1 /* 2 * Copyright (c) 1999, 2016, 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 "classfile/systemDictionary.hpp" 28 #include "classfile/vmSymbols.hpp" 29 #include "compiler/compileBroker.hpp" 30 #include "compiler/compileLog.hpp" 31 #include "oops/objArrayKlass.hpp" 32 #include "opto/addnode.hpp" 33 #include "opto/arraycopynode.hpp" 34 #include "opto/c2compiler.hpp" 35 #include "opto/callGenerator.hpp" 36 #include "opto/castnode.hpp" 37 #include "opto/cfgnode.hpp" 38 #include "opto/convertnode.hpp" 39 #include "opto/countbitsnode.hpp" 40 #include "opto/intrinsicnode.hpp" 41 #include "opto/idealKit.hpp" 42 #include "opto/mathexactnode.hpp" 43 #include "opto/movenode.hpp" 44 #include "opto/mulnode.hpp" 45 #include "opto/narrowptrnode.hpp" 46 #include "opto/opaquenode.hpp" 47 #include "opto/parse.hpp" 48 #include "opto/runtime.hpp" 49 #include "opto/subnode.hpp" 50 #include "prims/nativeLookup.hpp" 51 #include "runtime/sharedRuntime.hpp" 52 #include "trace/traceMacros.hpp" 53 54 class LibraryIntrinsic : public InlineCallGenerator { 55 // Extend the set of intrinsics known to the runtime: 56 public: 57 private: 58 bool _is_virtual; 59 bool _does_virtual_dispatch; 60 int8_t _predicates_count; // Intrinsic is predicated by several conditions 61 int8_t _last_predicate; // Last generated predicate 62 vmIntrinsics::ID _intrinsic_id; 63 64 public: 65 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id) 66 : InlineCallGenerator(m), 67 _is_virtual(is_virtual), 68 _does_virtual_dispatch(does_virtual_dispatch), 69 _predicates_count((int8_t)predicates_count), 70 _last_predicate((int8_t)-1), 71 _intrinsic_id(id) 72 { 73 } 74 virtual bool is_intrinsic() const { return true; } 75 virtual bool is_virtual() const { return _is_virtual; } 76 virtual bool is_predicated() const { return _predicates_count > 0; } 77 virtual int predicates_count() const { return _predicates_count; } 78 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; } 79 virtual JVMState* generate(JVMState* jvms); 80 virtual Node* generate_predicate(JVMState* jvms, int predicate); 81 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; } 82 }; 83 84 85 // Local helper class for LibraryIntrinsic: 86 class LibraryCallKit : public GraphKit { 87 private: 88 LibraryIntrinsic* _intrinsic; // the library intrinsic being called 89 Node* _result; // the result node, if any 90 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted 91 92 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr = false); 93 94 public: 95 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic) 96 : GraphKit(jvms), 97 _intrinsic(intrinsic), 98 _result(NULL) 99 { 100 // Check if this is a root compile. In that case we don't have a caller. 101 if (!jvms->has_method()) { 102 _reexecute_sp = sp(); 103 } else { 104 // Find out how many arguments the interpreter needs when deoptimizing 105 // and save the stack pointer value so it can used by uncommon_trap. 106 // We find the argument count by looking at the declared signature. 107 bool ignored_will_link; 108 ciSignature* declared_signature = NULL; 109 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature); 110 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci())); 111 _reexecute_sp = sp() + nargs; // "push" arguments back on stack 112 } 113 } 114 115 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; } 116 117 ciMethod* caller() const { return jvms()->method(); } 118 int bci() const { return jvms()->bci(); } 119 LibraryIntrinsic* intrinsic() const { return _intrinsic; } 120 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); } 121 ciMethod* callee() const { return _intrinsic->method(); } 122 123 bool try_to_inline(int predicate); 124 Node* try_to_predicate(int predicate); 125 126 void push_result() { 127 // Push the result onto the stack. 128 if (!stopped() && result() != NULL) { 129 BasicType bt = result()->bottom_type()->basic_type(); 130 push_node(bt, result()); 131 } 132 } 133 134 private: 135 void fatal_unexpected_iid(vmIntrinsics::ID iid) { 136 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid)); 137 } 138 139 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; } 140 void set_result(RegionNode* region, PhiNode* value); 141 Node* result() { return _result; } 142 143 virtual int reexecute_sp() { return _reexecute_sp; } 144 145 // Helper functions to inline natives 146 Node* generate_guard(Node* test, RegionNode* region, float true_prob); 147 Node* generate_slow_guard(Node* test, RegionNode* region); 148 Node* generate_fair_guard(Node* test, RegionNode* region); 149 Node* generate_negative_guard(Node* index, RegionNode* region, 150 // resulting CastII of index: 151 Node* *pos_index = NULL); 152 Node* generate_limit_guard(Node* offset, Node* subseq_length, 153 Node* array_length, 154 RegionNode* region); 155 void generate_string_range_check(Node* array, Node* offset, 156 Node* length, bool char_count); 157 Node* generate_current_thread(Node* &tls_output); 158 Node* load_mirror_from_klass(Node* klass); 159 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null, 160 RegionNode* region, int null_path, 161 int offset); 162 Node* load_klass_from_mirror(Node* mirror, bool never_see_null, 163 RegionNode* region, int null_path) { 164 int offset = java_lang_Class::klass_offset_in_bytes(); 165 return load_klass_from_mirror_common(mirror, never_see_null, 166 region, null_path, 167 offset); 168 } 169 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null, 170 RegionNode* region, int null_path) { 171 int offset = java_lang_Class::array_klass_offset_in_bytes(); 172 return load_klass_from_mirror_common(mirror, never_see_null, 173 region, null_path, 174 offset); 175 } 176 Node* generate_access_flags_guard(Node* kls, 177 int modifier_mask, int modifier_bits, 178 RegionNode* region); 179 Node* generate_interface_guard(Node* kls, RegionNode* region); 180 Node* generate_array_guard(Node* kls, RegionNode* region) { 181 return generate_array_guard_common(kls, region, false, false); 182 } 183 Node* generate_non_array_guard(Node* kls, RegionNode* region) { 184 return generate_array_guard_common(kls, region, false, true); 185 } 186 Node* generate_objArray_guard(Node* kls, RegionNode* region) { 187 return generate_array_guard_common(kls, region, true, false); 188 } 189 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) { 190 return generate_array_guard_common(kls, region, true, true); 191 } 192 Node* generate_array_guard_common(Node* kls, RegionNode* region, 193 bool obj_array, bool not_array); 194 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region); 195 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id, 196 bool is_virtual = false, bool is_static = false); 197 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) { 198 return generate_method_call(method_id, false, true); 199 } 200 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) { 201 return generate_method_call(method_id, true, false); 202 } 203 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 204 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 205 206 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae); 207 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae); 208 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae); 209 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae); 210 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 211 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae); 212 bool inline_string_indexOfChar(); 213 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae); 214 bool inline_string_toBytesU(); 215 bool inline_string_getCharsU(); 216 bool inline_string_copy(bool compress); 217 bool inline_string_char_access(bool is_store); 218 Node* round_double_node(Node* n); 219 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName); 220 bool inline_math_native(vmIntrinsics::ID id); 221 bool inline_trig(vmIntrinsics::ID id); 222 bool inline_math(vmIntrinsics::ID id); 223 template <typename OverflowOp> 224 bool inline_math_overflow(Node* arg1, Node* arg2); 225 void inline_math_mathExact(Node* math, Node* test); 226 bool inline_math_addExactI(bool is_increment); 227 bool inline_math_addExactL(bool is_increment); 228 bool inline_math_multiplyExactI(); 229 bool inline_math_multiplyExactL(); 230 bool inline_math_negateExactI(); 231 bool inline_math_negateExactL(); 232 bool inline_math_subtractExactI(bool is_decrement); 233 bool inline_math_subtractExactL(bool is_decrement); 234 bool inline_min_max(vmIntrinsics::ID id); 235 bool inline_notify(vmIntrinsics::ID id); 236 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y); 237 // This returns Type::AnyPtr, RawPtr, or OopPtr. 238 int classify_unsafe_addr(Node* &base, Node* &offset); 239 Node* make_unsafe_address(Node* base, Node* offset); 240 // Helper for inline_unsafe_access. 241 // Generates the guards that check whether the result of 242 // Unsafe.getObject should be recorded in an SATB log buffer. 243 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar); 244 245 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind; 246 bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, AccessKind kind, bool is_unaligned); 247 static bool klass_needs_init_guard(Node* kls); 248 bool inline_unsafe_allocate(); 249 bool inline_unsafe_copyMemory(); 250 bool inline_native_currentThread(); 251 #ifdef TRACE_HAVE_INTRINSICS 252 bool inline_native_classID(); 253 bool inline_native_threadID(); 254 #endif 255 bool inline_native_time_funcs(address method, const char* funcName); 256 bool inline_native_isInterrupted(); 257 bool inline_native_Class_query(vmIntrinsics::ID id); 258 bool inline_native_subtype_check(); 259 260 bool inline_native_newArray(); 261 bool inline_native_getLength(); 262 bool inline_array_copyOf(bool is_copyOfRange); 263 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae); 264 bool inline_objects_checkIndex(); 265 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark); 266 bool inline_native_clone(bool is_virtual); 267 bool inline_native_Reflection_getCallerClass(); 268 // Helper function for inlining native object hash method 269 bool inline_native_hashcode(bool is_virtual, bool is_static); 270 bool inline_native_getClass(); 271 272 // Helper functions for inlining arraycopy 273 bool inline_arraycopy(); 274 AllocateArrayNode* tightly_coupled_allocation(Node* ptr, 275 RegionNode* slow_region); 276 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp); 277 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp); 278 279 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind; 280 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store); 281 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind); 282 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind); 283 bool inline_unsafe_fence(vmIntrinsics::ID id); 284 bool inline_fp_conversions(vmIntrinsics::ID id); 285 bool inline_number_methods(vmIntrinsics::ID id); 286 bool inline_reference_get(); 287 bool inline_Class_cast(); 288 bool inline_aescrypt_Block(vmIntrinsics::ID id); 289 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id); 290 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id); 291 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting); 292 Node* inline_counterMode_AESCrypt_predicate(); 293 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object); 294 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object); 295 bool inline_ghash_processBlocks(); 296 bool inline_sha_implCompress(vmIntrinsics::ID id); 297 bool inline_digestBase_implCompressMB(int predicate); 298 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA, 299 bool long_state, address stubAddr, const char *stubName, 300 Node* src_start, Node* ofs, Node* limit); 301 Node* get_state_from_sha_object(Node *sha_object); 302 Node* get_state_from_sha5_object(Node *sha_object); 303 Node* inline_digestBase_implCompressMB_predicate(int predicate); 304 bool inline_encodeISOArray(); 305 bool inline_updateCRC32(); 306 bool inline_updateBytesCRC32(); 307 bool inline_updateByteBufferCRC32(); 308 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class); 309 bool inline_updateBytesCRC32C(); 310 bool inline_updateDirectByteBufferCRC32C(); 311 bool inline_updateBytesAdler32(); 312 bool inline_updateByteBufferAdler32(); 313 bool inline_multiplyToLen(); 314 bool inline_hasNegatives(); 315 bool inline_squareToLen(); 316 bool inline_mulAdd(); 317 bool inline_montgomeryMultiply(); 318 bool inline_montgomerySquare(); 319 bool inline_vectorizedMismatch(); 320 321 bool inline_profileBoolean(); 322 bool inline_isCompileConstant(); 323 324 bool inline_deoptimize(); 325 }; 326 327 //---------------------------make_vm_intrinsic---------------------------- 328 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) { 329 vmIntrinsics::ID id = m->intrinsic_id(); 330 assert(id != vmIntrinsics::_none, "must be a VM intrinsic"); 331 332 if (!m->is_loaded()) { 333 // Do not attempt to inline unloaded methods. 334 return NULL; 335 } 336 337 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization); 338 bool is_available = false; 339 340 { 341 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag 342 // the compiler must transition to '_thread_in_vm' state because both 343 // methods access VM-internal data. 344 VM_ENTRY_MARK; 345 methodHandle mh(THREAD, m->get_Method()); 346 is_available = compiler->is_intrinsic_supported(mh, is_virtual) && 347 !C->directive()->is_intrinsic_disabled(mh) && 348 !vmIntrinsics::is_disabled_by_flags(mh); 349 350 } 351 352 if (is_available) { 353 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility"); 354 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?"); 355 return new LibraryIntrinsic(m, is_virtual, 356 vmIntrinsics::predicates_needed(id), 357 vmIntrinsics::does_virtual_dispatch(id), 358 (vmIntrinsics::ID) id); 359 } else { 360 return NULL; 361 } 362 } 363 364 //----------------------register_library_intrinsics----------------------- 365 // Initialize this file's data structures, for each Compile instance. 366 void Compile::register_library_intrinsics() { 367 // Nothing to do here. 368 } 369 370 JVMState* LibraryIntrinsic::generate(JVMState* jvms) { 371 LibraryCallKit kit(jvms, this); 372 Compile* C = kit.C; 373 int nodes = C->unique(); 374 #ifndef PRODUCT 375 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 376 char buf[1000]; 377 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 378 tty->print_cr("Intrinsic %s", str); 379 } 380 #endif 381 ciMethod* callee = kit.callee(); 382 const int bci = kit.bci(); 383 384 // Try to inline the intrinsic. 385 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) && 386 kit.try_to_inline(_last_predicate)) { 387 if (C->print_intrinsics() || C->print_inlining()) { 388 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)"); 389 } 390 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 391 if (C->log()) { 392 C->log()->elem("intrinsic id='%s'%s nodes='%d'", 393 vmIntrinsics::name_at(intrinsic_id()), 394 (is_virtual() ? " virtual='1'" : ""), 395 C->unique() - nodes); 396 } 397 // Push the result from the inlined method onto the stack. 398 kit.push_result(); 399 C->print_inlining_update(this); 400 return kit.transfer_exceptions_into_jvms(); 401 } 402 403 // The intrinsic bailed out 404 if (C->print_intrinsics() || C->print_inlining()) { 405 if (jvms->has_method()) { 406 // Not a root compile. 407 const char* msg; 408 if (callee->intrinsic_candidate()) { 409 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)"; 410 } else { 411 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated" 412 : "failed to inline (intrinsic), method not annotated"; 413 } 414 C->print_inlining(callee, jvms->depth() - 1, bci, msg); 415 } else { 416 // Root compile 417 tty->print("Did not generate intrinsic %s%s at bci:%d in", 418 vmIntrinsics::name_at(intrinsic_id()), 419 (is_virtual() ? " (virtual)" : ""), bci); 420 } 421 } 422 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 423 C->print_inlining_update(this); 424 return NULL; 425 } 426 427 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) { 428 LibraryCallKit kit(jvms, this); 429 Compile* C = kit.C; 430 int nodes = C->unique(); 431 _last_predicate = predicate; 432 #ifndef PRODUCT 433 assert(is_predicated() && predicate < predicates_count(), "sanity"); 434 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 435 char buf[1000]; 436 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 437 tty->print_cr("Predicate for intrinsic %s", str); 438 } 439 #endif 440 ciMethod* callee = kit.callee(); 441 const int bci = kit.bci(); 442 443 Node* slow_ctl = kit.try_to_predicate(predicate); 444 if (!kit.failing()) { 445 if (C->print_intrinsics() || C->print_inlining()) { 446 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)"); 447 } 448 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 449 if (C->log()) { 450 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'", 451 vmIntrinsics::name_at(intrinsic_id()), 452 (is_virtual() ? " virtual='1'" : ""), 453 C->unique() - nodes); 454 } 455 return slow_ctl; // Could be NULL if the check folds. 456 } 457 458 // The intrinsic bailed out 459 if (C->print_intrinsics() || C->print_inlining()) { 460 if (jvms->has_method()) { 461 // Not a root compile. 462 const char* msg = "failed to generate predicate for intrinsic"; 463 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg); 464 } else { 465 // Root compile 466 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in", 467 vmIntrinsics::name_at(intrinsic_id()), 468 (is_virtual() ? " (virtual)" : ""), bci); 469 } 470 } 471 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 472 return NULL; 473 } 474 475 bool LibraryCallKit::try_to_inline(int predicate) { 476 // Handle symbolic names for otherwise undistinguished boolean switches: 477 const bool is_store = true; 478 const bool is_compress = true; 479 const bool is_native_ptr = true; 480 const bool is_static = true; 481 const bool is_volatile = true; 482 483 if (!jvms()->has_method()) { 484 // Root JVMState has a null method. 485 assert(map()->memory()->Opcode() == Op_Parm, ""); 486 // Insert the memory aliasing node 487 set_all_memory(reset_memory()); 488 } 489 assert(merged_memory(), ""); 490 491 492 switch (intrinsic_id()) { 493 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static); 494 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static); 495 case vmIntrinsics::_getClass: return inline_native_getClass(); 496 497 case vmIntrinsics::_dsin: 498 case vmIntrinsics::_dcos: 499 case vmIntrinsics::_dtan: 500 case vmIntrinsics::_dabs: 501 case vmIntrinsics::_datan2: 502 case vmIntrinsics::_dsqrt: 503 case vmIntrinsics::_dexp: 504 case vmIntrinsics::_dlog: 505 case vmIntrinsics::_dlog10: 506 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id()); 507 508 case vmIntrinsics::_min: 509 case vmIntrinsics::_max: return inline_min_max(intrinsic_id()); 510 511 case vmIntrinsics::_notify: 512 case vmIntrinsics::_notifyAll: 513 if (InlineNotify) { 514 return inline_notify(intrinsic_id()); 515 } 516 return false; 517 518 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 519 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 520 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 521 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 522 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 523 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 524 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 525 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 526 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 527 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 528 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 529 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 530 531 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 532 533 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 534 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 535 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 536 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 537 538 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 539 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 540 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 541 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 542 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 543 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 544 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(); 545 546 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 547 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU); 548 549 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 550 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 551 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 552 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 553 554 case vmIntrinsics::_compressStringC: 555 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 556 case vmIntrinsics::_inflateStringC: 557 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 558 559 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Relaxed, false); 560 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Relaxed, false); 561 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Relaxed, false); 562 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, false); 563 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, false); 564 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, false); 565 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, false); 566 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Relaxed, false); 567 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Relaxed, false); 568 569 case vmIntrinsics::_putObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Relaxed, false); 570 case vmIntrinsics::_putBoolean: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Relaxed, false); 571 case vmIntrinsics::_putByte: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Relaxed, false); 572 case vmIntrinsics::_putShort: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, false); 573 case vmIntrinsics::_putChar: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, false); 574 case vmIntrinsics::_putInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, false); 575 case vmIntrinsics::_putLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, false); 576 case vmIntrinsics::_putFloat: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Relaxed, false); 577 case vmIntrinsics::_putDouble: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Relaxed, false); 578 579 case vmIntrinsics::_getByte_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE, Relaxed, false); 580 case vmIntrinsics::_getShort_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT, Relaxed, false); 581 case vmIntrinsics::_getChar_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR, Relaxed, false); 582 case vmIntrinsics::_getInt_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_INT, Relaxed, false); 583 case vmIntrinsics::_getLong_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_LONG, Relaxed, false); 584 case vmIntrinsics::_getFloat_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT, Relaxed, false); 585 case vmIntrinsics::_getDouble_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE, Relaxed, false); 586 case vmIntrinsics::_getAddress_raw: return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS, Relaxed, false); 587 588 case vmIntrinsics::_putByte_raw: return inline_unsafe_access( is_native_ptr, is_store, T_BYTE, Relaxed, false); 589 case vmIntrinsics::_putShort_raw: return inline_unsafe_access( is_native_ptr, is_store, T_SHORT, Relaxed, false); 590 case vmIntrinsics::_putChar_raw: return inline_unsafe_access( is_native_ptr, is_store, T_CHAR, Relaxed, false); 591 case vmIntrinsics::_putInt_raw: return inline_unsafe_access( is_native_ptr, is_store, T_INT, Relaxed, false); 592 case vmIntrinsics::_putLong_raw: return inline_unsafe_access( is_native_ptr, is_store, T_LONG, Relaxed, false); 593 case vmIntrinsics::_putFloat_raw: return inline_unsafe_access( is_native_ptr, is_store, T_FLOAT, Relaxed, false); 594 case vmIntrinsics::_putDouble_raw: return inline_unsafe_access( is_native_ptr, is_store, T_DOUBLE, Relaxed, false); 595 case vmIntrinsics::_putAddress_raw: return inline_unsafe_access( is_native_ptr, is_store, T_ADDRESS, Relaxed, false); 596 597 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Volatile, false); 598 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Volatile, false); 599 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Volatile, false); 600 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Volatile, false); 601 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Volatile, false); 602 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Volatile, false); 603 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Volatile, false); 604 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Volatile, false); 605 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Volatile, false); 606 607 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Volatile, false); 608 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Volatile, false); 609 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Volatile, false); 610 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Volatile, false); 611 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Volatile, false); 612 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Volatile, false); 613 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Volatile, false); 614 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Volatile, false); 615 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Volatile, false); 616 617 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Relaxed, true); 618 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Relaxed, true); 619 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Relaxed, true); 620 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Relaxed, true); 621 622 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Relaxed, true); 623 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Relaxed, true); 624 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Relaxed, true); 625 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Relaxed, true); 626 627 case vmIntrinsics::_putOrderedObject: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false); 628 case vmIntrinsics::_putOrderedInt: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false); 629 case vmIntrinsics::_putOrderedLong: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false); 630 631 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Acquire, false); 632 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Acquire, false); 633 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Acquire, false); 634 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Acquire, false); 635 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Acquire, false); 636 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Acquire, false); 637 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Acquire, false); 638 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Acquire, false); 639 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Acquire, false); 640 641 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Release, false); 642 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Release, false); 643 case vmIntrinsics::_putByteRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Release, false); 644 case vmIntrinsics::_putShortRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Release, false); 645 case vmIntrinsics::_putCharRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Release, false); 646 case vmIntrinsics::_putIntRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Release, false); 647 case vmIntrinsics::_putLongRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Release, false); 648 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Release, false); 649 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Release, false); 650 651 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, Opaque, false); 652 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, Opaque, false); 653 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE, Opaque, false); 654 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT, Opaque, false); 655 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR, Opaque, false); 656 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_INT, Opaque, false); 657 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG, Opaque, false); 658 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT, Opaque, false); 659 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE, Opaque, false); 660 661 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_OBJECT, Opaque, false); 662 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BOOLEAN, Opaque, false); 663 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_BYTE, Opaque, false); 664 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_SHORT, Opaque, false); 665 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_CHAR, Opaque, false); 666 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_INT, Opaque, false); 