirectByteBufferCRC32C(); 349 bool inline_updateBytesAdler32(); 350 bool inline_updateByteBufferAdler32(); 351 bool inline_multiplyToLen(); 352 bool inline_hasNegatives(); 353 bool inline_squareToLen(); 354 bool inline_mulAdd(); 355 bool inline_montgomeryMultiply(); 356 bool inline_montgomerySquare(); 357 bool inline_vectorizedMismatch(); 358 bool inline_fma(vmIntrinsics::ID id); 359 bool inline_character_compare(vmIntrinsics::ID id); 360 bool inline_fp_min_max(vmIntrinsics::ID id); 361 362 bool inline_profileBoolean(); 363 bool inline_isCompileConstant(); 364 void clear_upper_avx() { 365 #ifdef X86 366 if (UseAVX >= 2) { 367 C->set_clear_upper_avx(true); 368 } 369 #endif 370 } 371 }; 372 373 //---------------------------make_vm_intrinsic---------------------------- 374 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) { 375 vmIntrinsics::ID id = m->intrinsic_id(); 376 assert(id != vmIntrinsics::_none, "must be a VM intrinsic"); 377 378 if (!m->is_loaded()) { 379 // Do not attempt to inline unloaded methods. 380 return NULL; 381 } 382 383 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization); 384 bool is_available = false; 385 386 { 387 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag 388 // the compiler must transition to '_thread_in_vm' state because both 389 // methods access VM-internal data. 390 VM_ENTRY_MARK; 391 methodHandle mh(THREAD, m->get_Method()); 392 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) && 393 !C->directive()->is_intrinsic_disabled(mh) && 394 !vmIntrinsics::is_disabled_by_flags(mh); 395 396 } 397 398 if (is_available) { 399 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility"); 400 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?"); 401 return new LibraryIntrinsic(m, is_virtual, 402 vmIntrinsics::predicates_needed(id), 403 vmIntrinsics::does_virtual_dispatch(id), 404 (vmIntrinsics::ID) id); 405 } else { 406 return NULL; 407 } 408 } 409 410 //----------------------register_library_intrinsics----------------------- 411 // Initialize this file's data structures, for each Compile instance. 412 void Compile::register_library_intrinsics() { 413 // Nothing to do here. 414 } 415 416 JVMState* LibraryIntrinsic::generate(JVMState* jvms) { 417 LibraryCallKit kit(jvms, this); 418 Compile* C = kit.C; 419 int nodes = C->unique(); 420 #ifndef PRODUCT 421 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 422 char buf[1000]; 423 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 424 tty->print_cr("Intrinsic %s", str); 425 } 426 #endif 427 ciMethod* callee = kit.callee(); 428 const int bci = kit.bci(); 429 430 // Try to inline the intrinsic. 431 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) && 432 kit.try_to_inline(_last_predicate)) { 433 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)" 434 : "(intrinsic)"; 435 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 436 if (C->print_intrinsics() || C->print_inlining()) { 437 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 438 } 439 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 440 if (C->log()) { 441 C->log()->elem("intrinsic id='%s'%s nodes='%d'", 442 vmIntrinsics::name_at(intrinsic_id()), 443 (is_virtual() ? " virtual='1'" : ""), 444 C->unique() - nodes); 445 } 446 // Push the result from the inlined method onto the stack. 447 kit.push_result(); 448 C->print_inlining_update(this); 449 return kit.transfer_exceptions_into_jvms(); 450 } 451 452 // The intrinsic bailed out 453 if (jvms->has_method()) { 454 // Not a root compile. 455 const char* msg; 456 if (callee->intrinsic_candidate()) { 457 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)"; 458 } else { 459 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated" 460 : "failed to inline (intrinsic), method not annotated"; 461 } 462 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg); 463 if (C->print_intrinsics() || C->print_inlining()) { 464 C->print_inlining(callee, jvms->depth() - 1, bci, msg); 465 } 466 } else { 467 // Root compile 468 ResourceMark rm; 469 stringStream msg_stream; 470 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 471 vmIntrinsics::name_at(intrinsic_id()), 472 is_virtual() ? " (virtual)" : "", bci); 473 const char *msg = msg_stream.as_string(); 474 log_debug(jit, inlining)("%s", msg); 475 if (C->print_intrinsics() || C->print_inlining()) { 476 tty->print("%s", msg); 477 } 478 } 479 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 480 C->print_inlining_update(this); 481 return NULL; 482 } 483 484 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) { 485 LibraryCallKit kit(jvms, this); 486 Compile* C = kit.C; 487 int nodes = C->unique(); 488 _last_predicate = predicate; 489 #ifndef PRODUCT 490 assert(is_predicated() && predicate < predicates_count(), "sanity"); 491 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 492 char buf[1000]; 493 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 494 tty->print_cr("Predicate for intrinsic %s", str); 495 } 496 #endif 497 ciMethod* callee = kit.callee(); 498 const int bci = kit.bci(); 499 500 Node* slow_ctl = kit.try_to_predicate(predicate); 501 if (!kit.failing()) { 502 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)" 503 : "(intrinsic, predicate)"; 504 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 505 if (C->print_intrinsics() || C->print_inlining()) { 506 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 507 } 508 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 509 if (C->log()) { 510 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'", 511 vmIntrinsics::name_at(intrinsic_id()), 512 (is_virtual() ? " virtual='1'" : ""), 513 C->unique() - nodes); 514 } 515 return slow_ctl; // Could be NULL if the check folds. 516 } 517 518 // The intrinsic bailed out 519 if (jvms->has_method()) { 520 // Not a root compile. 521 const char* msg = "failed to generate predicate for intrinsic"; 522 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg); 523 if (C->print_intrinsics() || C->print_inlining()) { 524 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg); 525 } 526 } else { 527 // Root compile 528 ResourceMark rm; 529 stringStream msg_stream; 530 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 531 vmIntrinsics::name_at(intrinsic_id()), 532 is_virtual() ? " (virtual)" : "", bci); 533 const char *msg = msg_stream.as_string(); 534 log_debug(jit, inlining)("%s", msg); 535 if (C->print_intrinsics() || C->print_inlining()) { 536 C->print_inlining_stream()->print("%s", msg); 537 } 538 } 539 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 540 return NULL; 541 } 542 543 bool LibraryCallKit::try_to_inline(int predicate) { 544 // Handle symbolic names for otherwise undistinguished boolean switches: 545 const bool is_store = true; 546 const bool is_compress = true; 547 const bool is_static = true; 548 const bool is_volatile = true; 549 550 if (!jvms()->has_method()) { 551 // Root JVMState has a null method. 552 assert(map()->memory()->Opcode() == Op_Parm, ""); 553 // Insert the memory aliasing node 554 set_all_memory(reset_memory()); 555 } 556 assert(merged_memory(), ""); 557 558 559 switch (intrinsic_id()) { 560 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static); 561 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static); 562 case vmIntrinsics::_getClass: return inline_native_getClass(); 563 564 case vmIntrinsics::_ceil: 565 case vmIntrinsics::_floor: 566 case vmIntrinsics::_rint: 567 case vmIntrinsics::_dsin: 568 case vmIntrinsics::_dcos: 569 case vmIntrinsics::_dtan: 570 case vmIntrinsics::_dabs: 571 case vmIntrinsics::_fabs: 572 case vmIntrinsics::_iabs: 573 case vmIntrinsics::_labs: 574 case vmIntrinsics::_datan2: 575 case vmIntrinsics::_dsqrt: 576 case vmIntrinsics::_dexp: 577 case vmIntrinsics::_dlog: 578 case vmIntrinsics::_dlog10: 579 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id()); 580 581 case vmIntrinsics::_min: 582 case vmIntrinsics::_max: return inline_min_max(intrinsic_id()); 583 584 case vmIntrinsics::_notify: 585 case vmIntrinsics::_notifyAll: 586 return inline_notify(intrinsic_id()); 587 588 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 589 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 590 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 591 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 592 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 593 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 594 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 595 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 596 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh(); 597 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 598 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 599 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 600 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 601 602 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 603 604 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 605 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 606 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 607 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 608 609 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 610 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 611 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 612 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 613 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 614 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 615 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(); 616 617 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 618 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU); 619 620 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 621 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 622 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 623 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 624 625 case vmIntrinsics::_compressStringC: 626 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 627 case vmIntrinsics::_inflateStringC: 628 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 629 630 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer(); 631 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer(); 632 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false); 633 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false); 634 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false); 635 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false); 636 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false); 637 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false); 638 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false); 639 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false); 640 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false); 641 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_VALUETYPE,Relaxed, false); 642 643 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false); 644 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false); 645 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false); 646 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false); 647 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false); 648 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false); 649 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false); 650 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false); 651 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false); 652 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_VALUETYPE,Relaxed, false); 653 654 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false); 655 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false); 656 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false); 657 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false); 658 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false); 659 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false); 660 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false); 661 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false); 662 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false); 663 664 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false); 665 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false); 666 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false); 667 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false); 668 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false); 669 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false); 670 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false); 671 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false); 672 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false); 673 674 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true); 675 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true); 676 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true); 677 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true); 678 679 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true); 680 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true); 681 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true); 682 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true); 683 684 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false); 685 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false); 686 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false); 687 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false); 688 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false); 689 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false); 690 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false); 691 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false); 692 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false); 693 694 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false); 695 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false); 696 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false); 697 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false); 698 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false); 699 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false); 700 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false); 701 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false); 702 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false); 703 704 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false); 705 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false); 706 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false); 707 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false); 708 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false); 709 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false); 710 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false); 711 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false); 712 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false); 713 714 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false); 715 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false); 716 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false); 717 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false); 718 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false); 719 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false); 720 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false); 721 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false); 722 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false); 723 724 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 725 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile); 726 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile); 727 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 728 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 729 730 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 731 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 732 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 733 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile); 734 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed); 735 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire); 736 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release); 737 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile); 738 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed); 739 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire); 740 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release); 741 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile); 742 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 743 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 744 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 745 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile); 746 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 747 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 748 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 749 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile); 750 751 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 752 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 753 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 754 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile); 755 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire); 756 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release); 757 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile); 758 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire); 759 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release); 760 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 761 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 762 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 763 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 764 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 765 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 766 767 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile); 768 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile); 769 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 770 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 771 772 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile); 773 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile); 774 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 775 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 776 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 777 778 case vmIntrinsics::_loadFence: 779 case vmIntrinsics::_storeFence: 780 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 781 782 case vmIntrinsics::_onSpinWait: return inline_onspinwait(); 783 784 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 785 786 #ifdef JFR_HAVE_INTRINSICS 787 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime"); 788 case vmIntrinsics::_getClassId: return inline_native_classID(); 789 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter(); 790 #endif 791 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 792 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 793 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0(); 794 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true); 795 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false); 796 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 797 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 798 case vmIntrinsics::_getLength: return inline_native_getLength(); 799 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 800 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 801 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 802 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 803 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(); 804 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 805 806 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true); 807 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false); 808 809 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 810 811 case vmIntrinsics::_isInstance: 812 case vmIntrinsics::_getModifiers: 813 case vmIntrinsics::_isInterface: 814 case vmIntrinsics::_isArray: 815 case vmIntrinsics::_isPrimitive: 816 case vmIntrinsics::_getSuperclass: 817 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id()); 818 819 case vmIntrinsics::_asPrimaryType: 820 case vmIntrinsics::_asIndirectType: return inline_value_Class_conversion(intrinsic_id()); 821 822 case vmIntrinsics::_floatToRawIntBits: 823 case vmIntrinsics::_floatToIntBits: 824 case vmIntrinsics::_intBitsToFloat: 825 case vmIntrinsics::_doubleToRawLongBits: 826 case vmIntrinsics::_doubleToLongBits: 827 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id()); 828 829 case vmIntrinsics::_numberOfLeadingZeros_i: 830 case vmIntrinsics::_numberOfLeadingZeros_l: 831 case vmIntrinsics::_numberOfTrailingZeros_i: 832 case vmIntrinsics::_numberOfTrailingZeros_l: 833 case vmIntrinsics::_bitCount_i: 834 case vmIntrinsics::_bitCount_l: 835 case vmIntrinsics::_reverseBytes_i: 836 case vmIntrinsics::_reverseBytes_l: 837 case vmIntrinsics::_reverseBytes_s: 838 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id()); 839 840 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass(); 841 842 case vmIntrinsics::_Reference_get: return inline_reference_get(); 843 844 case vmIntrinsics::_Class_cast: return inline_Class_cast(); 845 846 case vmIntrinsics::_aescrypt_encryptBlock: 847 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id()); 848 849 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 850 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 851 return inline_cipherBlockChaining_AESCrypt(intrinsic_id()); 852 853 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 854 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 855 return inline_electronicCodeBook_AESCrypt(intrinsic_id()); 856 857 case vmIntrinsics::_counterMode_AESCrypt: 858 return inline_counterMode_AESCrypt(intrinsic_id()); 859 860 case vmIntrinsics::_sha_implCompress: 861 case vmIntrinsics::_sha2_implCompress: 862 case vmIntrinsics::_sha5_implCompress: 863 return inline_sha_implCompress(intrinsic_id()); 864 865 case vmIntrinsics::_digestBase_implCompressMB: 866 return inline_digestBase_implCompressMB(predicate); 867 868 case vmIntrinsics::_multiplyToLen: 869 return inline_multiplyToLen(); 870 871 case vmIntrinsics::_squareToLen: 872 return inline_squareToLen(); 873 874 case vmIntrinsics::_mulAdd: 875 return inline_mulAdd(); 876 877 case vmIntrinsics::_montgomeryMultiply: 878 return inline_montgomeryMultiply(); 879 case vmIntrinsics::_montgomerySquare: 880 return inline_montgomerySquare(); 881 882 case vmIntrinsics::_vectorizedMismatch: 883 return inline_vectorizedMismatch(); 884 885 case vmIntrinsics::_ghash_processBlocks: 886 return inline_ghash_processBlocks(); 887 case vmIntrinsics::_base64_encodeBlock: 888 return inline_base64_encodeBlock(); 889 890 case vmIntrinsics::_encodeISOArray: 891 case vmIntrinsics::_encodeByteISOArray: 892 return inline_encodeISOArray(); 893 894 case vmIntrinsics::_updateCRC32: 895 return inline_updateCRC32(); 896 case vmIntrinsics::_updateBytesCRC32: 897 return inline_updateBytesCRC32(); 898 case vmIntrinsics::_updateByteBufferCRC32: 899 return inline_updateByteBufferCRC32(); 900 901 case vmIntrinsics::_updateBytesCRC32C: 902 return inline_updateBytesCRC32C(); 903 case vmIntrinsics::_updateDirectByteBufferCRC32C: 904 return inline_updateDirectByteBufferCRC32C(); 905 906 case vmIntrinsics::_updateBytesAdler32: 907 return inline_updateBytesAdler32(); 908 case vmIntrinsics::_updateByteBufferAdler32: 909 return inline_updateByteBufferAdler32(); 910 911 case vmIntrinsics::_profileBoolean: 912 return inline_profileBoolean(); 913 case vmIntrinsics::_isCompileConstant: 914 return inline_isCompileConstant(); 915 916 case vmIntrinsics::_hasNegatives: 917 return inline_hasNegatives(); 918 919 case vmIntrinsics::_fmaD: 920 case vmIntrinsics::_fmaF: 921 return inline_fma(intrinsic_id()); 922 923 case vmIntrinsics::_isDigit: 924 case vmIntrinsics::_isLowerCase: 925 case vmIntrinsics::_isUpperCase: 926 case vmIntrinsics::_isWhitespace: 927 return inline_character_compare(intrinsic_id()); 928 929 case vmIntrinsics::_maxF: 930 case vmIntrinsics::_minF: 931 case vmIntrinsics::_maxD: 932 case vmIntrinsics::_minD: 933 return inline_fp_min_max(intrinsic_id()); 934 935 default: 936 // If you get here, it may be that someone has added a new intrinsic 937 // to the list in vmSymbols.hpp without implementing it here. 938 #ifndef PRODUCT 939 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 940 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 941 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 942 } 943 #endif 944 return false; 945 } 946 } 947 948 Node* LibraryCallKit::try_to_predicate(int predicate) { 949 if (!jvms()->has_method()) { 950 // Root JVMState has a null method. 951 assert(map()->memory()->Opcode() == Op_Parm, ""); 952 // Insert the memory aliasing node 953 set_all_memory(reset_memory()); 954 } 955 assert(merged_memory(), ""); 956 957 switch (intrinsic_id()) { 958 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 959 return inline_cipherBlockChaining_AESCrypt_predicate(false); 960 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 961 return inline_cipherBlockChaining_AESCrypt_predicate(true); 962 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 963 return inline_electronicCodeBook_AESCrypt_predicate(false); 964 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 965 return inline_electronicCodeBook_AESCrypt_predicate(true); 966 case vmIntrinsics::_counterMode_AESCrypt: 967 return inline_counterMode_AESCrypt_predicate(); 968 case vmIntrinsics::_digestBase_implCompressMB: 969 return inline_digestBase_implCompressMB_predicate(predicate); 970 971 default: 972 // If you get here, it may be that someone has added a new intrinsic 973 // to the list in vmSymbols.hpp without implementing it here. 974 #ifndef PRODUCT 975 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 976 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 977 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 978 } 979 #endif 980 Node* slow_ctl = control(); 981 set_control(top()); // No fast path instrinsic 982 return slow_ctl; 983 } 984 } 985 986 //------------------------------set_result------------------------------- 987 // Helper function for finishing intrinsics. 988 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 989 record_for_igvn(region); 990 set_control(_gvn.transform(region)); 991 set_result( _gvn.transform(value)); 992 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 993 } 994 995 //------------------------------generate_guard--------------------------- 996 // Helper function for generating guarded fast-slow graph structures. 997 // The given 'test', if true, guards a slow path. If the test fails 998 // then a fast path can be taken. (We generally hope it fails.) 999 // In all cases, GraphKit::control() is updated to the fast path. 1000 // The returned value represents the control for the slow path. 1001 // The return value is never 'top'; it is either a valid control 1002 // or NULL if it is obvious that the slow path can never be taken. 1003 // Also, if region and the slow control are not NULL, the slow edge 1004 // is appended to the region. 1005 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 1006 if (stopped()) { 1007 // Already short circuited. 1008 return NULL; 1009 } 1010 1011 // Build an if node and its projections. 1012 // If test is true we take the slow path, which we assume is uncommon. 1013 if (_gvn.type(test) == TypeInt::ZERO) { 1014 // The slow branch is never taken. No need to build this guard. 1015 return NULL; 1016 } 1017 1018 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 1019 1020 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 1021 if (if_slow == top()) { 1022 // The slow branch is never taken. No need to build this guard. 1023 return NULL; 1024 } 1025 1026 if (region != NULL) 1027 region->add_req(if_slow); 1028 1029 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 1030 set_control(if_fast); 1031 1032 return if_slow; 1033 } 1034 1035 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 1036 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 1037 } 1038 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 1039 return generate_guard(test, region, PROB_FAIR); 1040 } 1041 1042 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 1043 Node* *pos_index) { 1044 if (stopped()) 1045 return NULL; // already stopped 1046 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 1047 return NULL; // index is already adequately typed 1048 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 1049 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1050 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 1051 if (is_neg != NULL && pos_index != NULL) { 1052 // Emulate effect of Parse::adjust_map_after_if. 1053 Node* ccast = new CastIINode(index, TypeInt::POS); 1054 ccast->set_req(0, control()); 1055 (*pos_index) = _gvn.transform(ccast); 1056 } 1057 return is_neg; 1058 } 1059 1060 // Make sure that 'position' is a valid limit index, in [0..length]. 1061 // There are two equivalent plans for checking this: 1062 // A. (offset + copyLength) unsigned<= arrayLength 1063 // B. offset <= (arrayLength - copyLength) 1064 // We require that all of the values above, except for the sum and 1065 // difference, are already known to be non-negative. 1066 // Plan A is robust in the face of overflow, if offset and copyLength 1067 // are both hugely positive. 1068 // 1069 // Plan B is less direct and intuitive, but it does not overflow at 1070 // all, since the difference of two non-negatives is always 1071 // representable. Whenever Java methods must perform the equivalent 1072 // check they generally use Plan B instead of Plan A. 1073 // For the moment we use Plan A. 1074 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 1075 Node* subseq_length, 1076 Node* array_length, 1077 RegionNode* region) { 1078 if (stopped()) 1079 return NULL; // already stopped 1080 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 1081 if (zero_offset && subseq_length->eqv_uncast(array_length)) 1082 return NULL; // common case of whole-array copy 1083 Node* last = subseq_length; 1084 if (!zero_offset) // last += offset 1085 last = _gvn.transform(new AddINode(last, offset)); 1086 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 1087 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1088 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 1089 return is_over; 1090 } 1091 1092 // Emit range checks for the given String.value byte array 1093 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) { 1094 if (stopped()) { 1095 return; // already stopped 1096 } 1097 RegionNode* bailout = new RegionNode(1); 1098 record_for_igvn(bailout); 1099 if (char_count) { 1100 // Convert char count to byte count 1101 count = _gvn.transform(new LShiftINode(count, intcon(1))); 1102 } 1103 1104 // Offset and count must not be negative 1105 generate_negative_guard(offset, bailout); 1106 generate_negative_guard(count, bailout); 1107 // Offset + count must not exceed length of array 1108 generate_limit_guard(offset, count, load_array_length(array), bailout); 1109 1110 if (bailout->req() > 1) { 1111 PreserveJVMState pjvms(this); 1112 set_control(_gvn.transform(bailout)); 1113 uncommon_trap(Deoptimization::Reason_intrinsic, 1114 Deoptimization::Action_maybe_recompile); 1115 } 1116 } 1117 1118 //--------------------------generate_current_thread-------------------- 1119 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 1120 ciKlass* thread_klass = env()->Thread_klass(); 1121 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 1122 Node* thread = _gvn.transform(new ThreadLocalNode()); 1123 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset())); 1124 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered); 1125 tls_output = thread; 1126 return threadObj; 1127 } 1128 1129 1130 //------------------------------make_string_method_node------------------------ 1131 // Helper method for String intrinsic functions. This version is called with 1132 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 1133 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 1134 // containing the lengths of str1 and str2. 