667 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_LONG, Opaque, false); 668 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_FLOAT, Opaque, false); 669 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access(!is_native_ptr, is_store, T_DOUBLE, Opaque, false); 670 671 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 672 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 673 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 674 675 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 676 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 677 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 678 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 679 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 680 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 681 case vmIntrinsics::_weakCompareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 682 case vmIntrinsics::_weakCompareAndSwapLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 683 case vmIntrinsics::_weakCompareAndSwapLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 684 685 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 686 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 687 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 688 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 689 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 690 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 691 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 692 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 693 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 694 695 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 696 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 697 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 698 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 699 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 700 701 case vmIntrinsics::_loadFence: 702 case vmIntrinsics::_storeFence: 703 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 704 705 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 706 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted(); 707 708 #ifdef TRACE_HAVE_INTRINSICS 709 case vmIntrinsics::_classID: return inline_native_classID(); 710 case vmIntrinsics::_threadID: return inline_native_threadID(); 711 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime"); 712 #endif 713 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 714 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 715 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 716 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 717 case vmIntrinsics::_newArray: return inline_native_newArray(); 718 case vmIntrinsics::_getLength: return inline_native_getLength(); 719 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 720 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 721 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 722 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 723 case vmIntrinsics::_Objects_checkIndex: return inline_objects_checkIndex(); 724 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 725 726 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 727 728 case vmIntrinsics::_isInstance: 729 case vmIntrinsics::_getModifiers: 730 case vmIntrinsics::_isInterface: 731 case vmIntrinsics::_isArray: 732 case vmIntrinsics::_isPrimitive: 733 case vmIntrinsics::_getSuperclass: 734 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id()); 735 736 case vmIntrinsics::_floatToRawIntBits: 737 case vmIntrinsics::_floatToIntBits: 738 case vmIntrinsics::_intBitsToFloat: 739 case vmIntrinsics::_doubleToRawLongBits: 740 case vmIntrinsics::_doubleToLongBits: 741 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id()); 742 743 case vmIntrinsics::_numberOfLeadingZeros_i: 744 case vmIntrinsics::_numberOfLeadingZeros_l: 745 case vmIntrinsics::_numberOfTrailingZeros_i: 746 case vmIntrinsics::_numberOfTrailingZeros_l: 747 case vmIntrinsics::_bitCount_i: 748 case vmIntrinsics::_bitCount_l: 749 case vmIntrinsics::_reverseBytes_i: 750 case vmIntrinsics::_reverseBytes_l: 751 case vmIntrinsics::_reverseBytes_s: 752 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id()); 753 754 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass(); 755 756 case vmIntrinsics::_Reference_get: return inline_reference_get(); 757 758 case vmIntrinsics::_Class_cast: return inline_Class_cast(); 759 760 case vmIntrinsics::_aescrypt_encryptBlock: 761 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id()); 762 763 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 764 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 765 return inline_cipherBlockChaining_AESCrypt(intrinsic_id()); 766 767 case vmIntrinsics::_counterMode_AESCrypt: 768 return inline_counterMode_AESCrypt(intrinsic_id()); 769 770 case vmIntrinsics::_sha_implCompress: 771 case vmIntrinsics::_sha2_implCompress: 772 case vmIntrinsics::_sha5_implCompress: 773 return inline_sha_implCompress(intrinsic_id()); 774 775 case vmIntrinsics::_digestBase_implCompressMB: 776 return inline_digestBase_implCompressMB(predicate); 777 778 case vmIntrinsics::_multiplyToLen: 779 return inline_multiplyToLen(); 780 781 case vmIntrinsics::_squareToLen: 782 return inline_squareToLen(); 783 784 case vmIntrinsics::_mulAdd: 785 return inline_mulAdd(); 786 787 case vmIntrinsics::_montgomeryMultiply: 788 return inline_montgomeryMultiply(); 789 case vmIntrinsics::_montgomerySquare: 790 return inline_montgomerySquare(); 791 792 case vmIntrinsics::_vectorizedMismatch: 793 return inline_vectorizedMismatch(); 794 795 case vmIntrinsics::_ghash_processBlocks: 796 return inline_ghash_processBlocks(); 797 798 case vmIntrinsics::_encodeISOArray: 799 case vmIntrinsics::_encodeByteISOArray: 800 return inline_encodeISOArray(); 801 802 case vmIntrinsics::_updateCRC32: 803 return inline_updateCRC32(); 804 case vmIntrinsics::_updateBytesCRC32: 805 return inline_updateBytesCRC32(); 806 case vmIntrinsics::_updateByteBufferCRC32: 807 return inline_updateByteBufferCRC32(); 808 809 case vmIntrinsics::_updateBytesCRC32C: 810 return inline_updateBytesCRC32C(); 811 case vmIntrinsics::_updateDirectByteBufferCRC32C: 812 return inline_updateDirectByteBufferCRC32C(); 813 814 case vmIntrinsics::_updateBytesAdler32: 815 return inline_updateBytesAdler32(); 816 case vmIntrinsics::_updateByteBufferAdler32: 817 return inline_updateByteBufferAdler32(); 818 819 case vmIntrinsics::_profileBoolean: 820 return inline_profileBoolean(); 821 case vmIntrinsics::_isCompileConstant: 822 return inline_isCompileConstant(); 823 824 case vmIntrinsics::_hasNegatives: 825 return inline_hasNegatives(); 826 827 case vmIntrinsics::_deoptimize: 828 return inline_deoptimize(); 829 830 default: 831 // If you get here, it may be that someone has added a new intrinsic 832 // to the list in vmSymbols.hpp without implementing it here. 833 #ifndef PRODUCT 834 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 835 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 836 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 837 } 838 #endif 839 return false; 840 } 841 } 842 843 Node* LibraryCallKit::try_to_predicate(int predicate) { 844 if (!jvms()->has_method()) { 845 // Root JVMState has a null method. 846 assert(map()->memory()->Opcode() == Op_Parm, ""); 847 // Insert the memory aliasing node 848 set_all_memory(reset_memory()); 849 } 850 assert(merged_memory(), ""); 851 852 switch (intrinsic_id()) { 853 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 854 return inline_cipherBlockChaining_AESCrypt_predicate(false); 855 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 856 return inline_cipherBlockChaining_AESCrypt_predicate(true); 857 case vmIntrinsics::_counterMode_AESCrypt: 858 return inline_counterMode_AESCrypt_predicate(); 859 case vmIntrinsics::_digestBase_implCompressMB: 860 return inline_digestBase_implCompressMB_predicate(predicate); 861 862 default: 863 // If you get here, it may be that someone has added a new intrinsic 864 // to the list in vmSymbols.hpp without implementing it here. 865 #ifndef PRODUCT 866 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 867 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 868 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 869 } 870 #endif 871 Node* slow_ctl = control(); 872 set_control(top()); // No fast path instrinsic 873 return slow_ctl; 874 } 875 } 876 877 //------------------------------set_result------------------------------- 878 // Helper function for finishing intrinsics. 879 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 880 record_for_igvn(region); 881 set_control(_gvn.transform(region)); 882 set_result( _gvn.transform(value)); 883 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 884 } 885 886 //------------------------------generate_guard--------------------------- 887 // Helper function for generating guarded fast-slow graph structures. 888 // The given 'test', if true, guards a slow path. If the test fails 889 // then a fast path can be taken. (We generally hope it fails.) 890 // In all cases, GraphKit::control() is updated to the fast path. 891 // The returned value represents the control for the slow path. 892 // The return value is never 'top'; it is either a valid control 893 // or NULL if it is obvious that the slow path can never be taken. 894 // Also, if region and the slow control are not NULL, the slow edge 895 // is appended to the region. 896 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 897 if (stopped()) { 898 // Already short circuited. 899 return NULL; 900 } 901 902 // Build an if node and its projections. 903 // If test is true we take the slow path, which we assume is uncommon. 904 if (_gvn.type(test) == TypeInt::ZERO) { 905 // The slow branch is never taken. No need to build this guard. 906 return NULL; 907 } 908 909 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 910 911 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 912 if (if_slow == top()) { 913 // The slow branch is never taken. No need to build this guard. 914 return NULL; 915 } 916 917 if (region != NULL) 918 region->add_req(if_slow); 919 920 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 921 set_control(if_fast); 922 923 return if_slow; 924 } 925 926 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 927 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 928 } 929 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 930 return generate_guard(test, region, PROB_FAIR); 931 } 932 933 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 934 Node* *pos_index) { 935 if (stopped()) 936 return NULL; // already stopped 937 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 938 return NULL; // index is already adequately typed 939 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 940 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 941 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 942 if (is_neg != NULL && pos_index != NULL) { 943 // Emulate effect of Parse::adjust_map_after_if. 944 Node* ccast = new CastIINode(index, TypeInt::POS); 945 ccast->set_req(0, control()); 946 (*pos_index) = _gvn.transform(ccast); 947 } 948 return is_neg; 949 } 950 951 // Make sure that 'position' is a valid limit index, in [0..length]. 952 // There are two equivalent plans for checking this: 953 // A. (offset + copyLength) unsigned<= arrayLength 954 // B. offset <= (arrayLength - copyLength) 955 // We require that all of the values above, except for the sum and 956 // difference, are already known to be non-negative. 957 // Plan A is robust in the face of overflow, if offset and copyLength 958 // are both hugely positive. 959 // 960 // Plan B is less direct and intuitive, but it does not overflow at 961 // all, since the difference of two non-negatives is always 962 // representable. Whenever Java methods must perform the equivalent 963 // check they generally use Plan B instead of Plan A. 964 // For the moment we use Plan A. 965 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 966 Node* subseq_length, 967 Node* array_length, 968 RegionNode* region) { 969 if (stopped()) 970 return NULL; // already stopped 971 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 972 if (zero_offset && subseq_length->eqv_uncast(array_length)) 973 return NULL; // common case of whole-array copy 974 Node* last = subseq_length; 975 if (!zero_offset) // last += offset 976 last = _gvn.transform(new AddINode(last, offset)); 977 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 978 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 979 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 980 return is_over; 981 } 982 983 // Emit range checks for the given String.value byte array 984 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) { 985 if (stopped()) { 986 return; // already stopped 987 } 988 RegionNode* bailout = new RegionNode(1); 989 record_for_igvn(bailout); 990 if (char_count) { 991 // Convert char count to byte count 992 count = _gvn.transform(new LShiftINode(count, intcon(1))); 993 } 994 995 // Offset and count must not be negative 996 generate_negative_guard(offset, bailout); 997 generate_negative_guard(count, bailout); 998 // Offset + count must not exceed length of array 999 generate_limit_guard(offset, count, load_array_length(array), bailout); 1000 1001 if (bailout->req() > 1) { 1002 PreserveJVMState pjvms(this); 1003 set_control(_gvn.transform(bailout)); 1004 uncommon_trap(Deoptimization::Reason_intrinsic, 1005 Deoptimization::Action_maybe_recompile); 1006 } 1007 } 1008 1009 //--------------------------generate_current_thread-------------------- 1010 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 1011 ciKlass* thread_klass = env()->Thread_klass(); 1012 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 1013 Node* thread = _gvn.transform(new ThreadLocalNode()); 1014 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset())); 1015 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered); 1016 tls_output = thread; 1017 return threadObj; 1018 } 1019 1020 1021 //------------------------------make_string_method_node------------------------ 1022 // Helper method for String intrinsic functions. This version is called with 1023 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 1024 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 1025 // containing the lengths of str1 and str2. 1026 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 1027 Node* result = NULL; 1028 switch (opcode) { 1029 case Op_StrIndexOf: 1030 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 1031 str1_start, cnt1, str2_start, cnt2, ae); 1032 break; 1033 case Op_StrComp: 1034 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 1035 str1_start, cnt1, str2_start, cnt2, ae); 1036 break; 1037 case Op_StrEquals: 1038 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1039 // Use the constant length if there is one because optimized match rule may exist. 1040 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1041 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1042 break; 1043 default: 1044 ShouldNotReachHere(); 1045 return NULL; 1046 } 1047 1048 // All these intrinsics have checks. 1049 C->set_has_split_ifs(true); // Has chance for split-if optimization 1050 1051 return _gvn.transform(result); 1052 } 1053 1054 //------------------------------inline_string_compareTo------------------------ 1055 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1056 Node* arg1 = argument(0); 1057 Node* arg2 = argument(1); 1058 1059 // Get start addr and length of first argument 1060 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1061 Node* arg1_cnt = load_array_length(arg1); 1062 1063 // Get start addr and length of second argument 1064 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1065 Node* arg2_cnt = load_array_length(arg2); 1066 1067 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1068 set_result(result); 1069 return true; 1070 } 1071 1072 //------------------------------inline_string_equals------------------------ 1073 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1074 Node* arg1 = argument(0); 1075 Node* arg2 = argument(1); 1076 1077 // paths (plus control) merge 1078 RegionNode* region = new RegionNode(3); 1079 Node* phi = new PhiNode(region, TypeInt::BOOL); 1080 1081 if (!stopped()) { 1082 // Get start addr and length of first argument 1083 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1084 Node* arg1_cnt = load_array_length(arg1); 1085 1086 // Get start addr and length of second argument 1087 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1088 Node* arg2_cnt = load_array_length(arg2); 1089 1090 // Check for arg1_cnt != arg2_cnt 1091 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1092 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1093 Node* if_ne = generate_slow_guard(bol, NULL); 1094 if (if_ne != NULL) { 1095 phi->init_req(2, intcon(0)); 1096 region->init_req(2, if_ne); 1097 } 1098 1099 // Check for count == 0 is done by assembler code for StrEquals. 1100 1101 if (!stopped()) { 1102 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1103 phi->init_req(1, equals); 1104 region->init_req(1, control()); 1105 } 1106 } 1107 1108 // post merge 1109 set_control(_gvn.transform(region)); 1110 record_for_igvn(region); 1111 1112 set_result(_gvn.transform(phi)); 1113 return true; 1114 } 1115 1116 //------------------------------inline_array_equals---------------------------- 1117 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1118 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1119 Node* arg1 = argument(0); 1120 Node* arg2 = argument(1); 1121 1122 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1123 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1124 return true; 1125 } 1126 1127 //------------------------------inline_hasNegatives------------------------------ 1128 bool LibraryCallKit::inline_hasNegatives() { 1129 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1130 return false; 1131 } 1132 1133 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters"); 1134 // no receiver since it is static method 1135 Node* ba = argument(0); 1136 Node* offset = argument(1); 1137 Node* len = argument(2); 1138 1139 // Range checks 1140 generate_string_range_check(ba, offset, len, false); 1141 if (stopped()) { 1142 return true; 1143 } 1144 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1145 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1146 set_result(_gvn.transform(result)); 1147 return true; 1148 } 1149 1150 bool LibraryCallKit::inline_objects_checkIndex() { 1151 Node* index = argument(0); 1152 Node* length = argument(1); 1153 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1154 return false; 1155 } 1156 1157 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0))); 1158 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1159 1160 { 1161 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1162 uncommon_trap(Deoptimization::Reason_intrinsic, 1163 Deoptimization::Action_make_not_entrant); 1164 } 1165 1166 if (stopped()) { 1167 return false; 1168 } 1169 1170 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length)); 1171 BoolTest::mask btest = BoolTest::lt; 1172 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1173 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1174 _gvn.set_type(rc, rc->Value(&_gvn)); 1175 if (!rc_bool->is_Con()) { 1176 record_for_igvn(rc); 1177 } 1178 set_control(_gvn.transform(new IfTrueNode(rc))); 1179 { 1180 PreserveJVMState pjvms(this); 1181 set_control(_gvn.transform(new IfFalseNode(rc))); 1182 uncommon_trap(Deoptimization::Reason_range_check, 1183 Deoptimization::Action_make_not_entrant); 1184 } 1185 1186 if (stopped()) { 1187 return false; 1188 } 1189 1190 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax)); 1191 result->set_req(0, control()); 1192 result = _gvn.transform(result); 1193 set_result(result); 1194 replace_in_map(index, result); 1195 return true; 1196 } 1197 1198 //------------------------------inline_string_indexOf------------------------ 1199 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1200 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1201 return false; 1202 } 1203 Node* src = argument(0); 1204 Node* tgt = argument(1); 1205 1206 // Make the merge point 1207 RegionNode* result_rgn = new RegionNode(4); 1208 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 1209 1210 // Get start addr and length of source string 1211 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1212 Node* src_count = load_array_length(src); 1213 1214 // Get start addr and length of substring 1215 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1216 Node* tgt_count = load_array_length(tgt); 1217 1218 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1219 // Divide src size by 2 if String is UTF16 encoded 1220 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1221 } 1222 if (ae == StrIntrinsicNode::UU) { 1223 // Divide substring size by 2 if String is UTF16 encoded 1224 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1225 } 1226 1227 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae); 1228 if (result != NULL) { 1229 result_phi->init_req(3, result); 1230 result_rgn->init_req(3, control()); 1231 } 1232 set_control(_gvn.transform(result_rgn)); 1233 record_for_igvn(result_rgn); 1234 set_result(_gvn.transform(result_phi)); 1235 1236 return true; 1237 } 1238 1239 //-----------------------------inline_string_indexOf----------------------- 1240 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1241 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1242 return false; 1243 } 1244 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1245 return false; 1246 } 1247 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1248 Node* src = argument(0); // byte[] 1249 Node* src_count = argument(1); // char count 1250 Node* tgt = argument(2); // byte[] 1251 Node* tgt_count = argument(3); // char count 1252 Node* from_index = argument(4); // char index 1253 1254 // Multiply byte array index by 2 if String is UTF16 encoded 1255 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1256 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1257 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1258 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1259 1260 // Range checks 1261 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1262 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1263 if (stopped()) { 1264 return true; 1265 } 1266 1267 RegionNode* region = new RegionNode(5); 1268 Node* phi = new PhiNode(region, TypeInt::INT); 1269 1270 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae); 1271 if (result != NULL) { 1272 // The result is index relative to from_index if substring was found, -1 otherwise. 1273 // Generate code which will fold into cmove. 1274 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1275 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1276 1277 Node* if_lt = generate_slow_guard(bol, NULL); 1278 if (if_lt != NULL) { 1279 // result == -1 1280 phi->init_req(3, result); 1281 region->init_req(3, if_lt); 1282 } 1283 if (!stopped()) { 1284 result = _gvn.transform(new AddINode(result, from_index)); 1285 phi->init_req(4, result); 1286 region->init_req(4, control()); 1287 } 1288 } 1289 1290 set_control(_gvn.transform(region)); 1291 record_for_igvn(region); 1292 set_result(_gvn.transform(phi)); 1293 1294 return true; 1295 } 1296 1297 // Create StrIndexOfNode with fast path checks 1298 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1299 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1300 // Check for substr count > string count 1301 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1302 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1303 Node* if_gt = generate_slow_guard(bol, NULL); 1304 if (if_gt != NULL) { 1305 phi->init_req(1, intcon(-1)); 1306 region->init_req(1, if_gt); 1307 } 1308 if (!stopped()) { 1309 // Check for substr count == 0 1310 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1311 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1312 Node* if_zero = generate_slow_guard(bol, NULL); 1313 if (if_zero != NULL) { 1314 phi->init_req(2, intcon(0)); 1315 region->init_req(2, if_zero); 1316 } 1317 } 1318 if (!stopped()) { 1319 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1320 } 1321 return NULL; 1322 } 1323 1324 //-----------------------------inline_string_indexOfChar----------------------- 1325 bool LibraryCallKit::inline_string_indexOfChar() { 1326 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1327 return false; 1328 } 1329 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1330 return false; 1331 } 1332 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1333 Node* src = argument(0); // byte[] 1334 Node* tgt = argument(1); // tgt is int ch 1335 Node* from_index = argument(2); 1336 Node* max = argument(3); 1337 1338 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1))); 1339 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1340 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1341 1342 // Range checks 1343 generate_string_range_check(src, src_offset, src_count, true); 1344 if (stopped()) { 1345 return true; 1346 } 1347 1348 RegionNode* region = new RegionNode(3); 1349 Node* phi = new PhiNode(region, TypeInt::INT); 1350 1351 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none); 1352 C->set_has_split_ifs(true); // Has chance for split-if optimization 1353 _gvn.transform(result); 1354 1355 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1356 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1357 1358 Node* if_lt = generate_slow_guard(bol, NULL); 1359 if (if_lt != NULL) { 1360 // result == -1 1361 phi->init_req(2, result); 1362 region->init_req(2, if_lt); 1363 } 1364 if (!stopped()) { 1365 result = _gvn.transform(new AddINode(result, from_index)); 1366 phi->init_req(1, result); 1367 region->init_req(1, control()); 1368 } 1369 set_control(_gvn.transform(region)); 1370 record_for_igvn(region); 1371 set_result(_gvn.transform(phi)); 1372 1373 return true; 1374 } 1375 //---------------------------inline_string_copy--------------------- 1376 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1377 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1378 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1379 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1380 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1381 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1382 bool LibraryCallKit::inline_string_copy(bool compress) { 1383 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1384 return false; 1385 } 1386 int nargs = 5; // 2 oops, 3 ints 1387 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1388 1389 Node* src = argument(0); 1390 Node* src_offset = argument(1); 1391 Node* dst = argument(2); 1392 Node* dst_offset = argument(3); 1393 Node* length = argument(4); 1394 1395 // Check for allocation before we add nodes that would confuse 1396 // tightly_coupled_allocation() 1397 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1398 1399 // Figure out the size and type of the elements we will be copying. 1400 const Type* src_type = src->Value(&_gvn); 1401 const Type* dst_type = dst->Value(&_gvn); 1402 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1403 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1404 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1405 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1406 "Unsupported array types for inline_string_copy"); 1407 1408 // Range checks 1409 generate_string_range_check(src, src_offset, length, compress && src_elem == T_BYTE); 1410 generate_string_range_check(dst, dst_offset, length, !compress && dst_elem == T_BYTE); 1411 if (stopped()) { 1412 return true; 1413 } 1414 1415 // Convert char[] offsets to byte[] offsets 1416 if (compress && src_elem == T_BYTE) { 1417 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1418 } else if (!compress && dst_elem == T_BYTE) { 1419 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1420 } 1421 1422 Node* src_start = array_element_address(src, src_offset, src_elem); 1423 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1424 // 'src_start' points to src array + scaled offset 1425 // 'dst_start' points to dst array + scaled offset 1426 Node* count = NULL; 1427 if (compress) { 1428 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1429 } else { 1430 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1431 } 1432 1433 if (alloc != NULL) { 1434 if (alloc->maybe_set_complete(&_gvn)) { 1435 // "You break it, you buy it." 1436 InitializeNode* init = alloc->initialization(); 1437 assert(init->is_complete(), "we just did this"); 1438 init->set_complete_with_arraycopy(); 1439 assert(dst->is_CheckCastPP(), "sanity"); 1440 assert(dst->in(0)->in(0) == init, "dest pinned"); 1441 } 1442 // Do not let stores that initialize this object be reordered with 1443 // a subsequent store that would make this object accessible by 1444 // other threads. 1445 // Record what AllocateNode this StoreStore protects so that 1446 // escape analysis can go from the MemBarStoreStoreNode to the 1447 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1448 // based on the escape status of the AllocateNode. 1449 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress)); 1450 } 1451 if (compress) { 1452 set_result(_gvn.transform(count)); 1453 } 1454 return true; 1455 } 1456 1457 #ifdef _LP64 1458 #define XTOP ,top() /*additional argument*/ 1459 #else //_LP64 1460 #define XTOP /*no additional argument*/ 1461 #endif //_LP64 1462 1463 //------------------------inline_string_toBytesU-------------------------- 1464 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1465 bool LibraryCallKit::inline_string_toBytesU() { 1466 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1467 return false; 1468 } 1469 // Get the arguments. 1470 Node* value = argument(0); 1471 Node* offset = argument(1); 1472 Node* length = argument(2); 1473 1474 Node* newcopy = NULL; 1475 1476 // Set the original stack and the reexecute bit for the interpreter to reexecute 1477 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1478 { PreserveReexecuteState preexecs(this); 1479 jvms()->set_should_reexecute(true); 1480 1481 // Check if a null path was taken unconditionally. 1482 value = null_check(value); 1483 1484 RegionNode* bailout = new RegionNode(1); 1485 record_for_igvn(bailout); 1486 1487 // Range checks 1488 generate_negative_guard(offset, bailout); 1489 generate_negative_guard(length, bailout); 1490 generate_limit_guard(offset, length, load_array_length(value), bailout); 1491 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1492 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1493 1494 if (bailout->req() > 1) { 1495 PreserveJVMState pjvms(this); 1496 set_control(_gvn.transform(bailout)); 1497 uncommon_trap(Deoptimization::Reason_intrinsic, 1498 Deoptimization::Action_maybe_recompile); 1499 } 1500 if (stopped()) { 1501 return true; 1502 } 1503 1504 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1505 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1506 newcopy = new_array(klass_node, size, 0); // no arguments to push 1507 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL); 1508 1509 // Calculate starting addresses. 1510 Node* src_start = array_element_address(value, offset, T_CHAR); 1511 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1512 1513 // Check if src array address is aligned to HeapWordSize (dst is always aligned) 1514 const TypeInt* toffset = gvn().type(offset)->is_int(); 1515 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1516 1517 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1518 const char* copyfunc_name = "arraycopy"; 1519 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1520 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1521 OptoRuntime::fast_arraycopy_Type(), 1522 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1523 src_start, dst_start, ConvI2X(length) XTOP); 1524 // Do not let reads from the cloned object float above the arraycopy. 1525 if (alloc != NULL) { 1526 if (alloc->maybe_set_complete(&_gvn)) { 1527 // "You break it, you buy it." 1528 InitializeNode* init = alloc->initialization(); 1529 assert(init->is_complete(), "we just did this"); 1530 init->set_complete_with_arraycopy(); 1531 assert(newcopy->is_CheckCastPP(), "sanity"); 1532 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1533 } 1534 // Do not let stores that initialize this object be reordered with 1535 // a subsequent store that would make this object accessible by 1536 // other threads. 1537 // Record what AllocateNode this StoreStore protects so that 1538 // escape analysis can go from the MemBarStoreStoreNode to the 1539 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1540 // based on the escape status of the AllocateNode. 1541 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress)); 1542 } else { 1543 insert_mem_bar(Op_MemBarCPUOrder); 1544 } 1545 } // original reexecute is set back here 1546 1547 C->set_has_split_ifs(true); // Has chance for split-if optimization 1548 if (!stopped()) { 1549 set_result(newcopy); 1550 } 1551 return true; 1552 } 1553 1554 //------------------------inline_string_getCharsU-------------------------- 1555 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1556 bool LibraryCallKit::inline_string_getCharsU() { 1557 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1558 return false; 1559 } 1560 1561 // Get the arguments. 1562 Node* src = argument(0); 1563 Node* src_begin = argument(1); 1564 Node* src_end = argument(2); // exclusive offset (i < src_end) 1565 Node* dst = argument(3); 1566 Node* dst_begin = argument(4); 1567 1568 // Check for allocation before we add nodes that would confuse 1569 // tightly_coupled_allocation() 1570 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1571 1572 // Check if a null path was taken unconditionally. 1573 src = null_check(src); 1574 dst = null_check(dst); 1575 if (stopped()) { 1576 return true; 1577 } 1578 1579 // Get length and convert char[] offset to byte[] offset 1580 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1581 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1582 1583 // Range checks 1584 generate_string_range_check(src, src_begin, length, true); 1585 generate_string_range_check(dst, dst_begin, length, false); 1586 if (stopped()) { 1587 return true; 1588 } 1589 1590 if (!stopped()) { 1591 // Calculate starting addresses. 1592 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1593 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1594 1595 // Check if array addresses are aligned to HeapWordSize 1596 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1597 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1598 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1599 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1600 1601 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1602 const char* copyfunc_name = "arraycopy"; 1603 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1604 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1605 OptoRuntime::fast_arraycopy_Type(), 1606 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1607 src_start, dst_start, ConvI2X(length) XTOP); 1608 // Do not let reads from the cloned object float above the arraycopy. 1609 if (alloc != NULL) { 1610 if (alloc->maybe_set_complete(&_gvn)) { 1611 // "You break it, you buy it." 1612 InitializeNode* init = alloc->initialization(); 1613 assert(init->is_complete(), "we just did this"); 1614 init->set_complete_with_arraycopy(); 1615 assert(dst->is_CheckCastPP(), "sanity"); 1616 assert(dst->in(0)->in(0) == init, "dest pinned"); 1617 } 1618 // Do not let stores that initialize this object be reordered with 1619 // a subsequent store that would make this object accessible by 1620 // other threads. 1621 // Record what AllocateNode this StoreStore protects so that 1622 // escape analysis can go from the MemBarStoreStoreNode to the 1623 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1624 // based on the escape status of the AllocateNode. 1625 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress)); 1626 } else { 1627 insert_mem_bar(Op_MemBarCPUOrder); 1628 } 1629 } 1630 1631 C->set_has_split_ifs(true); // Has chance for split-if optimization 1632 return true; 1633 } 1634 1635 //----------------------inline_string_char_access---------------------------- 1636 // Store/Load char to/from byte[] array. 1637 // static void StringUTF16.putChar(byte[] val, int index, int c) 1638 // static char StringUTF16.getChar(byte[] val, int index) 1639 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1640 Node* value = argument(0); 1641 Node* index = argument(1); 1642 Node* ch = is_store ? argument(2) : NULL; 1643 1644 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1645 // correctly requires matched array shapes. 1646 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1647 "sanity: byte[] and char[] bases agree"); 1648 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1649 "sanity: byte[] and char[] scales agree"); 1650 1651 // Bail when getChar over constants is requested: constant folding would 1652 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1653 // Java method would constant fold nicely instead. 1654 if (!is_store && value->is_Con() && index->is_Con()) { 1655 return false; 1656 } 1657 1658 Node* adr = array_element_address(value, index, T_CHAR); 1659 if (is_store) { 1660 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered, 1661 false, false, true /* mismatched */); 1662 } else { 1663 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered, 1664 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */); 1665 set_result(ch); 1666 } 1667 return true; 1668 } 1669 1670 //--------------------------round_double_node-------------------------------- 1671 // Round a double node if necessary. 1672 Node* LibraryCallKit::round_double_node(Node* n) { 1673 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1) 1674 n = _gvn.transform(new RoundDoubleNode(0, n)); 1675 return n; 1676 } 1677 1678 //------------------------------inline_math----------------------------------- 1679 // public static double Math.abs(double) 1680 // public static double Math.sqrt(double) 1681 // public static double Math.log(double) 1682 // public static double Math.log10(double) 1683 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1684 Node* arg = round_double_node(argument(0)); 1685 Node* n = NULL; 1686 switch (id) { 1687 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1688 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break; 1689 case vmIntrinsics::_dlog10: n = new Log10DNode(C, control(), arg); break; 1690 default: fatal_unexpected_iid(id); break; 1691 } 1692 set_result(_gvn.transform(n)); 1693 return true; 1694 } 1695 1696 //------------------------------inline_trig---------------------------------- 1697 // Inline sin/cos/tan instructions, if possible. If rounding is required, do 1698 // argument reduction which will turn into a fast/slow diamond. 1699 bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) { 1700 Node* arg = round_double_node(argument(0)); 1701 Node* n = NULL; 1702 1703 switch (id) { 1704 case vmIntrinsics::_dtan: n = new TanDNode(C, control(), arg); break; 1705 default: fatal_unexpected_iid(id); break; 1706 } 1707 n = _gvn.transform(n); 1708 1709 // Rounding required? Check for argument reduction! 1710 if (Matcher::strict_fp_requires_explicit_rounding) { 1711 static const double pi_4 = 0.7853981633974483; 1712 static const double neg_pi_4 = -0.7853981633974483; 1713 // pi/2 in 80-bit extended precision 1714 // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00}; 1715 // -pi/2 in 80-bit extended precision 1716 // static const unsigned char neg_pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0xbf,0x00,0x00,0x00,0x00,0x00,0x00}; 1717 // Cutoff value for using this argument reduction technique 1718 //static const double pi_2_minus_epsilon = 1.564660403643354; 1719 //static const double neg_pi_2_plus_epsilon = -1.564660403643354; 1720 1721 // Pseudocode for sin: 1722 // if (x <= Math.PI / 4.0) { 1723 // if (x >= -Math.PI / 4.0) return fsin(x); 1724 // if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0); 1725 // } else { 1726 // if (x <= Math.PI / 2.0) return fcos(x - Math.PI / 2.0); 1727 // } 1728 // return StrictMath.sin(x); 1729 1730 // Pseudocode for cos: 1731 // if (x <= Math.PI / 4.0) { 1732 // if (x >= -Math.PI / 4.0) return fcos(x); 1733 // if (x >= -Math.PI / 2.0) return fsin(x + Math.PI / 2.0); 1734 // } else { 1735 // if (x <= Math.PI / 2.0) return -fsin(x - Math.PI / 2.0); 1736 // } 1737 // return StrictMath.cos(x); 1738 1739 // Actually, sticking in an 80-bit Intel value into C2 will be tough; it 1740 // requires a special machine instruction to load it. Instead we'll try 1741 // the 'easy' case. If we really need the extra range +/- PI/2 we'll 1742 // probably do the math inside the SIN encoding. 1743 1744 // Make the merge point 1745 RegionNode* r = new RegionNode(3); 1746 Node* phi = new PhiNode(r, Type::DOUBLE); 1747 1748 // Flatten arg so we need only 1 test 1749 Node *abs = _gvn.transform(new AbsDNode(arg)); 1750 // Node for PI/4 constant 1751 Node *pi4 = makecon(TypeD::make(pi_4)); 1752 // Check PI/4 : abs(arg) 1753 Node *cmp = _gvn.transform(new CmpDNode(pi4,abs)); 1754 // Check: If PI/4 < abs(arg) then go slow 1755 Node *bol = _gvn.transform(new BoolNode( cmp, BoolTest::lt )); 1756 // Branch either way 1757 IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 1758 set_control(opt_iff(r,iff)); 1759 1760 // Set fast path result 1761 phi->init_req(2, n); 1762 1763 // Slow path - non-blocking leaf call 1764 Node* call = NULL; 1765 switch (id) { 1766 case vmIntrinsics::_dtan: 1767 call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(), 1768 CAST_FROM_FN_PTR(address, SharedRuntime::dtan), 1769 "Tan", NULL, arg, top()); 1770 break; 1771 } 1772 assert(control()->in(0) == call, ""); 1773 Node* slow_result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 1774 r->init_req(1, control()); 1775 phi->init_req(1, slow_result); 1776 1777 // Post-merge 1778 set_control(_gvn.transform(r)); 1779 record_for_igvn(r); 1780 n = _gvn.transform(phi); 1781 1782 C->set_has_split_ifs(true); // Has chance for split-if optimization 1783 } 1784 set_result(n); 1785 return true; 1786 } 1787 1788 //------------------------------runtime_math----------------------------- 1789 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1790 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1791 "must be (DD)D or (D)D type"); 1792 1793 // Inputs 1794 Node* a = round_double_node(argument(0)); 1795 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL; 1796 1797 const TypePtr* no_memory_effects = NULL; 1798 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName, 1799 no_memory_effects, 1800 a, top(), b, b ? top() : NULL); 1801 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1802 #ifdef ASSERT 1803 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1804 assert(value_top == top(), "second value must be top"); 1805 #endif 1806 1807 set_result(value); 1808 return true; 1809 } 1810 1811 //------------------------------inline_math_native----------------------------- 1812 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1813 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f) 1814 switch (id) { 1815 // These intrinsics are not properly supported on all hardware 1816 case vmIntrinsics::_dsin: 1817 return StubRoutines::dsin() != NULL ? 1818 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1819 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN"); 1820 case vmIntrinsics::_dcos: 1821 return StubRoutines::dcos() != NULL ? 1822 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1823 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS"); 1824 case vmIntrinsics::_dtan: return Matcher::has_match_rule(Op_TanD) ? inline_trig(id) : 1825 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN"); 1826 1827 case vmIntrinsics::_dlog: 1828 return StubRoutines::dlog() != NULL ? 1829 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1830 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG"); 1831 case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_math(id) : 1832 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10"); 1833 1834 // These intrinsics are supported on all hardware 1835 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false; 1836 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false; 1837 1838 case vmIntrinsics::_dexp: 1839 return StubRoutines::dexp() != NULL ? 1840 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1841 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP"); 1842 case vmIntrinsics::_dpow: 1843 return StubRoutines::dpow() != NULL ? 1844 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1845 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW"); 1846 #undef FN_PTR 1847 1848 // These intrinsics are not yet correctly implemented 1849 case vmIntrinsics::_datan2: 1850 return false; 1851 1852 default: 1853 fatal_unexpected_iid(id); 1854 return false; 1855 } 1856 } 1857 1858 static bool is_simple_name(Node* n) { 1859 return (n->req() == 1 // constant 1860 || (n->is_Type() && n->as_Type()->type()->singleton()) 1861 || n->is_Proj() // parameter or return value 1862 || n->is_Phi() // local of some sort 1863 ); 1864 } 1865 1866 //----------------------------inline_notify-----------------------------------* 1867 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1868 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1869 address func; 1870 if (id == vmIntrinsics::_notify) { 1871 func = OptoRuntime::monitor_notify_Java(); 1872 } else { 1873 func = OptoRuntime::monitor_notifyAll_Java(); 1874 } 1875 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0)); 1876 make_slow_call_ex(call, env()->Throwable_klass(), false); 1877 return true; 1878 } 1879 1880 1881 //----------------------------inline_min_max----------------------------------- 1882 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1883 set_result(generate_min_max(id, argument(0), argument(1))); 1884 return true; 1885 } 1886 1887 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) { 1888 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 1889 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 1890 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 1891 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 1892 1893 { 1894 PreserveJVMState pjvms(this); 1895 PreserveReexecuteState preexecs(this); 1896 jvms()->set_should_reexecute(true); 1897 1898 set_control(slow_path); 1899 set_i_o(i_o()); 1900 1901 uncommon_trap(Deoptimization::Reason_intrinsic, 1902 Deoptimization::Action_none); 1903 } 1904 1905 set_control(fast_path); 1906 set_result(math); 1907 } 1908 1909 template <typename OverflowOp> 1910 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 1911 typedef typename OverflowOp::MathOp MathOp; 1912 1913 MathOp* mathOp = new MathOp(arg1, arg2); 1914 Node* operation = _gvn.transform( mathOp ); 1915 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 1916 inline_math_mathExact(operation, ofcheck); 1917 return true; 1918 } 1919 1920 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 1921 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 1922 } 1923 1924 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 1925 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 1926 } 1927 1928 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 1929 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 1930 } 1931 1932 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 1933 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 1934 } 1935 1936 bool LibraryCallKit::inline_math_negateExactI() { 1937 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 1938 } 1939 1940 bool LibraryCallKit::inline_math_negateExactL() { 1941 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 1942 } 1943 1944 bool LibraryCallKit::inline_math_multiplyExactI() { 1945 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 1946 } 1947 1948 bool LibraryCallKit::inline_math_multiplyExactL() { 1949 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 1950 } 1951 1952 Node* 1953 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) { 1954 // These are the candidate return value: 1955 Node* xvalue = x0; 1956 Node* yvalue = y0; 1957 1958 if (xvalue == yvalue) { 1959 return xvalue; 1960 } 1961 1962 bool want_max = (id == vmIntrinsics::_max); 1963 1964 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int(); 1965 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int(); 1966 if (txvalue == NULL || tyvalue == NULL) return top(); 1967 // This is not really necessary, but it is consistent with a 1968 // hypothetical MaxINode::Value method: 1969 int widen = MAX2(txvalue->_widen, tyvalue->_widen); 1970 1971 // %%% This folding logic should (ideally) be in a different place. 1972 // Some should be inside IfNode, and there to be a more reliable 1973 // transformation of ?: style patterns into cmoves. We also want 1974 // more powerful optimizations around cmove and min/max. 1975 1976 // Try to find a dominating comparison of these guys. 1977 // It can simplify the index computation for Arrays.copyOf 1978 // and similar uses of System.arraycopy. 1979 // First, compute the normalized version of CmpI(x, y). 1980 int cmp_op = Op_CmpI; 1981 Node* xkey = xvalue; 1982 Node* ykey = yvalue; 1983 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey)); 1984 if (ideal_cmpxy->is_Cmp()) { 1985 // E.g., if we have CmpI(length - offset, count), 1986 // it might idealize to CmpI(length, count + offset) 1987 cmp_op = ideal_cmpxy->Opcode(); 1988 xkey = ideal_cmpxy->in(1); 1989 ykey = ideal_cmpxy->in(2); 1990 } 1991 1992 // Start by locating any relevant comparisons. 1993 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey; 1994 Node* cmpxy = NULL; 1995 Node* cmpyx = NULL; 1996 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) { 1997 Node* cmp = start_from->fast_out(k); 1998 if (cmp->outcnt() > 0 && // must have prior uses 1999 cmp->in(0) == NULL && // must be context-independent 2000 cmp->Opcode() == cmp_op) { // right kind of compare 2001 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp; 2002 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp; 2003 } 2004 } 2005 2006 const int NCMPS = 2; 2007 Node* cmps[NCMPS] = { cmpxy, cmpyx }; 2008 int cmpn; 2009 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2010 if (cmps[cmpn] != NULL) break; // find a result 2011 } 2012 if (cmpn < NCMPS) { 2013 // Look for a dominating test that tells us the min and max. 2014 int depth = 0; // Limit search depth for speed 2015 Node* dom = control(); 2016 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) { 2017 if (++depth >= 100) break; 2018 Node* ifproj = dom; 2019 if (!ifproj->is_Proj()) continue; 2020 Node* iff = ifproj->in(0); 2021 if (!iff->is_If()) continue; 2022 Node* bol = iff->in(1); 2023 if (!bol->is_Bool()) continue; 2024 Node* cmp = bol->in(1); 2025 if (cmp == NULL) continue; 2026 for (cmpn = 0; cmpn < NCMPS; cmpn++) 2027 if (cmps[cmpn] == cmp) break; 2028 if (cmpn == NCMPS) continue; 2029 BoolTest::mask btest = bol->as_Bool()->_test._test; 2030 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate(); 2031 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2032 // At this point, we know that 'x btest y' is true. 2033 switch (btest) { 2034 case BoolTest::eq: 2035 // They are proven equal, so we can collapse the min/max. 2036 // Either value is the answer. Choose the simpler. 2037 if (is_simple_name(yvalue) && !is_simple_name(xvalue)) 2038 return yvalue; 2039 return xvalue; 2040 case BoolTest::lt: // x < y 2041 case BoolTest::le: // x <= y 2042 return (want_max ? yvalue : xvalue); 2043 case BoolTest::gt: // x > y 2044 case BoolTest::ge: // x >= y 2045 return (want_max ? xvalue : yvalue); 2046 } 2047 } 2048 } 2049 2050 // We failed to find a dominating test. 2051 // Let's pick a test that might GVN with prior tests. 2052 Node* best_bol = NULL; 2053 BoolTest::mask best_btest = BoolTest::illegal; 2054 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2055 Node* cmp = cmps[cmpn]; 2056 if (cmp == NULL) continue; 2057 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) { 2058 Node* bol = cmp->fast_out(j); 2059 if (!bol->is_Bool()) continue; 2060 BoolTest::mask btest = bol->as_Bool()->_test._test; 2061 if (btest == BoolTest::eq || btest == BoolTest::ne) continue; 2062 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2063 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) { 2064 best_bol = bol->as_Bool(); 2065 best_btest = btest; 2066 } 2067 } 2068 } 2069 2070 Node* answer_if_true = NULL; 2071 Node* answer_if_false = NULL; 2072 switch (best_btest) { 2073 default: 2074 if (cmpxy == NULL) 2075 cmpxy = ideal_cmpxy; 2076 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt)); 2077 // and fall through: 2078 case BoolTest::lt: // x < y 2079 case BoolTest::le: // x <= y 2080 answer_if_true = (want_max ? yvalue : xvalue); 2081 answer_if_false = (want_max ? xvalue : yvalue); 2082 break; 2083 case BoolTest::gt: // x > y 2084 case BoolTest::ge: // x >= y 2085 answer_if_true = (want_max ? xvalue : yvalue); 2086 answer_if_false = (want_max ? yvalue : xvalue); 2087 break; 2088 } 2089 2090 jint hi, lo; 2091 if (want_max) { 2092 // We can sharpen the minimum. 2093 hi = MAX2(txvalue->_hi, tyvalue->_hi); 2094 lo = MAX2(txvalue->_lo, tyvalue->_lo); 2095 } else { 2096 // We can sharpen the maximum. 2097 hi = MIN2(txvalue->_hi, tyvalue->_hi); 2098 lo = MIN2(txvalue->_lo, tyvalue->_lo); 2099 } 2100 2101 // Use a flow-free graph structure, to avoid creating excess control edges 2102 // which could hinder other optimizations. 2103 // Since Math.min/max is often used with arraycopy, we want 2104 // tightly_coupled_allocation to be able to see beyond min/max expressions. 2105 Node* cmov = CMoveNode::make(NULL, best_bol, 2106 answer_if_false, answer_if_true, 2107 TypeInt::make(lo, hi, widen)); 2108 2109 return _gvn.transform(cmov); 2110 2111 /* 2112 // This is not as desirable as it may seem, since Min and Max 2113 // nodes do not have a full set of optimizations. 2114 // And they would interfere, anyway, with 'if' optimizations 2115 // and with CMoveI canonical forms. 2116 switch (id) { 2117 case vmIntrinsics::_min: 2118 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break; 2119 case vmIntrinsics::_max: 2120 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break; 2121 default: 2122 ShouldNotReachHere(); 2123 } 2124 */ 2125 } 2126 2127 inline int 2128 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) { 2129 const TypePtr* base_type = TypePtr::NULL_PTR; 2130 if (base != NULL) base_type = _gvn.type(base)->isa_ptr(); 2131 if (base_type == NULL) { 2132 // Unknown type. 2133 return Type::AnyPtr; 2134 } else if (base_type == TypePtr::NULL_PTR) { 2135 // Since this is a NULL+long form, we have to switch to a rawptr. 2136 base = _gvn.transform(new CastX2PNode(offset)); 2137 offset = MakeConX(0); 2138 return Type::RawPtr; 2139 } else if (base_type->base() == Type::RawPtr) { 2140 return Type::RawPtr; 2141 } else if (base_type->isa_oopptr()) { 2142 // Base is never null => always a heap address. 2143 if (base_type->ptr() == TypePtr::NotNull) { 2144 return Type::OopPtr; 2145 } 2146 // Offset is small => always a heap address. 2147 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2148 if (offset_type != NULL && 2149 base_type->offset() == 0 && // (should always be?) 2150 offset_type->_lo >= 0 && 2151 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2152 return Type::OopPtr; 2153 } 2154 // Otherwise, it might either be oop+off or NULL+addr. 2155 return Type::AnyPtr; 2156 } else { 2157 // No information: 2158 return Type::AnyPtr; 2159 } 2160 } 2161 2162 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) { 2163 int kind = classify_unsafe_addr(base, offset); 2164 if (kind == Type::RawPtr) { 2165 return basic_plus_adr(top(), base, offset); 2166 } else { 2167 return basic_plus_adr(base, offset); 2168 } 2169 } 2170 2171 //--------------------------inline_number_methods----------------------------- 2172 // inline int Integer.numberOfLeadingZeros(int) 2173 // inline int Long.numberOfLeadingZeros(long) 2174 // 2175 // inline int Integer.numberOfTrailingZeros(int) 2176 // inline int Long.numberOfTrailingZeros(long) 2177 // 2178 // inline int Integer.bitCount(int) 2179 // inline int Long.bitCount(long) 2180 // 2181 // inline char Character.reverseBytes(char) 2182 // inline short Short.reverseBytes(short) 2183 // inline int Integer.reverseBytes(int) 2184 // inline long Long.reverseBytes(long) 2185 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2186 Node* arg = argument(0); 2187 Node* n = NULL; 2188 switch (id) { 2189 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2190 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2191 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2192 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2193 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2194 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2195 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break; 2196 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break; 2197 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break; 2198 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break; 2199 default: fatal_unexpected_iid(id); break; 2200 } 2201 set_result(_gvn.transform(n)); 2202 return true; 2203 } 2204 2205 //----------------------------inline_unsafe_access---------------------------- 2206 2207 const static BasicType T_ADDRESS_HOLDER = T_LONG; 2208 2209 // Helper that guards and inserts a pre-barrier. 2210 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset, 2211 Node* pre_val, bool need_mem_bar) { 2212 // We could be accessing the referent field of a reference object. If so, when G1 2213 // is enabled, we need to log the value in the referent field in an SATB buffer. 2214 // This routine performs some compile time filters and generates suitable 2215 // runtime filters that guard the pre-barrier code. 2216 // Also add memory barrier for non volatile load from the referent field 2217 // to prevent commoning of loads across safepoint. 