1135 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 1136 Node* result = NULL; 1137 switch (opcode) { 1138 case Op_StrIndexOf: 1139 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 1140 str1_start, cnt1, str2_start, cnt2, ae); 1141 break; 1142 case Op_StrComp: 1143 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 1144 str1_start, cnt1, str2_start, cnt2, ae); 1145 break; 1146 case Op_StrEquals: 1147 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1148 // Use the constant length if there is one because optimized match rule may exist. 1149 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1150 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1151 break; 1152 default: 1153 ShouldNotReachHere(); 1154 return NULL; 1155 } 1156 1157 // All these intrinsics have checks. 1158 C->set_has_split_ifs(true); // Has chance for split-if optimization 1159 clear_upper_avx(); 1160 1161 return _gvn.transform(result); 1162 } 1163 1164 //------------------------------inline_string_compareTo------------------------ 1165 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1166 Node* arg1 = argument(0); 1167 Node* arg2 = argument(1); 1168 1169 arg1 = must_be_not_null(arg1, true); 1170 arg2 = must_be_not_null(arg2, true); 1171 1172 arg1 = access_resolve(arg1, ACCESS_READ); 1173 arg2 = access_resolve(arg2, ACCESS_READ); 1174 1175 // Get start addr and length of first argument 1176 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1177 Node* arg1_cnt = load_array_length(arg1); 1178 1179 // Get start addr and length of second argument 1180 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1181 Node* arg2_cnt = load_array_length(arg2); 1182 1183 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1184 set_result(result); 1185 return true; 1186 } 1187 1188 //------------------------------inline_string_equals------------------------ 1189 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1190 Node* arg1 = argument(0); 1191 Node* arg2 = argument(1); 1192 1193 // paths (plus control) merge 1194 RegionNode* region = new RegionNode(3); 1195 Node* phi = new PhiNode(region, TypeInt::BOOL); 1196 1197 if (!stopped()) { 1198 1199 arg1 = must_be_not_null(arg1, true); 1200 arg2 = must_be_not_null(arg2, true); 1201 1202 arg1 = access_resolve(arg1, ACCESS_READ); 1203 arg2 = access_resolve(arg2, ACCESS_READ); 1204 1205 // Get start addr and length of first argument 1206 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1207 Node* arg1_cnt = load_array_length(arg1); 1208 1209 // Get start addr and length of second argument 1210 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1211 Node* arg2_cnt = load_array_length(arg2); 1212 1213 // Check for arg1_cnt != arg2_cnt 1214 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1215 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1216 Node* if_ne = generate_slow_guard(bol, NULL); 1217 if (if_ne != NULL) { 1218 phi->init_req(2, intcon(0)); 1219 region->init_req(2, if_ne); 1220 } 1221 1222 // Check for count == 0 is done by assembler code for StrEquals. 1223 1224 if (!stopped()) { 1225 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1226 phi->init_req(1, equals); 1227 region->init_req(1, control()); 1228 } 1229 } 1230 1231 // post merge 1232 set_control(_gvn.transform(region)); 1233 record_for_igvn(region); 1234 1235 set_result(_gvn.transform(phi)); 1236 return true; 1237 } 1238 1239 //------------------------------inline_array_equals---------------------------- 1240 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1241 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1242 Node* arg1 = argument(0); 1243 Node* arg2 = argument(1); 1244 1245 arg1 = access_resolve(arg1, ACCESS_READ); 1246 arg2 = access_resolve(arg2, ACCESS_READ); 1247 1248 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1249 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1250 clear_upper_avx(); 1251 1252 return true; 1253 } 1254 1255 //------------------------------inline_hasNegatives------------------------------ 1256 bool LibraryCallKit::inline_hasNegatives() { 1257 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1258 return false; 1259 } 1260 1261 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters"); 1262 // no receiver since it is static method 1263 Node* ba = argument(0); 1264 Node* offset = argument(1); 1265 Node* len = argument(2); 1266 1267 ba = must_be_not_null(ba, true); 1268 1269 // Range checks 1270 generate_string_range_check(ba, offset, len, false); 1271 if (stopped()) { 1272 return true; 1273 } 1274 ba = access_resolve(ba, ACCESS_READ); 1275 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1276 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1277 set_result(_gvn.transform(result)); 1278 return true; 1279 } 1280 1281 bool LibraryCallKit::inline_preconditions_checkIndex() { 1282 Node* index = argument(0); 1283 Node* length = argument(1); 1284 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1285 return false; 1286 } 1287 1288 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0))); 1289 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1290 1291 { 1292 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1293 uncommon_trap(Deoptimization::Reason_intrinsic, 1294 Deoptimization::Action_make_not_entrant); 1295 } 1296 1297 if (stopped()) { 1298 return false; 1299 } 1300 1301 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length)); 1302 BoolTest::mask btest = BoolTest::lt; 1303 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1304 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1305 _gvn.set_type(rc, rc->Value(&_gvn)); 1306 if (!rc_bool->is_Con()) { 1307 record_for_igvn(rc); 1308 } 1309 set_control(_gvn.transform(new IfTrueNode(rc))); 1310 { 1311 PreserveJVMState pjvms(this); 1312 set_control(_gvn.transform(new IfFalseNode(rc))); 1313 uncommon_trap(Deoptimization::Reason_range_check, 1314 Deoptimization::Action_make_not_entrant); 1315 } 1316 1317 if (stopped()) { 1318 return false; 1319 } 1320 1321 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax)); 1322 result->set_req(0, control()); 1323 result = _gvn.transform(result); 1324 set_result(result); 1325 replace_in_map(index, result); 1326 clear_upper_avx(); 1327 return true; 1328 } 1329 1330 //------------------------------inline_string_indexOf------------------------ 1331 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1332 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1333 return false; 1334 } 1335 Node* src = argument(0); 1336 Node* tgt = argument(1); 1337 1338 // Make the merge point 1339 RegionNode* result_rgn = new RegionNode(4); 1340 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 1341 1342 src = must_be_not_null(src, true); 1343 tgt = must_be_not_null(tgt, true); 1344 1345 src = access_resolve(src, ACCESS_READ); 1346 tgt = access_resolve(tgt, ACCESS_READ); 1347 1348 // Get start addr and length of source string 1349 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1350 Node* src_count = load_array_length(src); 1351 1352 // Get start addr and length of substring 1353 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1354 Node* tgt_count = load_array_length(tgt); 1355 1356 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1357 // Divide src size by 2 if String is UTF16 encoded 1358 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1359 } 1360 if (ae == StrIntrinsicNode::UU) { 1361 // Divide substring size by 2 if String is UTF16 encoded 1362 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1363 } 1364 1365 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae); 1366 if (result != NULL) { 1367 result_phi->init_req(3, result); 1368 result_rgn->init_req(3, control()); 1369 } 1370 set_control(_gvn.transform(result_rgn)); 1371 record_for_igvn(result_rgn); 1372 set_result(_gvn.transform(result_phi)); 1373 1374 return true; 1375 } 1376 1377 //-----------------------------inline_string_indexOf----------------------- 1378 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1379 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1380 return false; 1381 } 1382 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1383 return false; 1384 } 1385 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1386 Node* src = argument(0); // byte[] 1387 Node* src_count = argument(1); // char count 1388 Node* tgt = argument(2); // byte[] 1389 Node* tgt_count = argument(3); // char count 1390 Node* from_index = argument(4); // char index 1391 1392 src = must_be_not_null(src, true); 1393 tgt = must_be_not_null(tgt, true); 1394 1395 src = access_resolve(src, ACCESS_READ); 1396 tgt = access_resolve(tgt, ACCESS_READ); 1397 1398 // Multiply byte array index by 2 if String is UTF16 encoded 1399 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1400 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1401 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1402 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1403 1404 // Range checks 1405 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1406 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1407 if (stopped()) { 1408 return true; 1409 } 1410 1411 RegionNode* region = new RegionNode(5); 1412 Node* phi = new PhiNode(region, TypeInt::INT); 1413 1414 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae); 1415 if (result != NULL) { 1416 // The result is index relative to from_index if substring was found, -1 otherwise. 1417 // Generate code which will fold into cmove. 1418 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1419 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1420 1421 Node* if_lt = generate_slow_guard(bol, NULL); 1422 if (if_lt != NULL) { 1423 // result == -1 1424 phi->init_req(3, result); 1425 region->init_req(3, if_lt); 1426 } 1427 if (!stopped()) { 1428 result = _gvn.transform(new AddINode(result, from_index)); 1429 phi->init_req(4, result); 1430 region->init_req(4, control()); 1431 } 1432 } 1433 1434 set_control(_gvn.transform(region)); 1435 record_for_igvn(region); 1436 set_result(_gvn.transform(phi)); 1437 clear_upper_avx(); 1438 1439 return true; 1440 } 1441 1442 // Create StrIndexOfNode with fast path checks 1443 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1444 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1445 // Check for substr count > string count 1446 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1447 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1448 Node* if_gt = generate_slow_guard(bol, NULL); 1449 if (if_gt != NULL) { 1450 phi->init_req(1, intcon(-1)); 1451 region->init_req(1, if_gt); 1452 } 1453 if (!stopped()) { 1454 // Check for substr count == 0 1455 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1456 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1457 Node* if_zero = generate_slow_guard(bol, NULL); 1458 if (if_zero != NULL) { 1459 phi->init_req(2, intcon(0)); 1460 region->init_req(2, if_zero); 1461 } 1462 } 1463 if (!stopped()) { 1464 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1465 } 1466 return NULL; 1467 } 1468 1469 //-----------------------------inline_string_indexOfChar----------------------- 1470 bool LibraryCallKit::inline_string_indexOfChar() { 1471 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1472 return false; 1473 } 1474 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1475 return false; 1476 } 1477 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1478 Node* src = argument(0); // byte[] 1479 Node* tgt = argument(1); // tgt is int ch 1480 Node* from_index = argument(2); 1481 Node* max = argument(3); 1482 1483 src = must_be_not_null(src, true); 1484 src = access_resolve(src, ACCESS_READ); 1485 1486 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1))); 1487 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1488 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1489 1490 // Range checks 1491 generate_string_range_check(src, src_offset, src_count, true); 1492 if (stopped()) { 1493 return true; 1494 } 1495 1496 RegionNode* region = new RegionNode(3); 1497 Node* phi = new PhiNode(region, TypeInt::INT); 1498 1499 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none); 1500 C->set_has_split_ifs(true); // Has chance for split-if optimization 1501 _gvn.transform(result); 1502 1503 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1504 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1505 1506 Node* if_lt = generate_slow_guard(bol, NULL); 1507 if (if_lt != NULL) { 1508 // result == -1 1509 phi->init_req(2, result); 1510 region->init_req(2, if_lt); 1511 } 1512 if (!stopped()) { 1513 result = _gvn.transform(new AddINode(result, from_index)); 1514 phi->init_req(1, result); 1515 region->init_req(1, control()); 1516 } 1517 set_control(_gvn.transform(region)); 1518 record_for_igvn(region); 1519 set_result(_gvn.transform(phi)); 1520 1521 return true; 1522 } 1523 //---------------------------inline_string_copy--------------------- 1524 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1525 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1526 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1527 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1528 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1529 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1530 bool LibraryCallKit::inline_string_copy(bool compress) { 1531 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1532 return false; 1533 } 1534 int nargs = 5; // 2 oops, 3 ints 1535 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1536 1537 Node* src = argument(0); 1538 Node* src_offset = argument(1); 1539 Node* dst = argument(2); 1540 Node* dst_offset = argument(3); 1541 Node* length = argument(4); 1542 1543 // Check for allocation before we add nodes that would confuse 1544 // tightly_coupled_allocation() 1545 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1546 1547 // Figure out the size and type of the elements we will be copying. 1548 const Type* src_type = src->Value(&_gvn); 1549 const Type* dst_type = dst->Value(&_gvn); 1550 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1551 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1552 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1553 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1554 "Unsupported array types for inline_string_copy"); 1555 1556 src = must_be_not_null(src, true); 1557 dst = must_be_not_null(dst, true); 1558 1559 // Convert char[] offsets to byte[] offsets 1560 bool convert_src = (compress && src_elem == T_BYTE); 1561 bool convert_dst = (!compress && dst_elem == T_BYTE); 1562 if (convert_src) { 1563 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1564 } else if (convert_dst) { 1565 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1566 } 1567 1568 // Range checks 1569 generate_string_range_check(src, src_offset, length, convert_src); 1570 generate_string_range_check(dst, dst_offset, length, convert_dst); 1571 if (stopped()) { 1572 return true; 1573 } 1574 1575 src = access_resolve(src, ACCESS_READ); 1576 dst = access_resolve(dst, ACCESS_WRITE); 1577 1578 Node* src_start = array_element_address(src, src_offset, src_elem); 1579 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1580 // 'src_start' points to src array + scaled offset 1581 // 'dst_start' points to dst array + scaled offset 1582 Node* count = NULL; 1583 if (compress) { 1584 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1585 } else { 1586 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1587 } 1588 1589 if (alloc != NULL) { 1590 if (alloc->maybe_set_complete(&_gvn)) { 1591 // "You break it, you buy it." 1592 InitializeNode* init = alloc->initialization(); 1593 assert(init->is_complete(), "we just did this"); 1594 init->set_complete_with_arraycopy(); 1595 assert(dst->is_CheckCastPP(), "sanity"); 1596 assert(dst->in(0)->in(0) == init, "dest pinned"); 1597 } 1598 // Do not let stores that initialize this object be reordered with 1599 // a subsequent store that would make this object accessible by 1600 // other threads. 1601 // Record what AllocateNode this StoreStore protects so that 1602 // escape analysis can go from the MemBarStoreStoreNode to the 1603 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1604 // based on the escape status of the AllocateNode. 1605 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1606 } 1607 if (compress) { 1608 set_result(_gvn.transform(count)); 1609 } 1610 clear_upper_avx(); 1611 1612 return true; 1613 } 1614 1615 #ifdef _LP64 1616 #define XTOP ,top() /*additional argument*/ 1617 #else //_LP64 1618 #define XTOP /*no additional argument*/ 1619 #endif //_LP64 1620 1621 //------------------------inline_string_toBytesU-------------------------- 1622 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1623 bool LibraryCallKit::inline_string_toBytesU() { 1624 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1625 return false; 1626 } 1627 // Get the arguments. 1628 Node* value = argument(0); 1629 Node* offset = argument(1); 1630 Node* length = argument(2); 1631 1632 Node* newcopy = NULL; 1633 1634 // Set the original stack and the reexecute bit for the interpreter to reexecute 1635 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1636 { PreserveReexecuteState preexecs(this); 1637 jvms()->set_should_reexecute(true); 1638 1639 // Check if a null path was taken unconditionally. 1640 value = null_check(value); 1641 1642 RegionNode* bailout = new RegionNode(1); 1643 record_for_igvn(bailout); 1644 1645 // Range checks 1646 generate_negative_guard(offset, bailout); 1647 generate_negative_guard(length, bailout); 1648 generate_limit_guard(offset, length, load_array_length(value), bailout); 1649 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1650 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1651 1652 if (bailout->req() > 1) { 1653 PreserveJVMState pjvms(this); 1654 set_control(_gvn.transform(bailout)); 1655 uncommon_trap(Deoptimization::Reason_intrinsic, 1656 Deoptimization::Action_maybe_recompile); 1657 } 1658 if (stopped()) { 1659 return true; 1660 } 1661 1662 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1663 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1664 newcopy = new_array(klass_node, size, 0); // no arguments to push 1665 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL); 1666 1667 // Calculate starting addresses. 1668 value = access_resolve(value, ACCESS_READ); 1669 Node* src_start = array_element_address(value, offset, T_CHAR); 1670 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1671 1672 // Check if src array address is aligned to HeapWordSize (dst is always aligned) 1673 const TypeInt* toffset = gvn().type(offset)->is_int(); 1674 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1675 1676 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1677 const char* copyfunc_name = "arraycopy"; 1678 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1679 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1680 OptoRuntime::fast_arraycopy_Type(), 1681 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1682 src_start, dst_start, ConvI2X(length) XTOP); 1683 // Do not let reads from the cloned object float above the arraycopy. 1684 if (alloc != NULL) { 1685 if (alloc->maybe_set_complete(&_gvn)) { 1686 // "You break it, you buy it." 1687 InitializeNode* init = alloc->initialization(); 1688 assert(init->is_complete(), "we just did this"); 1689 init->set_complete_with_arraycopy(); 1690 assert(newcopy->is_CheckCastPP(), "sanity"); 1691 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1692 } 1693 // Do not let stores that initialize this object be reordered with 1694 // a subsequent store that would make this object accessible by 1695 // other threads. 1696 // Record what AllocateNode this StoreStore protects so that 1697 // escape analysis can go from the MemBarStoreStoreNode to the 1698 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1699 // based on the escape status of the AllocateNode. 1700 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1701 } else { 1702 insert_mem_bar(Op_MemBarCPUOrder); 1703 } 1704 } // original reexecute is set back here 1705 1706 C->set_has_split_ifs(true); // Has chance for split-if optimization 1707 if (!stopped()) { 1708 set_result(newcopy); 1709 } 1710 clear_upper_avx(); 1711 1712 return true; 1713 } 1714 1715 //------------------------inline_string_getCharsU-------------------------- 1716 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1717 bool LibraryCallKit::inline_string_getCharsU() { 1718 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1719 return false; 1720 } 1721 1722 // Get the arguments. 1723 Node* src = argument(0); 1724 Node* src_begin = argument(1); 1725 Node* src_end = argument(2); // exclusive offset (i < src_end) 1726 Node* dst = argument(3); 1727 Node* dst_begin = argument(4); 1728 1729 // Check for allocation before we add nodes that would confuse 1730 // tightly_coupled_allocation() 1731 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1732 1733 // Check if a null path was taken unconditionally. 1734 src = null_check(src); 1735 dst = null_check(dst); 1736 if (stopped()) { 1737 return true; 1738 } 1739 1740 // Get length and convert char[] offset to byte[] offset 1741 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1742 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1743 1744 // Range checks 1745 generate_string_range_check(src, src_begin, length, true); 1746 generate_string_range_check(dst, dst_begin, length, false); 1747 if (stopped()) { 1748 return true; 1749 } 1750 1751 if (!stopped()) { 1752 src = access_resolve(src, ACCESS_READ); 1753 dst = access_resolve(dst, ACCESS_WRITE); 1754 1755 // Calculate starting addresses. 1756 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1757 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1758 1759 // Check if array addresses are aligned to HeapWordSize 1760 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1761 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1762 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1763 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1764 1765 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1766 const char* copyfunc_name = "arraycopy"; 1767 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1768 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1769 OptoRuntime::fast_arraycopy_Type(), 1770 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1771 src_start, dst_start, ConvI2X(length) XTOP); 1772 // Do not let reads from the cloned object float above the arraycopy. 1773 if (alloc != NULL) { 1774 if (alloc->maybe_set_complete(&_gvn)) { 1775 // "You break it, you buy it." 1776 InitializeNode* init = alloc->initialization(); 1777 assert(init->is_complete(), "we just did this"); 1778 init->set_complete_with_arraycopy(); 1779 assert(dst->is_CheckCastPP(), "sanity"); 1780 assert(dst->in(0)->in(0) == init, "dest pinned"); 1781 } 1782 // Do not let stores that initialize this object be reordered with 1783 // a subsequent store that would make this object accessible by 1784 // other threads. 1785 // Record what AllocateNode this StoreStore protects so that 1786 // escape analysis can go from the MemBarStoreStoreNode to the 1787 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1788 // based on the escape status of the AllocateNode. 1789 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1790 } else { 1791 insert_mem_bar(Op_MemBarCPUOrder); 1792 } 1793 } 1794 1795 C->set_has_split_ifs(true); // Has chance for split-if optimization 1796 return true; 1797 } 1798 1799 //----------------------inline_string_char_access---------------------------- 1800 // Store/Load char to/from byte[] array. 1801 // static void StringUTF16.putChar(byte[] val, int index, int c) 1802 // static char StringUTF16.getChar(byte[] val, int index) 1803 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1804 Node* value = argument(0); 1805 Node* index = argument(1); 1806 Node* ch = is_store ? argument(2) : NULL; 1807 1808 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1809 // correctly requires matched array shapes. 1810 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1811 "sanity: byte[] and char[] bases agree"); 1812 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1813 "sanity: byte[] and char[] scales agree"); 1814 1815 // Bail when getChar over constants is requested: constant folding would 1816 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1817 // Java method would constant fold nicely instead. 1818 if (!is_store && value->is_Con() && index->is_Con()) { 1819 return false; 1820 } 1821 1822 value = must_be_not_null(value, true); 1823 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ); 1824 1825 Node* adr = array_element_address(value, index, T_CHAR); 1826 if (adr->is_top()) { 1827 return false; 1828 } 1829 if (is_store) { 1830 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED); 1831 } else { 1832 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 1833 set_result(ch); 1834 } 1835 return true; 1836 } 1837 1838 //--------------------------round_double_node-------------------------------- 1839 // Round a double node if necessary. 1840 Node* LibraryCallKit::round_double_node(Node* n) { 1841 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1) 1842 n = _gvn.transform(new RoundDoubleNode(0, n)); 1843 return n; 1844 } 1845 1846 //------------------------------inline_math----------------------------------- 1847 // public static double Math.abs(double) 1848 // public static double Math.sqrt(double) 1849 // public static double Math.log(double) 1850 // public static double Math.log10(double) 1851 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) { 1852 Node* arg = round_double_node(argument(0)); 1853 Node* n = NULL; 1854 switch (id) { 1855 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1856 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break; 1857 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break; 1858 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break; 1859 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break; 1860 default: fatal_unexpected_iid(id); break; 1861 } 1862 set_result(_gvn.transform(n)); 1863 return true; 1864 } 1865 1866 //------------------------------inline_math----------------------------------- 1867 // public static float Math.abs(float) 1868 // public static int Math.abs(int) 1869 // public static long Math.abs(long) 1870 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1871 Node* arg = argument(0); 1872 Node* n = NULL; 1873 switch (id) { 1874 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break; 1875 case vmIntrinsics::_iabs: n = new AbsINode( arg); break; 1876 case vmIntrinsics::_labs: n = new AbsLNode( arg); break; 1877 default: fatal_unexpected_iid(id); break; 1878 } 1879 set_result(_gvn.transform(n)); 1880 return true; 1881 } 1882 1883 //------------------------------runtime_math----------------------------- 1884 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1885 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1886 "must be (DD)D or (D)D type"); 1887 1888 // Inputs 1889 Node* a = round_double_node(argument(0)); 1890 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL; 1891 1892 const TypePtr* no_memory_effects = NULL; 1893 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName, 1894 no_memory_effects, 1895 a, top(), b, b ? top() : NULL); 1896 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1897 #ifdef ASSERT 1898 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1899 assert(value_top == top(), "second value must be top"); 1900 #endif 1901 1902 set_result(value); 1903 return true; 1904 } 1905 1906 //------------------------------inline_math_native----------------------------- 1907 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1908 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f) 1909 switch (id) { 1910 // These intrinsics are not properly supported on all hardware 1911 case vmIntrinsics::_dsin: 1912 return StubRoutines::dsin() != NULL ? 1913 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1914 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN"); 1915 case vmIntrinsics::_dcos: 1916 return StubRoutines::dcos() != NULL ? 1917 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1918 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS"); 1919 case vmIntrinsics::_dtan: 1920 return StubRoutines::dtan() != NULL ? 1921 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") : 1922 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN"); 1923 case vmIntrinsics::_dlog: 1924 return StubRoutines::dlog() != NULL ? 1925 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1926 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG"); 1927 case vmIntrinsics::_dlog10: 1928 return StubRoutines::dlog10() != NULL ? 1929 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") : 1930 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10"); 1931 1932 // These intrinsics are supported on all hardware 1933 case vmIntrinsics::_ceil: 1934 case vmIntrinsics::_floor: 1935 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false; 1936 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false; 1937 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false; 1938 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false; 1939 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false; 1940 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false; 1941 1942 case vmIntrinsics::_dexp: 1943 return StubRoutines::dexp() != NULL ? 1944 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1945 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP"); 1946 case vmIntrinsics::_dpow: { 1947 Node* exp = round_double_node(argument(2)); 1948 const TypeD* d = _gvn.type(exp)->isa_double_constant(); 1949 if (d != NULL && d->getd() == 2.0) { 1950 // Special case: pow(x, 2.0) => x * x 1951 Node* base = round_double_node(argument(0)); 1952 set_result(_gvn.transform(new MulDNode(base, base))); 1953 return true; 1954 } 1955 return StubRoutines::dpow() != NULL ? 1956 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1957 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW"); 1958 } 1959 #undef FN_PTR 1960 1961 // These intrinsics are not yet correctly implemented 1962 case vmIntrinsics::_datan2: 1963 return false; 1964 1965 default: 1966 fatal_unexpected_iid(id); 1967 return false; 1968 } 1969 } 1970 1971 static bool is_simple_name(Node* n) { 1972 return (n->req() == 1 // constant 1973 || (n->is_Type() && n->as_Type()->type()->singleton()) 1974 || n->is_Proj() // parameter or return value 1975 || n->is_Phi() // local of some sort 1976 ); 1977 } 1978 1979 //----------------------------inline_notify-----------------------------------* 1980 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1981 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1982 address func; 1983 if (id == vmIntrinsics::_notify) { 1984 func = OptoRuntime::monitor_notify_Java(); 1985 } else { 1986 func = OptoRuntime::monitor_notifyAll_Java(); 1987 } 1988 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0)); 1989 make_slow_call_ex(call, env()->Throwable_klass(), false); 1990 return true; 1991 } 1992 1993 1994 //----------------------------inline_min_max----------------------------------- 1995 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1996 set_result(generate_min_max(id, argument(0), argument(1))); 1997 return true; 1998 } 1999 2000 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) { 2001 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 2002 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 2003 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 2004 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 2005 2006 { 2007 PreserveJVMState pjvms(this); 2008 PreserveReexecuteState preexecs(this); 2009 jvms()->set_should_reexecute(true); 2010 2011 set_control(slow_path); 2012 set_i_o(i_o()); 2013 2014 uncommon_trap(Deoptimization::Reason_intrinsic, 2015 Deoptimization::Action_none); 2016 } 2017 2018 set_control(fast_path); 2019 set_result(math); 2020 } 2021 2022 template <typename OverflowOp> 2023 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 2024 typedef typename OverflowOp::MathOp MathOp; 2025 2026 MathOp* mathOp = new MathOp(arg1, arg2); 2027 Node* operation = _gvn.transform( mathOp ); 2028 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 2029 inline_math_mathExact(operation, ofcheck); 2030 return true; 2031 } 2032 2033 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 2034 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 2035 } 2036 2037 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 2038 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 2039 } 2040 2041 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 2042 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 2043 } 2044 2045 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 2046 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 2047 } 2048 2049 bool LibraryCallKit::inline_math_negateExactI() { 2050 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 2051 } 2052 2053 bool LibraryCallKit::inline_math_negateExactL() { 2054 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 2055 } 2056 2057 bool LibraryCallKit::inline_math_multiplyExactI() { 2058 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 2059 } 2060 2061 bool LibraryCallKit::inline_math_multiplyExactL() { 2062 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 2063 } 2064 2065 bool LibraryCallKit::inline_math_multiplyHigh() { 2066 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2)))); 2067 return true; 2068 } 2069 2070 Node* 2071 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) { 2072 // These are the candidate return value: 2073 Node* xvalue = x0; 2074 Node* yvalue = y0; 2075 2076 if (xvalue == yvalue) { 2077 return xvalue; 2078 } 2079 2080 bool want_max = (id == vmIntrinsics::_max); 2081 2082 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int(); 2083 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int(); 2084 if (txvalue == NULL || tyvalue == NULL) return top(); 2085 // This is not really necessary, but it is consistent with a 2086 // hypothetical MaxINode::Value method: 2087 int widen = MAX2(txvalue->_widen, tyvalue->_widen); 2088 2089 // %%% This folding logic should (ideally) be in a different place. 