2218 if (!UseG1GC && !need_mem_bar) 2219 return; 2220 2221 // Some compile time checks. 2222 2223 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset? 2224 const TypeX* otype = offset->find_intptr_t_type(); 2225 if (otype != NULL && otype->is_con() && 2226 otype->get_con() != java_lang_ref_Reference::referent_offset) { 2227 // Constant offset but not the reference_offset so just return 2228 return; 2229 } 2230 2231 // We only need to generate the runtime guards for instances. 2232 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr(); 2233 if (btype != NULL) { 2234 if (btype->isa_aryptr()) { 2235 // Array type so nothing to do 2236 return; 2237 } 2238 2239 const TypeInstPtr* itype = btype->isa_instptr(); 2240 if (itype != NULL) { 2241 // Can the klass of base_oop be statically determined to be 2242 // _not_ a sub-class of Reference and _not_ Object? 2243 ciKlass* klass = itype->klass(); 2244 if ( klass->is_loaded() && 2245 !klass->is_subtype_of(env()->Reference_klass()) && 2246 !env()->Object_klass()->is_subtype_of(klass)) { 2247 return; 2248 } 2249 } 2250 } 2251 2252 // The compile time filters did not reject base_oop/offset so 2253 // we need to generate the following runtime filters 2254 // 2255 // if (offset == java_lang_ref_Reference::_reference_offset) { 2256 // if (instance_of(base, java.lang.ref.Reference)) { 2257 // pre_barrier(_, pre_val, ...); 2258 // } 2259 // } 2260 2261 float likely = PROB_LIKELY( 0.999); 2262 float unlikely = PROB_UNLIKELY(0.999); 2263 2264 IdealKit ideal(this); 2265 #define __ ideal. 2266 2267 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset); 2268 2269 __ if_then(offset, BoolTest::eq, referent_off, unlikely); { 2270 // Update graphKit memory and control from IdealKit. 2271 sync_kit(ideal); 2272 2273 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass())); 2274 Node* is_instof = gen_instanceof(base_oop, ref_klass_con); 2275 2276 // Update IdealKit memory and control from graphKit. 2277 __ sync_kit(this); 2278 2279 Node* one = __ ConI(1); 2280 // is_instof == 0 if base_oop == NULL 2281 __ if_then(is_instof, BoolTest::eq, one, unlikely); { 2282 2283 // Update graphKit from IdeakKit. 2284 sync_kit(ideal); 2285 2286 // Use the pre-barrier to record the value in the referent field 2287 pre_barrier(false /* do_load */, 2288 __ ctrl(), 2289 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */, 2290 pre_val /* pre_val */, 2291 T_OBJECT); 2292 if (need_mem_bar) { 2293 // Add memory barrier to prevent commoning reads from this field 2294 // across safepoint since GC can change its value. 2295 insert_mem_bar(Op_MemBarCPUOrder); 2296 } 2297 // Update IdealKit from graphKit. 2298 __ sync_kit(this); 2299 2300 } __ end_if(); // _ref_type != ref_none 2301 } __ end_if(); // offset == referent_offset 2302 2303 // Final sync IdealKit and GraphKit. 2304 final_sync(ideal); 2305 #undef __ 2306 } 2307 2308 2309 // Interpret Unsafe.fieldOffset cookies correctly: 2310 extern jlong Unsafe_field_offset_to_byte_offset(jlong field_offset); 2311 2312 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) { 2313 // Attempt to infer a sharper value type from the offset and base type. 2314 ciKlass* sharpened_klass = NULL; 2315 2316 // See if it is an instance field, with an object type. 2317 if (alias_type->field() != NULL) { 2318 assert(!is_native_ptr, "native pointer op cannot use a java address"); 2319 if (alias_type->field()->type()->is_klass()) { 2320 sharpened_klass = alias_type->field()->type()->as_klass(); 2321 } 2322 } 2323 2324 // See if it is a narrow oop array. 2325 if (adr_type->isa_aryptr()) { 2326 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2327 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr(); 2328 if (elem_type != NULL) { 2329 sharpened_klass = elem_type->klass(); 2330 } 2331 } 2332 } 2333 2334 // The sharpened class might be unloaded if there is no class loader 2335 // contraint in place. 2336 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) { 2337 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass); 2338 2339 #ifndef PRODUCT 2340 if (C->print_intrinsics() || C->print_inlining()) { 2341 tty->print(" from base type: "); adr_type->dump(); 2342 tty->print(" sharpened value: "); tjp->dump(); 2343 } 2344 #endif 2345 // Sharpen the value type. 2346 return tjp; 2347 } 2348 return NULL; 2349 } 2350 2351 bool LibraryCallKit::inline_unsafe_access(const bool is_native_ptr, bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) { 2352 if (callee()->is_static()) return false; // caller must have the capability! 2353 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2354 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2355 2356 #ifndef PRODUCT 2357 { 2358 ResourceMark rm; 2359 // Check the signatures. 2360 ciSignature* sig = callee()->signature(); 2361 #ifdef ASSERT 2362 if (!is_store) { 2363 // Object getObject(Object base, int/long offset), etc. 2364 BasicType rtype = sig->return_type()->basic_type(); 2365 if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name()) 2366 rtype = T_ADDRESS; // it is really a C void* 2367 assert(rtype == type, "getter must return the expected value"); 2368 if (!is_native_ptr) { 2369 assert(sig->count() == 2, "oop getter has 2 arguments"); 2370 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2371 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2372 } else { 2373 assert(sig->count() == 1, "native getter has 1 argument"); 2374 assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long"); 2375 } 2376 } else { 2377 // void putObject(Object base, int/long offset, Object x), etc. 2378 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2379 if (!is_native_ptr) { 2380 assert(sig->count() == 3, "oop putter has 3 arguments"); 2381 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2382 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2383 } else { 2384 assert(sig->count() == 2, "native putter has 2 arguments"); 2385 assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long"); 2386 } 2387 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2388 if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name()) 2389 vtype = T_ADDRESS; // it is really a C void* 2390 assert(vtype == type, "putter must accept the expected value"); 2391 } 2392 #endif // ASSERT 2393 } 2394 #endif //PRODUCT 2395 2396 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2397 2398 Node* receiver = argument(0); // type: oop 2399 2400 // Build address expression. 2401 Node* adr; 2402 Node* heap_base_oop = top(); 2403 Node* offset = top(); 2404 Node* val; 2405 2406 if (!is_native_ptr) { 2407 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2408 Node* base = argument(1); // type: oop 2409 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2410 offset = argument(2); // type: long 2411 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2412 // to be plain byte offsets, which are also the same as those accepted 2413 // by oopDesc::field_base. 2414 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2415 "fieldOffset must be byte-scaled"); 2416 // 32-bit machines ignore the high half! 2417 offset = ConvL2X(offset); 2418 adr = make_unsafe_address(base, offset); 2419 heap_base_oop = base; 2420 val = is_store ? argument(4) : NULL; 2421 } else { 2422 Node* ptr = argument(1); // type: long 2423 ptr = ConvL2X(ptr); // adjust Java long to machine word 2424 adr = make_unsafe_address(NULL, ptr); 2425 val = is_store ? argument(3) : NULL; 2426 } 2427 2428 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2429 2430 // First guess at the value type. 2431 const Type *value_type = Type::get_const_basic_type(type); 2432 2433 // Try to categorize the address. If it comes up as TypeJavaPtr::BOTTOM, 2434 // there was not enough information to nail it down. 2435 Compile::AliasType* alias_type = C->alias_type(adr_type); 2436 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2437 2438 // We will need memory barriers unless we can determine a unique 2439 // alias category for this reference. (Note: If for some reason 2440 // the barriers get omitted and the unsafe reference begins to "pollute" 2441 // the alias analysis of the rest of the graph, either Compile::can_alias 2442 // or Compile::must_alias will throw a diagnostic assert.) 2443 bool need_mem_bar; 2444 switch (kind) { 2445 case Relaxed: 2446 need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM); 2447 break; 2448 case Opaque: 2449 // Opaque uses CPUOrder membars for protection against code movement. 2450 case Acquire: 2451 case Release: 2452 case Volatile: 2453 need_mem_bar = true; 2454 break; 2455 default: 2456 ShouldNotReachHere(); 2457 } 2458 2459 // Some accesses require access atomicity for all types, notably longs and doubles. 2460 // When AlwaysAtomicAccesses is enabled, all accesses are atomic. 2461 bool requires_atomic_access = false; 2462 switch (kind) { 2463 case Relaxed: 2464 case Opaque: 2465 requires_atomic_access = AlwaysAtomicAccesses; 2466 break; 2467 case Acquire: 2468 case Release: 2469 case Volatile: 2470 requires_atomic_access = true; 2471 break; 2472 default: 2473 ShouldNotReachHere(); 2474 } 2475 2476 // Figure out the memory ordering. 2477 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd 2478 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store); 2479 2480 // If we are reading the value of the referent field of a Reference 2481 // object (either by using Unsafe directly or through reflection) 2482 // then, if G1 is enabled, we need to record the referent in an 2483 // SATB log buffer using the pre-barrier mechanism. 2484 // Also we need to add memory barrier to prevent commoning reads 2485 // from this field across safepoint since GC can change its value. 2486 bool need_read_barrier = !is_native_ptr && !is_store && 2487 offset != top() && heap_base_oop != top(); 2488 2489 if (!is_store && type == T_OBJECT) { 2490 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr); 2491 if (tjp != NULL) { 2492 value_type = tjp; 2493 } 2494 } 2495 2496 receiver = null_check(receiver); 2497 if (stopped()) { 2498 return true; 2499 } 2500 // Heap pointers get a null-check from the interpreter, 2501 // as a courtesy. However, this is not guaranteed by Unsafe, 2502 // and it is not possible to fully distinguish unintended nulls 2503 // from intended ones in this API. 2504 2505 // We need to emit leading and trailing CPU membars (see below) in 2506 // addition to memory membars for special access modes. This is a little 2507 // too strong, but avoids the need to insert per-alias-type 2508 // volatile membars (for stores; compare Parse::do_put_xxx), which 2509 // we cannot do effectively here because we probably only have a 2510 // rough approximation of type. 2511 2512 switch(kind) { 2513 case Relaxed: 2514 case Opaque: 2515 case Acquire: 2516 break; 2517 case Release: 2518 case Volatile: 2519 if (is_store) { 2520 insert_mem_bar(Op_MemBarRelease); 2521 } else { 2522 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { 2523 insert_mem_bar(Op_MemBarVolatile); 2524 } 2525 } 2526 break; 2527 default: 2528 ShouldNotReachHere(); 2529 } 2530 2531 // Memory barrier to prevent normal and 'unsafe' accesses from 2532 // bypassing each other. Happens after null checks, so the 2533 // exception paths do not take memory state from the memory barrier, 2534 // so there's no problems making a strong assert about mixing users 2535 // of safe & unsafe memory. 2536 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder); 2537 2538 assert(alias_type->adr_type() == TypeRawPtr::BOTTOM || alias_type->adr_type() == TypeOopPtr::BOTTOM || 2539 alias_type->field() != NULL || alias_type->element() != NULL, "field, array element or unknown"); 2540 bool mismatched = false; 2541 if (alias_type->element() != NULL || alias_type->field() != NULL) { 2542 BasicType bt; 2543 if (alias_type->element() != NULL) { 2544 // Use address type to get the element type. Alias type doesn't provide 2545 // enough information (e.g., doesn't differentiate between byte[] and boolean[]). 2546 const Type* element = adr_type->is_aryptr()->elem(); 2547 bt = element->isa_narrowoop() ? T_OBJECT : element->array_element_basic_type(); 2548 } else { 2549 bt = alias_type->field()->layout_type(); 2550 } 2551 if (bt == T_ARRAY) { 2552 // accessing an array field with getObject is not a mismatch 2553 bt = T_OBJECT; 2554 } 2555 if (bt != type) { 2556 mismatched = true; 2557 } 2558 } 2559 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2560 2561 if (!is_store) { 2562 Node* p = NULL; 2563 // Try to constant fold a load from a constant field 2564 ciField* field = alias_type->field(); 2565 if (heap_base_oop != top() && 2566 field != NULL && field->is_constant() && !mismatched) { 2567 // final or stable field 2568 const Type* con_type = Type::make_constant(alias_type->field(), heap_base_oop); 2569 if (con_type != NULL) { 2570 p = makecon(con_type); 2571 } 2572 } 2573 if (p == NULL) { 2574 // To be valid, unsafe loads may depend on other conditions than 2575 // the one that guards them: pin the Load node 2576 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched); 2577 // load value 2578 switch (type) { 2579 case T_BOOLEAN: 2580 case T_CHAR: 2581 case T_BYTE: 2582 case T_SHORT: 2583 case T_INT: 2584 case T_LONG: 2585 case T_FLOAT: 2586 case T_DOUBLE: 2587 break; 2588 case T_OBJECT: 2589 if (need_read_barrier) { 2590 // We do not require a mem bar inside pre_barrier if need_mem_bar 2591 // is set: the barriers would be emitted by us. 2592 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar); 2593 } 2594 break; 2595 case T_ADDRESS: 2596 // Cast to an int type. 2597 p = _gvn.transform(new CastP2XNode(NULL, p)); 2598 p = ConvX2UL(p); 2599 break; 2600 default: 2601 fatal("unexpected type %d: %s", type, type2name(type)); 2602 break; 2603 } 2604 } 2605 // The load node has the control of the preceding MemBarCPUOrder. All 2606 // following nodes will have the control of the MemBarCPUOrder inserted at 2607 // the end of this method. So, pushing the load onto the stack at a later 2608 // point is fine. 2609 set_result(p); 2610 } else { 2611 // place effect of store into memory 2612 switch (type) { 2613 case T_DOUBLE: 2614 val = dstore_rounding(val); 2615 break; 2616 case T_ADDRESS: 2617 // Repackage the long as a pointer. 2618 val = ConvL2X(val); 2619 val = _gvn.transform(new CastX2PNode(val)); 2620 break; 2621 } 2622 2623 if (type != T_OBJECT) { 2624 (void) store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched); 2625 } else { 2626 // Possibly an oop being stored to Java heap or native memory 2627 if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) { 2628 // oop to Java heap. 2629 (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched); 2630 } else { 2631 // We can't tell at compile time if we are storing in the Java heap or outside 2632 // of it. So we need to emit code to conditionally do the proper type of 2633 // store. 2634 2635 IdealKit ideal(this); 2636 #define __ ideal. 2637 // QQQ who knows what probability is here?? 2638 __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); { 2639 // Sync IdealKit and graphKit. 2640 sync_kit(ideal); 2641 Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched); 2642 // Update IdealKit memory. 2643 __ sync_kit(this); 2644 } __ else_(); { 2645 __ store(__ ctrl(), adr, val, type, alias_type->index(), mo, requires_atomic_access, mismatched); 2646 } __ end_if(); 2647 // Final sync IdealKit and GraphKit. 2648 final_sync(ideal); 2649 #undef __ 2650 } 2651 } 2652 } 2653 2654 switch(kind) { 2655 case Relaxed: 2656 case Opaque: 2657 case Release: 2658 break; 2659 case Acquire: 2660 case Volatile: 2661 if (!is_store) { 2662 insert_mem_bar(Op_MemBarAcquire); 2663 } else { 2664 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { 2665 insert_mem_bar(Op_MemBarVolatile); 2666 } 2667 } 2668 break; 2669 default: 2670 ShouldNotReachHere(); 2671 } 2672 2673 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder); 2674 2675 return true; 2676 } 2677 2678 //----------------------------inline_unsafe_load_store---------------------------- 2679 // This method serves a couple of different customers (depending on LoadStoreKind): 2680 // 2681 // LS_cmp_swap: 2682 // 2683 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x); 2684 // boolean compareAndSwapInt( Object o, long offset, int expected, int x); 2685 // boolean compareAndSwapLong( Object o, long offset, long expected, long x); 2686 // 2687 // LS_cmp_swap_weak: 2688 // 2689 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x); 2690 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x); 2691 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x); 2692 // 2693 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x); 2694 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x); 2695 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x); 2696 // 2697 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x); 2698 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x); 2699 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x); 2700 // 2701 // LS_cmp_exchange: 2702 // 2703 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x); 2704 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x); 2705 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x); 2706 // 2707 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2708 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2709 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2710 // 2711 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2712 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2713 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2714 // 2715 // LS_get_add: 2716 // 2717 // int getAndAddInt( Object o, long offset, int delta) 2718 // long getAndAddLong(Object o, long offset, long delta) 2719 // 2720 // LS_get_set: 2721 // 2722 // int getAndSet(Object o, long offset, int newValue) 2723 // long getAndSet(Object o, long offset, long newValue) 2724 // Object getAndSet(Object o, long offset, Object newValue) 2725 // 2726 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2727 // This basic scheme here is the same as inline_unsafe_access, but 2728 // differs in enough details that combining them would make the code 2729 // overly confusing. (This is a true fact! I originally combined 2730 // them, but even I was confused by it!) As much code/comments as 2731 // possible are retained from inline_unsafe_access though to make 2732 // the correspondences clearer. - dl 2733 2734 if (callee()->is_static()) return false; // caller must have the capability! 2735 2736 #ifndef PRODUCT 2737 BasicType rtype; 2738 { 2739 ResourceMark rm; 2740 // Check the signatures. 2741 ciSignature* sig = callee()->signature(); 2742 rtype = sig->return_type()->basic_type(); 2743 switch(kind) { 2744 case LS_get_add: 2745 case LS_get_set: { 2746 // Check the signatures. 2747 #ifdef ASSERT 2748 assert(rtype == type, "get and set must return the expected type"); 2749 assert(sig->count() == 3, "get and set has 3 arguments"); 2750 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2751 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2752 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2753 #endif // ASSERT 2754 break; 2755 } 2756 case LS_cmp_swap: 2757 case LS_cmp_swap_weak: { 2758 // Check the signatures. 2759 #ifdef ASSERT 2760 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2761 assert(sig->count() == 4, "CAS has 4 arguments"); 2762 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2763 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2764 #endif // ASSERT 2765 break; 2766 } 2767 case LS_cmp_exchange: { 2768 // Check the signatures. 2769 #ifdef ASSERT 2770 assert(rtype == type, "CAS must return the expected type"); 2771 assert(sig->count() == 4, "CAS has 4 arguments"); 2772 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2773 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2774 #endif // ASSERT 2775 break; 2776 } 2777 default: 2778 ShouldNotReachHere(); 2779 } 2780 } 2781 #endif //PRODUCT 2782 2783 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2784 2785 // Get arguments: 2786 Node* receiver = NULL; 2787 Node* base = NULL; 2788 Node* offset = NULL; 2789 Node* oldval = NULL; 2790 Node* newval = NULL; 2791 switch(kind) { 2792 case LS_cmp_swap: 2793 case LS_cmp_swap_weak: 2794 case LS_cmp_exchange: { 2795 const bool two_slot_type = type2size[type] == 2; 2796 receiver = argument(0); // type: oop 2797 base = argument(1); // type: oop 2798 offset = argument(2); // type: long 2799 oldval = argument(4); // type: oop, int, or long 2800 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2801 break; 2802 } 2803 case LS_get_add: 2804 case LS_get_set: { 2805 receiver = argument(0); // type: oop 2806 base = argument(1); // type: oop 2807 offset = argument(2); // type: long 2808 oldval = NULL; 2809 newval = argument(4); // type: oop, int, or long 2810 break; 2811 } 2812 default: 2813 ShouldNotReachHere(); 2814 } 2815 2816 // Null check receiver. 2817 receiver = null_check(receiver); 2818 if (stopped()) { 2819 return true; 2820 } 2821 2822 // Build field offset expression. 2823 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2824 // to be plain byte offsets, which are also the same as those accepted 2825 // by oopDesc::field_base. 2826 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2827 // 32-bit machines ignore the high half of long offsets 2828 offset = ConvL2X(offset); 2829 Node* adr = make_unsafe_address(base, offset); 2830 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2831 2832 // For CAS, unlike inline_unsafe_access, there seems no point in 2833 // trying to refine types. Just use the coarse types here. 2834 const Type *value_type = Type::get_const_basic_type(type); 2835 Compile::AliasType* alias_type = C->alias_type(adr_type); 2836 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2837 2838 switch (kind) { 2839 case LS_get_set: 2840 case LS_cmp_exchange: { 2841 if (type == T_OBJECT) { 2842 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2843 if (tjp != NULL) { 2844 value_type = tjp; 2845 } 2846 } 2847 break; 2848 } 2849 case LS_cmp_swap: 2850 case LS_cmp_swap_weak: 2851 case LS_get_add: 2852 break; 2853 default: 2854 ShouldNotReachHere(); 2855 } 2856 2857 int alias_idx = C->get_alias_index(adr_type); 2858 2859 // Memory-model-wise, a LoadStore acts like a little synchronized 2860 // block, so needs barriers on each side. These don't translate 2861 // into actual barriers on most machines, but we still need rest of 2862 // compiler to respect ordering. 2863 2864 switch (access_kind) { 2865 case Relaxed: 2866 case Acquire: 2867 break; 2868 case Release: 2869 case Volatile: 2870 insert_mem_bar(Op_MemBarRelease); 2871 break; 2872 default: 2873 ShouldNotReachHere(); 2874 } 2875 insert_mem_bar(Op_MemBarCPUOrder); 2876 2877 // Figure out the memory ordering. 2878 MemNode::MemOrd mo = access_kind_to_memord(access_kind); 2879 2880 // 4984716: MemBars must be inserted before this 2881 // memory node in order to avoid a false 2882 // dependency which will confuse the scheduler. 2883 Node *mem = memory(alias_idx); 2884 2885 // For now, we handle only those cases that actually exist: ints, 2886 // longs, and Object. Adding others should be straightforward. 2887 Node* load_store = NULL; 2888 switch(type) { 2889 case T_INT: 2890 switch(kind) { 2891 case LS_get_add: 2892 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type)); 2893 break; 2894 case LS_get_set: 2895 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type)); 2896 break; 2897 case LS_cmp_swap_weak: 2898 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo)); 2899 break; 2900 case LS_cmp_swap: 2901 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo)); 2902 break; 2903 case LS_cmp_exchange: 2904 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo)); 2905 break; 2906 default: 2907 ShouldNotReachHere(); 2908 } 2909 break; 2910 case T_LONG: 2911 switch(kind) { 2912 case LS_get_add: 2913 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type)); 2914 break; 2915 case LS_get_set: 2916 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type)); 2917 break; 2918 case LS_cmp_swap_weak: 2919 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo)); 2920 break; 2921 case LS_cmp_swap: 2922 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo)); 2923 break; 2924 case LS_cmp_exchange: 2925 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo)); 2926 break; 2927 default: 2928 ShouldNotReachHere(); 2929 } 2930 break; 2931 case T_OBJECT: 2932 // Transformation of a value which could be NULL pointer (CastPP #NULL) 2933 // could be delayed during Parse (for example, in adjust_map_after_if()). 2934 // Execute transformation here to avoid barrier generation in such case. 2935 if (_gvn.type(newval) == TypePtr::NULL_PTR) 2936 newval = _gvn.makecon(TypePtr::NULL_PTR); 2937 2938 // Reference stores need a store barrier. 2939 switch(kind) { 2940 case LS_get_set: { 2941 // If pre-barrier must execute before the oop store, old value will require do_load here. 2942 if (!can_move_pre_barrier()) { 2943 pre_barrier(true /* do_load*/, 2944 control(), base, adr, alias_idx, newval, value_type->make_oopptr(), 2945 NULL /* pre_val*/, 2946 T_OBJECT); 2947 } // Else move pre_barrier to use load_store value, see below. 2948 break; 2949 } 2950 case LS_cmp_swap_weak: 2951 case LS_cmp_swap: 2952 case LS_cmp_exchange: { 2953 // Same as for newval above: 2954 if (_gvn.type(oldval) == TypePtr::NULL_PTR) { 2955 oldval = _gvn.makecon(TypePtr::NULL_PTR); 2956 } 2957 // The only known value which might get overwritten is oldval. 2958 pre_barrier(false /* do_load */, 2959 control(), NULL, NULL, max_juint, NULL, NULL, 2960 oldval /* pre_val */, 2961 T_OBJECT); 2962 break; 2963 } 2964 default: 2965 ShouldNotReachHere(); 2966 } 2967 2968 #ifdef _LP64 2969 if (adr->bottom_type()->is_ptr_to_narrowoop()) { 2970 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop())); 2971 2972 switch(kind) { 2973 case LS_get_set: 2974 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop())); 2975 break; 2976 case LS_cmp_swap_weak: { 2977 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop())); 2978 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo)); 2979 break; 2980 } 2981 case LS_cmp_swap: { 2982 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop())); 2983 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo)); 2984 break; 2985 } 2986 case LS_cmp_exchange: { 2987 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop())); 2988 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo)); 2989 break; 2990 } 2991 default: 2992 ShouldNotReachHere(); 2993 } 2994 } else 2995 #endif 2996 switch (kind) { 2997 case LS_get_set: 2998 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr())); 2999 break; 3000 case LS_cmp_swap_weak: 3001 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo)); 3002 break; 3003 case LS_cmp_swap: 3004 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo)); 3005 break; 3006 case LS_cmp_exchange: 3007 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo)); 3008 break; 3009 default: 3010 ShouldNotReachHere(); 3011 } 3012 3013 // Emit the post barrier only when the actual store happened. This makes sense 3014 // to check only for LS_cmp_* that can fail to set the value. 3015 // LS_cmp_exchange does not produce any branches by default, so there is no 3016 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with 3017 // CompareAndExchange and move branches here, it would make sense to conditionalize 3018 // post_barriers for LS_cmp_exchange as well. 3019 // 3020 // CAS success path is marked more likely since we anticipate this is a performance 3021 // critical path, while CAS failure path can use the penalty for going through unlikely 3022 // path as backoff. Which is still better than doing a store barrier there. 3023 switch (kind) { 3024 case LS_get_set: 3025 case LS_cmp_exchange: { 3026 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true); 3027 break; 3028 } 3029 case LS_cmp_swap_weak: 3030 case LS_cmp_swap: { 3031 IdealKit ideal(this); 3032 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); { 3033 sync_kit(ideal); 3034 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true); 3035 ideal.sync_kit(this); 3036 } ideal.end_if(); 3037 final_sync(ideal); 3038 break; 3039 } 3040 default: 3041 ShouldNotReachHere(); 3042 } 3043 break; 3044 default: 3045 fatal("unexpected type %d: %s", type, type2name(type)); 3046 break; 3047 } 3048 3049 // SCMemProjNodes represent the memory state of a LoadStore. Their 3050 // main role is to prevent LoadStore nodes from being optimized away 3051 // when their results aren't used. 3052 Node* proj = _gvn.transform(new SCMemProjNode(load_store)); 3053 set_memory(proj, alias_idx); 3054 3055 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) { 3056 #ifdef _LP64 3057 if (adr->bottom_type()->is_ptr_to_narrowoop()) { 3058 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type())); 3059 } 3060 #endif 3061 if (can_move_pre_barrier()) { 3062 // Don't need to load pre_val. The old value is returned by load_store. 3063 // The pre_barrier can execute after the xchg as long as no safepoint 3064 // gets inserted between them. 3065 pre_barrier(false /* do_load */, 3066 control(), NULL, NULL, max_juint, NULL, NULL, 3067 load_store /* pre_val */, 3068 T_OBJECT); 3069 } 3070 } 3071 3072 // Add the trailing membar surrounding the access 3073 insert_mem_bar(Op_MemBarCPUOrder); 3074 3075 switch (access_kind) { 3076 case Relaxed: 3077 case Release: 3078 break; // do nothing 3079 case Acquire: 3080 case Volatile: 3081 insert_mem_bar(Op_MemBarAcquire); 3082 break; 3083 default: 3084 ShouldNotReachHere(); 3085 } 3086 3087 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 3088 set_result(load_store); 3089 return true; 3090 } 3091 3092 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) { 3093 MemNode::MemOrd mo = MemNode::unset; 3094 switch(kind) { 3095 case Opaque: 3096 case Relaxed: mo = MemNode::unordered; break; 3097 case Acquire: mo = MemNode::acquire; break; 3098 case Release: mo = MemNode::release; break; 3099 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break; 3100 default: 3101 ShouldNotReachHere(); 3102 } 3103 guarantee(mo != MemNode::unset, "Should select memory ordering"); 3104 return mo; 3105 } 3106 3107 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) { 3108 MemNode::MemOrd mo = MemNode::unset; 3109 switch(kind) { 3110 case Opaque: 3111 case Relaxed: mo = MemNode::unordered; break; 3112 case Acquire: mo = MemNode::acquire; break; 3113 case Release: mo = MemNode::release; break; 3114 case Volatile: mo = MemNode::seqcst; break; 3115 default: 3116 ShouldNotReachHere(); 3117 } 3118 guarantee(mo != MemNode::unset, "Should select memory ordering"); 3119 return mo; 3120 } 3121 3122 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 3123 // Regardless of form, don't allow previous ld/st to move down, 3124 // then issue acquire, release, or volatile mem_bar. 3125 insert_mem_bar(Op_MemBarCPUOrder); 3126 switch(id) { 3127 case vmIntrinsics::_loadFence: 3128 insert_mem_bar(Op_LoadFence); 3129 return true; 3130 case vmIntrinsics::_storeFence: 3131 insert_mem_bar(Op_StoreFence); 3132 return true; 3133 case vmIntrinsics::_fullFence: 3134 insert_mem_bar(Op_MemBarVolatile); 3135 return true; 3136 default: 3137 fatal_unexpected_iid(id); 3138 return false; 3139 } 3140 } 3141 3142 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 3143 if (!kls->is_Con()) { 3144 return true; 3145 } 3146 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr(); 3147 if (klsptr == NULL) { 3148 return true; 3149 } 3150 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass(); 3151 // don't need a guard for a klass that is already initialized 3152 return !ik->is_initialized(); 3153 } 3154 3155 //----------------------------inline_unsafe_allocate--------------------------- 3156 // public native Object Unsafe.allocateInstance(Class<?> cls); 3157 bool LibraryCallKit::inline_unsafe_allocate() { 3158 if (callee()->is_static()) return false; // caller must have the capability! 3159 3160 null_check_receiver(); // null-check, then ignore 3161 Node* cls = null_check(argument(1)); 3162 if (stopped()) return true; 3163 3164 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 3165 kls = null_check(kls); 3166 if (stopped()) return true; // argument was like int.class 3167 3168 Node* test = NULL; 3169 if (LibraryCallKit::klass_needs_init_guard(kls)) { 3170 // Note: The argument might still be an illegal value like 3171 // Serializable.class or Object[].class. The runtime will handle it. 3172 // But we must make an explicit check for initialization. 3173 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 3174 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 3175 // can generate code to load it as unsigned byte. 3176 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered); 3177 Node* bits = intcon(InstanceKlass::fully_initialized); 3178 test = _gvn.transform(new SubINode(inst, bits)); 3179 // The 'test' is non-zero if we need to take a slow path. 3180 } 3181 3182 Node* obj = new_instance(kls, test); 3183 set_result(obj); 3184 return true; 3185 } 3186 3187 #ifdef TRACE_HAVE_INTRINSICS 3188 /* 3189 * oop -> myklass 3190 * myklass->trace_id |= USED 3191 * return myklass->trace_id & ~0x3 3192 */ 3193 bool LibraryCallKit::inline_native_classID() { 3194 null_check_receiver(); // null-check, then ignore 3195 Node* cls = null_check(argument(1), T_OBJECT); 3196 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 3197 kls = null_check(kls, T_OBJECT); 3198 ByteSize offset = TRACE_ID_OFFSET; 3199 Node* insp = basic_plus_adr(kls, in_bytes(offset)); 3200 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered); 3201 Node* bits = longcon(~0x03l); // ignore bit 0 & 1 3202 Node* andl = _gvn.transform(new AndLNode(tvalue, bits)); 3203 Node* clsused = longcon(0x01l); // set the class bit 3204 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused)); 3205 3206 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr(); 3207 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered); 3208 set_result(andl); 3209 return true; 3210 } 3211 3212 bool LibraryCallKit::inline_native_threadID() { 3213 Node* tls_ptr = NULL; 3214 Node* cur_thr = generate_current_thread(tls_ptr); 3215 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset())); 3216 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3217 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::thread_id_offset())); 3218 3219 Node* threadid = NULL; 3220 size_t thread_id_size = OSThread::thread_id_size(); 3221 if (thread_id_size == (size_t) BytesPerLong) { 3222 threadid = ConvL2I(make_load(control(), p, TypeLong::LONG, T_LONG, MemNode::unordered)); 3223 } else if (thread_id_size == (size_t) BytesPerInt) { 3224 threadid = make_load(control(), p, TypeInt::INT, T_INT, MemNode::unordered); 3225 } else { 3226 ShouldNotReachHere(); 3227 } 3228 set_result(threadid); 3229 return true; 3230 } 3231 #endif 3232 3233 //------------------------inline_native_time_funcs-------------- 3234 // inline code for System.currentTimeMillis() and System.nanoTime() 3235 // these have the same type and signature 3236 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 3237 const TypeFunc* tf = OptoRuntime::void_long_Type(); 3238 const TypePtr* no_memory_effects = NULL; 3239 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 3240 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 3241 #ifdef ASSERT 3242 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 3243 assert(value_top == top(), "second value must be top"); 3244 #endif 3245 set_result(value); 3246 return true; 3247 } 3248 3249 //------------------------inline_native_currentThread------------------ 3250 bool LibraryCallKit::inline_native_currentThread() { 3251 Node* junk = NULL; 3252 set_result(generate_current_thread(junk)); 3253 return true; 3254 } 3255 3256 //------------------------inline_native_isInterrupted------------------ 3257 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted); 3258 bool LibraryCallKit::inline_native_isInterrupted() { 3259 // Add a fast path to t.isInterrupted(clear_int): 3260 // (t == Thread.current() && 3261 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int))) 3262 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int) 3263 // So, in the common case that the interrupt bit is false, 3264 // we avoid making a call into the VM. Even if the interrupt bit 3265 // is true, if the clear_int argument is false, we avoid the VM call. 3266 // However, if the receiver is not currentThread, we must call the VM, 3267 // because there must be some locking done around the operation. 3268 3269 // We only go to the fast case code if we pass two guards. 3270 // Paths which do not pass are accumulated in the slow_region. 3271 3272 enum { 3273 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted 3274 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int 3275 slow_result_path = 3, // slow path: t.isInterrupted(clear_int) 3276 PATH_LIMIT 3277 }; 3278 3279 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag 3280 // out of the function. 3281 insert_mem_bar(Op_MemBarCPUOrder); 3282 3283 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3284 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL); 3285 3286 RegionNode* slow_region = new RegionNode(1); 3287 record_for_igvn(slow_region); 3288 3289 // (a) Receiving thread must be the current thread. 3290 Node* rec_thr = argument(0); 3291 Node* tls_ptr = NULL; 3292 Node* cur_thr = generate_current_thread(tls_ptr); 3293 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr)); 3294 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne)); 3295 3296 generate_slow_guard(bol_thr, slow_region); 3297 3298 // (b) Interrupt bit on TLS must be false. 3299 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset())); 3300 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3301 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset())); 3302 3303 // Set the control input on the field _interrupted read to prevent it floating up. 3304 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered); 3305 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0))); 3306 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne)); 3307 3308 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 3309 3310 // First fast path: if (!TLS._interrupted) return false; 3311 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit)); 3312 result_rgn->init_req(no_int_result_path, false_bit); 3313 result_val->init_req(no_int_result_path, intcon(0)); 3314 3315 // drop through to next case 3316 set_control( _gvn.transform(new IfTrueNode(iff_bit))); 3317 3318 #ifndef TARGET_OS_FAMILY_windows 3319 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path. 3320 Node* clr_arg = argument(1); 3321 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0))); 3322 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne)); 3323 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN); 3324 3325 // Second fast path: ... else if (!clear_int) return true; 3326 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg)); 3327 result_rgn->init_req(no_clear_result_path, false_arg); 3328 result_val->init_req(no_clear_result_path, intcon(1)); 3329 3330 // drop through to next case 3331 set_control( _gvn.transform(new IfTrueNode(iff_arg))); 3332 #else 3333 // To return true on Windows you must read the _interrupted field 3334 // and check the event state i.e. take the slow path. 3335 #endif // TARGET_OS_FAMILY_windows 3336 3337 // (d) Otherwise, go to the slow path. 3338 slow_region->add_req(control()); 3339 set_control( _gvn.transform(slow_region)); 3340 3341 if (stopped()) { 3342 // There is no slow path. 3343 result_rgn->init_req(slow_result_path, top()); 3344 result_val->init_req(slow_result_path, top()); 3345 } else { 3346 // non-virtual because it is a private non-static 3347 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted); 3348 3349 Node* slow_val = set_results_for_java_call(slow_call); 3350 // this->control() comes from set_results_for_java_call 3351 3352 Node* fast_io = slow_call->in(TypeFunc::I_O); 3353 Node* fast_mem = slow_call->in(TypeFunc::Memory); 3354 3355 // These two phis are pre-filled with copies of of the fast IO and Memory 3356 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM); 3357 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO); 3358 3359 result_rgn->init_req(slow_result_path, control()); 3360 result_io ->init_req(slow_result_path, i_o()); 3361 result_mem->init_req(slow_result_path, reset_memory()); 3362 result_val->init_req(slow_result_path, slow_val); 3363 3364 set_all_memory(_gvn.transform(result_mem)); 3365 set_i_o( _gvn.transform(result_io)); 3366 } 3367 3368 C->set_has_split_ifs(true); // Has chance for split-if optimization 3369 set_result(result_rgn, result_val); 3370 return true; 3371 } 3372 3373 //---------------------------load_mirror_from_klass---------------------------- 3374 // Given a klass oop, load its java mirror (a java.lang.Class oop). 3375 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) { 3376 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset())); 3377 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered); 3378 } 3379 3380 //-----------------------load_klass_from_mirror_common------------------------- 3381 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3382 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3383 // and branch to the given path on the region. 3384 // If never_see_null, take an uncommon trap on null, so we can optimistically 3385 // compile for the non-null case. 3386 // If the region is NULL, force never_see_null = true. 3387 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3388 bool never_see_null, 3389 RegionNode* region, 3390 int null_path, 3391 int offset) { 3392 if (region == NULL) never_see_null = true; 3393 Node* p = basic_plus_adr(mirror, offset); 3394 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3395 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3396 Node* null_ctl = top(); 3397 kls = null_check_oop(kls, &null_ctl, never_see_null); 3398 if (region != NULL) { 3399 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3400 region->init_req(null_path, null_ctl); 3401 } else { 3402 assert(null_ctl == top(), "no loose ends"); 3403 } 3404 return kls; 3405 } 3406 3407 //--------------------(inline_native_Class_query helpers)--------------------- 3408 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER. 3409 // Fall through if (mods & mask) == bits, take the guard otherwise. 3410 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 3411 // Branch around if the given klass has the given modifier bit set. 3412 // Like generate_guard, adds a new path onto the region. 3413 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3414 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered); 3415 Node* mask = intcon(modifier_mask); 3416 Node* bits = intcon(modifier_bits); 3417 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3418 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3419 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3420 return generate_fair_guard(bol, region); 3421 } 3422 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3423 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region); 3424 } 3425 3426 //-------------------------inline_native_Class_query------------------- 3427 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 3428 const Type* return_type = TypeInt::BOOL; 3429 Node* prim_return_value = top(); // what happens if it's a primitive class? 3430 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3431 bool expect_prim = false; // most of these guys expect to work on refs 3432 3433 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 3434 3435 Node* mirror = argument(0); 3436 Node* obj = top(); 3437 3438 switch (id) { 3439 case vmIntrinsics::_isInstance: 3440 // nothing is an instance of a primitive type 3441 prim_return_value = intcon(0); 3442 obj = argument(1); 3443 break; 3444 case vmIntrinsics::_getModifiers: 3445 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3446 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line"); 3447 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin); 3448 break; 3449 case vmIntrinsics::_isInterface: 3450 prim_return_value = intcon(0); 3451 break; 3452 case vmIntrinsics::_isArray: 3453 prim_return_value = intcon(0); 3454 expect_prim = true; // cf. ObjectStreamClass.getClassSignature 3455 break; 3456 case vmIntrinsics::_isPrimitive: 3457 prim_return_value = intcon(1); 3458 expect_prim = true; // obviously 3459 break; 3460 case vmIntrinsics::_getSuperclass: 3461 prim_return_value = null(); 3462 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 3463 break; 3464 case vmIntrinsics::_getClassAccessFlags: 3465 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3466 return_type = TypeInt::INT; // not bool! 6297094 3467 break; 3468 default: 3469 fatal_unexpected_iid(id); 3470 break; 3471 } 3472 3473 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3474 if (mirror_con == NULL) return false; // cannot happen? 3475 3476 #ifndef PRODUCT 3477 if (C->print_intrinsics() || C->print_inlining()) { 3478 ciType* k = mirror_con->java_mirror_type(); 3479 if (k) { 3480 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 3481 k->print_name(); 3482 tty->cr(); 3483 } 3484 } 3485 #endif 3486 3487 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 3488 RegionNode* region = new RegionNode(PATH_LIMIT); 3489 record_for_igvn(region); 3490 PhiNode* phi = new PhiNode(region, return_type); 3491 3492 // The mirror will never be null of Reflection.getClassAccessFlags, however 3493 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 3494 // if it is. See bug 4774291. 3495 3496 // For Reflection.getClassAccessFlags(), the null check occurs in 3497 // the wrong place; see inline_unsafe_access(), above, for a similar 3498 // situation. 3499 mirror = null_check(mirror); 3500 // If mirror or obj is dead, only null-path is taken. 3501 if (stopped()) return true; 3502 3503 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 3504 3505 // Now load the mirror's klass metaobject, and null-check it. 3506 // Side-effects region with the control path if the klass is null. 3507 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 3508 // If kls is null, we have a primitive mirror. 3509 phi->init_req(_prim_path, prim_return_value); 3510 if (stopped()) { set_result(region, phi); return true; } 3511 bool safe_for_replace = (region->in(_prim_path) == top()); 3512 3513 Node* p; // handy temp 3514 Node* null_ctl; 3515 3516 // Now that we have the non-null klass, we can perform the real query. 3517 // For constant classes, the query will constant-fold in LoadNode::Value. 3518 Node* query_value = top(); 3519 switch (id) { 3520 case vmIntrinsics::_isInstance: 3521 // nothing is an instance of a primitive type 3522 query_value = gen_instanceof(obj, kls, safe_for_replace); 3523 break; 3524 3525 case vmIntrinsics::_getModifiers: 3526 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset())); 3527 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3528 break; 3529 3530 case vmIntrinsics::_isInterface: 3531 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3532 if (generate_interface_guard(kls, region) != NULL) 3533 // A guard was added. If the guard is taken, it was an interface. 3534 phi->add_req(intcon(1)); 3535 // If we fall through, it's a plain class. 3536 query_value = intcon(0); 3537 break; 3538 3539 case vmIntrinsics::_isArray: 3540 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.) 3541 if (generate_array_guard(kls, region) != NULL) 3542 // A guard was added. If the guard is taken, it was an array. 3543 phi->add_req(intcon(1)); 3544 // If we fall through, it's a plain class. 3545 query_value = intcon(0); 3546 break; 3547 3548 case vmIntrinsics::_isPrimitive: 3549 query_value = intcon(0); // "normal" path produces false 3550 break; 3551 3552 case vmIntrinsics::_getSuperclass: 3553 // The rules here are somewhat unfortunate, but we can still do better 3554 // with random logic than with a JNI call. 3555 // Interfaces store null or Object as _super, but must report null. 3556 // Arrays store an intermediate super as _super, but must report Object. 3557 // Other types can report the actual _super. 3558 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3559 if (generate_interface_guard(kls, region) != NULL) 3560 // A guard was added. If the guard is taken, it was an interface. 3561 phi->add_req(null()); 3562 if (generate_array_guard(kls, region) != NULL) 3563 // A guard was added. If the guard is taken, it was an array. 3564 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 3565 // If we fall through, it's a plain class. Get its _super. 3566 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 3567 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL)); 3568 null_ctl = top(); 3569 kls = null_check_oop(kls, &null_ctl); 3570 if (null_ctl != top()) { 3571 // If the guard is taken, Object.superClass is null (both klass and mirror). 3572 region->add_req(null_ctl); 3573 phi ->add_req(null()); 3574 } 3575 if (!stopped()) { 3576 query_value = load_mirror_from_klass(kls); 3577 } 3578 break; 3579 3580 case vmIntrinsics::_getClassAccessFlags: 3581 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3582 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3583 break; 3584 3585 default: 3586 fatal_unexpected_iid(id); 3587 break; 3588 } 3589 3590 // Fall-through is the normal case of a query to a real class. 3591 phi->init_req(1, query_value); 3592 region->init_req(1, control()); 3593 3594 C->set_has_split_ifs(true); // Has chance for split-if optimization 3595 set_result(region, phi); 3596 return true; 3597 } 3598 3599 //-------------------------inline_Class_cast------------------- 3600 bool LibraryCallKit::inline_Class_cast() { 3601 Node* mirror = argument(0); // Class 3602 Node* obj = argument(1); 3603 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3604 if (mirror_con == NULL) { 3605 return false; // dead path (mirror->is_top()). 3606 } 3607 if (obj == NULL || obj->is_top()) { 3608 return false; // dead path 3609 } 3610 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr(); 3611 3612 // First, see if Class.cast() can be folded statically. 3613 // java_mirror_type() returns non-null for compile-time Class constants. 3614 ciType* tm = mirror_con->java_mirror_type(); 3615 if (tm != NULL && tm->is_klass() && 3616 tp != NULL && tp->klass() != NULL) { 3617 if (!tp->klass()->is_loaded()) { 3618 // Don't use intrinsic when class is not loaded. 3619 return false; 3620 } else { 3621 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass()); 3622 if (static_res == Compile::SSC_always_true) { 3623 // isInstance() is true - fold the code. 3624 set_result(obj); 3625 return true; 3626 } else if (static_res == Compile::SSC_always_false) { 3627 // Don't use intrinsic, have to throw ClassCastException. 3628 // If the reference is null, the non-intrinsic bytecode will 3629 // be optimized appropriately. 3630 return false; 3631 } 3632 } 3633 } 3634 3635 // Bailout intrinsic and do normal inlining if exception path is frequent. 3636 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 3637 return false; 3638 } 3639 3640 // Generate dynamic checks. 3641 // Class.cast() is java implementation of _checkcast bytecode. 3642 // Do checkcast (Parse::do_checkcast()) optimizations here. 3643 3644 mirror = null_check(mirror); 3645 // If mirror is dead, only null-path is taken. 3646 if (stopped()) { 3647 return true; 3648 } 3649 3650 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive). 3651 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT }; 3652 RegionNode* region = new RegionNode(PATH_LIMIT); 3653 record_for_igvn(region); 3654 3655 // Now load the mirror's klass metaobject, and null-check it. 3656 // If kls is null, we have a primitive mirror and 3657 // nothing is an instance of a primitive type. 3658 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 3659 3660 Node* res = top(); 3661 if (!stopped()) { 3662 Node* bad_type_ctrl = top(); 3663 // Do checkcast optimizations. 3664 res = gen_checkcast(obj, kls, &bad_type_ctrl); 3665 region->init_req(_bad_type_path, bad_type_ctrl); 3666 } 3667 if (region->in(_prim_path) != top() || 3668 region->in(_bad_type_path) != top()) { 3669 // Let Interpreter throw ClassCastException. 3670 PreserveJVMState pjvms(this); 3671 set_control(_gvn.transform(region)); 3672 uncommon_trap(Deoptimization::Reason_intrinsic, 3673 Deoptimization::Action_maybe_recompile); 3674 } 3675 if (!stopped()) { 3676 set_result(res); 3677 } 3678 return true; 3679 } 3680 3681 3682 //--------------------------inline_native_subtype_check------------------------ 3683 // This intrinsic takes the JNI calls out of the heart of 3684 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 3685 bool LibraryCallKit::inline_native_subtype_check() { 3686 // Pull both arguments off the stack. 3687 Node* args[2]; // two java.lang.Class mirrors: superc, subc 3688 args[0] = argument(0); 3689 args[1] = argument(1); 3690 Node* klasses[2]; // corresponding Klasses: superk, subk 3691 klasses[0] = klasses[1] = top(); 3692 3693 enum { 3694 // A full decision tree on {superc is prim, subc is prim}: 3695 _prim_0_path = 1, // {P,N} => false 3696 // {P,P} & superc!=subc => false 3697 _prim_same_path, // {P,P} & superc==subc => true 3698 _prim_1_path, // {N,P} => false 3699 _ref_subtype_path, // {N,N} & subtype check wins => true 3700 _both_ref_path, // {N,N} & subtype check loses => false 3701 PATH_LIMIT 3702 }; 3703 3704 RegionNode* region = new RegionNode(PATH_LIMIT); 3705 Node* phi = new PhiNode(region, TypeInt::BOOL); 3706 record_for_igvn(region); 3707 3708 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 3709 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3710 int class_klass_offset = java_lang_Class::klass_offset_in_bytes(); 3711 3712 // First null-check both mirrors and load each mirror's klass metaobject. 3713 int which_arg; 3714 for (which_arg = 0; which_arg <= 1; which_arg++) { 3715 Node* arg = args[which_arg]; 3716 arg = null_check(arg); 3717 if (stopped()) break; 3718 args[which_arg] = arg; 3719 3720 Node* p = basic_plus_adr(arg, class_klass_offset); 3721 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type); 3722 klasses[which_arg] = _gvn.transform(kls); 3723 } 3724 3725 // Having loaded both klasses, test each for null. 3726 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3727 for (which_arg = 0; which_arg <= 1; which_arg++) { 3728 Node* kls = klasses[which_arg]; 3729 Node* null_ctl = top(); 3730 kls = null_check_oop(kls, &null_ctl, never_see_null); 3731 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path); 3732 region->init_req(prim_path, null_ctl); 3733 if (stopped()) break; 3734 klasses[which_arg] = kls; 3735 } 3736 3737 if (!stopped()) { 3738 // now we have two reference types, in klasses[0..1] 3739 Node* subk = klasses[1]; // the argument to isAssignableFrom 3740 Node* superk = klasses[0]; // the receiver 3741 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 3742 // now we have a successful reference subtype check 3743 region->set_req(_ref_subtype_path, control()); 3744 } 3745 3746 // If both operands are primitive (both klasses null), then 3747 // we must return true when they are identical primitives. 3748 // It is convenient to test this after the first null klass check. 3749 set_control(region->in(_prim_0_path)); // go back to first null check 3750 if (!stopped()) { 3751 // Since superc is primitive, make a guard for the superc==subc case. 3752 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 3753 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 3754 generate_guard(bol_eq, region, PROB_FAIR); 3755 if (region->req() == PATH_LIMIT+1) { 3756 // A guard was added. If the added guard is taken, superc==subc. 3757 region->swap_edges(PATH_LIMIT, _prim_same_path); 3758 region->del_req(PATH_LIMIT); 3759 } 3760 region->set_req(_prim_0_path, control()); // Not equal after all. 3761 } 3762 3763 // these are the only paths that produce 'true': 3764 phi->set_req(_prim_same_path, intcon(1)); 3765 phi->set_req(_ref_subtype_path, intcon(1)); 3766 3767 // pull together the cases: 3768 assert(region->req() == PATH_LIMIT, "sane region"); 3769 for (uint i = 1; i < region->req(); i++) { 3770 Node* ctl = region->in(i); 3771 if (ctl == NULL || ctl == top()) { 3772 region->set_req(i, top()); 3773 phi ->set_req(i, top()); 3774 } else if (phi->in(i) == NULL) { 3775 phi->set_req(i, intcon(0)); // all other paths produce 'false' 3776 } 3777 } 3778 3779 set_control(_gvn.transform(region)); 3780 set_result(_gvn.transform(phi)); 3781 return true; 3782 } 3783 3784 //---------------------generate_array_guard_common------------------------ 3785 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, 3786 bool obj_array, bool not_array) { 3787 3788 if (stopped()) { 3789 return NULL; 3790 } 3791 3792 // If obj_array/non_array==false/false: 3793 // Branch around if the given klass is in fact an array (either obj or prim). 3794 // If obj_array/non_array==false/true: 3795 // Branch around if the given klass is not an array klass of any kind. 3796 // If obj_array/non_array==true/true: 3797 // Branch around if the kls is not an oop array (kls is int[], String, etc.) 3798 // If obj_array/non_array==true/false: 3799 // Branch around if the kls is an oop array (Object[] or subtype) 3800 // 3801 // Like generate_guard, adds a new path onto the region. 3802 jint layout_con = 0; 3803 Node* layout_val = get_layout_helper(kls, layout_con); 3804 if (layout_val == NULL) { 3805 bool query = (obj_array 3806 ? Klass::layout_helper_is_objArray(layout_con) 3807 : Klass::layout_helper_is_array(layout_con)); 3808 if (query == not_array) { 3809 return NULL; // never a branch 3810 } else { // always a branch 3811 Node* always_branch = control(); 3812 if (region != NULL) 3813 region->add_req(always_branch); 3814 set_control(top()); 3815 return always_branch; 3816 } 3817 } 3818 // Now test the correct condition. 