2090 // Some should be inside IfNode, and there to be a more reliable 2091 // transformation of ?: style patterns into cmoves. We also want 2092 // more powerful optimizations around cmove and min/max. 2093 2094 // Try to find a dominating comparison of these guys. 2095 // It can simplify the index computation for Arrays.copyOf 2096 // and similar uses of System.arraycopy. 2097 // First, compute the normalized version of CmpI(x, y). 2098 int cmp_op = Op_CmpI; 2099 Node* xkey = xvalue; 2100 Node* ykey = yvalue; 2101 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey)); 2102 if (ideal_cmpxy->is_Cmp()) { 2103 // E.g., if we have CmpI(length - offset, count), 2104 // it might idealize to CmpI(length, count + offset) 2105 cmp_op = ideal_cmpxy->Opcode(); 2106 xkey = ideal_cmpxy->in(1); 2107 ykey = ideal_cmpxy->in(2); 2108 } 2109 2110 // Start by locating any relevant comparisons. 2111 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey; 2112 Node* cmpxy = NULL; 2113 Node* cmpyx = NULL; 2114 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) { 2115 Node* cmp = start_from->fast_out(k); 2116 if (cmp->outcnt() > 0 && // must have prior uses 2117 cmp->in(0) == NULL && // must be context-independent 2118 cmp->Opcode() == cmp_op) { // right kind of compare 2119 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp; 2120 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp; 2121 } 2122 } 2123 2124 const int NCMPS = 2; 2125 Node* cmps[NCMPS] = { cmpxy, cmpyx }; 2126 int cmpn; 2127 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2128 if (cmps[cmpn] != NULL) break; // find a result 2129 } 2130 if (cmpn < NCMPS) { 2131 // Look for a dominating test that tells us the min and max. 2132 int depth = 0; // Limit search depth for speed 2133 Node* dom = control(); 2134 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) { 2135 if (++depth >= 100) break; 2136 Node* ifproj = dom; 2137 if (!ifproj->is_Proj()) continue; 2138 Node* iff = ifproj->in(0); 2139 if (!iff->is_If()) continue; 2140 Node* bol = iff->in(1); 2141 if (!bol->is_Bool()) continue; 2142 Node* cmp = bol->in(1); 2143 if (cmp == NULL) continue; 2144 for (cmpn = 0; cmpn < NCMPS; cmpn++) 2145 if (cmps[cmpn] == cmp) break; 2146 if (cmpn == NCMPS) continue; 2147 BoolTest::mask btest = bol->as_Bool()->_test._test; 2148 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate(); 2149 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2150 // At this point, we know that 'x btest y' is true. 2151 switch (btest) { 2152 case BoolTest::eq: 2153 // They are proven equal, so we can collapse the min/max. 2154 // Either value is the answer. Choose the simpler. 2155 if (is_simple_name(yvalue) && !is_simple_name(xvalue)) 2156 return yvalue; 2157 return xvalue; 2158 case BoolTest::lt: // x < y 2159 case BoolTest::le: // x <= y 2160 return (want_max ? yvalue : xvalue); 2161 case BoolTest::gt: // x > y 2162 case BoolTest::ge: // x >= y 2163 return (want_max ? xvalue : yvalue); 2164 default: 2165 break; 2166 } 2167 } 2168 } 2169 2170 // We failed to find a dominating test. 2171 // Let's pick a test that might GVN with prior tests. 2172 Node* best_bol = NULL; 2173 BoolTest::mask best_btest = BoolTest::illegal; 2174 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2175 Node* cmp = cmps[cmpn]; 2176 if (cmp == NULL) continue; 2177 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) { 2178 Node* bol = cmp->fast_out(j); 2179 if (!bol->is_Bool()) continue; 2180 BoolTest::mask btest = bol->as_Bool()->_test._test; 2181 if (btest == BoolTest::eq || btest == BoolTest::ne) continue; 2182 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2183 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) { 2184 best_bol = bol->as_Bool(); 2185 best_btest = btest; 2186 } 2187 } 2188 } 2189 2190 Node* answer_if_true = NULL; 2191 Node* answer_if_false = NULL; 2192 switch (best_btest) { 2193 default: 2194 if (cmpxy == NULL) 2195 cmpxy = ideal_cmpxy; 2196 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt)); 2197 // and fall through: 2198 case BoolTest::lt: // x < y 2199 case BoolTest::le: // x <= y 2200 answer_if_true = (want_max ? yvalue : xvalue); 2201 answer_if_false = (want_max ? xvalue : yvalue); 2202 break; 2203 case BoolTest::gt: // x > y 2204 case BoolTest::ge: // x >= y 2205 answer_if_true = (want_max ? xvalue : yvalue); 2206 answer_if_false = (want_max ? yvalue : xvalue); 2207 break; 2208 } 2209 2210 jint hi, lo; 2211 if (want_max) { 2212 // We can sharpen the minimum. 2213 hi = MAX2(txvalue->_hi, tyvalue->_hi); 2214 lo = MAX2(txvalue->_lo, tyvalue->_lo); 2215 } else { 2216 // We can sharpen the maximum. 2217 hi = MIN2(txvalue->_hi, tyvalue->_hi); 2218 lo = MIN2(txvalue->_lo, tyvalue->_lo); 2219 } 2220 2221 // Use a flow-free graph structure, to avoid creating excess control edges 2222 // which could hinder other optimizations. 2223 // Since Math.min/max is often used with arraycopy, we want 2224 // tightly_coupled_allocation to be able to see beyond min/max expressions. 2225 Node* cmov = CMoveNode::make(NULL, best_bol, 2226 answer_if_false, answer_if_true, 2227 TypeInt::make(lo, hi, widen)); 2228 2229 return _gvn.transform(cmov); 2230 2231 /* 2232 // This is not as desirable as it may seem, since Min and Max 2233 // nodes do not have a full set of optimizations. 2234 // And they would interfere, anyway, with 'if' optimizations 2235 // and with CMoveI canonical forms. 2236 switch (id) { 2237 case vmIntrinsics::_min: 2238 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break; 2239 case vmIntrinsics::_max: 2240 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break; 2241 default: 2242 ShouldNotReachHere(); 2243 } 2244 */ 2245 } 2246 2247 inline int 2248 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) { 2249 const TypePtr* base_type = TypePtr::NULL_PTR; 2250 if (base != NULL) base_type = _gvn.type(base)->isa_ptr(); 2251 if (base_type == NULL) { 2252 // Unknown type. 2253 return Type::AnyPtr; 2254 } else if (base_type == TypePtr::NULL_PTR) { 2255 // Since this is a NULL+long form, we have to switch to a rawptr. 2256 base = _gvn.transform(new CastX2PNode(offset)); 2257 offset = MakeConX(0); 2258 return Type::RawPtr; 2259 } else if (base_type->base() == Type::RawPtr) { 2260 return Type::RawPtr; 2261 } else if (base_type->isa_oopptr()) { 2262 // Base is never null => always a heap address. 2263 if (!TypePtr::NULL_PTR->higher_equal(base_type)) { 2264 return Type::OopPtr; 2265 } 2266 // Offset is small => always a heap address. 2267 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2268 if (offset_type != NULL && 2269 base_type->offset() == 0 && // (should always be?) 2270 offset_type->_lo >= 0 && 2271 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2272 return Type::OopPtr; 2273 } else if (type == T_OBJECT) { 2274 // off heap access to an oop doesn't make any sense. Has to be on 2275 // heap. 2276 return Type::OopPtr; 2277 } 2278 // Otherwise, it might either be oop+off or NULL+addr. 2279 return Type::AnyPtr; 2280 } else { 2281 // No information: 2282 return Type::AnyPtr; 2283 } 2284 } 2285 2286 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) { 2287 Node* uncasted_base = base; 2288 int kind = classify_unsafe_addr(uncasted_base, offset, type); 2289 if (kind == Type::RawPtr) { 2290 return basic_plus_adr(top(), uncasted_base, offset); 2291 } else if (kind == Type::AnyPtr) { 2292 assert(base == uncasted_base, "unexpected base change"); 2293 if (can_cast) { 2294 if (!_gvn.type(base)->speculative_maybe_null() && 2295 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 2296 // According to profiling, this access is always on 2297 // heap. Casting the base to not null and thus avoiding membars 2298 // around the access should allow better optimizations 2299 Node* null_ctl = top(); 2300 base = null_check_oop(base, &null_ctl, true, true, true); 2301 assert(null_ctl->is_top(), "no null control here"); 2302 return basic_plus_adr(base, offset); 2303 } else if (_gvn.type(base)->speculative_always_null() && 2304 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 2305 // According to profiling, this access is always off 2306 // heap. 2307 base = null_assert(base); 2308 Node* raw_base = _gvn.transform(new CastX2PNode(offset)); 2309 offset = MakeConX(0); 2310 return basic_plus_adr(top(), raw_base, offset); 2311 } 2312 } 2313 // We don't know if it's an on heap or off heap access. Fall back 2314 // to raw memory access. 2315 base = access_resolve(base, decorators); 2316 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM)); 2317 return basic_plus_adr(top(), raw, offset); 2318 } else { 2319 assert(base == uncasted_base, "unexpected base change"); 2320 // We know it's an on heap access so base can't be null 2321 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) { 2322 base = must_be_not_null(base, true); 2323 } 2324 return basic_plus_adr(base, offset); 2325 } 2326 } 2327 2328 //--------------------------inline_number_methods----------------------------- 2329 // inline int Integer.numberOfLeadingZeros(int) 2330 // inline int Long.numberOfLeadingZeros(long) 2331 // 2332 // inline int Integer.numberOfTrailingZeros(int) 2333 // inline int Long.numberOfTrailingZeros(long) 2334 // 2335 // inline int Integer.bitCount(int) 2336 // inline int Long.bitCount(long) 2337 // 2338 // inline char Character.reverseBytes(char) 2339 // inline short Short.reverseBytes(short) 2340 // inline int Integer.reverseBytes(int) 2341 // inline long Long.reverseBytes(long) 2342 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2343 Node* arg = argument(0); 2344 Node* n = NULL; 2345 switch (id) { 2346 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2347 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2348 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2349 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2350 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2351 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2352 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break; 2353 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break; 2354 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break; 2355 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break; 2356 default: fatal_unexpected_iid(id); break; 2357 } 2358 set_result(_gvn.transform(n)); 2359 return true; 2360 } 2361 2362 //----------------------------inline_unsafe_access---------------------------- 2363 2364 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) { 2365 // Attempt to infer a sharper value type from the offset and base type. 2366 ciKlass* sharpened_klass = NULL; 2367 2368 // See if it is an instance field, with an object type. 2369 if (alias_type->field() != NULL) { 2370 if (alias_type->field()->type()->is_klass()) { 2371 sharpened_klass = alias_type->field()->type()->as_klass(); 2372 } 2373 } 2374 2375 // See if it is a narrow oop array. 2376 if (adr_type->isa_aryptr()) { 2377 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2378 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr(); 2379 if (elem_type != NULL) { 2380 sharpened_klass = elem_type->klass(); 2381 } 2382 } 2383 } 2384 2385 // The sharpened class might be unloaded if there is no class loader 2386 // contraint in place. 2387 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) { 2388 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass); 2389 2390 #ifndef PRODUCT 2391 if (C->print_intrinsics() || C->print_inlining()) { 2392 tty->print(" from base type: "); adr_type->dump(); tty->cr(); 2393 tty->print(" sharpened value: "); tjp->dump(); tty->cr(); 2394 } 2395 #endif 2396 // Sharpen the value type. 2397 return tjp; 2398 } 2399 return NULL; 2400 } 2401 2402 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) { 2403 switch (kind) { 2404 case Relaxed: 2405 return MO_UNORDERED; 2406 case Opaque: 2407 return MO_RELAXED; 2408 case Acquire: 2409 return MO_ACQUIRE; 2410 case Release: 2411 return MO_RELEASE; 2412 case Volatile: 2413 return MO_SEQ_CST; 2414 default: 2415 ShouldNotReachHere(); 2416 return 0; 2417 } 2418 } 2419 2420 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) { 2421 if (callee()->is_static()) return false; // caller must have the capability! 2422 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2423 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2424 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2425 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2426 2427 if (is_reference_type(type)) { 2428 decorators |= ON_UNKNOWN_OOP_REF; 2429 } 2430 2431 if (unaligned) { 2432 decorators |= C2_UNALIGNED; 2433 } 2434 2435 #ifndef PRODUCT 2436 { 2437 ResourceMark rm; 2438 // Check the signatures. 2439 ciSignature* sig = callee()->signature(); 2440 #ifdef ASSERT 2441 if (!is_store) { 2442 // Object getReference(Object base, int/long offset), etc. 2443 BasicType rtype = sig->return_type()->basic_type(); 2444 assert(rtype == type || (rtype == T_OBJECT && type == T_VALUETYPE), "getter must return the expected value"); 2445 assert(sig->count() == 2 || (type == T_VALUETYPE && sig->count() == 3), "oop getter has 2 or 3 arguments"); 2446 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2447 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2448 } else { 2449 // void putReference(Object base, int/long offset, Object x), etc. 2450 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2451 assert(sig->count() == 3 || (type == T_VALUETYPE && sig->count() == 4), "oop putter has 3 arguments"); 2452 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2453 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2454 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2455 assert(vtype == type || (type == T_VALUETYPE && vtype == T_OBJECT), "putter must accept the expected value"); 2456 } 2457 #endif // ASSERT 2458 } 2459 #endif //PRODUCT 2460 2461 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2462 2463 Node* receiver = argument(0); // type: oop 2464 2465 // Build address expression. 2466 Node* adr; 2467 Node* heap_base_oop = top(); 2468 Node* offset = top(); 2469 Node* val; 2470 2471 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2472 Node* base = argument(1); // type: oop 2473 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2474 offset = argument(2); // type: long 2475 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2476 // to be plain byte offsets, which are also the same as those accepted 2477 // by oopDesc::field_addr. 2478 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2479 "fieldOffset must be byte-scaled"); 2480 2481 ciValueKlass* value_klass = NULL; 2482 if (type == T_VALUETYPE) { 2483 Node* cls = null_check(argument(4)); 2484 if (stopped()) { 2485 return true; 2486 } 2487 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 2488 const TypeKlassPtr* kls_t = _gvn.type(kls)->isa_klassptr(); 2489 if (!kls_t->klass_is_exact()) { 2490 return false; 2491 } 2492 ciKlass* klass = kls_t->klass(); 2493 if (!klass->is_valuetype()) { 2494 return false; 2495 } 2496 value_klass = klass->as_value_klass(); 2497 } 2498 2499 receiver = null_check(receiver); 2500 if (stopped()) { 2501 return true; 2502 } 2503 2504 if (base->is_ValueType()) { 2505 ValueTypeNode* vt = base->as_ValueType(); 2506 2507 if (is_store) { 2508 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) { 2509 return false; 2510 } 2511 base = vt->get_oop(); 2512 } else { 2513 if (offset->is_Con()) { 2514 long off = find_long_con(offset, 0); 2515 ciValueKlass* vk = vt->type()->value_klass(); 2516 if ((long)(int)off != off || !vk->contains_field_offset(off)) { 2517 return false; 2518 } 2519 2520 ciField* f = vk->get_non_flattened_field_by_offset((int)off); 2521 2522 if (f != NULL) { 2523 BasicType bt = f->layout_type(); 2524 if (bt == T_ARRAY || bt == T_NARROWOOP) { 2525 bt = T_OBJECT; 2526 } 2527 if (bt == type) { 2528 if (bt != T_VALUETYPE || f->type() == value_klass) { 2529 set_result(vt->field_value_by_offset((int)off, false)); 2530 return true; 2531 } 2532 } 2533 } 2534 } 2535 // Re-execute the unsafe access if allocation triggers deoptimization. 2536 PreserveReexecuteState preexecs(this); 2537 jvms()->set_should_reexecute(true); 2538 vt = vt->allocate(this)->as_ValueType(); 2539 base = vt->get_oop(); 2540 } 2541 } 2542 2543 // 32-bit machines ignore the high half! 2544 offset = ConvL2X(offset); 2545 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed); 2546 2547 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) { 2548 heap_base_oop = base; 2549 } else if (type == T_OBJECT || (value_klass != NULL && value_klass->has_object_fields())) { 2550 return false; // off-heap oop accesses are not supported 2551 } 2552 2553 // Can base be NULL? Otherwise, always on-heap access. 2554 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base)); 2555 2556 if (!can_access_non_heap) { 2557 decorators |= IN_HEAP; 2558 } 2559 2560 val = is_store ? argument(4 + (type == T_VALUETYPE ? 1 : 0)) : NULL; 2561 2562 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr(); 2563 if (adr_type == TypePtr::NULL_PTR) { 2564 return false; // off-heap access with zero address 2565 } 2566 2567 // Try to categorize the address. 2568 Compile::AliasType* alias_type = C->alias_type(adr_type); 2569 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2570 2571 if (alias_type->adr_type() == TypeInstPtr::KLASS || 2572 alias_type->adr_type() == TypeAryPtr::RANGE) { 2573 return false; // not supported 2574 } 2575 2576 bool mismatched = false; 2577 BasicType bt = T_ILLEGAL; 2578 ciField* field = NULL; 2579 if (adr_type->isa_instptr()) { 2580 const TypeInstPtr* instptr = adr_type->is_instptr(); 2581 ciInstanceKlass* k = instptr->klass()->as_instance_klass(); 2582 int off = instptr->offset(); 2583 if (instptr->const_oop() != NULL && 2584 instptr->klass() == ciEnv::current()->Class_klass() && 2585 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) { 2586 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass(); 2587 field = k->get_field_by_offset(off, true); 2588 } else { 2589 field = k->get_non_flattened_field_by_offset(off); 2590 } 2591 if (field != NULL) { 2592 bt = field->layout_type(); 2593 } 2594 assert(bt == alias_type->basic_type() || bt == T_VALUETYPE, "should match"); 2595 if (field != NULL && bt == T_VALUETYPE && !field->is_flattened()) { 2596 bt = T_OBJECT; 2597 } 2598 } else { 2599 bt = alias_type->basic_type(); 2600 } 2601 2602 if (bt != T_ILLEGAL) { 2603 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access"); 2604 if (bt == T_BYTE && adr_type->isa_aryptr()) { 2605 // Alias type doesn't differentiate between byte[] and boolean[]). 2606 // Use address type to get the element type. 2607 bt = adr_type->is_aryptr()->elem()->array_element_basic_type(); 2608 } 2609 if (bt == T_ARRAY || bt == T_NARROWOOP) { 2610 // accessing an array field with getReference is not a mismatch 2611 bt = T_OBJECT; 2612 } 2613 if ((bt == T_OBJECT) != (type == T_OBJECT)) { 2614 // Don't intrinsify mismatched object accesses 2615 return false; 2616 } 2617 mismatched = (bt != type); 2618 } else if (alias_type->adr_type()->isa_oopptr()) { 2619 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched 2620 } 2621 2622 if (type == T_VALUETYPE) { 2623 if (adr_type->isa_instptr()) { 2624 if (field == NULL || field->type() != value_klass) { 2625 mismatched = true; 2626 } 2627 } else if (adr_type->isa_aryptr()) { 2628 const Type* elem = adr_type->is_aryptr()->elem(); 2629 if (!elem->isa_valuetype()) { 2630 mismatched = true; 2631 } else if (elem->value_klass() != value_klass) { 2632 mismatched = true; 2633 } 2634 } 2635 if (is_store) { 2636 const Type* val_t = _gvn.type(val); 2637 if (!val_t->isa_valuetype() || val_t->value_klass() != value_klass) { 2638 return false; 2639 } 2640 } 2641 } 2642 2643 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched"); 2644 2645 if (mismatched) { 2646 decorators |= C2_MISMATCHED; 2647 } 2648 2649 // First guess at the value type. 2650 const Type *value_type = Type::get_const_basic_type(type); 2651 2652 // Figure out the memory ordering. 2653 decorators |= mo_decorator_for_access_kind(kind); 2654 2655 if (!is_store) { 2656 if (type == T_OBJECT) { 2657 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2658 if (tjp != NULL) { 2659 value_type = tjp; 2660 } 2661 } else if (type == T_VALUETYPE) { 2662 value_type = NULL; 2663 } 2664 } 2665 2666 // Heap pointers get a null-check from the interpreter, 2667 // as a courtesy. However, this is not guaranteed by Unsafe, 2668 // and it is not possible to fully distinguish unintended nulls 2669 // from intended ones in this API. 2670 2671 if (!is_store) { 2672 Node* p = NULL; 2673 // Try to constant fold a load from a constant field 2674 2675 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) { 2676 // final or stable field 2677 p = make_constant_from_field(field, heap_base_oop); 2678 } 2679 2680 if (p == NULL) { // Could not constant fold the load 2681 if (type == T_VALUETYPE) { 2682 if (adr_type->isa_instptr() && !mismatched) { 2683 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass(); 2684 int offset = adr_type->is_instptr()->offset(); 2685 p = ValueTypeNode::make_from_flattened(this, value_klass, base, base, holder, offset, decorators); 2686 } else { 2687 p = ValueTypeNode::make_from_flattened(this, value_klass, base, adr, NULL, 0, decorators); 2688 } 2689 } else { 2690 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators); 2691 } 2692 // Normalize the value returned by getBoolean in the following cases 2693 if (type == T_BOOLEAN && 2694 (mismatched || 2695 heap_base_oop == top() || // - heap_base_oop is NULL or 2696 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL 2697 // and the unsafe access is made to large offset 2698 // (i.e., larger than the maximum offset necessary for any 2699 // field access) 2700 ) { 2701 IdealKit ideal = IdealKit(this); 2702 #define __ ideal. 2703 IdealVariable normalized_result(ideal); 2704 __ declarations_done(); 2705 __ set(normalized_result, p); 2706 __ if_then(p, BoolTest::ne, ideal.ConI(0)); 2707 __ set(normalized_result, ideal.ConI(1)); 2708 ideal.end_if(); 2709 final_sync(ideal); 2710 p = __ value(normalized_result); 2711 #undef __ 2712 } 2713 } 2714 if (type == T_ADDRESS) { 2715 p = gvn().transform(new CastP2XNode(NULL, p)); 2716 p = ConvX2UL(p); 2717 } 2718 if (field != NULL && field->is_flattenable() && !field->is_flattened()) { 2719 // Load a non-flattened but flattenable value type from memory 2720 if (value_type->value_klass()->is_scalarizable()) { 2721 p = ValueTypeNode::make_from_oop(this, p, value_type->value_klass()); 2722 } else { 2723 p = null2default(p, value_type->value_klass()); 2724 } 2725 } 2726 // The load node has the control of the preceding MemBarCPUOrder. All 2727 // following nodes will have the control of the MemBarCPUOrder inserted at 2728 // the end of this method. So, pushing the load onto the stack at a later 2729 // point is fine. 2730 set_result(p); 2731 } else { 2732 if (bt == T_ADDRESS) { 2733 // Repackage the long as a pointer. 2734 val = ConvL2X(val); 2735 val = gvn().transform(new CastX2PNode(val)); 2736 } 2737 if (type == T_VALUETYPE) { 2738 if (adr_type->isa_instptr() && !mismatched) { 2739 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass(); 2740 int offset = adr_type->is_instptr()->offset(); 2741 val->as_ValueType()->store_flattened(this, base, base, holder, offset, decorators); 2742 } else { 2743 val->as_ValueType()->store_flattened(this, base, adr, NULL, 0, decorators); 2744 } 2745 } else { 2746 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators); 2747 } 2748 } 2749 2750 if (argument(1)->is_ValueType() && is_store) { 2751 Node* value = ValueTypeNode::make_from_oop(this, base, _gvn.type(base)->value_klass()); 2752 value = value->as_ValueType()->make_larval(this, false); 2753 replace_in_map(argument(1), value); 2754 } 2755 2756 return true; 2757 } 2758 2759 bool LibraryCallKit::inline_unsafe_make_private_buffer() { 2760 Node* receiver = argument(0); 2761 Node* value = argument(1); 2762 2763 receiver = null_check(receiver); 2764 if (stopped()) { 2765 return true; 2766 } 2767 2768 if (!value->is_ValueType()) { 2769 return false; 2770 } 2771 2772 set_result(value->as_ValueType()->make_larval(this, true)); 2773 2774 return true; 2775 } 2776 2777 bool LibraryCallKit::inline_unsafe_finish_private_buffer() { 2778 Node* receiver = argument(0); 2779 Node* buffer = argument(1); 2780 2781 receiver = null_check(receiver); 2782 if (stopped()) { 2783 return true; 2784 } 2785 2786 if (!buffer->is_ValueType()) { 2787 return false; 2788 } 2789 2790 ValueTypeNode* vt = buffer->as_ValueType(); 2791 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) { 2792 return false; 2793 } 2794 2795 set_result(vt->finish_larval(this)); 2796 2797 return true; 2798 } 2799 2800 //----------------------------inline_unsafe_load_store---------------------------- 2801 // This method serves a couple of different customers (depending on LoadStoreKind): 2802 // 2803 // LS_cmp_swap: 2804 // 2805 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x); 2806 // boolean compareAndSetInt( Object o, long offset, int expected, int x); 2807 // boolean compareAndSetLong( Object o, long offset, long expected, long x); 2808 // 2809 // LS_cmp_swap_weak: 2810 // 2811 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x); 2812 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x); 2813 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x); 2814 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x); 2815 // 2816 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x); 2817 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x); 2818 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x); 2819 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x); 2820 // 2821 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x); 2822 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x); 2823 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x); 2824 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x); 2825 // 2826 // LS_cmp_exchange: 2827 // 2828 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x); 2829 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x); 2830 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x); 2831 // 2832 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2833 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2834 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2835 // 2836 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2837 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2838 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2839 // 2840 // LS_get_add: 2841 // 2842 // int getAndAddInt( Object o, long offset, int delta) 2843 // long getAndAddLong(Object o, long offset, long delta) 2844 // 2845 // LS_get_set: 2846 // 2847 // int getAndSet(Object o, long offset, int newValue) 2848 // long getAndSet(Object o, long offset, long newValue) 2849 // Object getAndSet(Object o, long offset, Object newValue) 2850 // 2851 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2852 // This basic scheme here is the same as inline_unsafe_access, but 2853 // differs in enough details that combining them would make the code 2854 // overly confusing. (This is a true fact! I originally combined 2855 // them, but even I was confused by it!) As much code/comments as 2856 // possible are retained from inline_unsafe_access though to make 2857 // the correspondences clearer. - dl 2858 2859 if (callee()->is_static()) return false; // caller must have the capability! 2860 2861 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2862 decorators |= mo_decorator_for_access_kind(access_kind); 2863 2864 #ifndef PRODUCT 2865 BasicType rtype; 2866 { 2867 ResourceMark rm; 2868 // Check the signatures. 2869 ciSignature* sig = callee()->signature(); 2870 rtype = sig->return_type()->basic_type(); 2871 switch(kind) { 2872 case LS_get_add: 2873 case LS_get_set: { 2874 // Check the signatures. 2875 #ifdef ASSERT 2876 assert(rtype == type, "get and set must return the expected type"); 2877 assert(sig->count() == 3, "get and set has 3 arguments"); 2878 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2879 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2880 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2881 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation"); 2882 #endif // ASSERT 2883 break; 2884 } 2885 case LS_cmp_swap: 2886 case LS_cmp_swap_weak: { 2887 // Check the signatures. 2888 #ifdef ASSERT 2889 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2890 assert(sig->count() == 4, "CAS has 4 arguments"); 2891 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2892 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2893 #endif // ASSERT 2894 break; 2895 } 2896 case LS_cmp_exchange: { 2897 // Check the signatures. 2898 #ifdef ASSERT 2899 assert(rtype == type, "CAS must return the expected type"); 2900 assert(sig->count() == 4, "CAS has 4 arguments"); 2901 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2902 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2903 #endif // ASSERT 2904 break; 2905 } 2906 default: 2907 ShouldNotReachHere(); 2908 } 2909 } 2910 #endif //PRODUCT 2911 2912 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2913 2914 // Get arguments: 2915 Node* receiver = NULL; 2916 Node* base = NULL; 2917 Node* offset = NULL; 2918 Node* oldval = NULL; 2919 Node* newval = NULL; 2920 switch(kind) { 2921 case LS_cmp_swap: 2922 case LS_cmp_swap_weak: 2923 case LS_cmp_exchange: { 2924 const bool two_slot_type = type2size[type] == 2; 2925 receiver = argument(0); // type: oop 2926 base = argument(1); // type: oop 2927 offset = argument(2); // type: long 2928 oldval = argument(4); // type: oop, int, or long 2929 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2930 break; 2931 } 2932 case LS_get_add: 2933 case LS_get_set: { 2934 receiver = argument(0); // type: oop 2935 base = argument(1); // type: oop 2936 offset = argument(2); // type: long 2937 oldval = NULL; 2938 newval = argument(4); // type: oop, int, or long 2939 break; 2940 } 2941 default: 2942 ShouldNotReachHere(); 2943 } 2944 2945 // Build field offset expression. 2946 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2947 // to be plain byte offsets, which are also the same as those accepted 2948 // by oopDesc::field_addr. 2949 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2950 // 32-bit machines ignore the high half of long offsets 2951 offset = ConvL2X(offset); 2952 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false); 2953 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2954 2955 Compile::AliasType* alias_type = C->alias_type(adr_type); 2956 BasicType bt = alias_type->basic_type(); 2957 if (bt != T_ILLEGAL && 2958 (is_reference_type(bt) != (type == T_OBJECT))) { 2959 // Don't intrinsify mismatched object accesses. 