3819 jint nval = (obj_array 3820 ? ((jint)Klass::_lh_array_tag_type_value 3821 << Klass::_lh_array_tag_shift) 3822 : Klass::_lh_neutral_value); 3823 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 3824 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array 3825 // invert the test if we are looking for a non-array 3826 if (not_array) btest = BoolTest(btest).negate(); 3827 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 3828 return generate_fair_guard(bol, region); 3829 } 3830 3831 3832 //-----------------------inline_native_newArray-------------------------- 3833 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length); 3834 bool LibraryCallKit::inline_native_newArray() { 3835 Node* mirror = argument(0); 3836 Node* count_val = argument(1); 3837 3838 mirror = null_check(mirror); 3839 // If mirror or obj is dead, only null-path is taken. 3840 if (stopped()) return true; 3841 3842 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 3843 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 3844 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 3845 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 3846 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 3847 3848 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3849 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 3850 result_reg, _slow_path); 3851 Node* normal_ctl = control(); 3852 Node* no_array_ctl = result_reg->in(_slow_path); 3853 3854 // Generate code for the slow case. We make a call to newArray(). 3855 set_control(no_array_ctl); 3856 if (!stopped()) { 3857 // Either the input type is void.class, or else the 3858 // array klass has not yet been cached. Either the 3859 // ensuing call will throw an exception, or else it 3860 // will cache the array klass for next time. 3861 PreserveJVMState pjvms(this); 3862 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray); 3863 Node* slow_result = set_results_for_java_call(slow_call); 3864 // this->control() comes from set_results_for_java_call 3865 result_reg->set_req(_slow_path, control()); 3866 result_val->set_req(_slow_path, slow_result); 3867 result_io ->set_req(_slow_path, i_o()); 3868 result_mem->set_req(_slow_path, reset_memory()); 3869 } 3870 3871 set_control(normal_ctl); 3872 if (!stopped()) { 3873 // Normal case: The array type has been cached in the java.lang.Class. 3874 // The following call works fine even if the array type is polymorphic. 3875 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3876 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push 3877 result_reg->init_req(_normal_path, control()); 3878 result_val->init_req(_normal_path, obj); 3879 result_io ->init_req(_normal_path, i_o()); 3880 result_mem->init_req(_normal_path, reset_memory()); 3881 } 3882 3883 // Return the combined state. 3884 set_i_o( _gvn.transform(result_io) ); 3885 set_all_memory( _gvn.transform(result_mem)); 3886 3887 C->set_has_split_ifs(true); // Has chance for split-if optimization 3888 set_result(result_reg, result_val); 3889 return true; 3890 } 3891 3892 //----------------------inline_native_getLength-------------------------- 3893 // public static native int java.lang.reflect.Array.getLength(Object array); 3894 bool LibraryCallKit::inline_native_getLength() { 3895 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3896 3897 Node* array = null_check(argument(0)); 3898 // If array is dead, only null-path is taken. 3899 if (stopped()) return true; 3900 3901 // Deoptimize if it is a non-array. 3902 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL); 3903 3904 if (non_array != NULL) { 3905 PreserveJVMState pjvms(this); 3906 set_control(non_array); 3907 uncommon_trap(Deoptimization::Reason_intrinsic, 3908 Deoptimization::Action_maybe_recompile); 3909 } 3910 3911 // If control is dead, only non-array-path is taken. 3912 if (stopped()) return true; 3913 3914 // The works fine even if the array type is polymorphic. 3915 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3916 Node* result = load_array_length(array); 3917 3918 C->set_has_split_ifs(true); // Has chance for split-if optimization 3919 set_result(result); 3920 return true; 3921 } 3922 3923 //------------------------inline_array_copyOf---------------------------- 3924 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 3925 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 3926 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 3927 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3928 3929 // Get the arguments. 3930 Node* original = argument(0); 3931 Node* start = is_copyOfRange? argument(1): intcon(0); 3932 Node* end = is_copyOfRange? argument(2): argument(1); 3933 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 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 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout); 3959 if (not_objArray != NULL) { 3960 // Improve the klass node's type from the new optimistic assumption: 3961 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 3962 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/); 3963 Node* cast = new CastPPNode(klass_node, akls); 3964 cast->init_req(0, control()); 3965 klass_node = _gvn.transform(cast); 3966 } 3967 3968 // Bail out if either start or end is negative. 3969 generate_negative_guard(start, bailout, &start); 3970 generate_negative_guard(end, bailout, &end); 3971 3972 Node* length = end; 3973 if (_gvn.type(start) != TypeInt::ZERO) { 3974 length = _gvn.transform(new SubINode(end, start)); 3975 } 3976 3977 // Bail out if length is negative. 3978 // Without this the new_array would throw 3979 // NegativeArraySizeException but IllegalArgumentException is what 3980 // should be thrown 3981 generate_negative_guard(length, bailout, &length); 3982 3983 if (bailout->req() > 1) { 3984 PreserveJVMState pjvms(this); 3985 set_control(_gvn.transform(bailout)); 3986 uncommon_trap(Deoptimization::Reason_intrinsic, 3987 Deoptimization::Action_maybe_recompile); 3988 } 3989 3990 if (!stopped()) { 3991 // How many elements will we copy from the original? 3992 // The answer is MinI(orig_length - start, length). 3993 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 3994 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length); 3995 3996 // Generate a direct call to the right arraycopy function(s). 3997 // We know the copy is disjoint but we might not know if the 3998 // oop stores need checking. 3999 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 4000 // This will fail a store-check if x contains any non-nulls. 4001 4002 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 4003 // loads/stores but it is legal only if we're sure the 4004 // Arrays.copyOf would succeed. So we need all input arguments 4005 // to the copyOf to be validated, including that the copy to the 4006 // new array won't trigger an ArrayStoreException. That subtype 4007 // check can be optimized if we know something on the type of 4008 // the input array from type speculation. 4009 if (_gvn.type(klass_node)->singleton()) { 4010 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass(); 4011 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass(); 4012 4013 int test = C->static_subtype_check(superk, subk); 4014 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 4015 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 4016 if (t_original->speculative_type() != NULL) { 4017 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 4018 } 4019 } 4020 } 4021 4022 bool validated = false; 4023 // Reason_class_check rather than Reason_intrinsic because we 4024 // want to intrinsify even if this traps. 4025 if (!too_many_traps(Deoptimization::Reason_class_check)) { 4026 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original), 4027 klass_node); 4028 4029 if (not_subtype_ctrl != top()) { 4030 PreserveJVMState pjvms(this); 4031 set_control(not_subtype_ctrl); 4032 uncommon_trap(Deoptimization::Reason_class_check, 4033 Deoptimization::Action_make_not_entrant); 4034 assert(stopped(), "Should be stopped"); 4035 } 4036 validated = true; 4037 } 4038 4039 if (!stopped()) { 4040 newcopy = new_array(klass_node, length, 0); // no arguments to push 4041 4042 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, 4043 load_object_klass(original), klass_node); 4044 if (!is_copyOfRange) { 4045 ac->set_copyof(validated); 4046 } else { 4047 ac->set_copyofrange(validated); 4048 } 4049 Node* n = _gvn.transform(ac); 4050 if (n == ac) { 4051 ac->connect_outputs(this); 4052 } else { 4053 assert(validated, "shouldn't transform if all arguments not validated"); 4054 set_all_memory(n); 4055 } 4056 } 4057 } 4058 } // original reexecute is set back here 4059 4060 C->set_has_split_ifs(true); // Has chance for split-if optimization 4061 if (!stopped()) { 4062 set_result(newcopy); 4063 } 4064 return true; 4065 } 4066 4067 4068 //----------------------generate_virtual_guard--------------------------- 4069 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 4070 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 4071 RegionNode* slow_region) { 4072 ciMethod* method = callee(); 4073 int vtable_index = method->vtable_index(); 4074 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4075 "bad index %d", vtable_index); 4076 // Get the Method* out of the appropriate vtable entry. 4077 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 4078 vtable_index*vtableEntry::size_in_bytes() + 4079 vtableEntry::method_offset_in_bytes(); 4080 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 4081 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 4082 4083 // Compare the target method with the expected method (e.g., Object.hashCode). 4084 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 4085 4086 Node* native_call = makecon(native_call_addr); 4087 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 4088 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 4089 4090 return generate_slow_guard(test_native, slow_region); 4091 } 4092 4093 //-----------------------generate_method_call---------------------------- 4094 // Use generate_method_call to make a slow-call to the real 4095 // method if the fast path fails. An alternative would be to 4096 // use a stub like OptoRuntime::slow_arraycopy_Java. 4097 // This only works for expanding the current library call, 4098 // not another intrinsic. (E.g., don't use this for making an 4099 // arraycopy call inside of the copyOf intrinsic.) 4100 CallJavaNode* 4101 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) { 4102 // When compiling the intrinsic method itself, do not use this technique. 4103 guarantee(callee() != C->method(), "cannot make slow-call to self"); 4104 4105 ciMethod* method = callee(); 4106 // ensure the JVMS we have will be correct for this call 4107 guarantee(method_id == method->intrinsic_id(), "must match"); 4108 4109 const TypeFunc* tf = TypeFunc::make(method); 4110 CallJavaNode* slow_call; 4111 if (is_static) { 4112 assert(!is_virtual, ""); 4113 slow_call = new CallStaticJavaNode(C, tf, 4114 SharedRuntime::get_resolve_static_call_stub(), 4115 method, bci()); 4116 } else if (is_virtual) { 4117 null_check_receiver(); 4118 int vtable_index = Method::invalid_vtable_index; 4119 if (UseInlineCaches) { 4120 // Suppress the vtable call 4121 } else { 4122 // hashCode and clone are not a miranda methods, 4123 // so the vtable index is fixed. 4124 // No need to use the linkResolver to get it. 4125 vtable_index = method->vtable_index(); 4126 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4127 "bad index %d", vtable_index); 4128 } 4129 slow_call = new CallDynamicJavaNode(tf, 4130 SharedRuntime::get_resolve_virtual_call_stub(), 4131 method, vtable_index, bci()); 4132 } else { // neither virtual nor static: opt_virtual 4133 null_check_receiver(); 4134 slow_call = new CallStaticJavaNode(C, tf, 4135 SharedRuntime::get_resolve_opt_virtual_call_stub(), 4136 method, bci()); 4137 slow_call->set_optimized_virtual(true); 4138 } 4139 set_arguments_for_java_call(slow_call); 4140 set_edges_for_java_call(slow_call); 4141 return slow_call; 4142 } 4143 4144 4145 /** 4146 * Build special case code for calls to hashCode on an object. This call may 4147 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 4148 * slightly different code. 4149 */ 4150 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 4151 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 4152 assert(!(is_virtual && is_static), "either virtual, special, or static"); 4153 4154 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 4155 4156 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4157 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 4158 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4159 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4160 Node* obj = NULL; 4161 if (!is_static) { 4162 // Check for hashing null object 4163 obj = null_check_receiver(); 4164 if (stopped()) return true; // unconditionally null 4165 result_reg->init_req(_null_path, top()); 4166 result_val->init_req(_null_path, top()); 4167 } else { 4168 // Do a null check, and return zero if null. 4169 // System.identityHashCode(null) == 0 4170 obj = argument(0); 4171 Node* null_ctl = top(); 4172 obj = null_check_oop(obj, &null_ctl); 4173 result_reg->init_req(_null_path, null_ctl); 4174 result_val->init_req(_null_path, _gvn.intcon(0)); 4175 } 4176 4177 // Unconditionally null? Then return right away. 4178 if (stopped()) { 4179 set_control( result_reg->in(_null_path)); 4180 if (!stopped()) 4181 set_result(result_val->in(_null_path)); 4182 return true; 4183 } 4184 4185 // We only go to the fast case code if we pass a number of guards. The 4186 // paths which do not pass are accumulated in the slow_region. 4187 RegionNode* slow_region = new RegionNode(1); 4188 record_for_igvn(slow_region); 4189 4190 // If this is a virtual call, we generate a funny guard. We pull out 4191 // the vtable entry corresponding to hashCode() from the target object. 4192 // If the target method which we are calling happens to be the native 4193 // Object hashCode() method, we pass the guard. We do not need this 4194 // guard for non-virtual calls -- the caller is known to be the native 4195 // Object hashCode(). 4196 if (is_virtual) { 4197 // After null check, get the object's klass. 4198 Node* obj_klass = load_object_klass(obj); 4199 generate_virtual_guard(obj_klass, slow_region); 4200 } 4201 4202 // Get the header out of the object, use LoadMarkNode when available 4203 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 4204 // The control of the load must be NULL. Otherwise, the load can move before 4205 // the null check after castPP removal. 4206 Node* no_ctrl = NULL; 4207 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 4208 4209 // Test the header to see if it is unlocked. 4210 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place); 4211 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 4212 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value); 4213 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val)); 4214 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne)); 4215 4216 generate_slow_guard(test_unlocked, slow_region); 4217 4218 // Get the hash value and check to see that it has been properly assigned. 4219 // We depend on hash_mask being at most 32 bits and avoid the use of 4220 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 4221 // vm: see markOop.hpp. 4222 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask); 4223 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift); 4224 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 4225 // This hack lets the hash bits live anywhere in the mark object now, as long 4226 // as the shift drops the relevant bits into the low 32 bits. Note that 4227 // Java spec says that HashCode is an int so there's no point in capturing 4228 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 4229 hshifted_header = ConvX2I(hshifted_header); 4230 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 4231 4232 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash); 4233 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 4234 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 4235 4236 generate_slow_guard(test_assigned, slow_region); 4237 4238 Node* init_mem = reset_memory(); 4239 // fill in the rest of the null path: 4240 result_io ->init_req(_null_path, i_o()); 4241 result_mem->init_req(_null_path, init_mem); 4242 4243 result_val->init_req(_fast_path, hash_val); 4244 result_reg->init_req(_fast_path, control()); 4245 result_io ->init_req(_fast_path, i_o()); 4246 result_mem->init_req(_fast_path, init_mem); 4247 4248 // Generate code for the slow case. We make a call to hashCode(). 4249 set_control(_gvn.transform(slow_region)); 4250 if (!stopped()) { 4251 // No need for PreserveJVMState, because we're using up the present state. 4252 set_all_memory(init_mem); 4253 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 4254 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static); 4255 Node* slow_result = set_results_for_java_call(slow_call); 4256 // this->control() comes from set_results_for_java_call 4257 result_reg->init_req(_slow_path, control()); 4258 result_val->init_req(_slow_path, slow_result); 4259 result_io ->set_req(_slow_path, i_o()); 4260 result_mem ->set_req(_slow_path, reset_memory()); 4261 } 4262 4263 // Return the combined state. 4264 set_i_o( _gvn.transform(result_io) ); 4265 set_all_memory( _gvn.transform(result_mem)); 4266 4267 set_result(result_reg, result_val); 4268 return true; 4269 } 4270 4271 //---------------------------inline_native_getClass---------------------------- 4272 // public final native Class<?> java.lang.Object.getClass(); 4273 // 4274 // Build special case code for calls to getClass on an object. 4275 bool LibraryCallKit::inline_native_getClass() { 4276 Node* obj = null_check_receiver(); 4277 if (stopped()) return true; 4278 set_result(load_mirror_from_klass(load_object_klass(obj))); 4279 return true; 4280 } 4281 4282 //-----------------inline_native_Reflection_getCallerClass--------------------- 4283 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 4284 // 4285 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 4286 // 4287 // NOTE: This code must perform the same logic as JVM_GetCallerClass 4288 // in that it must skip particular security frames and checks for 4289 // caller sensitive methods. 4290 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 4291 #ifndef PRODUCT 4292 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4293 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 4294 } 4295 #endif 4296 4297 if (!jvms()->has_method()) { 4298 #ifndef PRODUCT 4299 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4300 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 4301 } 4302 #endif 4303 return false; 4304 } 4305 4306 // Walk back up the JVM state to find the caller at the required 4307 // depth. 4308 JVMState* caller_jvms = jvms(); 4309 4310 // Cf. JVM_GetCallerClass 4311 // NOTE: Start the loop at depth 1 because the current JVM state does 4312 // not include the Reflection.getCallerClass() frame. 4313 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) { 4314 ciMethod* m = caller_jvms->method(); 4315 switch (n) { 4316 case 0: 4317 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 4318 break; 4319 case 1: 4320 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 4321 if (!m->caller_sensitive()) { 4322 #ifndef PRODUCT 4323 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4324 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 4325 } 4326 #endif 4327 return false; // bail-out; let JVM_GetCallerClass do the work 4328 } 4329 break; 4330 default: 4331 if (!m->is_ignored_by_security_stack_walk()) { 4332 // We have reached the desired frame; return the holder class. 4333 // Acquire method holder as java.lang.Class and push as constant. 4334 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 4335 ciInstance* caller_mirror = caller_klass->java_mirror(); 4336 set_result(makecon(TypeInstPtr::make(caller_mirror))); 4337 4338 #ifndef PRODUCT 4339 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4340 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()); 4341 tty->print_cr(" JVM state at this point:"); 4342 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4343 ciMethod* m = jvms()->of_depth(i)->method(); 4344 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4345 } 4346 } 4347 #endif 4348 return true; 4349 } 4350 break; 4351 } 4352 } 4353 4354 #ifndef PRODUCT 4355 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4356 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 4357 tty->print_cr(" JVM state at this point:"); 4358 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4359 ciMethod* m = jvms()->of_depth(i)->method(); 4360 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4361 } 4362 } 4363 #endif 4364 4365 return false; // bail-out; let JVM_GetCallerClass do the work 4366 } 4367 4368 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 4369 Node* arg = argument(0); 4370 Node* result = NULL; 4371 4372 switch (id) { 4373 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 4374 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 4375 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 4376 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 4377 4378 case vmIntrinsics::_doubleToLongBits: { 4379 // two paths (plus control) merge in a wood 4380 RegionNode *r = new RegionNode(3); 4381 Node *phi = new PhiNode(r, TypeLong::LONG); 4382 4383 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 4384 // Build the boolean node 4385 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4386 4387 // Branch either way. 4388 // NaN case is less traveled, which makes all the difference. 4389 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4390 Node *opt_isnan = _gvn.transform(ifisnan); 4391 assert( opt_isnan->is_If(), "Expect an IfNode"); 4392 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4393 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4394 4395 set_control(iftrue); 4396 4397 static const jlong nan_bits = CONST64(0x7ff8000000000000); 4398 Node *slow_result = longcon(nan_bits); // return NaN 4399 phi->init_req(1, _gvn.transform( slow_result )); 4400 r->init_req(1, iftrue); 4401 4402 // Else fall through 4403 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4404 set_control(iffalse); 4405 4406 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 4407 r->init_req(2, iffalse); 4408 4409 // Post merge 4410 set_control(_gvn.transform(r)); 4411 record_for_igvn(r); 4412 4413 C->set_has_split_ifs(true); // Has chance for split-if optimization 4414 result = phi; 4415 assert(result->bottom_type()->isa_long(), "must be"); 4416 break; 4417 } 4418 4419 case vmIntrinsics::_floatToIntBits: { 4420 // two paths (plus control) merge in a wood 4421 RegionNode *r = new RegionNode(3); 4422 Node *phi = new PhiNode(r, TypeInt::INT); 4423 4424 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 4425 // Build the boolean node 4426 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4427 4428 // Branch either way. 4429 // NaN case is less traveled, which makes all the difference. 4430 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4431 Node *opt_isnan = _gvn.transform(ifisnan); 4432 assert( opt_isnan->is_If(), "Expect an IfNode"); 4433 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4434 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4435 4436 set_control(iftrue); 4437 4438 static const jint nan_bits = 0x7fc00000; 4439 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 4440 phi->init_req(1, _gvn.transform( slow_result )); 4441 r->init_req(1, iftrue); 4442 4443 // Else fall through 4444 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4445 set_control(iffalse); 4446 4447 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 4448 r->init_req(2, iffalse); 4449 4450 // Post merge 4451 set_control(_gvn.transform(r)); 4452 record_for_igvn(r); 4453 4454 C->set_has_split_ifs(true); // Has chance for split-if optimization 4455 result = phi; 4456 assert(result->bottom_type()->isa_int(), "must be"); 4457 break; 4458 } 4459 4460 default: 4461 fatal_unexpected_iid(id); 4462 break; 4463 } 4464 set_result(_gvn.transform(result)); 4465 return true; 4466 } 4467 4468 //----------------------inline_unsafe_copyMemory------------------------- 4469 // public native void Unsafe.copyMemory(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 4470 bool LibraryCallKit::inline_unsafe_copyMemory() { 4471 if (callee()->is_static()) return false; // caller must have the capability! 4472 null_check_receiver(); // null-check receiver 4473 if (stopped()) return true; 4474 4475 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 4476 4477 Node* src_ptr = argument(1); // type: oop 4478 Node* src_off = ConvL2X(argument(2)); // type: long 4479 Node* dst_ptr = argument(4); // type: oop 4480 Node* dst_off = ConvL2X(argument(5)); // type: long 4481 Node* size = ConvL2X(argument(7)); // type: long 4482 4483 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 4484 "fieldOffset must be byte-scaled"); 4485 4486 Node* src = make_unsafe_address(src_ptr, src_off); 4487 Node* dst = make_unsafe_address(dst_ptr, dst_off); 4488 4489 // Conservatively insert a memory barrier on all memory slices. 4490 // Do not let writes of the copy source or destination float below the copy. 4491 insert_mem_bar(Op_MemBarCPUOrder); 4492 4493 // Call it. Note that the length argument is not scaled. 4494 make_runtime_call(RC_LEAF|RC_NO_FP, 4495 OptoRuntime::fast_arraycopy_Type(), 4496 StubRoutines::unsafe_arraycopy(), 4497 "unsafe_arraycopy", 4498 TypeRawPtr::BOTTOM, 4499 src, dst, size XTOP); 4500 4501 // Do not let reads of the copy destination float above the copy. 4502 insert_mem_bar(Op_MemBarCPUOrder); 4503 4504 return true; 4505 } 4506 4507 //------------------------clone_coping----------------------------------- 4508 // Helper function for inline_native_clone. 4509 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) { 4510 assert(obj_size != NULL, ""); 4511 Node* raw_obj = alloc_obj->in(1); 4512 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 4513 4514 AllocateNode* alloc = NULL; 4515 if (ReduceBulkZeroing) { 4516 // We will be completely responsible for initializing this object - 4517 // mark Initialize node as complete. 4518 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn); 4519 // The object was just allocated - there should be no any stores! 4520 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), ""); 4521 // Mark as complete_with_arraycopy so that on AllocateNode 4522 // expansion, we know this AllocateNode is initialized by an array 4523 // copy and a StoreStore barrier exists after the array copy. 4524 alloc->initialization()->set_complete_with_arraycopy(); 4525 } 4526 4527 // Copy the fastest available way. 4528 // TODO: generate fields copies for small objects instead. 4529 Node* src = obj; 4530 Node* dest = alloc_obj; 4531 Node* size = _gvn.transform(obj_size); 4532 4533 // Exclude the header but include array length to copy by 8 bytes words. 4534 // Can't use base_offset_in_bytes(bt) since basic type is unknown. 4535 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() : 4536 instanceOopDesc::base_offset_in_bytes(); 4537 // base_off: 4538 // 8 - 32-bit VM 4539 // 12 - 64-bit VM, compressed klass 4540 // 16 - 64-bit VM, normal klass 4541 if (base_off % BytesPerLong != 0) { 4542 assert(UseCompressedClassPointers, ""); 4543 if (is_array) { 4544 // Exclude length to copy by 8 bytes words. 4545 base_off += sizeof(int); 4546 } else { 4547 // Include klass to copy by 8 bytes words. 4548 base_off = instanceOopDesc::klass_offset_in_bytes(); 4549 } 4550 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment"); 4551 } 4552 src = basic_plus_adr(src, base_off); 4553 dest = basic_plus_adr(dest, base_off); 4554 4555 // Compute the length also, if needed: 4556 Node* countx = size; 4557 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off))); 4558 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) )); 4559 4560 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM; 4561 4562 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false); 4563 ac->set_clonebasic(); 4564 Node* n = _gvn.transform(ac); 4565 if (n == ac) { 4566 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type); 4567 } else { 4568 set_all_memory(n); 4569 } 4570 4571 // If necessary, emit some card marks afterwards. (Non-arrays only.) 4572 if (card_mark) { 4573 assert(!is_array, ""); 4574 // Put in store barrier for any and all oops we are sticking 4575 // into this object. (We could avoid this if we could prove 4576 // that the object type contains no oop fields at all.) 4577 Node* no_particular_value = NULL; 4578 Node* no_particular_field = NULL; 4579 int raw_adr_idx = Compile::AliasIdxRaw; 4580 post_barrier(control(), 4581 memory(raw_adr_type), 4582 alloc_obj, 4583 no_particular_field, 4584 raw_adr_idx, 4585 no_particular_value, 4586 T_OBJECT, 4587 false); 4588 } 4589 4590 // Do not let reads from the cloned object float above the arraycopy. 4591 if (alloc != NULL) { 4592 // Do not let stores that initialize this object be reordered with 4593 // a subsequent store that would make this object accessible by 4594 // other threads. 4595 // Record what AllocateNode this StoreStore protects so that 4596 // escape analysis can go from the MemBarStoreStoreNode to the 4597 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 4598 // based on the escape status of the AllocateNode. 4599 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress)); 4600 } else { 4601 insert_mem_bar(Op_MemBarCPUOrder); 4602 } 4603 } 4604 4605 //------------------------inline_native_clone---------------------------- 4606 // protected native Object java.lang.Object.clone(); 4607 // 4608 // Here are the simple edge cases: 4609 // null receiver => normal trap 4610 // virtual and clone was overridden => slow path to out-of-line clone 4611 // not cloneable or finalizer => slow path to out-of-line Object.clone 4612 // 4613 // The general case has two steps, allocation and copying. 4614 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 4615 // 4616 // Copying also has two cases, oop arrays and everything else. 4617 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 4618 // Everything else uses the tight inline loop supplied by CopyArrayNode. 4619 // 4620 // These steps fold up nicely if and when the cloned object's klass 4621 // can be sharply typed as an object array, a type array, or an instance. 4622 // 4623 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 4624 PhiNode* result_val; 4625 4626 // Set the reexecute bit for the interpreter to reexecute 4627 // the bytecode that invokes Object.clone if deoptimization happens. 4628 { PreserveReexecuteState preexecs(this); 4629 jvms()->set_should_reexecute(true); 4630 4631 Node* obj = null_check_receiver(); 4632 if (stopped()) return true; 4633 4634 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 4635 4636 // If we are going to clone an instance, we need its exact type to 4637 // know the number and types of fields to convert the clone to 4638 // loads/stores. Maybe a speculative type can help us. 4639 if (!obj_type->klass_is_exact() && 4640 obj_type->speculative_type() != NULL && 4641 obj_type->speculative_type()->is_instance_klass()) { 4642 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 4643 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 4644 !spec_ik->has_injected_fields()) { 4645 ciKlass* k = obj_type->klass(); 4646 if (!k->is_instance_klass() || 4647 k->as_instance_klass()->is_interface() || 4648 k->as_instance_klass()->has_subklass()) { 4649 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 4650 } 4651 } 4652 } 4653 4654 Node* obj_klass = load_object_klass(obj); 4655 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr(); 4656 const TypeOopPtr* toop = ((tklass != NULL) 4657 ? tklass->as_instance_type() 4658 : TypeInstPtr::NOTNULL); 4659 4660 // Conservatively insert a memory barrier on all memory slices. 4661 // Do not let writes into the original float below the clone. 4662 insert_mem_bar(Op_MemBarCPUOrder); 4663 4664 // paths into result_reg: 4665 enum { 4666 _slow_path = 1, // out-of-line call to clone method (virtual or not) 4667 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 4668 _array_path, // plain array allocation, plus arrayof_long_arraycopy 4669 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 4670 PATH_LIMIT 4671 }; 4672 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4673 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4674 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 4675 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4676 record_for_igvn(result_reg); 4677 4678 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM; 4679 int raw_adr_idx = Compile::AliasIdxRaw; 4680 4681 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL); 4682 if (array_ctl != NULL) { 4683 // It's an array. 4684 PreserveJVMState pjvms(this); 4685 set_control(array_ctl); 4686 Node* obj_length = load_array_length(obj); 4687 Node* obj_size = NULL; 4688 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push 4689 4690 if (!use_ReduceInitialCardMarks()) { 4691 // If it is an oop array, it requires very special treatment, 4692 // because card marking is required on each card of the array. 4693 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL); 4694 if (is_obja != NULL) { 4695 PreserveJVMState pjvms2(this); 4696 set_control(is_obja); 4697 // Generate a direct call to the right arraycopy function(s). 4698 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL); 4699 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL); 4700 ac->set_cloneoop(); 4701 Node* n = _gvn.transform(ac); 4702 assert(n == ac, "cannot disappear"); 4703 ac->connect_outputs(this); 4704 4705 result_reg->init_req(_objArray_path, control()); 4706 result_val->init_req(_objArray_path, alloc_obj); 4707 result_i_o ->set_req(_objArray_path, i_o()); 4708 result_mem ->set_req(_objArray_path, reset_memory()); 4709 } 4710 } 4711 // Otherwise, there are no card marks to worry about. 4712 // (We can dispense with card marks if we know the allocation 4713 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 4714 // causes the non-eden paths to take compensating steps to 4715 // simulate a fresh allocation, so that no further 4716 // card marks are required in compiled code to initialize 4717 // the object.) 4718 4719 if (!stopped()) { 4720 copy_to_clone(obj, alloc_obj, obj_size, true, false); 4721 4722 // Present the results of the copy. 4723 result_reg->init_req(_array_path, control()); 4724 result_val->init_req(_array_path, alloc_obj); 4725 result_i_o ->set_req(_array_path, i_o()); 4726 result_mem ->set_req(_array_path, reset_memory()); 4727 } 4728 } 4729 4730 // We only go to the instance fast case code if we pass a number of guards. 4731 // The paths which do not pass are accumulated in the slow_region. 4732 RegionNode* slow_region = new RegionNode(1); 4733 record_for_igvn(slow_region); 4734 if (!stopped()) { 4735 // It's an instance (we did array above). Make the slow-path tests. 4736 // If this is a virtual call, we generate a funny guard. We grab 4737 // the vtable entry corresponding to clone() from the target object. 4738 // If the target method which we are calling happens to be the 4739 // Object clone() method, we pass the guard. We do not need this 4740 // guard for non-virtual calls; the caller is known to be the native 4741 // Object clone(). 4742 if (is_virtual) { 4743 generate_virtual_guard(obj_klass, slow_region); 4744 } 4745 4746 // The object must be easily cloneable and must not have a finalizer. 4747 // Both of these conditions may be checked in a single test. 4748 // We could optimize the test further, but we don't care. 4749 generate_access_flags_guard(obj_klass, 4750 // Test both conditions: 4751 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER, 4752 // Must be cloneable but not finalizer: 4753 JVM_ACC_IS_CLONEABLE_FAST, 4754 slow_region); 4755 } 4756 4757 if (!stopped()) { 4758 // It's an instance, and it passed the slow-path tests. 4759 PreserveJVMState pjvms(this); 4760 Node* obj_size = NULL; 4761 // Need to deoptimize on exception from allocation since Object.clone intrinsic 4762 // is reexecuted if deoptimization occurs and there could be problems when merging 4763 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 4764 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true); 4765 4766 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks()); 4767 4768 // Present the results of the slow call. 4769 result_reg->init_req(_instance_path, control()); 4770 result_val->init_req(_instance_path, alloc_obj); 4771 result_i_o ->set_req(_instance_path, i_o()); 4772 result_mem ->set_req(_instance_path, reset_memory()); 4773 } 4774 4775 // Generate code for the slow case. We make a call to clone(). 4776 set_control(_gvn.transform(slow_region)); 4777 if (!stopped()) { 4778 PreserveJVMState pjvms(this); 4779 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual); 4780 Node* slow_result = set_results_for_java_call(slow_call); 4781 // this->control() comes from set_results_for_java_call 4782 result_reg->init_req(_slow_path, control()); 4783 result_val->init_req(_slow_path, slow_result); 4784 result_i_o ->set_req(_slow_path, i_o()); 4785 result_mem ->set_req(_slow_path, reset_memory()); 4786 } 4787 4788 // Return the combined state. 4789 set_control( _gvn.transform(result_reg)); 4790 set_i_o( _gvn.transform(result_i_o)); 4791 set_all_memory( _gvn.transform(result_mem)); 4792 } // original reexecute is set back here 4793 4794 set_result(_gvn.transform(result_val)); 4795 return true; 4796 } 4797 4798 // If we have a tighly coupled allocation, the arraycopy may take care 4799 // of the array initialization. If one of the guards we insert between 4800 // the allocation and the arraycopy causes a deoptimization, an 4801 // unitialized array will escape the compiled method. To prevent that 4802 // we set the JVM state for uncommon traps between the allocation and 4803 // the arraycopy to the state before the allocation so, in case of 4804 // deoptimization, we'll reexecute the allocation and the 4805 // initialization. 4806 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 4807 if (alloc != NULL) { 4808 ciMethod* trap_method = alloc->jvms()->method(); 4809 int trap_bci = alloc->jvms()->bci(); 4810 4811 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) & 4812 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 4813 // Make sure there's no store between the allocation and the 4814 // arraycopy otherwise visible side effects could be rexecuted 4815 // in case of deoptimization and cause incorrect execution. 4816 bool no_interfering_store = true; 4817 Node* mem = alloc->in(TypeFunc::Memory); 4818 if (mem->is_MergeMem()) { 4819 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 4820 Node* n = mms.memory(); 4821 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4822 assert(n->is_Store(), "what else?"); 4823 no_interfering_store = false; 4824 break; 4825 } 4826 } 4827 } else { 4828 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 4829 Node* n = mms.memory(); 4830 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4831 assert(n->is_Store(), "what else?"); 4832 no_interfering_store = false; 4833 break; 4834 } 4835 } 4836 } 4837 4838 if (no_interfering_store) { 4839 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 4840 uint size = alloc->req(); 4841 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 4842 old_jvms->set_map(sfpt); 4843 for (uint i = 0; i < size; i++) { 4844 sfpt->init_req(i, alloc->in(i)); 4845 } 4846 // re-push array length for deoptimization 4847 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength)); 4848 old_jvms->set_sp(old_jvms->sp()+1); 4849 old_jvms->set_monoff(old_jvms->monoff()+1); 4850 old_jvms->set_scloff(old_jvms->scloff()+1); 4851 old_jvms->set_endoff(old_jvms->endoff()+1); 4852 old_jvms->set_should_reexecute(true); 4853 4854 sfpt->set_i_o(map()->i_o()); 4855 sfpt->set_memory(map()->memory()); 4856 sfpt->set_control(map()->control()); 4857 4858 JVMState* saved_jvms = jvms(); 4859 saved_reexecute_sp = _reexecute_sp; 4860 4861 set_jvms(sfpt->jvms()); 4862 _reexecute_sp = jvms()->sp(); 4863 4864 return saved_jvms; 4865 } 4866 } 4867 } 4868 return NULL; 4869 } 4870 4871 // In case of a deoptimization, we restart execution at the 4872 // allocation, allocating a new array. We would leave an uninitialized 4873 // array in the heap that GCs wouldn't expect. Move the allocation 4874 // after the traps so we don't allocate the array if we 4875 // deoptimize. This is possible because tightly_coupled_allocation() 4876 // guarantees there's no observer of the allocated array at this point 4877 // and the control flow is simple enough. 4878 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp) { 4879 if (saved_jvms != NULL && !stopped()) { 4880 assert(alloc != NULL, "only with a tightly coupled allocation"); 4881 // restore JVM state to the state at the arraycopy 4882 saved_jvms->map()->set_control(map()->control()); 4883 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?"); 4884 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?"); 4885 // If we've improved the types of some nodes (null check) while 4886 // emitting the guards, propagate them to the current state 4887 map()->replaced_nodes().apply(saved_jvms->map()); 4888 set_jvms(saved_jvms); 4889 _reexecute_sp = saved_reexecute_sp; 4890 4891 // Remove the allocation from above the guards 4892 CallProjections callprojs; 4893 alloc->extract_projections(&callprojs, true); 4894 InitializeNode* init = alloc->initialization(); 4895 Node* alloc_mem = alloc->in(TypeFunc::Memory); 4896 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 4897 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 4898 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 4899 4900 // move the allocation here (after the guards) 4901 _gvn.hash_delete(alloc); 4902 alloc->set_req(TypeFunc::Control, control()); 4903 alloc->set_req(TypeFunc::I_O, i_o()); 4904 Node *mem = reset_memory(); 4905 set_all_memory(mem); 4906 alloc->set_req(TypeFunc::Memory, mem); 4907 set_control(init->proj_out(TypeFunc::Control)); 4908 set_i_o(callprojs.fallthrough_ioproj); 4909 4910 // Update memory as done in GraphKit::set_output_for_allocation() 4911 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 4912 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 4913 if (ary_type->isa_aryptr() && length_type != NULL) { 4914 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 4915 } 4916 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 4917 int elemidx = C->get_alias_index(telemref); 4918 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw); 4919 set_memory(init->proj_out(TypeFunc::Memory), elemidx); 4920 4921 Node* allocx = _gvn.transform(alloc); 4922 assert(allocx == alloc, "where has the allocation gone?"); 4923 assert(dest->is_CheckCastPP(), "not an allocation result?"); 4924 4925 _gvn.hash_delete(dest); 4926 dest->set_req(0, control()); 4927 Node* destx = _gvn.transform(dest); 4928 assert(destx == dest, "where has the allocation result gone?"); 4929 } 4930 } 4931 4932 4933 //------------------------------inline_arraycopy----------------------- 4934 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 4935 // Object dest, int destPos, 4936 // int length); 4937 bool LibraryCallKit::inline_arraycopy() { 4938 // Get the arguments. 4939 Node* src = argument(0); // type: oop 4940 Node* src_offset = argument(1); // type: int 4941 Node* dest = argument(2); // type: oop 4942 Node* dest_offset = argument(3); // type: int 4943 Node* length = argument(4); // type: int 4944 4945 4946 // Check for allocation before we add nodes that would confuse 4947 // tightly_coupled_allocation() 4948 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL); 4949 4950 int saved_reexecute_sp = -1; 4951 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 4952 // See arraycopy_restore_alloc_state() comment 4953 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards 4954 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation 4955 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards 4956 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL); 4957 4958 // The following tests must be performed 4959 // (1) src and dest are arrays. 4960 // (2) src and dest arrays must have elements of the same BasicType 4961 // (3) src and dest must not be null. 4962 // (4) src_offset must not be negative. 4963 // (5) dest_offset must not be negative. 4964 // (6) length must not be negative. 4965 // (7) src_offset + length must not exceed length of src. 4966 // (8) dest_offset + length must not exceed length of dest. 4967 // (9) each element of an oop array must be assignable 4968 4969 // (3) src and dest must not be null. 4970 // always do this here because we need the JVM state for uncommon traps 4971 Node* null_ctl = top(); 4972 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 4973 assert(null_ctl->is_top(), "no null control here"); 4974 dest = null_check(dest, T_ARRAY); 4975 4976 if (!can_emit_guards) { 4977 // if saved_jvms == NULL and alloc != NULL, we don't emit any 4978 // guards but the arraycopy node could still take advantage of a 4979 // tightly allocated allocation. tightly_coupled_allocation() is 4980 // called again to make sure it takes the null check above into 4981 // account: the null check is mandatory and if it caused an 4982 // uncommon trap to be emitted then the allocation can't be 4983 // considered tightly coupled in this context. 4984 alloc = tightly_coupled_allocation(dest, NULL); 4985 } 4986 4987 bool validated = false; 4988 4989 const Type* src_type = _gvn.type(src); 4990 const Type* dest_type = _gvn.type(dest); 4991 const TypeAryPtr* top_src = src_type->isa_aryptr(); 4992 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 4993 4994 // Do we have the type of src? 4995 bool has_src = (top_src != NULL && top_src->klass() != NULL); 4996 // Do we have the type of dest? 4997 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL); 4998 // Is the type for src from speculation? 4999 bool src_spec = false; 5000 // Is the type for dest from speculation? 5001 bool dest_spec = false; 5002 5003 if ((!has_src || !has_dest) && can_emit_guards) { 5004 // We don't have sufficient type information, let's see if 5005 // speculative types can help. We need to have types for both src 5006 // and dest so that it pays off. 5007 5008 // Do we already have or could we have type information for src 5009 bool could_have_src = has_src; 5010 // Do we already have or could we have type information for dest 5011 bool could_have_dest = has_dest; 5012 5013 ciKlass* src_k = NULL; 5014 if (!has_src) { 5015 src_k = src_type->speculative_type_not_null(); 5016 if (src_k != NULL && src_k->is_array_klass()) { 5017 could_have_src = true; 5018 } 5019 } 5020 5021 ciKlass* dest_k = NULL; 5022 if (!has_dest) { 5023 dest_k = dest_type->speculative_type_not_null(); 5024 if (dest_k != NULL && dest_k->is_array_klass()) { 5025 could_have_dest = true; 5026 } 5027 } 5028 5029 if (could_have_src && could_have_dest) { 5030 // This is going to pay off so emit the required guards 5031 if (!has_src) { 5032 src = maybe_cast_profiled_obj(src, src_k, true); 5033 src_type = _gvn.type(src); 5034 top_src = src_type->isa_aryptr(); 5035 has_src = (top_src != NULL && top_src->klass() != NULL); 5036 src_spec = true; 5037 } 5038 if (!has_dest) { 5039 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5040 dest_type = _gvn.type(dest); 5041 top_dest = dest_type->isa_aryptr(); 5042 has_dest = (top_dest != NULL && top_dest->klass() != NULL); 5043 dest_spec = true; 5044 } 5045 } 5046 } 5047 5048 if (has_src && has_dest && can_emit_guards) { 5049 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type(); 5050 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type(); 5051 if (src_elem == T_ARRAY) src_elem = T_OBJECT; 5052 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT; 5053 5054 if (src_elem == dest_elem && src_elem == T_OBJECT) { 5055 // If both arrays are object arrays then having the exact types 5056 // for both will remove the need for a subtype check at runtime 5057 // before the call and may make it possible to pick a faster copy 5058 // routine (without a subtype check on every element) 5059 // Do we have the exact type of src? 5060 bool could_have_src = src_spec; 5061 // Do we have the exact type of dest? 5062 bool could_have_dest = dest_spec; 5063 ciKlass* src_k = top_src->klass(); 5064 ciKlass* dest_k = top_dest->klass(); 5065 if (!src_spec) { 5066 src_k = src_type->speculative_type_not_null(); 5067 if (src_k != NULL && src_k->is_array_klass()) { 5068 could_have_src = true; 5069 } 5070 } 5071 if (!dest_spec) { 5072 dest_k = dest_type->speculative_type_not_null(); 5073 if (dest_k != NULL && dest_k->is_array_klass()) { 5074 could_have_dest = true; 5075 } 5076 } 5077 if (could_have_src && could_have_dest) { 5078 // If we can have both exact types, emit the missing guards 5079 if (could_have_src && !src_spec) { 5080 src = maybe_cast_profiled_obj(src, src_k, true); 5081 } 5082 if (could_have_dest && !dest_spec) { 5083 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5084 } 5085 } 5086 } 5087 } 5088 5089 ciMethod* trap_method = method(); 5090 int trap_bci = bci(); 5091 if (saved_jvms != NULL) { 5092 trap_method = alloc->jvms()->method(); 5093 trap_bci = alloc->jvms()->bci(); 5094 } 5095 5096 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5097 can_emit_guards && 5098 !src->is_top() && !dest->is_top()) { 5099 // validate arguments: enables transformation the ArrayCopyNode 5100 validated = true; 5101 5102 RegionNode* slow_region = new RegionNode(1); 5103 record_for_igvn(slow_region); 5104 5105 // (1) src and dest are arrays. 5106 generate_non_array_guard(load_object_klass(src), slow_region); 5107 generate_non_array_guard(load_object_klass(dest), slow_region); 5108 5109 // (2) src and dest arrays must have elements of the same BasicType 5110 // done at macro expansion or at Ideal transformation time 5111 5112 // (4) src_offset must not be negative. 5113 generate_negative_guard(src_offset, slow_region); 5114 5115 // (5) dest_offset must not be negative. 5116 generate_negative_guard(dest_offset, slow_region); 5117 5118 // (7) src_offset + length must not exceed length of src. 5119 generate_limit_guard(src_offset, length, 5120 load_array_length(src), 5121 slow_region); 5122 5123 // (8) dest_offset + length must not exceed length of dest. 5124 generate_limit_guard(dest_offset, length, 5125 load_array_length(dest), 5126 slow_region); 5127 5128 // (9) each element of an oop array must be assignable 5129 Node* src_klass = load_object_klass(src); 5130 Node* dest_klass = load_object_klass(dest); 5131 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass); 5132 5133 if (not_subtype_ctrl != top()) { 5134 PreserveJVMState pjvms(this); 5135 set_control(not_subtype_ctrl); 5136 uncommon_trap(Deoptimization::Reason_intrinsic, 5137 Deoptimization::Action_make_not_entrant); 5138 assert(stopped(), "Should be stopped"); 5139 } 5140 { 5141 PreserveJVMState pjvms(this); 5142 set_control(_gvn.transform(slow_region)); 5143 uncommon_trap(Deoptimization::Reason_intrinsic, 5144 Deoptimization::Action_make_not_entrant); 5145 assert(stopped(), "Should be stopped"); 5146 } 5147 } 5148 5149 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp); 5150 5151 if (stopped()) { 5152 return true; 5153 } 5154 5155 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, 5156 // Create LoadRange and LoadKlass nodes for use during macro expansion here 5157 // so the compiler has a chance to eliminate them: during macro expansion, 5158 // we have to set their control (CastPP nodes are eliminated). 5159 load_object_klass(src), load_object_klass(dest), 5160 load_array_length(src), load_array_length(dest)); 5161 5162 ac->set_arraycopy(validated); 5163 5164 Node* n = _gvn.transform(ac); 5165 if (n == ac) { 5166 ac->connect_outputs(this); 5167 } else { 5168 assert(validated, "shouldn't transform if all arguments not validated"); 5169 set_all_memory(n); 5170 } 5171 5172 return true; 5173 } 5174 5175 5176 // Helper function which determines if an arraycopy immediately follows 5177 // an allocation, with no intervening tests or other escapes for the object. 5178 AllocateArrayNode* 5179 LibraryCallKit::tightly_coupled_allocation(Node* ptr, 5180 RegionNode* slow_region) { 5181 if (stopped()) return NULL; // no fast path 5182 if (C->AliasLevel() == 0) return NULL; // no MergeMems around 5183 5184 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn); 5185 if (alloc == NULL) return NULL; 5186 5187 Node* rawmem = memory(Compile::AliasIdxRaw); 5188 // Is the allocation's memory state untouched? 5189 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 5190 // Bail out if there have been raw-memory effects since the allocation. 5191 // (Example: There might have been a call or safepoint.) 5192 return NULL; 5193 } 5194 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 5195 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 5196 return NULL; 5197 } 5198 5199 // There must be no unexpected observers of this allocation. 5200 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 5201 Node* obs = ptr->fast_out(i); 5202 if (obs != this->map()) { 5203 return NULL; 5204 } 5205 } 5206 5207 // This arraycopy must unconditionally follow the allocation of the ptr. 5208 Node* alloc_ctl = ptr->in(0); 5209 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo"); 5210 5211 Node* ctl = control(); 5212 while (ctl != alloc_ctl) { 5213 // There may be guards which feed into the slow_region. 5214 // Any other control flow means that we might not get a chance 5215 // to finish initializing the allocated object. 5216 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) { 5217 IfNode* iff = ctl->in(0)->as_If(); 5218 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con); 5219 assert(not_ctl != NULL && not_ctl != ctl, "found alternate"); 5220 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) { 5221 ctl = iff->in(0); // This test feeds the known slow_region. 5222 continue; 5223 } 5224 // One more try: Various low-level checks bottom out in 5225 // uncommon traps. If the debug-info of the trap omits 5226 // any reference to the allocation, as we've already 5227 // observed, then there can be no objection to the trap. 5228 bool found_trap = false; 5229 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) { 5230 Node* obs = not_ctl->fast_out(j); 5231 if (obs->in(0) == not_ctl && obs->is_Call() && 5232 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) { 5233 found_trap = true; break; 5234 } 5235 } 5236 if (found_trap) { 5237 ctl = iff->in(0); // This test feeds a harmless uncommon trap. 5238 continue; 5239 } 5240 } 5241 return NULL; 5242 } 5243 5244 // If we get this far, we have an allocation which immediately 5245 // precedes the arraycopy, and we can take over zeroing the new object. 5246 // The arraycopy will finish the initialization, and provide 5247 // a new control state to which we will anchor the destination pointer. 5248 5249 return alloc; 5250 } 5251 5252 //-------------inline_encodeISOArray----------------------------------- 5253 // encode char[] to byte[] in ISO_8859_1 5254 bool LibraryCallKit::inline_encodeISOArray() { 5255 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 5256 // no receiver since it is static method 5257 Node *src = argument(0); 5258 Node *src_offset = argument(1); 5259 Node *dst = argument(2); 5260 Node *dst_offset = argument(3); 5261 Node *length = argument(4); 5262 5263 const Type* src_type = src->Value(&_gvn); 5264 const Type* dst_type = dst->Value(&_gvn); 5265 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5266 const TypeAryPtr* top_dest = dst_type->isa_aryptr(); 5267 if (top_src == NULL || top_src->klass() == NULL || 5268 top_dest == NULL || top_dest->klass() == NULL) { 5269 // failed array check 5270 return false; 5271 } 5272 5273 // Figure out the size and type of the elements we will be copying. 5274 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5275 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5276 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 5277 return false; 5278 } 5279 5280 Node* src_start = array_element_address(src, src_offset, T_CHAR); 5281 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 5282 // 'src_start' points to src array + scaled offset 5283 // 'dst_start' points to dst array + scaled offset 5284 5285 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 5286 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length); 5287 enc = _gvn.transform(enc); 5288 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 5289 set_memory(res_mem, mtype); 5290 set_result(enc); 5291 return true; 5292 } 5293 5294 //-------------inline_multiplyToLen----------------------------------- 5295 bool LibraryCallKit::inline_multiplyToLen() { 5296 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 5297 5298 address stubAddr = StubRoutines::multiplyToLen(); 5299 if (stubAddr == NULL) { 5300 return false; // Intrinsic's stub is not implemented on this platform 5301 } 5302 const char* stubName = "multiplyToLen"; 5303 5304 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 5305 5306 // no receiver because it is a static method 5307 Node* x = argument(0); 5308 Node* xlen = argument(1); 5309 Node* y = argument(2); 5310 Node* ylen = argument(3); 5311 Node* z = argument(4); 5312 5313 const Type* x_type = x->Value(&_gvn); 5314 const Type* y_type = y->Value(&_gvn); 5315 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5316 const TypeAryPtr* top_y = y_type->isa_aryptr(); 5317 if (top_x == NULL || top_x->klass() == NULL || 5318 top_y == NULL || top_y->klass() == NULL) { 5319 // failed array check 5320 return false; 5321 } 5322 5323 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5324 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5325 if (x_elem != T_INT || y_elem != T_INT) { 5326 return false; 5327 } 5328 5329 // Set the original stack and the reexecute bit for the interpreter to reexecute 5330 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens 5331 // on the return from z array allocation in runtime. 5332 { PreserveReexecuteState preexecs(this); 5333 jvms()->set_should_reexecute(true); 5334 5335 Node* x_start = array_element_address(x, intcon(0), x_elem); 5336 Node* y_start = array_element_address(y, intcon(0), y_elem); 5337 // 'x_start' points to x array + scaled xlen 5338 // 'y_start' points to y array + scaled ylen 5339 5340 // Allocate the result array 5341 Node* zlen = _gvn.transform(new AddINode(xlen, ylen)); 5342 ciKlass* klass = ciTypeArrayKlass::make(T_INT); 5343 Node* klass_node = makecon(TypeKlassPtr::make(klass)); 5344 5345 IdealKit ideal(this); 5346 5347 #define __ ideal. 