2960 return false; 2961 } 2962 2963 // For CAS, unlike inline_unsafe_access, there seems no point in 2964 // trying to refine types. Just use the coarse types here. 2965 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2966 const Type *value_type = Type::get_const_basic_type(type); 2967 2968 switch (kind) { 2969 case LS_get_set: 2970 case LS_cmp_exchange: { 2971 if (type == T_OBJECT) { 2972 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2973 if (tjp != NULL) { 2974 value_type = tjp; 2975 } 2976 } 2977 break; 2978 } 2979 case LS_cmp_swap: 2980 case LS_cmp_swap_weak: 2981 case LS_get_add: 2982 break; 2983 default: 2984 ShouldNotReachHere(); 2985 } 2986 2987 // Null check receiver. 2988 receiver = null_check(receiver); 2989 if (stopped()) { 2990 return true; 2991 } 2992 2993 int alias_idx = C->get_alias_index(adr_type); 2994 2995 if (is_reference_type(type)) { 2996 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF; 2997 2998 // Transformation of a value which could be NULL pointer (CastPP #NULL) 2999 // could be delayed during Parse (for example, in adjust_map_after_if()). 3000 // Execute transformation here to avoid barrier generation in such case. 3001 if (_gvn.type(newval) == TypePtr::NULL_PTR) 3002 newval = _gvn.makecon(TypePtr::NULL_PTR); 3003 3004 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) { 3005 // Refine the value to a null constant, when it is known to be null 3006 oldval = _gvn.makecon(TypePtr::NULL_PTR); 3007 } 3008 } 3009 3010 Node* result = NULL; 3011 switch (kind) { 3012 case LS_cmp_exchange: { 3013 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx, 3014 oldval, newval, value_type, type, decorators); 3015 break; 3016 } 3017 case LS_cmp_swap_weak: 3018 decorators |= C2_WEAK_CMPXCHG; 3019 case LS_cmp_swap: { 3020 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx, 3021 oldval, newval, value_type, type, decorators); 3022 break; 3023 } 3024 case LS_get_set: { 3025 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx, 3026 newval, value_type, type, decorators); 3027 break; 3028 } 3029 case LS_get_add: { 3030 result = access_atomic_add_at(base, adr, adr_type, alias_idx, 3031 newval, value_type, type, decorators); 3032 break; 3033 } 3034 default: 3035 ShouldNotReachHere(); 3036 } 3037 3038 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 3039 set_result(result); 3040 return true; 3041 } 3042 3043 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 3044 // Regardless of form, don't allow previous ld/st to move down, 3045 // then issue acquire, release, or volatile mem_bar. 3046 insert_mem_bar(Op_MemBarCPUOrder); 3047 switch(id) { 3048 case vmIntrinsics::_loadFence: 3049 insert_mem_bar(Op_LoadFence); 3050 return true; 3051 case vmIntrinsics::_storeFence: 3052 insert_mem_bar(Op_StoreFence); 3053 return true; 3054 case vmIntrinsics::_fullFence: 3055 insert_mem_bar(Op_MemBarVolatile); 3056 return true; 3057 default: 3058 fatal_unexpected_iid(id); 3059 return false; 3060 } 3061 } 3062 3063 bool LibraryCallKit::inline_onspinwait() { 3064 insert_mem_bar(Op_OnSpinWait); 3065 return true; 3066 } 3067 3068 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 3069 if (!kls->is_Con()) { 3070 return true; 3071 } 3072 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr(); 3073 if (klsptr == NULL) { 3074 return true; 3075 } 3076 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass(); 3077 // don't need a guard for a klass that is already initialized 3078 return !ik->is_initialized(); 3079 } 3080 3081 //----------------------------inline_unsafe_writeback0------------------------- 3082 // public native void Unsafe.writeback0(long address) 3083 bool LibraryCallKit::inline_unsafe_writeback0() { 3084 if (!Matcher::has_match_rule(Op_CacheWB)) { 3085 return false; 3086 } 3087 #ifndef PRODUCT 3088 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync"); 3089 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync"); 3090 ciSignature* sig = callee()->signature(); 3091 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!"); 3092 #endif 3093 null_check_receiver(); // null-check, then ignore 3094 Node *addr = argument(1); 3095 addr = new CastX2PNode(addr); 3096 addr = _gvn.transform(addr); 3097 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr); 3098 flush = _gvn.transform(flush); 3099 set_memory(flush, TypeRawPtr::BOTTOM); 3100 return true; 3101 } 3102 3103 //----------------------------inline_unsafe_writeback0------------------------- 3104 // public native void Unsafe.writeback0(long address) 3105 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) { 3106 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) { 3107 return false; 3108 } 3109 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) { 3110 return false; 3111 } 3112 #ifndef PRODUCT 3113 assert(Matcher::has_match_rule(Op_CacheWB), 3114 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB" 3115 : "found match rule for CacheWBPostSync but not CacheWB")); 3116 3117 #endif 3118 null_check_receiver(); // null-check, then ignore 3119 Node *sync; 3120 if (is_pre) { 3121 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 3122 } else { 3123 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 3124 } 3125 sync = _gvn.transform(sync); 3126 set_memory(sync, TypeRawPtr::BOTTOM); 3127 return true; 3128 } 3129 3130 //----------------------------inline_unsafe_allocate--------------------------- 3131 // public native Object Unsafe.allocateInstance(Class<?> cls); 3132 bool LibraryCallKit::inline_unsafe_allocate() { 3133 if (callee()->is_static()) return false; // caller must have the capability! 3134 3135 null_check_receiver(); // null-check, then ignore 3136 Node* cls = null_check(argument(1)); 3137 if (stopped()) return true; 3138 3139 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 3140 kls = null_check(kls); 3141 if (stopped()) return true; // argument was like int.class 3142 3143 Node* test = NULL; 3144 if (LibraryCallKit::klass_needs_init_guard(kls)) { 3145 // Note: The argument might still be an illegal value like 3146 // Serializable.class or Object[].class. The runtime will handle it. 3147 // But we must make an explicit check for initialization. 3148 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 3149 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 3150 // can generate code to load it as unsigned byte. 3151 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered); 3152 Node* bits = intcon(InstanceKlass::fully_initialized); 3153 test = _gvn.transform(new SubINode(inst, bits)); 3154 // The 'test' is non-zero if we need to take a slow path. 3155 } 3156 3157 Node* obj = new_instance(kls, test); 3158 set_result(obj); 3159 return true; 3160 } 3161 3162 //------------------------inline_native_time_funcs-------------- 3163 // inline code for System.currentTimeMillis() and System.nanoTime() 3164 // these have the same type and signature 3165 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 3166 const TypeFunc* tf = OptoRuntime::void_long_Type(); 3167 const TypePtr* no_memory_effects = NULL; 3168 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 3169 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 3170 #ifdef ASSERT 3171 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 3172 assert(value_top == top(), "second value must be top"); 3173 #endif 3174 set_result(value); 3175 return true; 3176 } 3177 3178 #ifdef JFR_HAVE_INTRINSICS 3179 3180 /* 3181 * oop -> myklass 3182 * myklass->trace_id |= USED 3183 * return myklass->trace_id & ~0x3 3184 */ 3185 bool LibraryCallKit::inline_native_classID() { 3186 Node* cls = null_check(argument(0), T_OBJECT); 3187 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 3188 kls = null_check(kls, T_OBJECT); 3189 3190 ByteSize offset = KLASS_TRACE_ID_OFFSET; 3191 Node* insp = basic_plus_adr(kls, in_bytes(offset)); 3192 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered); 3193 3194 Node* clsused = longcon(0x01l); // set the class bit 3195 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused)); 3196 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr(); 3197 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered); 3198 3199 #ifdef TRACE_ID_META_BITS 3200 Node* mbits = longcon(~TRACE_ID_META_BITS); 3201 tvalue = _gvn.transform(new AndLNode(tvalue, mbits)); 3202 #endif 3203 #ifdef TRACE_ID_SHIFT 3204 Node* cbits = intcon(TRACE_ID_SHIFT); 3205 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits)); 3206 #endif 3207 3208 set_result(tvalue); 3209 return true; 3210 3211 } 3212 3213 bool LibraryCallKit::inline_native_getEventWriter() { 3214 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3215 3216 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, 3217 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)); 3218 3219 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered); 3220 3221 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) ); 3222 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) ); 3223 3224 IfNode* iff_jobj_null = 3225 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN); 3226 3227 enum { _normal_path = 1, 3228 _null_path = 2, 3229 PATH_LIMIT }; 3230 3231 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3232 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM); 3233 3234 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null)); 3235 result_rgn->init_req(_null_path, jobj_is_null); 3236 result_val->init_req(_null_path, null()); 3237 3238 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null)); 3239 set_control(jobj_is_not_null); 3240 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT, 3241 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD); 3242 result_rgn->init_req(_normal_path, control()); 3243 result_val->init_req(_normal_path, res); 3244 3245 set_result(result_rgn, result_val); 3246 3247 return true; 3248 } 3249 3250 #endif // JFR_HAVE_INTRINSICS 3251 3252 //------------------------inline_native_currentThread------------------ 3253 bool LibraryCallKit::inline_native_currentThread() { 3254 Node* junk = NULL; 3255 set_result(generate_current_thread(junk)); 3256 return true; 3257 } 3258 3259 //-----------------------load_klass_from_mirror_common------------------------- 3260 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3261 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3262 // and branch to the given path on the region. 3263 // If never_see_null, take an uncommon trap on null, so we can optimistically 3264 // compile for the non-null case. 3265 // If the region is NULL, force never_see_null = true. 3266 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3267 bool never_see_null, 3268 RegionNode* region, 3269 int null_path, 3270 int offset) { 3271 if (region == NULL) never_see_null = true; 3272 Node* p = basic_plus_adr(mirror, offset); 3273 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3274 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3275 Node* null_ctl = top(); 3276 kls = null_check_oop(kls, &null_ctl, never_see_null); 3277 if (region != NULL) { 3278 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3279 region->init_req(null_path, null_ctl); 3280 } else { 3281 assert(null_ctl == top(), "no loose ends"); 3282 } 3283 return kls; 3284 } 3285 3286 //--------------------(inline_native_Class_query helpers)--------------------- 3287 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER. 3288 // Fall through if (mods & mask) == bits, take the guard otherwise. 3289 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 3290 // Branch around if the given klass has the given modifier bit set. 3291 // Like generate_guard, adds a new path onto the region. 3292 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3293 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered); 3294 Node* mask = intcon(modifier_mask); 3295 Node* bits = intcon(modifier_bits); 3296 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3297 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3298 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3299 return generate_fair_guard(bol, region); 3300 } 3301 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3302 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region); 3303 } 3304 3305 //-------------------------inline_native_Class_query------------------- 3306 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 3307 const Type* return_type = TypeInt::BOOL; 3308 Node* prim_return_value = top(); // what happens if it's a primitive class? 3309 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3310 bool expect_prim = false; // most of these guys expect to work on refs 3311 3312 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 3313 3314 Node* mirror = argument(0); 3315 Node* obj = top(); 3316 3317 switch (id) { 3318 case vmIntrinsics::_isInstance: 3319 // nothing is an instance of a primitive type 3320 prim_return_value = intcon(0); 3321 obj = argument(1); 3322 break; 3323 case vmIntrinsics::_getModifiers: 3324 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3325 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line"); 3326 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin); 3327 break; 3328 case vmIntrinsics::_isInterface: 3329 prim_return_value = intcon(0); 3330 break; 3331 case vmIntrinsics::_isArray: 3332 prim_return_value = intcon(0); 3333 expect_prim = true; // cf. ObjectStreamClass.getClassSignature 3334 break; 3335 case vmIntrinsics::_isPrimitive: 3336 prim_return_value = intcon(1); 3337 expect_prim = true; // obviously 3338 break; 3339 case vmIntrinsics::_getSuperclass: 3340 prim_return_value = null(); 3341 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 3342 break; 3343 case vmIntrinsics::_getClassAccessFlags: 3344 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3345 return_type = TypeInt::INT; // not bool! 6297094 3346 break; 3347 default: 3348 fatal_unexpected_iid(id); 3349 break; 3350 } 3351 3352 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3353 if (mirror_con == NULL) return false; // cannot happen? 3354 3355 #ifndef PRODUCT 3356 if (C->print_intrinsics() || C->print_inlining()) { 3357 ciType* k = mirror_con->java_mirror_type(); 3358 if (k) { 3359 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 3360 k->print_name(); 3361 tty->cr(); 3362 } 3363 } 3364 #endif 3365 3366 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 3367 RegionNode* region = new RegionNode(PATH_LIMIT); 3368 record_for_igvn(region); 3369 PhiNode* phi = new PhiNode(region, return_type); 3370 3371 // The mirror will never be null of Reflection.getClassAccessFlags, however 3372 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 3373 // if it is. See bug 4774291. 3374 3375 // For Reflection.getClassAccessFlags(), the null check occurs in 3376 // the wrong place; see inline_unsafe_access(), above, for a similar 3377 // situation. 3378 mirror = null_check(mirror); 3379 // If mirror or obj is dead, only null-path is taken. 3380 if (stopped()) return true; 3381 3382 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 3383 3384 // Now load the mirror's klass metaobject, and null-check it. 3385 // Side-effects region with the control path if the klass is null. 3386 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 3387 // If kls is null, we have a primitive mirror. 3388 phi->init_req(_prim_path, prim_return_value); 3389 if (stopped()) { set_result(region, phi); return true; } 3390 bool safe_for_replace = (region->in(_prim_path) == top()); 3391 3392 Node* p; // handy temp 3393 Node* null_ctl; 3394 3395 // Now that we have the non-null klass, we can perform the real query. 3396 // For constant classes, the query will constant-fold in LoadNode::Value. 3397 Node* query_value = top(); 3398 switch (id) { 3399 case vmIntrinsics::_isInstance: 3400 // nothing is an instance of a primitive type 3401 query_value = gen_instanceof(obj, kls, safe_for_replace); 3402 break; 3403 3404 case vmIntrinsics::_getModifiers: 3405 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset())); 3406 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3407 break; 3408 3409 case vmIntrinsics::_isInterface: 3410 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3411 if (generate_interface_guard(kls, region) != NULL) 3412 // A guard was added. If the guard is taken, it was an interface. 3413 phi->add_req(intcon(1)); 3414 // If we fall through, it's a plain class. 3415 query_value = intcon(0); 3416 break; 3417 3418 case vmIntrinsics::_isArray: 3419 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.) 3420 if (generate_array_guard(kls, region) != NULL) 3421 // A guard was added. If the guard is taken, it was an array. 3422 phi->add_req(intcon(1)); 3423 // If we fall through, it's a plain class. 3424 query_value = intcon(0); 3425 break; 3426 3427 case vmIntrinsics::_isPrimitive: 3428 query_value = intcon(0); // "normal" path produces false 3429 break; 3430 3431 case vmIntrinsics::_getSuperclass: 3432 // The rules here are somewhat unfortunate, but we can still do better 3433 // with random logic than with a JNI call. 3434 // Interfaces store null or Object as _super, but must report null. 3435 // Arrays store an intermediate super as _super, but must report Object. 3436 // Other types can report the actual _super. 3437 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3438 if (generate_interface_guard(kls, region) != NULL) 3439 // A guard was added. If the guard is taken, it was an interface. 3440 phi->add_req(null()); 3441 if (generate_array_guard(kls, region) != NULL) 3442 // A guard was added. If the guard is taken, it was an array. 3443 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 3444 // If we fall through, it's a plain class. Get its _super. 3445 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 3446 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL)); 3447 null_ctl = top(); 3448 kls = null_check_oop(kls, &null_ctl); 3449 if (null_ctl != top()) { 3450 // If the guard is taken, Object.superClass is null (both klass and mirror). 3451 region->add_req(null_ctl); 3452 phi ->add_req(null()); 3453 } 3454 if (!stopped()) { 3455 query_value = load_mirror_from_klass(kls); 3456 } 3457 break; 3458 3459 case vmIntrinsics::_getClassAccessFlags: 3460 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3461 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3462 break; 3463 3464 default: 3465 fatal_unexpected_iid(id); 3466 break; 3467 } 3468 3469 // Fall-through is the normal case of a query to a real class. 3470 phi->init_req(1, query_value); 3471 region->init_req(1, control()); 3472 3473 C->set_has_split_ifs(true); // Has chance for split-if optimization 3474 set_result(region, phi); 3475 return true; 3476 } 3477 3478 //-------------------------inline_value_Class_conversion------------------- 3479 // public Class<T> java.lang.Class.asPrimaryType(); 3480 // public Class<T> java.lang.Class.asIndirectType() 3481 bool LibraryCallKit::inline_value_Class_conversion(vmIntrinsics::ID id) { 3482 Node* mirror = argument(0); // Receiver Class 3483 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3484 if (mirror_con == NULL) { 3485 return false; 3486 } 3487 3488 bool is_indirect_type = true; 3489 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type); 3490 if (tm != NULL) { 3491 Node* result = mirror; 3492 if (tm->is_valuetype()) { 3493 if (id == vmIntrinsics::_asPrimaryType && is_indirect_type) { 3494 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->inline_mirror_instance())); 3495 } else if (id == vmIntrinsics::_asIndirectType && !is_indirect_type) { 3496 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->indirect_mirror_instance())); 3497 } 3498 } 3499 set_result(result); 3500 return true; 3501 } 3502 return false; 3503 } 3504 3505 //-------------------------inline_Class_cast------------------- 3506 bool LibraryCallKit::inline_Class_cast() { 3507 Node* mirror = argument(0); // Class 3508 Node* obj = argument(1); 3509 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3510 if (mirror_con == NULL) { 3511 return false; // dead path (mirror->is_top()). 3512 } 3513 if (obj == NULL || obj->is_top()) { 3514 return false; // dead path 3515 } 3516 3517 ciKlass* obj_klass = NULL; 3518 if (obj->is_ValueType()) { 3519 obj_klass = _gvn.type(obj)->value_klass(); 3520 } else { 3521 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr(); 3522 if (tp != NULL) { 3523 obj_klass = tp->klass(); 3524 } 3525 } 3526 3527 // First, see if Class.cast() can be folded statically. 3528 // java_mirror_type() returns non-null for compile-time Class constants. 3529 bool is_indirect_type = true; 3530 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type); 3531 if (!obj->is_ValueType() && !is_indirect_type) { 3532 obj = null_check(obj); 3533 if (stopped()) { 3534 return true; 3535 } 3536 } 3537 if (tm != NULL && tm->is_klass() && obj_klass != NULL) { 3538 if (!obj_klass->is_loaded()) { 3539 // Don't use intrinsic when class is not loaded. 3540 return false; 3541 } else { 3542 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass); 3543 if (static_res == Compile::SSC_always_true) { 3544 // isInstance() is true - fold the code. 3545 set_result(obj); 3546 return true; 3547 } else if (static_res == Compile::SSC_always_false) { 3548 // Don't use intrinsic, have to throw ClassCastException. 3549 // If the reference is null, the non-intrinsic bytecode will 3550 // be optimized appropriately. 3551 return false; 3552 } 3553 } 3554 } 3555 3556 // Bailout intrinsic and do normal inlining if exception path is frequent. 3557 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 3558 return false; 3559 } 3560 3561 // Generate dynamic checks. 3562 // Class.cast() is java implementation of _checkcast bytecode. 3563 // Do checkcast (Parse::do_checkcast()) optimizations here. 3564 3565 mirror = null_check(mirror); 3566 // If mirror is dead, only null-path is taken. 3567 if (stopped()) { 3568 return true; 3569 } 3570 3571 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive). 3572 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT }; 3573 RegionNode* region = new RegionNode(PATH_LIMIT); 3574 record_for_igvn(region); 3575 3576 // Now load the mirror's klass metaobject, and null-check it. 3577 // If kls is null, we have a primitive mirror and 3578 // nothing is an instance of a primitive type. 3579 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 3580 3581 Node* res = top(); 3582 if (!stopped()) { 3583 if (EnableValhalla && !obj->is_ValueType() && is_indirect_type) { 3584 // Check if (mirror == inline_mirror && obj == null) 3585 Node* is_val_mirror = generate_fair_guard(is_value_mirror(mirror), NULL); 3586 if (is_val_mirror != NULL) { 3587 RegionNode* r = new RegionNode(3); 3588 record_for_igvn(r); 3589 r->init_req(1, control()); 3590 3591 // Casting to .val, check for null 3592 set_control(is_val_mirror); 3593 Node *null_ctr = top(); 3594 null_check_oop(obj, &null_ctr); 3595 region->init_req(_npe_path, null_ctr); 3596 r->init_req(2, control()); 3597 3598 set_control(_gvn.transform(r)); 3599 } 3600 } 3601 3602 Node* bad_type_ctrl = top(); 3603 // Do checkcast optimizations. 3604 res = gen_checkcast(obj, kls, &bad_type_ctrl); 3605 region->init_req(_bad_type_path, bad_type_ctrl); 3606 } 3607 if (region->in(_prim_path) != top() || 3608 region->in(_bad_type_path) != top() || 3609 region->in(_npe_path) != top()) { 3610 // Let Interpreter throw ClassCastException. 3611 PreserveJVMState pjvms(this); 3612 set_control(_gvn.transform(region)); 3613 uncommon_trap(Deoptimization::Reason_intrinsic, 3614 Deoptimization::Action_maybe_recompile); 3615 } 3616 if (!stopped()) { 3617 set_result(res); 3618 } 3619 return true; 3620 } 3621 3622 3623 //--------------------------inline_native_subtype_check------------------------ 3624 // This intrinsic takes the JNI calls out of the heart of 3625 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 3626 bool LibraryCallKit::inline_native_subtype_check() { 3627 // Pull both arguments off the stack. 3628 Node* args[2]; // two java.lang.Class mirrors: superc, subc 3629 args[0] = argument(0); 3630 args[1] = argument(1); 3631 Node* klasses[2]; // corresponding Klasses: superk, subk 3632 klasses[0] = klasses[1] = top(); 3633 3634 enum { 3635 // A full decision tree on {superc is prim, subc is prim}: 3636 _prim_0_path = 1, // {P,N} => false 3637 // {P,P} & superc!=subc => false 3638 _prim_same_path, // {P,P} & superc==subc => true 3639 _prim_1_path, // {N,P} => false 3640 _ref_subtype_path, // {N,N} & subtype check wins => true 3641 _both_ref_path, // {N,N} & subtype check loses => false 3642 PATH_LIMIT 3643 }; 3644 3645 RegionNode* region = new RegionNode(PATH_LIMIT); 3646 RegionNode* prim_region = new RegionNode(2); 3647 Node* phi = new PhiNode(region, TypeInt::BOOL); 3648 record_for_igvn(region); 3649 record_for_igvn(prim_region); 3650 3651 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 3652 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3653 int class_klass_offset = java_lang_Class::klass_offset_in_bytes(); 3654 3655 // First null-check both mirrors and load each mirror's klass metaobject. 3656 int which_arg; 3657 for (which_arg = 0; which_arg <= 1; which_arg++) { 3658 Node* arg = args[which_arg]; 3659 arg = null_check(arg); 3660 if (stopped()) break; 3661 args[which_arg] = arg; 3662 3663 Node* p = basic_plus_adr(arg, class_klass_offset); 3664 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type); 3665 klasses[which_arg] = _gvn.transform(kls); 3666 } 3667 3668 // Resolve oops to stable for CmpP below. 3669 args[0] = access_resolve(args[0], 0); 3670 args[1] = access_resolve(args[1], 0); 3671 3672 // Having loaded both klasses, test each for null. 3673 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3674 for (which_arg = 0; which_arg <= 1; which_arg++) { 3675 Node* kls = klasses[which_arg]; 3676 Node* null_ctl = top(); 3677 kls = null_check_oop(kls, &null_ctl, never_see_null); 3678 if (which_arg == 0) { 3679 prim_region->init_req(1, null_ctl); 3680 } else { 3681 region->init_req(_prim_1_path, null_ctl); 3682 } 3683 if (stopped()) break; 3684 klasses[which_arg] = kls; 3685 } 3686 3687 if (!stopped()) { 3688 // now we have two reference types, in klasses[0..1] 3689 Node* subk = klasses[1]; // the argument to isAssignableFrom 3690 Node* superk = klasses[0]; // the receiver 3691 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 3692 // If superc is a value mirror, we also need to check if superc == subc because 3693 // V? is not a subtype of V but due to subk == superk the subtype check will pass. 3694 generate_fair_guard(is_value_mirror(args[0]), prim_region); 3695 // now we have a successful reference subtype check 3696 region->set_req(_ref_subtype_path, control()); 3697 } 3698 3699 // If both operands are primitive (both klasses null), then 3700 // we must return true when they are identical primitives. 3701 // It is convenient to test this after the first null klass check. 3702 // This path is also used if superc is a value mirror. 3703 set_control(_gvn.transform(prim_region)); 3704 if (!stopped()) { 3705 // Since superc is primitive, make a guard for the superc==subc case. 3706 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 3707 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 3708 generate_fair_guard(bol_eq, region); 3709 if (region->req() == PATH_LIMIT+1) { 3710 // A guard was added. If the added guard is taken, superc==subc. 3711 region->swap_edges(PATH_LIMIT, _prim_same_path); 3712 region->del_req(PATH_LIMIT); 3713 } 3714 region->set_req(_prim_0_path, control()); // Not equal after all. 3715 } 3716 3717 // these are the only paths that produce 'true': 3718 phi->set_req(_prim_same_path, intcon(1)); 3719 phi->set_req(_ref_subtype_path, intcon(1)); 3720 3721 // pull together the cases: 3722 assert(region->req() == PATH_LIMIT, "sane region"); 3723 for (uint i = 1; i < region->req(); i++) { 3724 Node* ctl = region->in(i); 3725 if (ctl == NULL || ctl == top()) { 3726 region->set_req(i, top()); 3727 phi ->set_req(i, top()); 3728 } else if (phi->in(i) == NULL) { 3729 phi->set_req(i, intcon(0)); // all other paths produce 'false' 3730 } 3731 } 3732 3733 set_control(_gvn.transform(region)); 3734 set_result(_gvn.transform(phi)); 3735 return true; 3736 } 3737 3738 //---------------------generate_array_guard_common------------------------ 3739 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) { 3740 3741 if (stopped()) { 3742 return NULL; 3743 } 3744 3745 // Like generate_guard, adds a new path onto the region. 3746 jint layout_con = 0; 3747 Node* layout_val = get_layout_helper(kls, layout_con); 3748 if (layout_val == NULL) { 3749 bool query = 0; 3750 switch(kind) { 3751 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break; 3752 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break; 3753 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break; 3754 case ValueArray: query = Klass::layout_helper_is_valueArray(layout_con); break; 3755 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break; 3756 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break; 3757 default: 3758 ShouldNotReachHere(); 3759 } 3760 if (!query) { 3761 return NULL; // never a branch 3762 } else { // always a branch 3763 Node* always_branch = control(); 3764 if (region != NULL) 3765 region->add_req(always_branch); 3766 set_control(top()); 3767 return always_branch; 3768 } 3769 } 3770 unsigned int value = 0; 3771 BoolTest::mask btest = BoolTest::illegal; 3772 switch(kind) { 3773 case ObjectArray: 3774 case NonObjectArray: { 3775 value = Klass::_lh_array_tag_obj_value; 3776 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 3777 btest = kind == ObjectArray ? BoolTest::eq : BoolTest::ne; 3778 break; 3779 } 3780 case TypeArray: { 3781 value = Klass::_lh_array_tag_type_value; 3782 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 3783 btest = BoolTest::eq; 3784 break; 3785 } 3786 case ValueArray: { 3787 value = Klass::_lh_array_tag_vt_value; 3788 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 3789 btest = BoolTest::eq; 3790 break; 3791 } 3792 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break; 3793 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break; 3794 default: 3795 ShouldNotReachHere(); 3796 } 3797 // Now test the correct condition. 3798 jint nval = (jint)value; 3799 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 3800 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 3801 return generate_fair_guard(bol, region); 3802 } 3803 3804 3805 //-----------------------inline_native_newArray-------------------------- 3806 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length); 3807 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size); 3808 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) { 3809 Node* mirror; 3810 Node* count_val; 3811 if (uninitialized) { 3812 mirror = argument(1); 3813 count_val = argument(2); 3814 } else { 3815 mirror = argument(0); 3816 count_val = argument(1); 3817 } 3818 3819 mirror = null_check(mirror); 3820 // If mirror or obj is dead, only null-path is taken. 3821 if (stopped()) return true; 3822 3823 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 3824 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 3825 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 3826 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 3827 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 3828 3829 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3830 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 3831 result_reg, _slow_path); 3832 Node* normal_ctl = control(); 3833 Node* no_array_ctl = result_reg->in(_slow_path); 3834 3835 // Generate code for the slow case. We make a call to newArray(). 3836 set_control(no_array_ctl); 3837 if (!stopped()) { 3838 // Either the input type is void.class, or else the 3839 // array klass has not yet been cached. Either the 3840 // ensuing call will throw an exception, or else it 3841 // will cache the array klass for next time. 3842 PreserveJVMState pjvms(this); 3843 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray); 3844 Node* slow_result = set_results_for_java_call(slow_call); 3845 // this->control() comes from set_results_for_java_call 3846 result_reg->set_req(_slow_path, control()); 3847 result_val->set_req(_slow_path, slow_result); 3848 result_io ->set_req(_slow_path, i_o()); 3849 result_mem->set_req(_slow_path, reset_memory()); 3850 } 3851 3852 set_control(normal_ctl); 3853 if (!stopped()) { 3854 // Normal case: The array type has been cached in the java.lang.Class. 3855 // The following call works fine even if the array type is polymorphic. 3856 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3857 Node* obj = new_array(klass_node, count_val, 0, NULL, false, mirror); // no arguments to push 3858 result_reg->init_req(_normal_path, control()); 3859 result_val->init_req(_normal_path, obj); 3860 result_io ->init_req(_normal_path, i_o()); 3861 result_mem->init_req(_normal_path, reset_memory()); 3862 3863 if (uninitialized) { 3864 // Mark the allocation so that zeroing is skipped 3865 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn); 3866 alloc->maybe_set_complete(&_gvn); 3867 } 3868 } 3869 3870 // Return the combined state. 3871 set_i_o( _gvn.transform(result_io) ); 3872 set_all_memory( _gvn.transform(result_mem)); 3873 3874 C->set_has_split_ifs(true); // Has chance for split-if optimization 3875 set_result(result_reg, result_val); 3876 return true; 3877 } 3878 3879 //----------------------inline_native_getLength-------------------------- 3880 // public static native int java.lang.reflect.Array.getLength(Object array); 3881 bool LibraryCallKit::inline_native_getLength() { 3882 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3883 3884 Node* array = null_check(argument(0)); 3885 // If array is dead, only null-path is taken. 3886 if (stopped()) return true; 3887 3888 // Deoptimize if it is a non-array. 3889 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL); 3890 3891 if (non_array != NULL) { 3892 PreserveJVMState pjvms(this); 3893 set_control(non_array); 3894 uncommon_trap(Deoptimization::Reason_intrinsic, 3895 Deoptimization::Action_maybe_recompile); 3896 } 3897 3898 // If control is dead, only non-array-path is taken. 3899 if (stopped()) return true; 3900 3901 // The works fine even if the array type is polymorphic. 3902 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3903 Node* result = load_array_length(array); 3904 3905 C->set_has_split_ifs(true); // Has chance for split-if optimization 3906 set_result(result); 3907 return true; 3908 } 3909 3910 //------------------------inline_array_copyOf---------------------------- 3911 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 3912 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 3913 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 3914 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3915 3916 // Get the arguments. 3917 Node* original = argument(0); 3918 Node* start = is_copyOfRange? argument(1): intcon(0); 3919 Node* end = is_copyOfRange? argument(2): argument(1); 3920 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 3921 3922 const TypeAryPtr* original_t = _gvn.type(original)->isa_aryptr(); 3923 const TypeInstPtr* mirror_t = _gvn.type(array_type_mirror)->isa_instptr(); 3924 if (EnableValhalla && ValueArrayFlatten && 3925 (original_t == NULL || mirror_t == NULL || 3926 (mirror_t->java_mirror_type() == NULL && 3927 (original_t->elem()->isa_valuetype() || 3928 (original_t->elem()->make_oopptr() != NULL && 3929 original_t->elem()->make_oopptr()->can_be_value_type()))))) { 3930 // We need to know statically if the copy is to a flattened array 3931 // or not but can't tell. 3932 return false; 3933 } 3934 3935 Node* newcopy = NULL; 3936 3937 // Set the original stack and the reexecute bit for the interpreter to reexecute 3938 // the bytecode that invokes Arrays.copyOf if deoptimization happens. 3939 { PreserveReexecuteState preexecs(this); 3940 jvms()->set_should_reexecute(true); 3941 3942 array_type_mirror = null_check(array_type_mirror); 3943 original = null_check(original); 3944 3945 // Check if a null path was taken unconditionally. 3946 if (stopped()) return true; 3947 3948 Node* orig_length = load_array_length(original); 3949 3950 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0); 3951 klass_node = null_check(klass_node); 3952 3953 RegionNode* bailout = new RegionNode(1); 3954 record_for_igvn(bailout); 3955 3956 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc. 3957 // Bail out if that is so. 3958 // Value type array may have object field that would require a 3959 // write barrier. Conservatively, go to slow path. 3960 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 3961 Node* not_objArray = !bs->array_copy_requires_gc_barriers(false, T_OBJECT, false, BarrierSetC2::Parsing) ? 3962 generate_typeArray_guard(klass_node, bailout) : generate_non_objArray_guard(klass_node, bailout); 3963 if (not_objArray != NULL) { 3964 // Improve the klass node's type from the new optimistic assumption: 3965 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 3966 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), false); 3967 Node* cast = new CastPPNode(klass_node, akls); 3968 cast->init_req(0, control()); 3969 klass_node = _gvn.transform(cast); 3970 } 3971 3972 Node* original_kls = load_object_klass(original); 3973 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 3974 // loads/stores but it is legal only if we're sure the 3975 // Arrays.copyOf would succeed. So we need all input arguments 3976 // to the copyOf to be validated, including that the copy to the 3977 // new array won't trigger an ArrayStoreException. That subtype 3978 // check can be optimized if we know something on the type of 3979 // the input array from type speculation. 3980 if (_gvn.type(klass_node)->singleton() && !stopped()) { 3981 ciKlass* subk = _gvn.type(original_kls)->is_klassptr()->klass(); 3982 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass(); 3983 3984 int test = C->static_subtype_check(superk, subk); 3985 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 3986 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 3987 if (t_original->speculative_type() != NULL) { 3988 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 3989 original_kls = load_object_klass(original); 3990 } 3991 } 3992 } 3993 3994 if (ValueArrayFlatten) { 3995 // Either both or neither new array klass and original array 3996 // klass must be flattened 3997 Node* is_flat = generate_valueArray_guard(klass_node, NULL); 3998 if (!original_t->is_not_flat()) { 3999 generate_valueArray_guard(original_kls, bailout); 4000 } 4001 if (is_flat != NULL) { 4002 RegionNode* r = new RegionNode(2); 4003 record_for_igvn(r); 4004 r->init_req(1, control()); 4005 set_control(is_flat); 4006 if (!original_t->is_not_flat()) { 4007 generate_valueArray_guard(original_kls, r); 4008 } 4009 bailout->add_req(control()); 4010 set_control(_gvn.transform(r)); 4011 } 4012 } 4013 4014 // Bail out if either start or end is negative. 4015 generate_negative_guard(start, bailout, &start); 4016 generate_negative_guard(end, bailout, &end); 4017 4018 Node* length = end; 4019 if (_gvn.type(start) != TypeInt::ZERO) { 4020 length = _gvn.transform(new SubINode(end, start)); 4021 } 4022 4023 // Bail out if length is negative. 4024 // Without this the new_array would throw 4025 // NegativeArraySizeException but IllegalArgumentException is what 4026 // should be thrown 4027 generate_negative_guard(length, bailout, &length); 4028 4029 if (bailout->req() > 1) { 4030 PreserveJVMState pjvms(this); 4031 set_control(_gvn.transform(bailout)); 4032 uncommon_trap(Deoptimization::Reason_intrinsic, 4033 Deoptimization::Action_maybe_recompile); 4034 } 4035 4036 if (!stopped()) { 4037 // How many elements will we copy from the original? 4038 // The answer is MinI(orig_length - start, length). 4039 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 4040 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length); 4041 4042 original = access_resolve(original, ACCESS_READ); 4043 4044 // Generate a direct call to the right arraycopy function(s). 4045 // We know the copy is disjoint but we might not know if the 4046 // oop stores need checking. 4047 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 4048 // This will fail a store-check if x contains any non-nulls. 4049 4050 bool validated = false; 4051 // Reason_class_check rather than Reason_intrinsic because we 4052 // want to intrinsify even if this traps. 4053 if (!too_many_traps(Deoptimization::Reason_class_check)) { 4054 Node* not_subtype_ctrl = gen_subtype_check(original_kls, 4055 klass_node); 4056 4057 if (not_subtype_ctrl != top()) { 4058 PreserveJVMState pjvms(this); 4059 set_control(not_subtype_ctrl); 4060 uncommon_trap(Deoptimization::Reason_class_check, 4061 Deoptimization::Action_make_not_entrant); 4062 assert(stopped(), "Should be stopped"); 4063 } 4064 validated = true; 4065 } 4066 4067 if (!stopped()) { 4068 // Load element mirror 4069 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes()); 4070 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP); 4071 4072 newcopy = new_array(klass_node, length, 0, NULL, false, elem_mirror); 4073 4074 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false, 4075 original_kls, klass_node); 4076 if (!is_copyOfRange) { 4077 ac->set_copyof(validated); 4078 } else { 4079 ac->set_copyofrange(validated); 4080 } 4081 Node* n = _gvn.transform(ac); 4082 if (n == ac) { 4083 ac->connect_outputs(this); 4084 } else { 4085 assert(validated, "shouldn't transform if all arguments not validated"); 4086 set_all_memory(n); 4087 } 4088 } 4089 } 4090 } // original reexecute is set back here 4091 4092 C->set_has_split_ifs(true); // Has chance for split-if optimization 4093 if (!stopped()) { 4094 set_result(newcopy); 4095 } 4096 return true; 4097 } 4098 4099 4100 //----------------------generate_virtual_guard--------------------------- 4101 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 4102 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 4103 RegionNode* slow_region) { 4104 ciMethod* method = callee(); 4105 int vtable_index = method->vtable_index(); 4106 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4107 "bad index %d", vtable_index); 4108 // Get the Method* out of the appropriate vtable entry. 4109 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 4110 vtable_index*vtableEntry::size_in_bytes() + 4111 vtableEntry::method_offset_in_bytes(); 4112 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 4113 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 4114 4115 // Compare the target method with the expected method (e.g., Object.hashCode). 4116 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 4117 4118 Node* native_call = makecon(native_call_addr); 4119 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 4120 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 4121 4122 return generate_slow_guard(test_native, slow_region); 4123 } 4124 4125 //-----------------------generate_method_call---------------------------- 4126 // Use generate_method_call to make a slow-call to the real 4127 // method if the fast path fails. An alternative would be to 4128 // use a stub like OptoRuntime::slow_arraycopy_Java. 4129 // This only works for expanding the current library call, 4130 // not another intrinsic. (E.g., don't use this for making an 4131 // arraycopy call inside of the copyOf intrinsic.) 4132 CallJavaNode* 4133 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) { 4134 // When compiling the intrinsic method itself, do not use this technique. 4135 guarantee(callee() != C->method(), "cannot make slow-call to self"); 4136 4137 ciMethod* method = callee(); 4138 // ensure the JVMS we have will be correct for this call 4139 guarantee(method_id == method->intrinsic_id(), "must match"); 4140 4141 const TypeFunc* tf = TypeFunc::make(method); 4142 CallJavaNode* slow_call; 4143 if (is_static) { 4144 assert(!is_virtual, ""); 4145 slow_call = new CallStaticJavaNode(C, tf, 4146 SharedRuntime::get_resolve_static_call_stub(), 4147 method, bci()); 4148 } else if (is_virtual) { 4149 null_check_receiver(); 4150 int vtable_index = Method::invalid_vtable_index; 4151 if (UseInlineCaches) { 4152 // Suppress the vtable call 4153 } else { 4154 // hashCode and clone are not a miranda methods, 4155 // so the vtable index is fixed. 4156 // No need to use the linkResolver to get it. 4157 vtable_index = method->vtable_index(); 4158 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4159 "bad index %d", vtable_index); 4160 } 4161 slow_call = new CallDynamicJavaNode(tf, 4162 SharedRuntime::get_resolve_virtual_call_stub(), 4163 method, vtable_index, bci()); 4164 } else { // neither virtual nor static: opt_virtual 4165 null_check_receiver(); 4166 slow_call = new CallStaticJavaNode(C, tf, 4167 SharedRuntime::get_resolve_opt_virtual_call_stub(), 4168 method, bci()); 4169 slow_call->set_optimized_virtual(true); 4170 } 4171 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) { 4172 // To be able to issue a direct call (optimized virtual or virtual) 4173 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information 4174 // about the method being invoked should be attached to the call site to 4175 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C). 4176 slow_call->set_override_symbolic_info(true); 4177 } 4178 set_arguments_for_java_call(slow_call); 4179 set_edges_for_java_call(slow_call); 4180 return slow_call; 4181 } 4182 4183 4184 /** 4185 * Build special case code for calls to hashCode on an object. This call may 4186 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 4187 * slightly different code. 4188 */ 4189 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 4190 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 4191 assert(!(is_virtual && is_static), "either virtual, special, or static"); 4192 4193 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 4194 4195 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4196 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 4197 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4198 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4199 Node* obj = argument(0); 4200 4201 if (obj->is_ValueType() || gvn().type(obj)->is_valuetypeptr()) { 4202 return false; 4203 } 4204 4205 if (!is_static) { 4206 // Check for hashing null object 4207 obj = null_check_receiver(); 4208 if (stopped()) return true; // unconditionally null 4209 result_reg->init_req(_null_path, top()); 4210 result_val->init_req(_null_path, top()); 4211 } else { 4212 // Do a null check, and return zero if null. 4213 // System.identityHashCode(null) == 0 4214 Node* null_ctl = top(); 4215 obj = null_check_oop(obj, &null_ctl); 4216 result_reg->init_req(_null_path, null_ctl); 4217 result_val->init_req(_null_path, _gvn.intcon(0)); 4218 } 4219 4220 // Unconditionally null? Then return right away. 4221 if (stopped()) { 4222 set_control( result_reg->in(_null_path)); 4223 if (!stopped()) 4224 set_result(result_val->in(_null_path)); 4225 return true; 4226 } 4227 4228 // We only go to the fast case code if we pass a number of guards. The 4229 // paths which do not pass are accumulated in the slow_region. 4230 RegionNode* slow_region = new RegionNode(1); 4231 record_for_igvn(slow_region); 4232 4233 // If this is a virtual call, we generate a funny guard. We pull out 4234 // the vtable entry corresponding to hashCode() from the target object. 4235 // If the target method which we are calling happens to be the native 4236 // Object hashCode() method, we pass the guard. We do not need this 4237 // guard for non-virtual calls -- the caller is known to be the native 4238 // Object hashCode(). 4239 if (is_virtual) { 4240 // After null check, get the object's klass. 4241 Node* obj_klass = load_object_klass(obj); 4242 generate_virtual_guard(obj_klass, slow_region); 4243 } 4244 4245 // Get the header out of the object, use LoadMarkNode when available 4246 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 4247 // The control of the load must be NULL. Otherwise, the load can move before 4248 // the null check after castPP removal. 4249 Node* no_ctrl = NULL; 4250 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 4251 4252 // Test the header to see if it is unlocked. 4253 // This also serves as guard against value types (they have the always_locked_pattern set). 4254 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place); 4255 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 4256 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value); 4257 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val)); 4258 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne)); 4259 4260 generate_slow_guard(test_unlocked, slow_region); 4261 4262 // Get the hash value and check to see that it has been properly assigned. 4263 // We depend on hash_mask being at most 32 bits and avoid the use of 4264 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 4265 // vm: see markWord.hpp. 4266 Node *hash_mask = _gvn.intcon(markWord::hash_mask); 4267 Node *hash_shift = _gvn.intcon(markWord::hash_shift); 4268 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 4269 // This hack lets the hash bits live anywhere in the mark object now, as long 4270 // as the shift drops the relevant bits into the low 32 bits. Note that 4271 // Java spec says that HashCode is an int so there's no point in capturing 4272 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 4273 hshifted_header = ConvX2I(hshifted_header); 4274 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 4275 4276 Node *no_hash_val = _gvn.intcon(markWord::no_hash); 4277 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 4278 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 4279 4280 generate_slow_guard(test_assigned, slow_region); 4281 4282 Node* init_mem = reset_memory(); 4283 // fill in the rest of the null path: 4284 result_io ->init_req(_null_path, i_o()); 4285 result_mem->init_req(_null_path, init_mem); 4286 4287 result_val->init_req(_fast_path, hash_val); 4288 result_reg->init_req(_fast_path, control()); 4289 result_io ->init_req(_fast_path, i_o()); 4290 result_mem->init_req(_fast_path, init_mem); 4291 4292 // Generate code for the slow case. We make a call to hashCode(). 4293 set_control(_gvn.transform(slow_region)); 4294 if (!stopped()) { 4295 // No need for PreserveJVMState, because we're using up the present state. 4296 set_all_memory(init_mem); 4297 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 4298 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static); 4299 Node* slow_result = set_results_for_java_call(slow_call); 4300 // this->control() comes from set_results_for_java_call 4301 result_reg->init_req(_slow_path, control()); 4302 result_val->init_req(_slow_path, slow_result); 4303 result_io ->set_req(_slow_path, i_o()); 4304 result_mem ->set_req(_slow_path, reset_memory()); 4305 } 4306 4307 // Return the combined state. 4308 set_i_o( _gvn.transform(result_io) ); 4309 set_all_memory( _gvn.transform(result_mem)); 4310 4311 set_result(result_reg, result_val); 4312 return true; 4313 } 4314 4315 //---------------------------inline_native_getClass---------------------------- 4316 // public final native Class<?> java.lang.Object.getClass(); 4317 // 4318 // Build special case code for calls to getClass on an object. 4319 bool LibraryCallKit::inline_native_getClass() { 4320 Node* obj = argument(0); 4321 if (obj->is_ValueType()) { 4322 ciKlass* vk = _gvn.type(obj)->value_klass(); 4323 set_result(makecon(TypeInstPtr::make(vk->java_mirror()))); 4324 return true; 4325 } 4326 obj = null_check_receiver(); 4327 if (stopped()) return true; 4328 set_result(load_mirror_from_klass(load_object_klass(obj))); 4329 return true; 4330 } 4331 4332 //-----------------inline_native_Reflection_getCallerClass--------------------- 4333 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 4334 // 4335 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 4336 // 4337 // NOTE: This code must perform the same logic as JVM_GetCallerClass 4338 // in that it must skip particular security frames and checks for 4339 // caller sensitive methods. 4340 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 4341 #ifndef PRODUCT 4342 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4343 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 4344 } 4345 #endif 4346 4347 if (!jvms()->has_method()) { 4348 #ifndef PRODUCT 4349 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4350 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 4351 } 4352 #endif 4353 return false; 4354 } 4355 4356 // Walk back up the JVM state to find the caller at the required 4357 // depth. 4358 JVMState* caller_jvms = jvms(); 4359 4360 // Cf. JVM_GetCallerClass 4361 // NOTE: Start the loop at depth 1 because the current JVM state does 4362 // not include the Reflection.getCallerClass() frame. 4363 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) { 4364 ciMethod* m = caller_jvms->method(); 4365 switch (n) { 4366 case 0: 4367 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 4368 break; 4369 case 1: 4370 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 4371 if (!m->caller_sensitive()) { 4372 #ifndef PRODUCT 4373 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4374 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 4375 } 4376 #endif 4377 return false; // bail-out; let JVM_GetCallerClass do the work 4378 } 4379 break; 4380 default: 4381 if (!m->is_ignored_by_security_stack_walk()) { 4382 // We have reached the desired frame; return the holder class. 4383 // Acquire method holder as java.lang.Class and push as constant. 4384 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 4385 ciInstance* caller_mirror = caller_klass->java_mirror(); 4386 set_result(makecon(TypeInstPtr::make(caller_mirror))); 4387 4388 #ifndef PRODUCT 4389 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4390 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()); 4391 tty->print_cr(" JVM state at this point:"); 4392 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4393 ciMethod* m = jvms()->of_depth(i)->method(); 4394 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4395 } 4396 } 4397 #endif 4398 return true; 4399 } 4400 break; 4401 } 4402 } 4403 4404 #ifndef PRODUCT 4405 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4406 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 4407 tty->print_cr(" JVM state at this point:"); 4408 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4409 ciMethod* m = jvms()->of_depth(i)->method(); 4410 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4411 } 4412 } 4413 #endif 4414 4415 return false; // bail-out; let JVM_GetCallerClass do the work 4416 } 4417 4418 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 4419 Node* arg = argument(0); 4420 Node* result = NULL; 4421 4422 switch (id) { 4423 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 4424 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 4425 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 4426 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 4427 4428 case vmIntrinsics::_doubleToLongBits: { 4429 // two paths (plus control) merge in a wood 4430 RegionNode *r = new RegionNode(3); 4431 Node *phi = new PhiNode(r, TypeLong::LONG); 4432 4433 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 4434 // Build the boolean node 4435 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4436 4437 // Branch either way. 4438 // NaN case is less traveled, which makes all the difference. 4439 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4440 Node *opt_isnan = _gvn.transform(ifisnan); 4441 assert( opt_isnan->is_If(), "Expect an IfNode"); 4442 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4443 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4444 4445 set_control(iftrue); 4446 4447 static const jlong nan_bits = CONST64(0x7ff8000000000000); 4448 Node *slow_result = longcon(nan_bits); // return NaN 4449 phi->init_req(1, _gvn.transform( slow_result )); 4450 r->init_req(1, iftrue); 4451 4452 // Else fall through 4453 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4454 set_control(iffalse); 4455 4456 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 4457 r->init_req(2, iffalse); 4458 4459 // Post merge 4460 set_control(_gvn.transform(r)); 4461 record_for_igvn(r); 4462 4463 C->set_has_split_ifs(true); // Has chance for split-if optimization 4464 result = phi; 4465 assert(result->bottom_type()->isa_long(), "must be"); 4466 break; 4467 } 4468 4469 case vmIntrinsics::_floatToIntBits: { 4470 // two paths (plus control) merge in a wood 4471 RegionNode *r = new RegionNode(3); 4472 Node *phi = new PhiNode(r, TypeInt::INT); 4473 4474 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 4475 // Build the boolean node 4476 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4477 4478 // Branch either way. 4479 // NaN case is less traveled, which makes all the difference. 4480 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4481 Node *opt_isnan = _gvn.transform(ifisnan); 4482 assert( opt_isnan->is_If(), "Expect an IfNode"); 4483 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4484 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4485 4486 set_control(iftrue); 4487 4488 static const jint nan_bits = 0x7fc00000; 4489 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 4490 phi->init_req(1, _gvn.transform( slow_result )); 4491 r->init_req(1, iftrue); 4492 4493 // Else fall through 4494 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4495 set_control(iffalse); 4496 4497 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 4498 r->init_req(2, iffalse); 4499 4500 // Post merge 4501 set_control(_gvn.transform(r)); 4502 record_for_igvn(r); 4503 4504 C->set_has_split_ifs(true); // Has chance for split-if optimization 4505 result = phi; 4506 assert(result->bottom_type()->isa_int(), "must be"); 4507 break; 4508 } 4509 4510 default: 4511 fatal_unexpected_iid(id); 4512 break; 4513 } 4514 set_result(_gvn.transform(result)); 4515 return true; 4516 } 4517 4518 //----------------------inline_unsafe_copyMemory------------------------- 4519 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 4520 bool LibraryCallKit::inline_unsafe_copyMemory() { 4521 if (callee()->is_static()) return false; // caller must have the capability! 4522 null_check_receiver(); // null-check receiver 4523 if (stopped()) return true; 4524 4525 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 4526 4527 Node* src_ptr = argument(1); // type: oop 4528 Node* src_off = ConvL2X(argument(2)); // type: long 4529 Node* dst_ptr = argument(4); // type: oop 4530 Node* dst_off = ConvL2X(argument(5)); // type: long 4531 Node* size = ConvL2X(argument(7)); // type: long 4532 4533 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 4534 "fieldOffset must be byte-scaled"); 4535 4536 src_ptr = access_resolve(src_ptr, ACCESS_READ); 4537 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE); 4538 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ); 4539 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE); 4540 4541 // Conservatively insert a memory barrier on all memory slices. 4542 // Do not let writes of the copy source or destination float below the copy. 4543 insert_mem_bar(Op_MemBarCPUOrder); 4544 4545 Node* thread = _gvn.transform(new ThreadLocalNode()); 4546 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 4547 BasicType doing_unsafe_access_bt = T_BYTE; 4548 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 4549 4550 // update volatile field 4551 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4552 4553 // Call it. Note that the length argument is not scaled. 4554 make_runtime_call(RC_LEAF|RC_NO_FP, 4555 OptoRuntime::fast_arraycopy_Type(), 4556 StubRoutines::unsafe_arraycopy(), 4557 "unsafe_arraycopy", 4558 TypeRawPtr::BOTTOM, 4559 src, dst, size XTOP); 4560 4561 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4562 4563 // Do not let reads of the copy destination float above the copy. 4564 insert_mem_bar(Op_MemBarCPUOrder); 4565 4566 return true; 4567 } 4568 4569 //------------------------clone_coping----------------------------------- 4570 // Helper function for inline_native_clone. 4571 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) { 4572 assert(obj_size != NULL, ""); 4573 Node* raw_obj = alloc_obj->in(1); 4574 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 4575 4576 AllocateNode* alloc = NULL; 4577 if (ReduceBulkZeroing) { 4578 // We will be completely responsible for initializing this object - 4579 // mark Initialize node as complete. 4580 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn); 4581 // The object was just allocated - there should be no any stores! 4582 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), ""); 4583 // Mark as complete_with_arraycopy so that on AllocateNode 4584 // expansion, we know this AllocateNode is initialized by an array 4585 // copy and a StoreStore barrier exists after the array copy. 4586 alloc->initialization()->set_complete_with_arraycopy(); 4587 } 4588 4589 Node* size = _gvn.transform(obj_size); 4590 // Exclude the header but include array length to copy by 8 bytes words. 4591 // Can't use base_offset_in_bytes(bt) since basic type is unknown. 4592 int base_off = BarrierSetC2::arraycopy_payload_base_offset(is_array); 4593 Node* src_base = basic_plus_adr(obj, base_off); 4594 Node* dst_base = basic_plus_adr(alloc_obj, base_off); 4595 4596 // Compute the length also, if needed: 4597 Node* countx = size; 4598 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off))); 4599 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong))); 4600 4601 access_clone(src_base, dst_base, countx, is_array); 4602 4603 // Do not let reads from the cloned object float above the arraycopy. 4604 if (alloc != NULL) { 4605 // Do not let stores that initialize this object be reordered with 4606 // a subsequent store that would make this object accessible by 4607 // other threads. 4608 // Record what AllocateNode this StoreStore protects so that 4609 // escape analysis can go from the MemBarStoreStoreNode to the 4610 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 4611 // based on the escape status of the AllocateNode. 4612 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 4613 } else { 4614 insert_mem_bar(Op_MemBarCPUOrder); 4615 } 4616 } 4617 4618 //------------------------inline_native_clone---------------------------- 4619 // protected native Object java.lang.Object.clone(); 4620 // 4621 // Here are the simple edge cases: 4622 // null receiver => normal trap 4623 // virtual and clone was overridden => slow path to out-of-line clone 4624 // not cloneable or finalizer => slow path to out-of-line Object.clone 4625 // 4626 // The general case has two steps, allocation and copying. 4627 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 4628 // 4629 // Copying also has two cases, oop arrays and everything else. 4630 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 4631 // Everything else uses the tight inline loop supplied by CopyArrayNode. 4632 // 4633 // These steps fold up nicely if and when the cloned object's klass 4634 // can be sharply typed as an object array, a type array, or an instance. 4635 // 4636 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 4637 PhiNode* result_val; 4638 4639 // Set the reexecute bit for the interpreter to reexecute 4640 // the bytecode that invokes Object.clone if deoptimization happens. 4641 { PreserveReexecuteState preexecs(this); 4642 jvms()->set_should_reexecute(true); 4643 4644 Node* obj = argument(0); 4645 if (obj->is_ValueType()) { 4646 return false; 4647 } 4648 4649 obj = null_check_receiver(); 4650 if (stopped()) return true; 4651 4652 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 4653 4654 // If we are going to clone an instance, we need its exact type to 4655 // know the number and types of fields to convert the clone to 4656 // loads/stores. Maybe a speculative type can help us. 4657 if (!obj_type->klass_is_exact() && 4658 obj_type->speculative_type() != NULL && 4659 obj_type->speculative_type()->is_instance_klass() && 4660 !obj_type->speculative_type()->is_valuetype()) { 4661 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 4662 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 4663 !spec_ik->has_injected_fields()) { 4664 ciKlass* k = obj_type->klass(); 4665 if (!k->is_instance_klass() || 4666 k->as_instance_klass()->is_interface() || 4667 k->as_instance_klass()->has_subklass()) { 4668 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 4669 } 4670 } 4671 } 4672 4673 // Conservatively insert a memory barrier on all memory slices. 4674 // Do not let writes into the original float below the clone. 4675 insert_mem_bar(Op_MemBarCPUOrder); 4676 4677 // paths into result_reg: 4678 enum { 4679 _slow_path = 1, // out-of-line call to clone method (virtual or not) 4680 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 4681 _array_path, // plain array allocation, plus arrayof_long_arraycopy 4682 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 4683 PATH_LIMIT 4684 }; 4685 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4686 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4687 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 4688 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4689 record_for_igvn(result_reg); 4690 4691 Node* obj_klass = load_object_klass(obj); 4692 // We only go to the fast case code if we pass a number of guards. 4693 // The paths which do not pass are accumulated in the slow_region. 4694 RegionNode* slow_region = new RegionNode(1); 4695 record_for_igvn(slow_region); 4696 4697 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL); 4698 if (array_ctl != NULL) { 4699 // It's an array. 4700 PreserveJVMState pjvms(this); 4701 set_control(array_ctl); 4702 4703 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4704 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing) && 4705 (!obj_type->isa_aryptr() || !obj_type->is_aryptr()->is_not_flat())) { 4706 // Flattened value type array may have object field that would require a 4707 // write barrier. Conservatively, go to slow path. 4708 generate_valueArray_guard(obj_klass, slow_region); 4709 } 4710 4711 if (!stopped()) { 4712 Node* obj_length = load_array_length(obj); 4713 Node* obj_size = NULL; 4714 // Load element mirror 4715 Node* array_type_mirror = load_mirror_from_klass(obj_klass); 4716 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes()); 4717 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP); 4718 4719 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, false, elem_mirror); 4720 4721 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4722 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) { 4723 // If it is an oop array, it requires very special treatment, 4724 // because gc barriers are required when accessing the array. 4725 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL); 4726 if (is_obja != NULL) { 4727 PreserveJVMState pjvms2(this); 4728 set_control(is_obja); 4729 // Generate a direct call to the right arraycopy function(s). 4730 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL); 4731 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false); 4732 ac->set_clone_oop_array(); 4733 Node* n = _gvn.transform(ac); 4734 assert(n == ac, "cannot disappear"); 4735 ac->connect_outputs(this); 4736 4737 result_reg->init_req(_objArray_path, control()); 4738 result_val->init_req(_objArray_path, alloc_obj); 4739 result_i_o ->set_req(_objArray_path, i_o()); 4740 result_mem ->set_req(_objArray_path, reset_memory()); 4741 } 4742 } 4743 4744 // Otherwise, there are no barriers to worry about. 4745 // (We can dispense with card marks if we know the allocation 4746 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 4747 // causes the non-eden paths to take compensating steps to 4748 // simulate a fresh allocation, so that no further 4749 // card marks are required in compiled code to initialize 4750 // the object.) 4751 4752 if (!stopped()) { 4753 copy_to_clone(obj, alloc_obj, obj_size, true); 4754 4755 // Present the results of the copy. 4756 result_reg->init_req(_array_path, control()); 4757 result_val->init_req(_array_path, alloc_obj); 4758 result_i_o ->set_req(_array_path, i_o()); 4759 result_mem ->set_req(_array_path, reset_memory()); 4760 } 4761 } 4762 } 4763 4764 if (!stopped()) { 4765 // It's an instance (we did array above). Make the slow-path tests. 4766 // If this is a virtual call, we generate a funny guard. We grab 4767 // the vtable entry corresponding to clone() from the target object. 4768 // If the target method which we are calling happens to be the 4769 // Object clone() method, we pass the guard. We do not need this 4770 // guard for non-virtual calls; the caller is known to be the native 4771 // Object clone(). 4772 if (is_virtual) { 4773 generate_virtual_guard(obj_klass, slow_region); 4774 } 4775 4776 // The object must be easily cloneable and must not have a finalizer. 4777 // Both of these conditions may be checked in a single test. 4778 // We could optimize the test further, but we don't care. 4779 generate_access_flags_guard(obj_klass, 4780 // Test both conditions: 4781 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER, 4782 // Must be cloneable but not finalizer: 4783 JVM_ACC_IS_CLONEABLE_FAST, 4784 slow_region); 4785 } 4786 4787 if (!stopped()) { 4788 // It's an instance, and it passed the slow-path tests. 4789 PreserveJVMState pjvms(this); 4790 Node* obj_size = NULL; 4791 // Need to deoptimize on exception from allocation since Object.clone intrinsic 4792 // is reexecuted if deoptimization occurs and there could be problems when merging 4793 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 4794 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true); 4795 4796 copy_to_clone(obj, alloc_obj, obj_size, false); 4797 4798 // Present the results of the slow call. 4799 result_reg->init_req(_instance_path, control()); 4800 result_val->init_req(_instance_path, alloc_obj); 4801 result_i_o ->set_req(_instance_path, i_o()); 4802 result_mem ->set_req(_instance_path, reset_memory()); 4803 } 4804 4805 // Generate code for the slow case. We make a call to clone(). 4806 set_control(_gvn.transform(slow_region)); 4807 if (!stopped()) { 4808 PreserveJVMState pjvms(this); 4809 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual); 4810 // We need to deoptimize on exception (see comment above) 4811 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true); 4812 // this->control() comes from set_results_for_java_call 4813 result_reg->init_req(_slow_path, control()); 4814 result_val->init_req(_slow_path, slow_result); 4815 result_i_o ->set_req(_slow_path, i_o()); 4816 result_mem ->set_req(_slow_path, reset_memory()); 4817 } 4818 4819 // Return the combined state. 4820 set_control( _gvn.transform(result_reg)); 4821 set_i_o( _gvn.transform(result_i_o)); 4822 set_all_memory( _gvn.transform(result_mem)); 4823 } // original reexecute is set back here 4824 4825 set_result(_gvn.transform(result_val)); 4826 return true; 4827 } 4828 4829 // If we have a tightly coupled allocation, the arraycopy may take care 4830 // of the array initialization. If one of the guards we insert between 4831 // the allocation and the arraycopy causes a deoptimization, an 4832 // unitialized array will escape the compiled method. To prevent that 4833 // we set the JVM state for uncommon traps between the allocation and 4834 // the arraycopy to the state before the allocation so, in case of 4835 // deoptimization, we'll reexecute the allocation and the 4836 // initialization. 4837 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 4838 if (alloc != NULL) { 4839 ciMethod* trap_method = alloc->jvms()->method(); 4840 int trap_bci = alloc->jvms()->bci(); 4841 4842 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 4843 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 4844 // Make sure there's no store between the allocation and the 4845 // arraycopy otherwise visible side effects could be rexecuted 4846 // in case of deoptimization and cause incorrect execution. 4847 bool no_interfering_store = true; 4848 Node* mem = alloc->in(TypeFunc::Memory); 4849 if (mem->is_MergeMem()) { 4850 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 4851 Node* n = mms.memory(); 4852 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4853 assert(n->is_Store(), "what else?"); 4854 no_interfering_store = false; 4855 break; 4856 } 4857 } 4858 } else { 4859 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 4860 Node* n = mms.memory(); 4861 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4862 assert(n->is_Store(), "what else?"); 4863 no_interfering_store = false; 4864 break; 4865 } 4866 } 4867 } 4868 4869 if (no_interfering_store) { 4870 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 4871 uint size = alloc->req(); 4872 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 4873 old_jvms->set_map(sfpt); 4874 for (uint i = 0; i < size; i++) { 4875 sfpt->init_req(i, alloc->in(i)); 4876 } 4877 // re-push array length for deoptimization 4878 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength)); 4879 old_jvms->set_sp(old_jvms->sp()+1); 4880 old_jvms->set_monoff(old_jvms->monoff()+1); 4881 old_jvms->set_scloff(old_jvms->scloff()+1); 4882 old_jvms->set_endoff(old_jvms->endoff()+1); 4883 old_jvms->set_should_reexecute(true); 4884 4885 sfpt->set_i_o(map()->i_o()); 4886 sfpt->set_memory(map()->memory()); 4887 sfpt->set_control(map()->control()); 4888 4889 JVMState* saved_jvms = jvms(); 4890 saved_reexecute_sp = _reexecute_sp; 4891 4892 set_jvms(sfpt->jvms()); 4893 _reexecute_sp = jvms()->sp(); 4894 4895 return saved_jvms; 4896 } 4897 } 4898 } 4899 return NULL; 4900 } 4901 4902 // In case of a deoptimization, we restart execution at the 4903 // allocation, allocating a new array. We would leave an uninitialized 4904 // array in the heap that GCs wouldn't expect. Move the allocation 4905 // after the traps so we don't allocate the array if we 4906 // deoptimize. This is possible because tightly_coupled_allocation() 4907 // guarantees there's no observer of the allocated array at this point 4908 // and the control flow is simple enough. 4909 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, 4910 int saved_reexecute_sp, uint new_idx) { 4911 if (saved_jvms != NULL && !stopped()) { 4912 assert(alloc != NULL, "only with a tightly coupled allocation"); 4913 // restore JVM state to the state at the arraycopy 4914 saved_jvms->map()->set_control(map()->control()); 4915 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?"); 4916 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?"); 4917 // If we've improved the types of some nodes (null check) while 4918 // emitting the guards, propagate them to the current state 4919 map()->replaced_nodes().apply(saved_jvms->map(), new_idx); 4920 set_jvms(saved_jvms); 4921 _reexecute_sp = saved_reexecute_sp; 4922 4923 // Remove the allocation from above the guards 4924 CallProjections* callprojs = alloc->extract_projections(true); 4925 InitializeNode* init = alloc->initialization(); 4926 Node* alloc_mem = alloc->in(TypeFunc::Memory); 4927 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 4928 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 4929 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 4930 4931 // move the allocation here (after the guards) 4932 _gvn.hash_delete(alloc); 4933 alloc->set_req(TypeFunc::Control, control()); 4934 alloc->set_req(TypeFunc::I_O, i_o()); 4935 Node *mem = reset_memory(); 4936 set_all_memory(mem); 4937 alloc->set_req(TypeFunc::Memory, mem); 4938 set_control(init->proj_out_or_null(TypeFunc::Control)); 4939 set_i_o(callprojs->fallthrough_ioproj); 4940 4941 // Update memory as done in GraphKit::set_output_for_allocation() 4942 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 4943 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 4944 if (ary_type->isa_aryptr() && length_type != NULL) { 4945 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 4946 } 4947 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 4948 int elemidx = C->get_alias_index(telemref); 4949 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw); 4950 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx); 4951 4952 Node* allocx = _gvn.transform(alloc); 4953 assert(allocx == alloc, "where has the allocation gone?"); 4954 assert(dest->is_CheckCastPP(), "not an allocation result?"); 4955 4956 _gvn.hash_delete(dest); 4957 dest->set_req(0, control()); 4958 Node* destx = _gvn.transform(dest); 4959 assert(destx == dest, "where has the allocation result gone?"); 4960 } 4961 } 4962 4963 4964 //------------------------------inline_arraycopy----------------------- 4965 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 4966 // Object dest, int destPos, 4967 // int length); 4968 bool LibraryCallKit::inline_arraycopy() { 4969 // Get the arguments. 4970 Node* src = argument(0); // type: oop 4971 Node* src_offset = argument(1); // type: int 4972 Node* dest = argument(2); // type: oop 4973 Node* dest_offset = argument(3); // type: int 4974 Node* length = argument(4); // type: int 4975 4976 uint new_idx = C->unique(); 4977 4978 // Check for allocation before we add nodes that would confuse 4979 // tightly_coupled_allocation() 4980 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL); 4981 4982 int saved_reexecute_sp = -1; 4983 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 4984 // See arraycopy_restore_alloc_state() comment 4985 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards 4986 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation 4987 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards 4988 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL); 4989 4990 // The following tests must be performed 4991 // (1) src and dest are arrays. 4992 // (2) src and dest arrays must have elements of the same BasicType 4993 // (3) src and dest must not be null. 4994 // (4) src_offset must not be negative. 4995 // (5) dest_offset must not be negative. 4996 // (6) length must not be negative. 4997 // (7) src_offset + length must not exceed length of src. 4998 // (8) dest_offset + length must not exceed length of dest. 4999 // (9) each element of an oop array must be assignable 5000 5001 // (3) src and dest must not be null. 5002 // always do this here because we need the JVM state for uncommon traps 5003 Node* null_ctl = top(); 5004 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 5005 assert(null_ctl->is_top(), "no null control here"); 5006 dest = null_check(dest, T_ARRAY); 5007 5008 if (!can_emit_guards) { 5009 // if saved_jvms == NULL and alloc != NULL, we don't emit any 5010 // guards but the arraycopy node could still take advantage of a 5011 // tightly allocated allocation. tightly_coupled_allocation() is 5012 // called again to make sure it takes the null check above into 5013 // account: the null check is mandatory and if it caused an 5014 // uncommon trap to be emitted then the allocation can't be 5015 // considered tightly coupled in this context. 5016 alloc = tightly_coupled_allocation(dest, NULL); 5017 } 5018 5019 bool validated = false; 5020 5021 const Type* src_type = _gvn.type(src); 5022 const Type* dest_type = _gvn.type(dest); 5023 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5024 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5025 5026 // Do we have the type of src? 5027 bool has_src = (top_src != NULL && top_src->klass() != NULL); 5028 // Do we have the type of dest? 5029 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL); 5030 // Is the type for src from speculation? 5031 bool src_spec = false; 5032 // Is the type for dest from speculation? 5033 bool dest_spec = false; 5034 5035 if ((!has_src || !has_dest) && can_emit_guards) { 5036 // We don't have sufficient type information, let's see if 5037 // speculative types can help. We need to have types for both src 5038 // and dest so that it pays off. 5039 5040 // Do we already have or could we have type information for src 5041 bool could_have_src = has_src; 5042 // Do we already have or could we have type information for dest 5043 bool could_have_dest = has_dest; 5044 5045 ciKlass* src_k = NULL; 5046 if (!has_src) { 5047 src_k = src_type->speculative_type_not_null(); 5048 if (src_k != NULL && src_k->is_array_klass()) { 5049 could_have_src = true; 5050 } 5051 } 5052 5053 ciKlass* dest_k = NULL; 5054 if (!has_dest) { 5055 dest_k = dest_type->speculative_type_not_null(); 5056 if (dest_k != NULL && dest_k->is_array_klass()) { 5057 could_have_dest = true; 5058 } 5059 } 5060 5061 if (could_have_src && could_have_dest) { 5062 // This is going to pay off so emit the required guards 5063 if (!has_src) { 5064 src = maybe_cast_profiled_obj(src, src_k, true); 5065 src_type = _gvn.type(src); 5066 top_src = src_type->isa_aryptr(); 5067 has_src = (top_src != NULL && top_src->klass() != NULL); 5068 src_spec = true; 5069 } 5070 if (!has_dest) { 5071 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5072 dest_type = _gvn.type(dest); 5073 top_dest = dest_type->isa_aryptr(); 5074 has_dest = (top_dest != NULL && top_dest->klass() != NULL); 5075 dest_spec = true; 5076 } 5077 } 5078 } 5079 5080 if (has_src && has_dest && can_emit_guards) { 5081 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type(); 5082 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type(); 5083 if (src_elem == T_ARRAY) src_elem = T_OBJECT; 5084 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT; 5085 5086 if (src_elem == dest_elem && src_elem == T_OBJECT) { 5087 // If both arrays are object arrays then having the exact types 5088 // for both will remove the need for a subtype check at runtime 5089 // before the call and may make it possible to pick a faster copy 5090 // routine (without a subtype check on every element) 5091 // Do we have the exact type of src? 5092 bool could_have_src = src_spec; 5093 // Do we have the exact type of dest? 5094 bool could_have_dest = dest_spec; 5095 ciKlass* src_k = top_src->klass(); 5096 ciKlass* dest_k = top_dest->klass(); 5097 if (!src_spec) { 5098 src_k = src_type->speculative_type_not_null(); 5099 if (src_k != NULL && src_k->is_array_klass()) { 5100 could_have_src = true; 5101 } 5102 } 5103 if (!dest_spec) { 5104 dest_k = dest_type->speculative_type_not_null(); 5105 if (dest_k != NULL && dest_k->is_array_klass()) { 5106 could_have_dest = true; 5107 } 5108 } 5109 if (could_have_src && could_have_dest) { 5110 // If we can have both exact types, emit the missing guards 5111 if (could_have_src && !src_spec) { 5112 src = maybe_cast_profiled_obj(src, src_k, true); 5113 } 5114 if (could_have_dest && !dest_spec) { 5115 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5116 } 5117 } 5118 } 5119 } 5120 5121 ciMethod* trap_method = method(); 5122 int trap_bci = bci(); 5123 if (saved_jvms != NULL) { 5124 trap_method = alloc->jvms()->method(); 5125 trap_bci = alloc->jvms()->bci(); 5126 } 5127 5128 bool negative_length_guard_generated = false; 5129 5130 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5131 can_emit_guards && 5132 !src->is_top() && !dest->is_top()) { 5133 // validate arguments: enables transformation the ArrayCopyNode 5134 validated = true; 5135 5136 RegionNode* slow_region = new RegionNode(1); 5137 record_for_igvn(slow_region); 5138 5139 // (1) src and dest are arrays. 5140 generate_non_array_guard(load_object_klass(src), slow_region); 5141 generate_non_array_guard(load_object_klass(dest), slow_region); 5142 5143 // (2) src and dest arrays must have elements of the same BasicType 5144 // done at macro expansion or at Ideal transformation time 5145 5146 // (4) src_offset must not be negative. 5147 generate_negative_guard(src_offset, slow_region); 5148 5149 // (5) dest_offset must not be negative. 5150 generate_negative_guard(dest_offset, slow_region); 5151 5152 // (7) src_offset + length must not exceed length of src. 5153 generate_limit_guard(src_offset, length, 5154 load_array_length(src), 5155 slow_region); 5156 5157 // (8) dest_offset + length must not exceed length of dest. 5158 generate_limit_guard(dest_offset, length, 5159 load_array_length(dest), 5160 slow_region); 5161 5162 // (6) length must not be negative. 5163 // This is also checked in generate_arraycopy() during macro expansion, but 5164 // we also have to check it here for the case where the ArrayCopyNode will 5165 // be eliminated by Escape Analysis. 5166 if (EliminateAllocations) { 5167 generate_negative_guard(length, slow_region); 5168 negative_length_guard_generated = true; 5169 } 5170 5171 // (9) each element of an oop array must be assignable 5172 Node* src_klass = load_object_klass(src); 5173 Node* dest_klass = load_object_klass(dest); 5174 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass); 5175 5176 if (not_subtype_ctrl != top()) { 5177 PreserveJVMState pjvms(this); 5178 set_control(not_subtype_ctrl); 5179 uncommon_trap(Deoptimization::Reason_intrinsic, 5180 Deoptimization::Action_make_not_entrant); 5181 assert(stopped(), "Should be stopped"); 5182 } 5183 5184 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr(); 5185 const Type* toop = TypeOopPtr::make_from_klass(dest_klass_t->klass()); 5186 src = _gvn.transform(new CheckCastPPNode(control(), src, toop)); 5187 src_type = _gvn.type(src); 5188 top_src = src_type->isa_aryptr(); 5189 5190 if (top_dest != NULL && !top_dest->elem()->isa_valuetype() && !top_dest->is_not_flat()) { 5191 generate_valueArray_guard(dest_klass, slow_region); 5192 } 5193 5194 if (top_src != NULL && !top_src->elem()->isa_valuetype() && !top_src->is_not_flat()) { 5195 generate_valueArray_guard(src_klass, slow_region); 5196 } 5197 5198 { 5199 PreserveJVMState pjvms(this); 5200 set_control(_gvn.transform(slow_region)); 5201 uncommon_trap(Deoptimization::Reason_intrinsic, 5202 Deoptimization::Action_make_not_entrant); 5203 assert(stopped(), "Should be stopped"); 5204 } 5205 } 5206 5207 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx); 5208 5209 if (stopped()) { 5210 return true; 5211 } 5212 5213 Node* new_src = access_resolve(src, ACCESS_READ); 5214 Node* new_dest = access_resolve(dest, ACCESS_WRITE); 5215 5216 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated, 5217 // Create LoadRange and LoadKlass nodes for use during macro expansion here 5218 // so the compiler has a chance to eliminate them: during macro expansion, 5219 // we have to set their control (CastPP nodes are eliminated). 5220 load_object_klass(src), load_object_klass(dest), 5221 load_array_length(src), load_array_length(dest)); 5222 5223 ac->set_arraycopy(validated); 5224 5225 Node* n = _gvn.transform(ac); 5226 if (n == ac) { 5227 ac->connect_outputs(this); 5228 } else { 5229 assert(validated, "shouldn't transform if all arguments not validated"); 5230 set_all_memory(n); 5231 } 5232 clear_upper_avx(); 5233 5234 5235 return true; 5236 } 5237 5238 5239 // Helper function which determines if an arraycopy immediately follows 5240 // an allocation, with no intervening tests or other escapes for the object. 5241 AllocateArrayNode* 5242 LibraryCallKit::tightly_coupled_allocation(Node* ptr, 5243 RegionNode* slow_region) { 5244 if (stopped()) return NULL; // no fast path 5245 if (C->AliasLevel() == 0) return NULL; // no MergeMems around 5246 5247 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn); 5248 if (alloc == NULL) return NULL; 5249 5250 Node* rawmem = memory(Compile::AliasIdxRaw); 5251 // Is the allocation's memory state untouched? 5252 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 5253 // Bail out if there have been raw-memory effects since the allocation. 5254 // (Example: There might have been a call or safepoint.) 5255 return NULL; 5256 } 5257 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 5258 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 5259 return NULL; 5260 } 5261 5262 // There must be no unexpected observers of this allocation. 5263 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 5264 Node* obs = ptr->fast_out(i); 5265 if (obs != this->map()) { 5266 return NULL; 5267 } 5268 } 5269 5270 // This arraycopy must unconditionally follow the allocation of the ptr. 5271 Node* alloc_ctl = ptr->in(0); 5272 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo"); 5273 5274 Node* ctl = control(); 5275 while (ctl != alloc_ctl) { 5276 // There may be guards which feed into the slow_region. 5277 // Any other control flow means that we might not get a chance 5278 // to finish initializing the allocated object. 5279 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) { 5280 IfNode* iff = ctl->in(0)->as_If(); 5281 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con); 5282 assert(not_ctl != NULL && not_ctl != ctl, "found alternate"); 5283 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) { 5284 ctl = iff->in(0); // This test feeds the known slow_region. 5285 continue; 5286 } 5287 // One more try: Various low-level checks bottom out in 5288 // uncommon traps. If the debug-info of the trap omits 5289 // any reference to the allocation, as we've already 5290 // observed, then there can be no objection to the trap. 5291 bool found_trap = false; 5292 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) { 5293 Node* obs = not_ctl->fast_out(j); 5294 if (obs->in(0) == not_ctl && obs->is_Call() && 5295 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) { 5296 found_trap = true; break; 5297 } 5298 } 5299 if (found_trap) { 5300 ctl = iff->in(0); // This test feeds a harmless uncommon trap. 5301 continue; 5302 } 5303 } 5304 return NULL; 5305 } 5306 5307 // If we get this far, we have an allocation which immediately 5308 // precedes the arraycopy, and we can take over zeroing the new object. 5309 // The arraycopy will finish the initialization, and provide 5310 // a new control state to which we will anchor the destination pointer. 5311 5312 return alloc; 5313 } 5314 5315 //-------------inline_encodeISOArray----------------------------------- 5316 // encode char[] to byte[] in ISO_8859_1 5317 bool LibraryCallKit::inline_encodeISOArray() { 5318 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 5319 // no receiver since it is static method 5320 Node *src = argument(0); 5321 Node *src_offset = argument(1); 5322 Node *dst = argument(2); 5323 Node *dst_offset = argument(3); 5324 Node *length = argument(4); 5325 5326 src = must_be_not_null(src, true); 5327 dst = must_be_not_null(dst, true); 5328 5329 src = access_resolve(src, ACCESS_READ); 5330 dst = access_resolve(dst, ACCESS_WRITE); 5331 5332 const Type* src_type = src->Value(&_gvn); 5333 const Type* dst_type = dst->Value(&_gvn); 5334 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5335 const TypeAryPtr* top_dest = dst_type->isa_aryptr(); 5336 if (top_src == NULL || top_src->klass() == NULL || 5337 top_dest == NULL || top_dest->klass() == NULL) { 5338 // failed array check 5339 return false; 5340 } 5341 5342 // Figure out the size and type of the elements we will be copying. 5343 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5344 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5345 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 5346 return false; 5347 } 5348 5349 Node* src_start = array_element_address(src, src_offset, T_CHAR); 5350 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 5351 // 'src_start' points to src array + scaled offset 5352 // 'dst_start' points to dst array + scaled offset 5353 5354 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 5355 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length); 5356 enc = _gvn.transform(enc); 5357 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 5358 set_memory(res_mem, mtype); 5359 set_result(enc); 5360 clear_upper_avx(); 5361 5362 return true; 5363 } 5364 5365 //-------------inline_multiplyToLen----------------------------------- 5366 bool LibraryCallKit::inline_multiplyToLen() { 5367 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 5368 5369 address stubAddr = StubRoutines::multiplyToLen(); 5370 if (stubAddr == NULL) { 5371 return false; // Intrinsic's stub is not implemented on this platform 5372 } 5373 const char* stubName = "multiplyToLen"; 5374 5375 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 5376 5377 // no receiver because it is a static method 5378 Node* x = argument(0); 5379 Node* xlen = argument(1); 5380 Node* y = argument(2); 5381 Node* ylen = argument(3); 5382 Node* z = argument(4); 5383 5384 x = must_be_not_null(x, true); 5385 y = must_be_not_null(y, true); 5386 5387 x = access_resolve(x, ACCESS_READ); 5388 y = access_resolve(y, ACCESS_READ); 5389 z = access_resolve(z, ACCESS_WRITE); 5390 5391 const Type* x_type = x->Value(&_gvn); 5392 const Type* y_type = y->Value(&_gvn); 5393 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5394 const TypeAryPtr* top_y = y_type->isa_aryptr(); 5395 if (top_x == NULL || top_x->klass() == NULL || 5396 top_y == NULL || top_y->klass() == NULL) { 5397 // failed array check 5398 return false; 5399 } 5400 5401 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5402 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5403 if (x_elem != T_INT || y_elem != T_INT) { 5404 return false; 5405 } 5406 5407 // Set the original stack and the reexecute bit for the interpreter to reexecute 5408 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens 5409 // on the return from z array allocation in runtime. 5410 { PreserveReexecuteState preexecs(this); 5411 jvms()->set_should_reexecute(true); 5412 5413 Node* x_start = array_element_address(x, intcon(0), x_elem); 5414 Node* y_start = array_element_address(y, intcon(0), y_elem); 5415 // 'x_start' points to x array + scaled xlen 5416 // 'y_start' points to y array + scaled ylen 5417 5418 // Allocate the result array 5419 Node* zlen = _gvn.transform(new AddINode(xlen, ylen)); 5420 ciKlass* klass = ciTypeArrayKlass::make(T_INT); 5421 Node* klass_node = makecon(TypeKlassPtr::make(klass)); 5422 5423 IdealKit ideal(this); 5424 5425 #define __ ideal. 5426 Node* one = __ ConI(1); 5427 Node* zero = __ ConI(0); 5428 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done(); 5429 __ set(need_alloc, zero); 5430 __ set(z_alloc, z); 5431 __ if_then(z, BoolTest::eq, null()); { 5432 __ increment (need_alloc, one); 5433 } __ else_(); { 5434 // Update graphKit memory and control from IdealKit. 5435 sync_kit(ideal); 5436 Node *cast = new CastPPNode(z, TypePtr::NOTNULL); 5437 cast->init_req(0, control()); 5438 _gvn.set_type(cast, cast->bottom_type()); 5439 C->record_for_igvn(cast); 5440 5441 Node* zlen_arg = load_array_length(cast); 5442 // Update IdealKit memory and control from graphKit. 5443 __ sync_kit(this); 5444 __ if_then(zlen_arg, BoolTest::lt, zlen); { 5445 __ increment (need_alloc, one); 5446 } __ end_if(); 5447 } __ end_if(); 5448 5449 __ if_then(__ value(need_alloc), BoolTest::ne, zero); { 5450 // Update graphKit memory and control from IdealKit. 5451 sync_kit(ideal); 5452 Node * narr = new_array(klass_node, zlen, 1); 5453 // Update IdealKit memory and control from graphKit. 5454 __ sync_kit(this); 5455 __ set(z_alloc, narr); 5456 } __ end_if(); 5457 5458 sync_kit(ideal); 5459 z = __ value(z_alloc); 5460 // Can't use TypeAryPtr::INTS which uses Bottom offset. 5461 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass)); 5462 // Final sync IdealKit and GraphKit. 5463 final_sync(ideal); 5464 #undef __ 5465 5466 Node* z_start = array_element_address(z, intcon(0), T_INT); 5467 5468 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5469 OptoRuntime::multiplyToLen_Type(), 5470 stubAddr, stubName, TypePtr::BOTTOM, 5471 x_start, xlen, y_start, ylen, z_start, zlen); 5472 } // original reexecute is set back here 5473 5474 C->set_has_split_ifs(true); // Has chance for split-if optimization 5475 set_result(z); 5476 return true; 5477 } 5478 5479 //-------------inline_squareToLen------------------------------------ 5480 bool LibraryCallKit::inline_squareToLen() { 5481 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 5482 5483 address stubAddr = StubRoutines::squareToLen(); 5484 if (stubAddr == NULL) { 5485 return false; // Intrinsic's stub is not implemented on this platform 5486 } 5487 const char* stubName = "squareToLen"; 5488 5489 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 5490 5491 Node* x = argument(0); 5492 Node* len = argument(1); 5493 Node* z = argument(2); 5494 Node* zlen = argument(3); 5495 5496 x = must_be_not_null(x, true); 5497 z = must_be_not_null(z, true); 5498 5499 x = access_resolve(x, ACCESS_READ); 5500 z = access_resolve(z, ACCESS_WRITE); 5501 5502 const Type* x_type = x->Value(&_gvn); 5503 const Type* z_type = z->Value(&_gvn); 5504 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5505 const TypeAryPtr* top_z = z_type->isa_aryptr(); 5506 if (top_x == NULL || top_x->klass() == NULL || 5507 top_z == NULL || top_z->klass() == NULL) { 5508 // failed array check 5509 return false; 5510 } 5511 5512 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5513 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5514 if (x_elem != T_INT || z_elem != T_INT) { 5515 return false; 5516 } 5517 5518 5519 Node* x_start = array_element_address(x, intcon(0), x_elem); 5520 Node* z_start = array_element_address(z, intcon(0), z_elem); 5521 5522 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5523 OptoRuntime::squareToLen_Type(), 5524 stubAddr, stubName, TypePtr::BOTTOM, 5525 x_start, len, z_start, zlen); 5526 5527 set_result(z); 5528 return true; 5529 } 5530 5531 //-------------inline_mulAdd------------------------------------------ 5532 bool LibraryCallKit::inline_mulAdd() { 5533 assert(UseMulAddIntrinsic, "not implemented on this platform"); 5534 5535 address stubAddr = StubRoutines::mulAdd(); 5536 if (stubAddr == NULL) { 5537 return false; // Intrinsic's stub is not implemented on this platform 5538 } 5539 const char* stubName = "mulAdd"; 5540 5541 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 5542 5543 Node* out = argument(0); 5544 Node* in = argument(1); 5545 Node* offset = argument(2); 5546 Node* len = argument(3); 5547 Node* k = argument(4); 5548 5549 out = must_be_not_null(out, true); 5550 5551 in = access_resolve(in, ACCESS_READ); 5552 out = access_resolve(out, ACCESS_WRITE); 5553 5554 const Type* out_type = out->Value(&_gvn); 5555 const Type* in_type = in->Value(&_gvn); 5556 const TypeAryPtr* top_out = out_type->isa_aryptr(); 5557 const TypeAryPtr* top_in = in_type->isa_aryptr(); 5558 if (top_out == NULL || top_out->klass() == NULL || 5559 top_in == NULL || top_in->klass() == NULL) { 5560 // failed array check 5561 return false; 5562 } 5563 5564 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5565 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5566 if (out_elem != T_INT || in_elem != T_INT) { 5567 return false; 5568 } 5569 5570 Node* outlen = load_array_length(out); 5571 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 5572 Node* out_start = array_element_address(out, intcon(0), out_elem); 5573 Node* in_start = array_element_address(in, intcon(0), in_elem); 5574 5575 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5576 OptoRuntime::mulAdd_Type(), 5577 stubAddr, stubName, TypePtr::BOTTOM, 5578 out_start,in_start, new_offset, len, k); 5579 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5580 set_result(result); 5581 return true; 5582 } 5583 5584 //-------------inline_montgomeryMultiply----------------------------------- 5585 bool LibraryCallKit::inline_montgomeryMultiply() { 5586 address stubAddr = StubRoutines::montgomeryMultiply(); 5587 if (stubAddr == NULL) { 5588 return false; // Intrinsic's stub is not implemented on this platform 5589 } 5590 5591 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 5592 const char* stubName = "montgomery_multiply"; 5593 5594 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 5595 5596 Node* a = argument(0); 5597 Node* b = argument(1); 5598 Node* n = argument(2); 5599 Node* len = argument(3); 5600 Node* inv = argument(4); 5601 Node* m = argument(6); 5602 5603 a = access_resolve(a, ACCESS_READ); 5604 b = access_resolve(b, ACCESS_READ); 5605 n = access_resolve(n, ACCESS_READ); 5606 m = access_resolve(m, ACCESS_WRITE); 5607 5608 const Type* a_type = a->Value(&_gvn); 5609 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5610 const Type* b_type = b->Value(&_gvn); 5611 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5612 const Type* n_type = a->Value(&_gvn); 5613 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5614 const Type* m_type = a->Value(&_gvn); 5615 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5616 if (top_a == NULL || top_a->klass() == NULL || 5617 top_b == NULL || top_b->klass() == NULL || 5618 top_n == NULL || top_n->klass() == NULL || 5619 top_m == NULL || top_m->klass() == NULL) { 5620 // failed array check 5621 return false; 5622 } 5623 5624 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5625 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5626 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5627 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5628 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5629 return false; 5630 } 5631 5632 // Make the call 5633 { 5634 Node* a_start = array_element_address(a, intcon(0), a_elem); 5635 Node* b_start = array_element_address(b, intcon(0), b_elem); 5636 Node* n_start = array_element_address(n, intcon(0), n_elem); 5637 Node* m_start = array_element_address(m, intcon(0), m_elem); 5638 5639 Node* call = make_runtime_call(RC_LEAF, 5640 OptoRuntime::montgomeryMultiply_Type(), 5641 stubAddr, stubName, TypePtr::BOTTOM, 5642 a_start, b_start, n_start, len, inv, top(), 5643 m_start); 5644 set_result(m); 5645 } 5646 5647 return true; 5648 } 5649 5650 bool LibraryCallKit::inline_montgomerySquare() { 5651 address stubAddr = StubRoutines::montgomerySquare(); 5652 if (stubAddr == NULL) { 5653 return false; // Intrinsic's stub is not implemented on this platform 5654 } 5655 5656 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 5657 const char* stubName = "montgomery_square"; 5658 5659 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 5660 5661 Node* a = argument(0); 5662 Node* n = argument(1); 5663 Node* len = argument(2); 5664 Node* inv = argument(3); 5665 Node* m = argument(5); 5666 5667 a = access_resolve(a, ACCESS_READ); 5668 n = access_resolve(n, ACCESS_READ); 5669 m = access_resolve(m, ACCESS_WRITE); 5670 5671 const Type* a_type = a->Value(&_gvn); 5672 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5673 const Type* n_type = a->Value(&_gvn); 5674 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5675 const Type* m_type = a->Value(&_gvn); 5676 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5677 if (top_a == NULL || top_a->klass() == NULL || 5678 top_n == NULL || top_n->klass() == NULL || 5679 top_m == NULL || top_m->klass() == NULL) { 5680 // failed array check 5681 return false; 5682 } 5683 5684 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5685 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5686 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5687 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5688 return false; 5689 } 5690 5691 // Make the call 5692 { 5693 Node* a_start = array_element_address(a, intcon(0), a_elem); 5694 Node* n_start = array_element_address(n, intcon(0), n_elem); 5695 Node* m_start = array_element_address(m, intcon(0), m_elem); 5696 5697 Node* call = make_runtime_call(RC_LEAF, 5698 OptoRuntime::montgomerySquare_Type(), 5699 stubAddr, stubName, TypePtr::BOTTOM, 5700 a_start, n_start, len, inv, top(), 5701 m_start); 5702 set_result(m); 5703 } 5704 5705 return true; 5706 } 5707 5708 //-------------inline_vectorizedMismatch------------------------------ 5709 bool LibraryCallKit::inline_vectorizedMismatch() { 5710 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform"); 5711 5712 address stubAddr = StubRoutines::vectorizedMismatch(); 5713 if (stubAddr == NULL) { 5714 return false; // Intrinsic's stub is not implemented on this platform 5715 } 5716 const char* stubName = "vectorizedMismatch"; 5717 int size_l = callee()->signature()->size(); 5718 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 5719 5720 Node* obja = argument(0); 5721 Node* aoffset = argument(1); 5722 Node* objb = argument(3); 5723 Node* boffset = argument(4); 5724 Node* length = argument(6); 5725 Node* scale = argument(7); 5726 5727 const Type* a_type = obja->Value(&_gvn); 5728 const Type* b_type = objb->Value(&_gvn); 5729 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5730 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5731 if (top_a == NULL || top_a->klass() == NULL || 5732 top_b == NULL || top_b->klass() == NULL) { 5733 // failed array check 5734 return false; 5735 } 5736 5737 Node* call; 5738 jvms()->set_should_reexecute(true); 5739 5740 obja = access_resolve(obja, ACCESS_READ); 5741 objb = access_resolve(objb, ACCESS_READ); 5742 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ); 5743 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ); 5744 5745 call = make_runtime_call(RC_LEAF, 5746 OptoRuntime::vectorizedMismatch_Type(), 5747 stubAddr, stubName, TypePtr::BOTTOM, 5748 obja_adr, objb_adr, length, scale); 5749 5750 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5751 set_result(result); 5752 return true; 5753 } 5754 5755 /** 5756 * Calculate CRC32 for byte. 5757 * int java.util.zip.CRC32.update(int crc, int b) 5758 */ 5759 bool LibraryCallKit::inline_updateCRC32() { 5760 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5761 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 5762 // no receiver since it is static method 5763 Node* crc = argument(0); // type: int 5764 Node* b = argument(1); // type: int 5765 5766 /* 5767 * int c = ~ crc; 5768 * b = timesXtoThe32[(b ^ c) & 0xFF]; 5769 * b = b ^ (c >>> 8); 5770 * crc = ~b; 5771 */ 5772 5773 Node* M1 = intcon(-1); 5774 crc = _gvn.transform(new XorINode(crc, M1)); 5775 Node* result = _gvn.transform(new XorINode(crc, b)); 5776 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 5777 5778 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 5779 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 5780 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 5781 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 5782 5783 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 5784 result = _gvn.transform(new XorINode(crc, result)); 5785 result = _gvn.transform(new XorINode(result, M1)); 5786 set_result(result); 5787 return true; 5788 } 5789 5790 /** 5791 * Calculate CRC32 for byte[] array. 5792 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 5793 */ 5794 bool LibraryCallKit::inline_updateBytesCRC32() { 5795 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5796 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5797 // no receiver since it is static method 5798 Node* crc = argument(0); // type: int 5799 Node* src = argument(1); // type: oop 5800 Node* offset = argument(2); // type: int 5801 Node* length = argument(3); // type: int 5802 5803 const Type* src_type = src->Value(&_gvn); 5804 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5805 if (top_src == NULL || top_src->klass() == NULL) { 5806 // failed array check 5807 return false; 5808 } 5809 5810 // Figure out the size and type of the elements we will be copying. 5811 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5812 if (src_elem != T_BYTE) { 5813 return false; 5814 } 5815 5816 // 'src_start' points to src array + scaled offset 5817 src = must_be_not_null(src, true); 5818 src = access_resolve(src, ACCESS_READ); 5819 Node* src_start = array_element_address(src, offset, src_elem); 5820 5821 // We assume that range check is done by caller. 5822 // TODO: generate range check (offset+length < src.length) in debug VM. 5823 5824 // Call the stub. 5825 address stubAddr = StubRoutines::updateBytesCRC32(); 5826 const char *stubName = "updateBytesCRC32"; 5827 5828 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5829 stubAddr, stubName, TypePtr::BOTTOM, 5830 crc, src_start, length); 5831 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5832 set_result(result); 5833 return true; 5834 } 5835 5836 /** 5837 * Calculate CRC32 for ByteBuffer. 5838 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 5839 */ 5840 bool LibraryCallKit::inline_updateByteBufferCRC32() { 5841 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5842 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5843 // no receiver since it is static method 5844 Node* crc = argument(0); // type: int 5845 Node* src = argument(1); // type: long 5846 Node* offset = argument(3); // type: int 5847 Node* length = argument(4); // type: int 5848 5849 src = ConvL2X(src); // adjust Java long to machine word 5850 Node* base = _gvn.transform(new CastX2PNode(src)); 5851 offset = ConvI2X(offset); 5852 5853 // 'src_start' points to src array + scaled offset 5854 Node* src_start = basic_plus_adr(top(), base, offset); 5855 5856 // Call the stub. 5857 address stubAddr = StubRoutines::updateBytesCRC32(); 5858 const char *stubName = "updateBytesCRC32"; 5859 5860 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5861 stubAddr, stubName, TypePtr::BOTTOM, 5862 crc, src_start, length); 5863 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5864 set_result(result); 5865 return true; 5866 } 5867 5868 //------------------------------get_table_from_crc32c_class----------------------- 5869 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 5870 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class); 5871 assert (table != NULL, "wrong version of java.util.zip.CRC32C"); 5872 5873 return table; 5874 } 5875 5876 //------------------------------inline_updateBytesCRC32C----------------------- 5877 // 5878 // Calculate CRC32C for byte[] array. 5879 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 5880 // 5881 bool LibraryCallKit::inline_updateBytesCRC32C() { 5882 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5883 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5884 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5885 // no receiver since it is a static method 5886 Node* crc = argument(0); // type: int 5887 Node* src = argument(1); // type: oop 5888 Node* offset = argument(2); // type: int 5889 Node* end = argument(3); // type: int 5890 5891 Node* length = _gvn.transform(new SubINode(end, offset)); 5892 5893 const Type* src_type = src->Value(&_gvn); 5894 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5895 if (top_src == NULL || top_src->klass() == NULL) { 5896 // failed array check 5897 return false; 5898 } 5899 5900 // Figure out the size and type of the elements we will be copying. 5901 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5902 if (src_elem != T_BYTE) { 5903 return false; 5904 } 5905 5906 // 'src_start' points to src array + scaled offset 5907 src = must_be_not_null(src, true); 5908 src = access_resolve(src, ACCESS_READ); 5909 Node* src_start = array_element_address(src, offset, src_elem); 5910 5911 // static final int[] byteTable in class CRC32C 5912 Node* table = get_table_from_crc32c_class(callee()->holder()); 5913 table = must_be_not_null(table, true); 5914 table = access_resolve(table, ACCESS_READ); 5915 Node* table_start = array_element_address(table, intcon(0), T_INT); 5916 5917 // We assume that range check is done by caller. 5918 // TODO: generate range check (offset+length < src.length) in debug VM. 5919 5920 // Call the stub. 5921 address stubAddr = StubRoutines::updateBytesCRC32C(); 5922 const char *stubName = "updateBytesCRC32C"; 5923 5924 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5925 stubAddr, stubName, TypePtr::BOTTOM, 5926 crc, src_start, length, table_start); 5927 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5928 set_result(result); 5929 return true; 5930 } 5931 5932 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 5933 // 5934 // Calculate CRC32C for DirectByteBuffer. 5935 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 5936 // 5937 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 5938 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5939 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 5940 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5941 // no receiver since it is a static method 5942 Node* crc = argument(0); // type: int 5943 Node* src = argument(1); // type: long 5944 Node* offset = argument(3); // type: int 5945 Node* end = argument(4); // type: int 5946 5947 Node* length = _gvn.transform(new SubINode(end, offset)); 5948 5949 src = ConvL2X(src); // adjust Java long to machine word 5950 Node* base = _gvn.transform(new CastX2PNode(src)); 5951 offset = ConvI2X(offset); 5952 5953 // 'src_start' points to src array + scaled offset 5954 Node* src_start = basic_plus_adr(top(), base, offset); 5955 5956 // static final int[] byteTable in class CRC32C 5957 Node* table = get_table_from_crc32c_class(callee()->holder()); 5958 table = must_be_not_null(table, true); 5959 table = access_resolve(table, ACCESS_READ); 5960 Node* table_start = array_element_address(table, intcon(0), T_INT); 5961 5962 // Call the stub. 5963 address stubAddr = StubRoutines::updateBytesCRC32C(); 5964 const char *stubName = "updateBytesCRC32C"; 5965 5966 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5967 stubAddr, stubName, TypePtr::BOTTOM, 5968 crc, src_start, length, table_start); 5969 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5970 set_result(result); 5971 return true; 5972 } 5973 5974 //------------------------------inline_updateBytesAdler32---------------------- 5975 // 5976 // Calculate Adler32 checksum for byte[] array. 5977 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 5978 // 5979 bool LibraryCallKit::inline_updateBytesAdler32() { 5980 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 5981 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5982 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 5983 // no receiver since it is static method 5984 Node* crc = argument(0); // type: int 5985 Node* src = argument(1); // type: oop 5986 Node* offset = argument(2); // type: int 5987 Node* length = argument(3); // type: int 5988 5989 const Type* src_type = src->Value(&_gvn); 5990 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5991 if (top_src == NULL || top_src->klass() == NULL) { 5992 // failed array check 5993 return false; 5994 } 5995 5996 // Figure out the size and type of the elements we will be copying. 5997 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5998 if (src_elem != T_BYTE) { 5999 return false; 6000 } 6001 6002 // 'src_start' points to src array + scaled offset 6003 src = access_resolve(src, ACCESS_READ); 6004 Node* src_start = array_element_address(src, offset, src_elem); 6005 6006 // We assume that range check is done by caller. 6007 // TODO: generate range check (offset+length < src.length) in debug VM. 6008 6009 // Call the stub. 6010 address stubAddr = StubRoutines::updateBytesAdler32(); 6011 const char *stubName = "updateBytesAdler32"; 6012 6013 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6014 stubAddr, stubName, TypePtr::BOTTOM, 6015 crc, src_start, length); 6016 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6017 set_result(result); 6018 return true; 6019 } 6020 6021 //------------------------------inline_updateByteBufferAdler32--------------- 6022 // 6023 // Calculate Adler32 checksum for DirectByteBuffer. 6024 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 6025 // 6026 bool LibraryCallKit::inline_updateByteBufferAdler32() { 6027 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 6028 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 6029 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 6030 // no receiver since it is static method 6031 Node* crc = argument(0); // type: int 6032 Node* src = argument(1); // type: long 6033 Node* offset = argument(3); // type: int 6034 Node* length = argument(4); // type: int 6035 6036 src = ConvL2X(src); // adjust Java long to machine word 6037 Node* base = _gvn.transform(new CastX2PNode(src)); 6038 offset = ConvI2X(offset); 6039 6040 // 'src_start' points to src array + scaled offset 6041 Node* src_start = basic_plus_adr(top(), base, offset); 6042 6043 // Call the stub. 6044 address stubAddr = StubRoutines::updateBytesAdler32(); 6045 const char *stubName = "updateBytesAdler32"; 6046 6047 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6048 stubAddr, stubName, TypePtr::BOTTOM, 6049 crc, src_start, length); 6050 6051 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6052 set_result(result); 6053 return true; 6054 } 6055 6056 //----------------------------inline_reference_get---------------------------- 6057 // public T java.lang.ref.Reference.get(); 6058 bool LibraryCallKit::inline_reference_get() { 6059 const int referent_offset = java_lang_ref_Reference::referent_offset; 6060 guarantee(referent_offset > 0, "should have already been set"); 6061 6062 // Get the argument: 6063 Node* reference_obj = null_check_receiver(); 6064 if (stopped()) return true; 6065 6066 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr(); 6067 assert(tinst != NULL, "obj is null"); 6068 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 6069 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass(); 6070 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"), 6071 ciSymbol::make("Ljava/lang/Object;"), 6072 false); 6073 assert (field != NULL, "undefined field"); 6074 6075 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset); 6076 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 6077 6078 ciInstanceKlass* klass = env()->Object_klass(); 6079 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass); 6080 6081 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF; 6082 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators); 6083 // Add memory barrier to prevent commoning reads from this field 6084 // across safepoint since GC can change its value. 6085 insert_mem_bar(Op_MemBarCPUOrder); 6086 6087 set_result(result); 6088 return true; 6089 } 6090 6091 6092 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 6093 bool is_exact=true, bool is_static=false, 6094 ciInstanceKlass * fromKls=NULL) { 6095 if (fromKls == NULL) { 6096 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 6097 assert(tinst != NULL, "obj is null"); 6098 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 6099 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 6100 fromKls = tinst->klass()->as_instance_klass(); 6101 } else { 6102 assert(is_static, "only for static field access"); 6103 } 6104 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 6105 ciSymbol::make(fieldTypeString), 6106 is_static); 6107 6108 assert (field != NULL, "undefined field"); 6109 if (field == NULL) return (Node *) NULL; 6110 6111 if (is_static) { 6112 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 6113 fromObj = makecon(tip); 6114 } 6115 6116 // Next code copied from Parse::do_get_xxx(): 6117 6118 // Compute address and memory type. 6119 int offset = field->offset_in_bytes(); 6120 bool is_vol = field->is_volatile(); 6121 ciType* field_klass = field->type(); 6122 assert(field_klass->is_loaded(), "should be loaded"); 6123 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 6124 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 6125 BasicType bt = field->layout_type(); 6126 6127 // Build the resultant type of the load 6128 const Type *type; 6129 if (bt == T_OBJECT) { 6130 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 6131 } else { 6132 type = Type::get_const_basic_type(bt); 6133 } 6134 6135 DecoratorSet decorators = IN_HEAP; 6136 6137 if (is_vol) { 6138 decorators |= MO_SEQ_CST; 6139 } 6140 6141 return access_load_at(fromObj, adr, adr_type, type, bt, decorators); 6142 } 6143 6144 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 6145 bool is_exact = true, bool is_static = false, 6146 ciInstanceKlass * fromKls = NULL) { 6147 if (fromKls == NULL) { 6148 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 6149 assert(tinst != NULL, "obj is null"); 6150 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 6151 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 6152 fromKls = tinst->klass()->as_instance_klass(); 6153 } 6154 else { 6155 assert(is_static, "only for static field access"); 6156 } 6157 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 6158 ciSymbol::make(fieldTypeString), 6159 is_static); 6160 6161 assert(field != NULL, "undefined field"); 6162 assert(!field->is_volatile(), "not defined for volatile fields"); 6163 6164 if (is_static) { 6165 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 6166 fromObj = makecon(tip); 6167 } 6168 6169 // Next code copied from Parse::do_get_xxx(): 6170 6171 // Compute address and memory type. 6172 int offset = field->offset_in_bytes(); 6173 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 6174 6175 return adr; 6176 } 6177 6178 //------------------------------inline_aescrypt_Block----------------------- 6179 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 6180 address stubAddr = NULL; 6181 const char *stubName; 6182 assert(UseAES, "need AES instruction support"); 6183 6184 switch(id) { 6185 case vmIntrinsics::_aescrypt_encryptBlock: 6186 stubAddr = StubRoutines::aescrypt_encryptBlock(); 6187 stubName = "aescrypt_encryptBlock"; 6188 break; 6189 case vmIntrinsics::_aescrypt_decryptBlock: 6190 stubAddr = StubRoutines::aescrypt_decryptBlock(); 6191 stubName = "aescrypt_decryptBlock"; 6192 break; 6193 default: 6194 break; 6195 } 6196 if (stubAddr == NULL) return false; 6197 6198 Node* aescrypt_object = argument(0); 6199 Node* src = argument(1); 6200 Node* src_offset = argument(2); 6201 Node* dest = argument(3); 6202 Node* dest_offset = argument(4); 6203 6204 src = must_be_not_null(src, true); 6205 dest = must_be_not_null(dest, true); 6206 6207 src = access_resolve(src, ACCESS_READ); 6208 dest = access_resolve(dest, ACCESS_WRITE); 6209 6210 // (1) src and dest are arrays. 6211 const Type* src_type = src->Value(&_gvn); 6212 const Type* dest_type = dest->Value(&_gvn); 6213 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6214 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6215 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6216 6217 // for the quick and dirty code we will skip all the checks. 6218 // we are just trying to get the call to be generated. 6219 Node* src_start = src; 6220 Node* dest_start = dest; 6221 if (src_offset != NULL || dest_offset != NULL) { 6222 assert(src_offset != NULL && dest_offset != NULL, ""); 6223 src_start = array_element_address(src, src_offset, T_BYTE); 6224 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6225 } 6226 6227 // now need to get the start of its expanded key array 6228 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6229 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6230 if (k_start == NULL) return false; 6231 6232 if (Matcher::pass_original_key_for_aes()) { 6233 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 6234 // compatibility issues between Java key expansion and SPARC crypto instructions 6235 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 6236 if (original_k_start == NULL) return false; 6237 6238 // Call the stub. 6239 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 6240 stubAddr, stubName, TypePtr::BOTTOM, 6241 src_start, dest_start, k_start, original_k_start); 6242 } else { 6243 // Call the stub. 6244 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 6245 stubAddr, stubName, TypePtr::BOTTOM, 6246 src_start, dest_start, k_start); 6247 } 6248 6249 return true; 6250 } 6251 6252 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 6253 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 6254 address stubAddr = NULL; 6255 const char *stubName = NULL; 6256 6257 assert(UseAES, "need AES instruction support"); 6258 6259 switch(id) { 6260 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 6261 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 6262 stubName = "cipherBlockChaining_encryptAESCrypt"; 6263 break; 6264 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 6265 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 6266 stubName = "cipherBlockChaining_decryptAESCrypt"; 6267 break; 6268 default: 6269 break; 6270 } 6271 if (stubAddr == NULL) return false; 6272 6273 Node* cipherBlockChaining_object = argument(0); 6274 Node* src = argument(1); 6275 Node* src_offset = argument(2); 6276 Node* len = argument(3); 6277 Node* dest = argument(4); 6278 Node* dest_offset = argument(5); 6279 6280 src = must_be_not_null(src, false); 6281 dest = must_be_not_null(dest, false); 6282 6283 src = access_resolve(src, ACCESS_READ); 6284 dest = access_resolve(dest, ACCESS_WRITE); 6285 6286 // (1) src and dest are arrays. 6287 const Type* src_type = src->Value(&_gvn); 6288 const Type* dest_type = dest->Value(&_gvn); 6289 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6290 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6291 assert (top_src != NULL && top_src->klass() != NULL 6292 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6293 6294 // checks are the responsibility of the caller 6295 Node* src_start = src; 6296 Node* dest_start = dest; 6297 if (src_offset != NULL || dest_offset != NULL) { 6298 assert(src_offset != NULL && dest_offset != NULL, ""); 6299 src_start = array_element_address(src, src_offset, T_BYTE); 6300 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6301 } 6302 6303 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6304 // (because of the predicated logic executed earlier). 6305 // so we cast it here safely. 6306 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6307 6308 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6309 if (embeddedCipherObj == NULL) return false; 6310 6311 // cast it to what we know it will be at runtime 6312 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 6313 assert(tinst != NULL, "CBC obj is null"); 6314 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded"); 6315 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6316 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6317 6318 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6319 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6320 const TypeOopPtr* xtype = aklass->as_instance_type(); 6321 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6322 aescrypt_object = _gvn.transform(aescrypt_object); 6323 6324 // we need to get the start of the aescrypt_object's expanded key array 6325 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6326 if (k_start == NULL) return false; 6327 6328 // similarly, get the start address of the r vector 6329 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false); 6330 if (objRvec == NULL) return false; 6331 objRvec = access_resolve(objRvec, ACCESS_WRITE); 6332 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 6333 6334 Node* cbcCrypt; 6335 if (Matcher::pass_original_key_for_aes()) { 6336 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 6337 // compatibility issues between Java key expansion and SPARC crypto instructions 6338 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 6339 if (original_k_start == NULL) return false; 6340 6341 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start 6342 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6343 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 6344 stubAddr, stubName, TypePtr::BOTTOM, 6345 src_start, dest_start, k_start, r_start, len, original_k_start); 6346 } else { 6347 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6348 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6349 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 6350 stubAddr, stubName, TypePtr::BOTTOM, 6351 src_start, dest_start, k_start, r_start, len); 6352 } 6353 6354 // return cipher length (int) 6355 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 6356 set_result(retvalue); 6357 return true; 6358 } 6359 6360 //------------------------------inline_electronicCodeBook_AESCrypt----------------------- 6361 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) { 6362 address stubAddr = NULL; 6363 const char *stubName = NULL; 6364 6365 assert(UseAES, "need AES instruction support"); 6366 6367 switch (id) { 6368 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 6369 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt(); 6370 stubName = "electronicCodeBook_encryptAESCrypt"; 6371 break; 6372 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 6373 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt(); 6374 stubName = "electronicCodeBook_decryptAESCrypt"; 6375 break; 6376 default: 6377 break; 6378 } 6379 6380 if (stubAddr == NULL) return false; 6381 6382 Node* electronicCodeBook_object = argument(0); 6383 Node* src = argument(1); 6384 Node* src_offset = argument(2); 6385 Node* len = argument(3); 6386 Node* dest = argument(4); 6387 Node* dest_offset = argument(5); 6388 6389 // (1) src and dest are arrays. 6390 const Type* src_type = src->Value(&_gvn); 6391 const Type* dest_type = dest->Value(&_gvn); 6392 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6393 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6394 assert(top_src != NULL && top_src->klass() != NULL 6395 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6396 6397 // checks are the responsibility of the caller 6398 Node* src_start = src; 6399 Node* dest_start = dest; 6400 if (src_offset != NULL || dest_offset != NULL) { 6401 assert(src_offset != NULL && dest_offset != NULL, ""); 6402 src_start = array_element_address(src, src_offset, T_BYTE); 6403 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6404 } 6405 6406 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6407 // (because of the predicated logic executed earlier). 6408 // so we cast it here safely. 6409 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6410 6411 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6412 if (embeddedCipherObj == NULL) return false; 6413 6414 // cast it to what we know it will be at runtime 6415 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr(); 6416 assert(tinst != NULL, "ECB obj is null"); 6417 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded"); 6418 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6419 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6420 6421 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6422 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6423 const TypeOopPtr* xtype = aklass->as_instance_type(); 6424 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6425 aescrypt_object = _gvn.transform(aescrypt_object); 6426 6427 // we need to get the start of the aescrypt_object's expanded key array 6428 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6429 if (k_start == NULL) return false; 6430 6431 Node* ecbCrypt; 6432 if (Matcher::pass_original_key_for_aes()) { 6433 // no SPARC version for AES/ECB intrinsics now. 6434 return false; 6435 } 6436 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6437 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP, 6438 OptoRuntime::electronicCodeBook_aescrypt_Type(), 6439 stubAddr, stubName, TypePtr::BOTTOM, 6440 src_start, dest_start, k_start, len); 6441 6442 // return cipher length (int) 6443 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms)); 6444 set_result(retvalue); 6445 return true; 6446 } 6447 6448 //------------------------------inline_counterMode_AESCrypt----------------------- 6449 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 6450 assert(UseAES, "need AES instruction support"); 6451 if (!UseAESCTRIntrinsics) return false; 6452 6453 address stubAddr = NULL; 6454 const char *stubName = NULL; 6455 if (id == vmIntrinsics::_counterMode_AESCrypt) { 6456 stubAddr = StubRoutines::counterMode_AESCrypt(); 6457 stubName = "counterMode_AESCrypt"; 6458 } 6459 if (stubAddr == NULL) return false; 6460 6461 Node* counterMode_object = argument(0); 6462 Node* src = argument(1); 6463 Node* src_offset = argument(2); 6464 Node* len = argument(3); 6465 Node* dest = argument(4); 6466 Node* dest_offset = argument(5); 6467 6468 src = access_resolve(src, ACCESS_READ); 6469 dest = access_resolve(dest, ACCESS_WRITE); 6470 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE); 6471 6472 // (1) src and dest are arrays. 6473 const Type* src_type = src->Value(&_gvn); 6474 const Type* dest_type = dest->Value(&_gvn); 6475 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6476 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6477 assert(top_src != NULL && top_src->klass() != NULL && 6478 top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6479 6480 // checks are the responsibility of the caller 6481 Node* src_start = src; 6482 Node* dest_start = dest; 6483 if (src_offset != NULL || dest_offset != NULL) { 6484 assert(src_offset != NULL && dest_offset != NULL, ""); 6485 src_start = array_element_address(src, src_offset, T_BYTE); 6486 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6487 } 6488 6489 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6490 // (because of the predicated logic executed earlier). 6491 // so we cast it here safely. 6492 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6493 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6494 if (embeddedCipherObj == NULL) return false; 6495 // cast it to what we know it will be at runtime 6496 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 6497 assert(tinst != NULL, "CTR obj is null"); 6498 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded"); 6499 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6500 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6501 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6502 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6503 const TypeOopPtr* xtype = aklass->as_instance_type(); 6504 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6505 aescrypt_object = _gvn.transform(aescrypt_object); 6506 // we need to get the start of the aescrypt_object's expanded key array 6507 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6508 if (k_start == NULL) return false; 6509 // similarly, get the start address of the r vector 6510 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false); 6511 if (obj_counter == NULL) return false; 6512 obj_counter = access_resolve(obj_counter, ACCESS_WRITE); 6513 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 6514 6515 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false); 6516 if (saved_encCounter == NULL) return false; 6517 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE); 6518 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 6519 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 6520 6521 Node* ctrCrypt; 6522 if (Matcher::pass_original_key_for_aes()) { 6523 // no SPARC version for AES/CTR intrinsics now. 6524 return false; 6525 } 6526 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6527 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6528 OptoRuntime::counterMode_aescrypt_Type(), 6529 stubAddr, stubName, TypePtr::BOTTOM, 6530 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 6531 6532 // return cipher length (int) 6533 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 6534 set_result(retvalue); 6535 return true; 6536 } 6537 6538 //------------------------------get_key_start_from_aescrypt_object----------------------- 6539 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 6540 #if defined(PPC64) || defined(S390) 6541 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys. 6542 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns. 6543 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption. 6544 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]). 6545 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false); 6546 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6547 if (objSessionK == NULL) { 6548 return (Node *) NULL; 6549 } 6550 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS); 6551 #else 6552 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false); 6553 #endif // PPC64 6554 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6555 if (objAESCryptKey == NULL) return (Node *) NULL; 6556 6557 // now have the array, need to get the start address of the K array 6558 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ); 6559 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 6560 return k_start; 6561 } 6562 6563 //------------------------------get_original_key_start_from_aescrypt_object----------------------- 6564 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) { 6565 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false); 6566 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6567 if (objAESCryptKey == NULL) return (Node *) NULL; 6568 6569 // now have the array, need to get the start address of the lastKey array 6570 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ); 6571 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE); 6572 return original_k_start; 6573 } 6574 6575 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 6576 // Return node representing slow path of predicate check. 6577 // the pseudo code we want to emulate with this predicate is: 6578 // for encryption: 6579 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6580 // for decryption: 6581 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6582 // note cipher==plain is more conservative than the original java code but that's OK 6583 // 6584 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 6585 // The receiver was checked for NULL already. 6586 Node* objCBC = argument(0); 6587 6588 Node* src = argument(1); 6589 Node* dest = argument(4); 6590 6591 // Load embeddedCipher field of CipherBlockChaining object. 6592 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6593 6594 // get AESCrypt klass for instanceOf check 6595 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6596 // will have same classloader as CipherBlockChaining object 6597 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 6598 assert(tinst != NULL, "CBCobj is null"); 6599 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded"); 6600 6601 // we want to do an instanceof comparison against the AESCrypt class 6602 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6603 if (!klass_AESCrypt->is_loaded()) { 6604 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6605 Node* ctrl = control(); 6606 set_control(top()); // no regular fast path 6607 return ctrl; 6608 } 6609 6610 src = must_be_not_null(src, true); 6611 dest = must_be_not_null(dest, true); 6612 6613 // Resolve oops to stable for CmpP below. 6614 src = access_resolve(src, 0); 6615 dest = access_resolve(dest, 0); 6616 6617 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6618 6619 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6620 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6621 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6622 6623 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6624 6625 // for encryption, we are done 6626 if (!decrypting) 6627 return instof_false; // even if it is NULL 6628 6629 // for decryption, we need to add a further check to avoid 6630 // taking the intrinsic path when cipher and plain are the same 6631 // see the original java code for why. 6632 RegionNode* region = new RegionNode(3); 6633 region->init_req(1, instof_false); 6634 6635 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6636 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6637 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6638 region->init_req(2, src_dest_conjoint); 6639 6640 record_for_igvn(region); 6641 return _gvn.transform(region); 6642 } 6643 6644 //----------------------------inline_electronicCodeBook_AESCrypt_predicate---------------------------- 6645 // Return node representing slow path of predicate check. 6646 // the pseudo code we want to emulate with this predicate is: 6647 // for encryption: 6648 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6649 // for decryption: 6650 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6651 // note cipher==plain is more conservative than the original java code but that's OK 6652 // 6653 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) { 6654 // The receiver was checked for NULL already. 6655 Node* objECB = argument(0); 6656 6657 // Load embeddedCipher field of ElectronicCodeBook object. 6658 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6659 6660 // get AESCrypt klass for instanceOf check 6661 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6662 // will have same classloader as ElectronicCodeBook object 6663 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr(); 6664 assert(tinst != NULL, "ECBobj is null"); 6665 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded"); 6666 6667 // we want to do an instanceof comparison against the AESCrypt class 6668 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6669 if (!klass_AESCrypt->is_loaded()) { 6670 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6671 Node* ctrl = control(); 6672 set_control(top()); // no regular fast path 6673 return ctrl; 6674 } 6675 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6676 6677 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6678 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6679 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6680 6681 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6682 6683 // for encryption, we are done 6684 if (!decrypting) 6685 return instof_false; // even if it is NULL 6686 6687 // for decryption, we need to add a further check to avoid 6688 // taking the intrinsic path when cipher and plain are the same 6689 // see the original java code for why. 6690 RegionNode* region = new RegionNode(3); 6691 region->init_req(1, instof_false); 6692 Node* src = argument(1); 6693 Node* dest = argument(4); 6694 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6695 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6696 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6697 region->init_req(2, src_dest_conjoint); 6698 6699 record_for_igvn(region); 6700 return _gvn.transform(region); 6701 } 6702 6703 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 6704 // Return node representing slow path of predicate check. 6705 // the pseudo code we want to emulate with this predicate is: 6706 // for encryption: 6707 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6708 // for decryption: 6709 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6710 // note cipher==plain is more conservative than the original java code but that's OK 6711 // 6712 6713 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 6714 // The receiver was checked for NULL already. 6715 Node* objCTR = argument(0); 6716 6717 // Load embeddedCipher field of CipherBlockChaining object. 6718 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6719 6720 // get AESCrypt klass for instanceOf check 6721 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6722 // will have same classloader as CipherBlockChaining object 6723 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 6724 assert(tinst != NULL, "CTRobj is null"); 6725 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded"); 6726 6727 // we want to do an instanceof comparison against the AESCrypt class 6728 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6729 if (!klass_AESCrypt->is_loaded()) { 6730 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6731 Node* ctrl = control(); 6732 set_control(top()); // no regular fast path 6733 return ctrl; 6734 } 6735 6736 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6737 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6738 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6739 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6740 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6741 6742 return instof_false; // even if it is NULL 6743 } 6744 6745 //------------------------------inline_ghash_processBlocks 6746 bool LibraryCallKit::inline_ghash_processBlocks() { 6747 address stubAddr; 6748 const char *stubName; 6749 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 6750 6751 stubAddr = StubRoutines::ghash_processBlocks(); 6752 stubName = "ghash_processBlocks"; 6753 6754 Node* data = argument(0); 6755 Node* offset = argument(1); 6756 Node* len = argument(2); 6757 Node* state = argument(3); 6758 Node* subkeyH = argument(4); 6759 6760 state = must_be_not_null(state, true); 6761 subkeyH = must_be_not_null(subkeyH, true); 6762 data = must_be_not_null(data, true); 6763 6764 state = access_resolve(state, ACCESS_WRITE); 6765 subkeyH = access_resolve(subkeyH, ACCESS_READ); 6766 data = access_resolve(data, ACCESS_READ); 6767 6768 Node* state_start = array_element_address(state, intcon(0), T_LONG); 6769 assert(state_start, "state is NULL"); 6770 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 6771 assert(subkeyH_start, "subkeyH is NULL"); 6772 Node* data_start = array_element_address(data, offset, T_BYTE); 6773 assert(data_start, "data is NULL"); 6774 6775 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 6776 OptoRuntime::ghash_processBlocks_Type(), 6777 stubAddr, stubName, TypePtr::BOTTOM, 6778 state_start, subkeyH_start, data_start, len); 6779 return true; 6780 } 6781 6782 bool LibraryCallKit::inline_base64_encodeBlock() { 6783 address stubAddr; 6784 const char *stubName; 6785 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 6786 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters"); 6787 stubAddr = StubRoutines::base64_encodeBlock(); 6788 stubName = "encodeBlock"; 6789 6790 if (!stubAddr) return false; 6791 Node* base64obj = argument(0); 6792 Node* src = argument(1); 6793 Node* offset = argument(2); 6794 Node* len = argument(3); 6795 Node* dest = argument(4); 6796 Node* dp = argument(5); 6797 Node* isURL = argument(6); 6798 6799 src = must_be_not_null(src, true); 6800 src = access_resolve(src, ACCESS_READ); 6801 dest = must_be_not_null(dest, true); 6802 dest = access_resolve(dest, ACCESS_WRITE); 6803 6804 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 6805 assert(src_start, "source array is NULL"); 6806 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 6807 assert(dest_start, "destination array is NULL"); 6808 6809 Node* base64 = make_runtime_call(RC_LEAF, 6810 OptoRuntime::base64_encodeBlock_Type(), 6811 stubAddr, stubName, TypePtr::BOTTOM, 6812 src_start, offset, len, dest_start, dp, isURL); 6813 return true; 6814 } 6815 6816 //------------------------------inline_sha_implCompress----------------------- 6817 // 6818 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 6819 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 6820 // 6821 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 6822 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 6823 // 6824 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 6825 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 6826 // 6827 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) { 6828 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 6829 6830 Node* sha_obj = argument(0); 6831 Node* src = argument(1); // type oop 6832 Node* ofs = argument(2); // type int 6833 6834 const Type* src_type = src->Value(&_gvn); 6835 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6836 if (top_src == NULL || top_src->klass() == NULL) { 6837 // failed array check 6838 return false; 6839 } 6840 // Figure out the size and type of the elements we will be copying. 6841 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6842 if (src_elem != T_BYTE) { 6843 return false; 6844 } 6845 // 'src_start' points to src array + offset 6846 src = must_be_not_null(src, true); 6847 src = access_resolve(src, ACCESS_READ); 6848 Node* src_start = array_element_address(src, ofs, src_elem); 6849 Node* state = NULL; 6850 address stubAddr; 6851 const char *stubName; 6852 6853 switch(id) { 6854 case vmIntrinsics::_sha_implCompress: 6855 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 6856 state = get_state_from_sha_object(sha_obj); 6857 stubAddr = StubRoutines::sha1_implCompress(); 6858 stubName = "sha1_implCompress"; 6859 break; 6860 case vmIntrinsics::_sha2_implCompress: 6861 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 6862 state = get_state_from_sha_object(sha_obj); 6863 stubAddr = StubRoutines::sha256_implCompress(); 6864 stubName = "sha256_implCompress"; 6865 break; 6866 case vmIntrinsics::_sha5_implCompress: 6867 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 6868 state = get_state_from_sha5_object(sha_obj); 6869 stubAddr = StubRoutines::sha512_implCompress(); 6870 stubName = "sha512_implCompress"; 6871 break; 6872 default: 6873 fatal_unexpected_iid(id); 6874 return false; 6875 } 6876 if (state == NULL) return false; 6877 6878 assert(stubAddr != NULL, "Stub is generated"); 6879 if (stubAddr == NULL) return false; 6880 6881 // Call the stub. 6882 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(), 6883 stubAddr, stubName, TypePtr::BOTTOM, 6884 src_start, state); 6885 6886 return true; 6887 } 6888 6889 //------------------------------inline_digestBase_implCompressMB----------------------- 6890 // 6891 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array. 6892 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 6893 // 6894 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 6895 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 6896 "need SHA1/SHA256/SHA512 instruction support"); 6897 assert((uint)predicate < 3, "sanity"); 6898 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 6899 6900 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already. 6901 Node* src = argument(1); // byte[] array 6902 Node* ofs = argument(2); // type int 6903 Node* limit = argument(3); // type int 6904 6905 const Type* src_type = src->Value(&_gvn); 6906 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6907 if (top_src == NULL || top_src->klass() == NULL) { 6908 // failed array check 6909 return false; 6910 } 6911 // Figure out the size and type of the elements we will be copying. 6912 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6913 if (src_elem != T_BYTE) { 6914 return false; 6915 } 6916 // 'src_start' points to src array + offset 6917 src = must_be_not_null(src, false); 6918 src = access_resolve(src, ACCESS_READ); 6919 Node* src_start = array_element_address(src, ofs, src_elem); 6920 6921 const char* klass_SHA_name = NULL; 6922 const char* stub_name = NULL; 6923 address stub_addr = NULL; 6924 bool long_state = false; 6925 6926 switch (predicate) { 6927 case 0: 6928 if (UseSHA1Intrinsics) { 6929 klass_SHA_name = "sun/security/provider/SHA"; 6930 stub_name = "sha1_implCompressMB"; 6931 stub_addr = StubRoutines::sha1_implCompressMB(); 6932 } 6933 break; 6934 case 1: 6935 if (UseSHA256Intrinsics) { 6936 klass_SHA_name = "sun/security/provider/SHA2"; 6937 stub_name = "sha256_implCompressMB"; 6938 stub_addr = StubRoutines::sha256_implCompressMB(); 6939 } 6940 break; 6941 case 2: 6942 if (UseSHA512Intrinsics) { 6943 klass_SHA_name = "sun/security/provider/SHA5"; 6944 stub_name = "sha512_implCompressMB"; 6945 stub_addr = StubRoutines::sha512_implCompressMB(); 6946 long_state = true; 6947 } 6948 break; 6949 default: 6950 fatal("unknown SHA intrinsic predicate: %d", predicate); 6951 } 6952 if (klass_SHA_name != NULL) { 6953 assert(stub_addr != NULL, "Stub is generated"); 6954 if (stub_addr == NULL) return false; 6955 6956 // get DigestBase klass to lookup for SHA klass 6957 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 6958 assert(tinst != NULL, "digestBase_obj is not instance???"); 6959 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 6960 6961 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 6962 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded"); 6963 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 6964 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit); 6965 } 6966 return false; 6967 } 6968 //------------------------------inline_sha_implCompressMB----------------------- 6969 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA, 6970 bool long_state, address stubAddr, const char *stubName, 6971 Node* src_start, Node* ofs, Node* limit) { 6972 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA); 6973 const TypeOopPtr* xtype = aklass->as_instance_type(); 6974 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 6975 sha_obj = _gvn.transform(sha_obj); 6976 6977 Node* state; 6978 if (long_state) { 6979 state = get_state_from_sha5_object(sha_obj); 6980 } else { 6981 state = get_state_from_sha_object(sha_obj); 6982 } 6983 if (state == NULL) return false; 6984 6985 // Call the stub. 6986 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6987 OptoRuntime::digestBase_implCompressMB_Type(), 6988 stubAddr, stubName, TypePtr::BOTTOM, 6989 src_start, state, ofs, limit); 6990 // return ofs (int) 6991 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6992 set_result(result); 6993 6994 return true; 6995 } 6996 6997 //------------------------------get_state_from_sha_object----------------------- 6998 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) { 6999 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false); 7000 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2"); 7001 if (sha_state == NULL) return (Node *) NULL; 7002 7003 // now have the array, need to get the start address of the state array 7004 sha_state = access_resolve(sha_state, ACCESS_WRITE); 7005 Node* state = array_element_address(sha_state, intcon(0), T_INT); 7006 return state; 7007 } 7008 7009 //------------------------------get_state_from_sha5_object----------------------- 7010 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) { 7011 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false); 7012 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5"); 7013 if (sha_state == NULL) return (Node *) NULL; 7014 7015 // now have the array, need to get the start address of the state array 7016 sha_state = access_resolve(sha_state, ACCESS_WRITE); 7017 Node* state = array_element_address(sha_state, intcon(0), T_LONG); 7018 return state; 7019 } 7020 7021 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 7022 // Return node representing slow path of predicate check. 7023 // the pseudo code we want to emulate with this predicate is: 7024 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath 7025 // 7026 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 7027 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 7028 "need SHA1/SHA256/SHA512 instruction support"); 7029 assert((uint)predicate < 3, "sanity"); 7030 7031 // The receiver was checked for NULL already. 7032 Node* digestBaseObj = argument(0); 7033 7034 // get DigestBase klass for instanceOf check 7035 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 7036 assert(tinst != NULL, "digestBaseObj is null"); 7037 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 7038 7039 const char* klass_SHA_name = NULL; 7040 switch (predicate) { 7041 case 0: 7042 if (UseSHA1Intrinsics) { 7043 // we want to do an instanceof comparison against the SHA class 7044 klass_SHA_name = "sun/security/provider/SHA"; 7045 } 7046 break; 7047 case 1: 7048 if (UseSHA256Intrinsics) { 7049 // we want to do an instanceof comparison against the SHA2 class 7050 klass_SHA_name = "sun/security/provider/SHA2"; 7051 } 7052 break; 7053 case 2: 7054 if (UseSHA512Intrinsics) { 7055 // we want to do an instanceof comparison against the SHA5 class 7056 klass_SHA_name = "sun/security/provider/SHA5"; 7057 } 7058 break; 7059 default: 7060 fatal("unknown SHA intrinsic predicate: %d", predicate); 7061 } 7062 7063 ciKlass* klass_SHA = NULL; 7064 if (klass_SHA_name != NULL) { 7065 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 7066 } 7067 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) { 7068 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 7069 Node* ctrl = control(); 7070 set_control(top()); // no intrinsic path 7071 return ctrl; 7072 } 7073 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 7074 7075 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA))); 7076 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1))); 7077 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7078 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 7079 7080 return instof_false; // even if it is NULL 7081 } 7082 7083 //-------------inline_fma----------------------------------- 7084 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) { 7085 Node *a = NULL; 7086 Node *b = NULL; 7087 Node *c = NULL; 7088 Node* result = NULL; 7089 switch (id) { 7090 case vmIntrinsics::_fmaD: 7091 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each."); 7092 // no receiver since it is static method 7093 a = round_double_node(argument(0)); 7094 b = round_double_node(argument(2)); 7095 c = round_double_node(argument(4)); 7096 result = _gvn.transform(new FmaDNode(control(), a, b, c)); 7097 break; 7098 case vmIntrinsics::_fmaF: 7099 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each."); 7100 a = argument(0); 7101 b = argument(1); 7102 c = argument(2); 7103 result = _gvn.transform(new FmaFNode(control(), a, b, c)); 7104 break; 7105 default: 7106 fatal_unexpected_iid(id); break; 7107 } 7108 set_result(result); 7109 return true; 7110 } 7111 7112 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) { 7113 // argument(0) is receiver 7114 Node* codePoint = argument(1); 7115 Node* n = NULL; 7116 7117 switch (id) { 7118 case vmIntrinsics::_isDigit : 7119 n = new DigitNode(control(), codePoint); 7120 break; 7121 case vmIntrinsics::_isLowerCase : 7122 n = new LowerCaseNode(control(), codePoint); 7123 break; 7124 case vmIntrinsics::_isUpperCase : 7125 n = new UpperCaseNode(control(), codePoint); 7126 break; 7127 case vmIntrinsics::_isWhitespace : 7128 n = new WhitespaceNode(control(), codePoint); 7129 break; 7130 default: 7131 fatal_unexpected_iid(id); 7132 } 7133 7134 set_result(_gvn.transform(n)); 7135 return true; 7136 } 7137 7138 //------------------------------inline_fp_min_max------------------------------ 7139 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) { 7140 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416. 7141 7142 // The intrinsic should be used only when the API branches aren't predictable, 7143 // the last one performing the most important comparison. The following heuristic 7144 // uses the branch statistics to eventually bail out if necessary. 7145 7146 ciMethodData *md = callee()->method_data(); 7147 7148 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) { 7149 ciCallProfile cp = caller()->call_profile_at_bci(bci()); 7150 7151 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) { 7152 // Bail out if the call-site didn't contribute enough to the statistics. 7153 return false; 7154 } 7155 7156 uint taken = 0, not_taken = 0; 7157 7158 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) { 7159 if (p->is_BranchData()) { 7160 taken = ((ciBranchData*)p)->taken(); 7161 not_taken = ((ciBranchData*)p)->not_taken(); 7162 } 7163 } 7164 7165 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count()); 7166 balance = balance < 0 ? -balance : balance; 7167 if ( balance > 0.2 ) { 7168 // Bail out if the most important branch is predictable enough. 7169 return false; 7170 } 7171 } 7172 */ 7173 7174 Node *a = NULL; 7175 Node *b = NULL; 7176 Node *n = NULL; 7177 switch (id) { 7178 case vmIntrinsics::_maxF: 7179 case vmIntrinsics::_minF: 7180 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each."); 7181 a = argument(0); 7182 b = argument(1); 7183 break; 7184 case vmIntrinsics::_maxD: 7185 case vmIntrinsics::_minD: 7186 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each."); 7187 a = round_double_node(argument(0)); 7188 b = round_double_node(argument(2)); 7189 break; 7190 default: 7191 fatal_unexpected_iid(id); 7192 break; 7193 } 7194 switch (id) { 7195 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break; 7196 case vmIntrinsics::_minF: n = new MinFNode(a, b); break; 7197 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break; 7198 case vmIntrinsics::_minD: n = new MinDNode(a, b); break; 7199 default: fatal_unexpected_iid(id); break; 7200 } 7201 set_result(_gvn.transform(n)); 7202 return true; 7203 } 7204 7205 bool LibraryCallKit::inline_profileBoolean() { 7206 Node* counts = argument(1); 7207 const TypeAryPtr* ary = NULL; 7208 ciArray* aobj = NULL; 7209 if (counts->is_Con() 7210 && (ary = counts->bottom_type()->isa_aryptr()) != NULL 7211 && (aobj = ary->const_oop()->as_array()) != NULL 7212 && (aobj->length() == 2)) { 7213 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 7214 jint false_cnt = aobj->element_value(0).as_int(); 7215 jint true_cnt = aobj->element_value(1).as_int(); 7216 7217 if (C->log() != NULL) { 7218 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 7219 false_cnt, true_cnt); 7220 } 7221 7222 if (false_cnt + true_cnt == 0) { 7223 // According to profile, never executed. 7224 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 7225 Deoptimization::Action_reinterpret); 7226 return true; 7227 } 7228 7229 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 7230 // is a number of each value occurrences. 7231 Node* result = argument(0); 7232 if (false_cnt == 0 || true_cnt == 0) { 7233 // According to profile, one value has been never seen. 7234 int expected_val = (false_cnt == 0) ? 1 : 0; 7235 7236 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 7237 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 7238 7239 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 7240 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 7241 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 7242 7243 { // Slow path: uncommon trap for never seen value and then reexecute 7244 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 7245 // the value has been seen at least once. 7246 PreserveJVMState pjvms(this); 7247 PreserveReexecuteState preexecs(this); 7248 jvms()->set_should_reexecute(true); 7249 7250 set_control(slow_path); 7251 set_i_o(i_o()); 7252 7253 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 7254 Deoptimization::Action_reinterpret); 7255 } 7256 // The guard for never seen value enables sharpening of the result and 7257 // returning a constant. It allows to eliminate branches on the same value 7258 // later on. 7259 set_control(fast_path); 7260 result = intcon(expected_val); 7261 } 7262 // Stop profiling. 7263 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 7264 // By replacing method body with profile data (represented as ProfileBooleanNode 7265 // on IR level) we effectively disable profiling. 7266 // It enables full speed execution once optimized code is generated. 7267 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 7268 C->record_for_igvn(profile); 7269 set_result(profile); 7270 return true; 7271 } else { 7272 // Continue profiling. 7273 // Profile data isn't available at the moment. So, execute method's bytecode version. 7274 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 7275 // is compiled and counters aren't available since corresponding MethodHandle 7276 // isn't a compile-time constant. 7277 return false; 7278 } 7279 } 7280 7281 bool LibraryCallKit::inline_isCompileConstant() { 7282 Node* n = argument(0); 7283 set_result(n->is_Con() ? intcon(1) : intcon(0)); 7284 return true; 7285 }