5348 Node* one = __ ConI(1); 5349 Node* zero = __ ConI(0); 5350 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done(); 5351 __ set(need_alloc, zero); 5352 __ set(z_alloc, z); 5353 __ if_then(z, BoolTest::eq, null()); { 5354 __ increment (need_alloc, one); 5355 } __ else_(); { 5356 // Update graphKit memory and control from IdealKit. 5357 sync_kit(ideal); 5358 Node* zlen_arg = load_array_length(z); 5359 // Update IdealKit memory and control from graphKit. 5360 __ sync_kit(this); 5361 __ if_then(zlen_arg, BoolTest::lt, zlen); { 5362 __ increment (need_alloc, one); 5363 } __ end_if(); 5364 } __ end_if(); 5365 5366 __ if_then(__ value(need_alloc), BoolTest::ne, zero); { 5367 // Update graphKit memory and control from IdealKit. 5368 sync_kit(ideal); 5369 Node * narr = new_array(klass_node, zlen, 1); 5370 // Update IdealKit memory and control from graphKit. 5371 __ sync_kit(this); 5372 __ set(z_alloc, narr); 5373 } __ end_if(); 5374 5375 sync_kit(ideal); 5376 z = __ value(z_alloc); 5377 // Can't use TypeAryPtr::INTS which uses Bottom offset. 5378 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass)); 5379 // Final sync IdealKit and GraphKit. 5380 final_sync(ideal); 5381 #undef __ 5382 5383 Node* z_start = array_element_address(z, intcon(0), T_INT); 5384 5385 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5386 OptoRuntime::multiplyToLen_Type(), 5387 stubAddr, stubName, TypePtr::BOTTOM, 5388 x_start, xlen, y_start, ylen, z_start, zlen); 5389 } // original reexecute is set back here 5390 5391 C->set_has_split_ifs(true); // Has chance for split-if optimization 5392 set_result(z); 5393 return true; 5394 } 5395 5396 //-------------inline_squareToLen------------------------------------ 5397 bool LibraryCallKit::inline_squareToLen() { 5398 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 5399 5400 address stubAddr = StubRoutines::squareToLen(); 5401 if (stubAddr == NULL) { 5402 return false; // Intrinsic's stub is not implemented on this platform 5403 } 5404 const char* stubName = "squareToLen"; 5405 5406 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 5407 5408 Node* x = argument(0); 5409 Node* len = argument(1); 5410 Node* z = argument(2); 5411 Node* zlen = argument(3); 5412 5413 const Type* x_type = x->Value(&_gvn); 5414 const Type* z_type = z->Value(&_gvn); 5415 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5416 const TypeAryPtr* top_z = z_type->isa_aryptr(); 5417 if (top_x == NULL || top_x->klass() == NULL || 5418 top_z == NULL || top_z->klass() == NULL) { 5419 // failed array check 5420 return false; 5421 } 5422 5423 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5424 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5425 if (x_elem != T_INT || z_elem != T_INT) { 5426 return false; 5427 } 5428 5429 5430 Node* x_start = array_element_address(x, intcon(0), x_elem); 5431 Node* z_start = array_element_address(z, intcon(0), z_elem); 5432 5433 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5434 OptoRuntime::squareToLen_Type(), 5435 stubAddr, stubName, TypePtr::BOTTOM, 5436 x_start, len, z_start, zlen); 5437 5438 set_result(z); 5439 return true; 5440 } 5441 5442 //-------------inline_mulAdd------------------------------------------ 5443 bool LibraryCallKit::inline_mulAdd() { 5444 assert(UseMulAddIntrinsic, "not implemented on this platform"); 5445 5446 address stubAddr = StubRoutines::mulAdd(); 5447 if (stubAddr == NULL) { 5448 return false; // Intrinsic's stub is not implemented on this platform 5449 } 5450 const char* stubName = "mulAdd"; 5451 5452 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 5453 5454 Node* out = argument(0); 5455 Node* in = argument(1); 5456 Node* offset = argument(2); 5457 Node* len = argument(3); 5458 Node* k = argument(4); 5459 5460 const Type* out_type = out->Value(&_gvn); 5461 const Type* in_type = in->Value(&_gvn); 5462 const TypeAryPtr* top_out = out_type->isa_aryptr(); 5463 const TypeAryPtr* top_in = in_type->isa_aryptr(); 5464 if (top_out == NULL || top_out->klass() == NULL || 5465 top_in == NULL || top_in->klass() == NULL) { 5466 // failed array check 5467 return false; 5468 } 5469 5470 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5471 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5472 if (out_elem != T_INT || in_elem != T_INT) { 5473 return false; 5474 } 5475 5476 Node* outlen = load_array_length(out); 5477 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 5478 Node* out_start = array_element_address(out, intcon(0), out_elem); 5479 Node* in_start = array_element_address(in, intcon(0), in_elem); 5480 5481 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5482 OptoRuntime::mulAdd_Type(), 5483 stubAddr, stubName, TypePtr::BOTTOM, 5484 out_start,in_start, new_offset, len, k); 5485 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5486 set_result(result); 5487 return true; 5488 } 5489 5490 //-------------inline_montgomeryMultiply----------------------------------- 5491 bool LibraryCallKit::inline_montgomeryMultiply() { 5492 address stubAddr = StubRoutines::montgomeryMultiply(); 5493 if (stubAddr == NULL) { 5494 return false; // Intrinsic's stub is not implemented on this platform 5495 } 5496 5497 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 5498 const char* stubName = "montgomery_square"; 5499 5500 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 5501 5502 Node* a = argument(0); 5503 Node* b = argument(1); 5504 Node* n = argument(2); 5505 Node* len = argument(3); 5506 Node* inv = argument(4); 5507 Node* m = argument(6); 5508 5509 const Type* a_type = a->Value(&_gvn); 5510 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5511 const Type* b_type = b->Value(&_gvn); 5512 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5513 const Type* n_type = a->Value(&_gvn); 5514 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5515 const Type* m_type = a->Value(&_gvn); 5516 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5517 if (top_a == NULL || top_a->klass() == NULL || 5518 top_b == NULL || top_b->klass() == NULL || 5519 top_n == NULL || top_n->klass() == NULL || 5520 top_m == NULL || top_m->klass() == NULL) { 5521 // failed array check 5522 return false; 5523 } 5524 5525 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5526 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5527 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5528 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5529 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5530 return false; 5531 } 5532 5533 // Make the call 5534 { 5535 Node* a_start = array_element_address(a, intcon(0), a_elem); 5536 Node* b_start = array_element_address(b, intcon(0), b_elem); 5537 Node* n_start = array_element_address(n, intcon(0), n_elem); 5538 Node* m_start = array_element_address(m, intcon(0), m_elem); 5539 5540 Node* call = make_runtime_call(RC_LEAF, 5541 OptoRuntime::montgomeryMultiply_Type(), 5542 stubAddr, stubName, TypePtr::BOTTOM, 5543 a_start, b_start, n_start, len, inv, top(), 5544 m_start); 5545 set_result(m); 5546 } 5547 5548 return true; 5549 } 5550 5551 bool LibraryCallKit::inline_montgomerySquare() { 5552 address stubAddr = StubRoutines::montgomerySquare(); 5553 if (stubAddr == NULL) { 5554 return false; // Intrinsic's stub is not implemented on this platform 5555 } 5556 5557 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 5558 const char* stubName = "montgomery_square"; 5559 5560 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 5561 5562 Node* a = argument(0); 5563 Node* n = argument(1); 5564 Node* len = argument(2); 5565 Node* inv = argument(3); 5566 Node* m = argument(5); 5567 5568 const Type* a_type = a->Value(&_gvn); 5569 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5570 const Type* n_type = a->Value(&_gvn); 5571 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5572 const Type* m_type = a->Value(&_gvn); 5573 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5574 if (top_a == NULL || top_a->klass() == NULL || 5575 top_n == NULL || top_n->klass() == NULL || 5576 top_m == NULL || top_m->klass() == NULL) { 5577 // failed array check 5578 return false; 5579 } 5580 5581 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5582 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5583 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5584 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5585 return false; 5586 } 5587 5588 // Make the call 5589 { 5590 Node* a_start = array_element_address(a, intcon(0), a_elem); 5591 Node* n_start = array_element_address(n, intcon(0), n_elem); 5592 Node* m_start = array_element_address(m, intcon(0), m_elem); 5593 5594 Node* call = make_runtime_call(RC_LEAF, 5595 OptoRuntime::montgomerySquare_Type(), 5596 stubAddr, stubName, TypePtr::BOTTOM, 5597 a_start, n_start, len, inv, top(), 5598 m_start); 5599 set_result(m); 5600 } 5601 5602 return true; 5603 } 5604 5605 //-------------inline_vectorizedMismatch------------------------------ 5606 bool LibraryCallKit::inline_vectorizedMismatch() { 5607 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform"); 5608 5609 address stubAddr = StubRoutines::vectorizedMismatch(); 5610 if (stubAddr == NULL) { 5611 return false; // Intrinsic's stub is not implemented on this platform 5612 } 5613 const char* stubName = "vectorizedMismatch"; 5614 int size_l = callee()->signature()->size(); 5615 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 5616 5617 Node* obja = argument(0); 5618 Node* aoffset = argument(1); 5619 Node* objb = argument(3); 5620 Node* boffset = argument(4); 5621 Node* length = argument(6); 5622 Node* scale = argument(7); 5623 5624 const Type* a_type = obja->Value(&_gvn); 5625 const Type* b_type = objb->Value(&_gvn); 5626 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5627 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5628 if (top_a == NULL || top_a->klass() == NULL || 5629 top_b == NULL || top_b->klass() == NULL) { 5630 // failed array check 5631 return false; 5632 } 5633 5634 Node* call; 5635 jvms()->set_should_reexecute(true); 5636 5637 Node* obja_adr = make_unsafe_address(obja, aoffset); 5638 Node* objb_adr = make_unsafe_address(objb, boffset); 5639 5640 call = make_runtime_call(RC_LEAF, 5641 OptoRuntime::vectorizedMismatch_Type(), 5642 stubAddr, stubName, TypePtr::BOTTOM, 5643 obja_adr, objb_adr, length, scale); 5644 5645 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5646 set_result(result); 5647 return true; 5648 } 5649 5650 /** 5651 * Calculate CRC32 for byte. 5652 * int java.util.zip.CRC32.update(int crc, int b) 5653 */ 5654 bool LibraryCallKit::inline_updateCRC32() { 5655 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5656 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 5657 // no receiver since it is static method 5658 Node* crc = argument(0); // type: int 5659 Node* b = argument(1); // type: int 5660 5661 /* 5662 * int c = ~ crc; 5663 * b = timesXtoThe32[(b ^ c) & 0xFF]; 5664 * b = b ^ (c >>> 8); 5665 * crc = ~b; 5666 */ 5667 5668 Node* M1 = intcon(-1); 5669 crc = _gvn.transform(new XorINode(crc, M1)); 5670 Node* result = _gvn.transform(new XorINode(crc, b)); 5671 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 5672 5673 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 5674 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 5675 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 5676 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 5677 5678 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 5679 result = _gvn.transform(new XorINode(crc, result)); 5680 result = _gvn.transform(new XorINode(result, M1)); 5681 set_result(result); 5682 return true; 5683 } 5684 5685 /** 5686 * Calculate CRC32 for byte[] array. 5687 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 5688 */ 5689 bool LibraryCallKit::inline_updateBytesCRC32() { 5690 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5691 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5692 // no receiver since it is static method 5693 Node* crc = argument(0); // type: int 5694 Node* src = argument(1); // type: oop 5695 Node* offset = argument(2); // type: int 5696 Node* length = argument(3); // type: int 5697 5698 const Type* src_type = src->Value(&_gvn); 5699 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5700 if (top_src == NULL || top_src->klass() == NULL) { 5701 // failed array check 5702 return false; 5703 } 5704 5705 // Figure out the size and type of the elements we will be copying. 5706 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5707 if (src_elem != T_BYTE) { 5708 return false; 5709 } 5710 5711 // 'src_start' points to src array + scaled offset 5712 Node* src_start = array_element_address(src, offset, src_elem); 5713 5714 // We assume that range check is done by caller. 5715 // TODO: generate range check (offset+length < src.length) in debug VM. 5716 5717 // Call the stub. 5718 address stubAddr = StubRoutines::updateBytesCRC32(); 5719 const char *stubName = "updateBytesCRC32"; 5720 5721 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5722 stubAddr, stubName, TypePtr::BOTTOM, 5723 crc, src_start, length); 5724 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5725 set_result(result); 5726 return true; 5727 } 5728 5729 /** 5730 * Calculate CRC32 for ByteBuffer. 5731 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 5732 */ 5733 bool LibraryCallKit::inline_updateByteBufferCRC32() { 5734 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5735 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5736 // no receiver since it is static method 5737 Node* crc = argument(0); // type: int 5738 Node* src = argument(1); // type: long 5739 Node* offset = argument(3); // type: int 5740 Node* length = argument(4); // type: int 5741 5742 src = ConvL2X(src); // adjust Java long to machine word 5743 Node* base = _gvn.transform(new CastX2PNode(src)); 5744 offset = ConvI2X(offset); 5745 5746 // 'src_start' points to src array + scaled offset 5747 Node* src_start = basic_plus_adr(top(), base, offset); 5748 5749 // Call the stub. 5750 address stubAddr = StubRoutines::updateBytesCRC32(); 5751 const char *stubName = "updateBytesCRC32"; 5752 5753 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5754 stubAddr, stubName, TypePtr::BOTTOM, 5755 crc, src_start, length); 5756 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5757 set_result(result); 5758 return true; 5759 } 5760 5761 //------------------------------get_table_from_crc32c_class----------------------- 5762 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 5763 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class); 5764 assert (table != NULL, "wrong version of java.util.zip.CRC32C"); 5765 5766 return table; 5767 } 5768 5769 //------------------------------inline_updateBytesCRC32C----------------------- 5770 // 5771 // Calculate CRC32C for byte[] array. 5772 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 5773 // 5774 bool LibraryCallKit::inline_updateBytesCRC32C() { 5775 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5776 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5777 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5778 // no receiver since it is a static method 5779 Node* crc = argument(0); // type: int 5780 Node* src = argument(1); // type: oop 5781 Node* offset = argument(2); // type: int 5782 Node* end = argument(3); // type: int 5783 5784 Node* length = _gvn.transform(new SubINode(end, offset)); 5785 5786 const Type* src_type = src->Value(&_gvn); 5787 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5788 if (top_src == NULL || top_src->klass() == NULL) { 5789 // failed array check 5790 return false; 5791 } 5792 5793 // Figure out the size and type of the elements we will be copying. 5794 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5795 if (src_elem != T_BYTE) { 5796 return false; 5797 } 5798 5799 // 'src_start' points to src array + scaled offset 5800 Node* src_start = array_element_address(src, offset, src_elem); 5801 5802 // static final int[] byteTable in class CRC32C 5803 Node* table = get_table_from_crc32c_class(callee()->holder()); 5804 Node* table_start = array_element_address(table, intcon(0), T_INT); 5805 5806 // We assume that range check is done by caller. 5807 // TODO: generate range check (offset+length < src.length) in debug VM. 5808 5809 // Call the stub. 5810 address stubAddr = StubRoutines::updateBytesCRC32C(); 5811 const char *stubName = "updateBytesCRC32C"; 5812 5813 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5814 stubAddr, stubName, TypePtr::BOTTOM, 5815 crc, src_start, length, table_start); 5816 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5817 set_result(result); 5818 return true; 5819 } 5820 5821 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 5822 // 5823 // Calculate CRC32C for DirectByteBuffer. 5824 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 5825 // 5826 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 5827 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5828 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 5829 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5830 // no receiver since it is a static method 5831 Node* crc = argument(0); // type: int 5832 Node* src = argument(1); // type: long 5833 Node* offset = argument(3); // type: int 5834 Node* end = argument(4); // type: int 5835 5836 Node* length = _gvn.transform(new SubINode(end, offset)); 5837 5838 src = ConvL2X(src); // adjust Java long to machine word 5839 Node* base = _gvn.transform(new CastX2PNode(src)); 5840 offset = ConvI2X(offset); 5841 5842 // 'src_start' points to src array + scaled offset 5843 Node* src_start = basic_plus_adr(top(), base, offset); 5844 5845 // static final int[] byteTable in class CRC32C 5846 Node* table = get_table_from_crc32c_class(callee()->holder()); 5847 Node* table_start = array_element_address(table, intcon(0), T_INT); 5848 5849 // Call the stub. 5850 address stubAddr = StubRoutines::updateBytesCRC32C(); 5851 const char *stubName = "updateBytesCRC32C"; 5852 5853 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5854 stubAddr, stubName, TypePtr::BOTTOM, 5855 crc, src_start, length, table_start); 5856 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5857 set_result(result); 5858 return true; 5859 } 5860 5861 //------------------------------inline_updateBytesAdler32---------------------- 5862 // 5863 // Calculate Adler32 checksum for byte[] array. 5864 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 5865 // 5866 bool LibraryCallKit::inline_updateBytesAdler32() { 5867 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 5868 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5869 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 5870 // no receiver since it is static method 5871 Node* crc = argument(0); // type: int 5872 Node* src = argument(1); // type: oop 5873 Node* offset = argument(2); // type: int 5874 Node* length = argument(3); // type: int 5875 5876 const Type* src_type = src->Value(&_gvn); 5877 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5878 if (top_src == NULL || top_src->klass() == NULL) { 5879 // failed array check 5880 return false; 5881 } 5882 5883 // Figure out the size and type of the elements we will be copying. 5884 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5885 if (src_elem != T_BYTE) { 5886 return false; 5887 } 5888 5889 // 'src_start' points to src array + scaled offset 5890 Node* src_start = array_element_address(src, offset, src_elem); 5891 5892 // We assume that range check is done by caller. 5893 // TODO: generate range check (offset+length < src.length) in debug VM. 5894 5895 // Call the stub. 5896 address stubAddr = StubRoutines::updateBytesAdler32(); 5897 const char *stubName = "updateBytesAdler32"; 5898 5899 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 5900 stubAddr, stubName, TypePtr::BOTTOM, 5901 crc, src_start, length); 5902 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5903 set_result(result); 5904 return true; 5905 } 5906 5907 //------------------------------inline_updateByteBufferAdler32--------------- 5908 // 5909 // Calculate Adler32 checksum for DirectByteBuffer. 5910 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 5911 // 5912 bool LibraryCallKit::inline_updateByteBufferAdler32() { 5913 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 5914 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5915 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 5916 // no receiver since it is static method 5917 Node* crc = argument(0); // type: int 5918 Node* src = argument(1); // type: long 5919 Node* offset = argument(3); // type: int 5920 Node* length = argument(4); // type: int 5921 5922 src = ConvL2X(src); // adjust Java long to machine word 5923 Node* base = _gvn.transform(new CastX2PNode(src)); 5924 offset = ConvI2X(offset); 5925 5926 // 'src_start' points to src array + scaled offset 5927 Node* src_start = basic_plus_adr(top(), base, offset); 5928 5929 // Call the stub. 5930 address stubAddr = StubRoutines::updateBytesAdler32(); 5931 const char *stubName = "updateBytesAdler32"; 5932 5933 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 5934 stubAddr, stubName, TypePtr::BOTTOM, 5935 crc, src_start, length); 5936 5937 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5938 set_result(result); 5939 return true; 5940 } 5941 5942 //----------------------------inline_reference_get---------------------------- 5943 // public T java.lang.ref.Reference.get(); 5944 bool LibraryCallKit::inline_reference_get() { 5945 const int referent_offset = java_lang_ref_Reference::referent_offset; 5946 guarantee(referent_offset > 0, "should have already been set"); 5947 5948 // Get the argument: 5949 Node* reference_obj = null_check_receiver(); 5950 if (stopped()) return true; 5951 5952 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset); 5953 5954 ciInstanceKlass* klass = env()->Object_klass(); 5955 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass); 5956 5957 Node* no_ctrl = NULL; 5958 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered); 5959 5960 // Use the pre-barrier to record the value in the referent field 5961 pre_barrier(false /* do_load */, 5962 control(), 5963 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */, 5964 result /* pre_val */, 5965 T_OBJECT); 5966 5967 // Add memory barrier to prevent commoning reads from this field 5968 // across safepoint since GC can change its value. 5969 insert_mem_bar(Op_MemBarCPUOrder); 5970 5971 set_result(result); 5972 return true; 5973 } 5974 5975 5976 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 5977 bool is_exact=true, bool is_static=false, 5978 ciInstanceKlass * fromKls=NULL) { 5979 if (fromKls == NULL) { 5980 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 5981 assert(tinst != NULL, "obj is null"); 5982 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 5983 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 5984 fromKls = tinst->klass()->as_instance_klass(); 5985 } else { 5986 assert(is_static, "only for static field access"); 5987 } 5988 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 5989 ciSymbol::make(fieldTypeString), 5990 is_static); 5991 5992 assert (field != NULL, "undefined field"); 5993 if (field == NULL) return (Node *) NULL; 5994 5995 if (is_static) { 5996 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 5997 fromObj = makecon(tip); 5998 } 5999 6000 // Next code copied from Parse::do_get_xxx(): 6001 6002 // Compute address and memory type. 6003 int offset = field->offset_in_bytes(); 6004 bool is_vol = field->is_volatile(); 6005 ciType* field_klass = field->type(); 6006 assert(field_klass->is_loaded(), "should be loaded"); 6007 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 6008 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 6009 BasicType bt = field->layout_type(); 6010 6011 // Build the resultant type of the load 6012 const Type *type; 6013 if (bt == T_OBJECT) { 6014 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 6015 } else { 6016 type = Type::get_const_basic_type(bt); 6017 } 6018 6019 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) { 6020 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier 6021 } 6022 // Build the load. 6023 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered; 6024 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol); 6025 // If reference is volatile, prevent following memory ops from 6026 // floating up past the volatile read. Also prevents commoning 6027 // another volatile read. 6028 if (is_vol) { 6029 // Memory barrier includes bogus read of value to force load BEFORE membar 6030 insert_mem_bar(Op_MemBarAcquire, loadedField); 6031 } 6032 return loadedField; 6033 } 6034 6035 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 6036 bool is_exact = true, bool is_static = false, 6037 ciInstanceKlass * fromKls = NULL) { 6038 if (fromKls == NULL) { 6039 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 6040 assert(tinst != NULL, "obj is null"); 6041 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 6042 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 6043 fromKls = tinst->klass()->as_instance_klass(); 6044 } 6045 else { 6046 assert(is_static, "only for static field access"); 6047 } 6048 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 6049 ciSymbol::make(fieldTypeString), 6050 is_static); 6051 6052 assert(field != NULL, "undefined field"); 6053 assert(!field->is_volatile(), "not defined for volatile fields"); 6054 6055 if (is_static) { 6056 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 6057 fromObj = makecon(tip); 6058 } 6059 6060 // Next code copied from Parse::do_get_xxx(): 6061 6062 // Compute address and memory type. 6063 int offset = field->offset_in_bytes(); 6064 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 6065 6066 return adr; 6067 } 6068 6069 //------------------------------inline_aescrypt_Block----------------------- 6070 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 6071 address stubAddr = NULL; 6072 const char *stubName; 6073 assert(UseAES, "need AES instruction support"); 6074 6075 switch(id) { 6076 case vmIntrinsics::_aescrypt_encryptBlock: 6077 stubAddr = StubRoutines::aescrypt_encryptBlock(); 6078 stubName = "aescrypt_encryptBlock"; 6079 break; 6080 case vmIntrinsics::_aescrypt_decryptBlock: 6081 stubAddr = StubRoutines::aescrypt_decryptBlock(); 6082 stubName = "aescrypt_decryptBlock"; 6083 break; 6084 } 6085 if (stubAddr == NULL) return false; 6086 6087 Node* aescrypt_object = argument(0); 6088 Node* src = argument(1); 6089 Node* src_offset = argument(2); 6090 Node* dest = argument(3); 6091 Node* dest_offset = argument(4); 6092 6093 // (1) src and dest are arrays. 6094 const Type* src_type = src->Value(&_gvn); 6095 const Type* dest_type = dest->Value(&_gvn); 6096 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6097 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6098 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6099 6100 // for the quick and dirty code we will skip all the checks. 6101 // we are just trying to get the call to be generated. 6102 Node* src_start = src; 6103 Node* dest_start = dest; 6104 if (src_offset != NULL || dest_offset != NULL) { 6105 assert(src_offset != NULL && dest_offset != NULL, ""); 6106 src_start = array_element_address(src, src_offset, T_BYTE); 6107 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6108 } 6109 6110 // now need to get the start of its expanded key array 6111 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6112 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6113 if (k_start == NULL) return false; 6114 6115 if (Matcher::pass_original_key_for_aes()) { 6116 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 6117 // compatibility issues between Java key expansion and SPARC crypto instructions 6118 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 6119 if (original_k_start == NULL) return false; 6120 6121 // Call the stub. 6122 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 6123 stubAddr, stubName, TypePtr::BOTTOM, 6124 src_start, dest_start, k_start, original_k_start); 6125 } else { 6126 // Call the stub. 6127 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 6128 stubAddr, stubName, TypePtr::BOTTOM, 6129 src_start, dest_start, k_start); 6130 } 6131 6132 return true; 6133 } 6134 6135 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 6136 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 6137 address stubAddr = NULL; 6138 const char *stubName = NULL; 6139 6140 assert(UseAES, "need AES instruction support"); 6141 6142 switch(id) { 6143 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 6144 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 6145 stubName = "cipherBlockChaining_encryptAESCrypt"; 6146 break; 6147 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 6148 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 6149 stubName = "cipherBlockChaining_decryptAESCrypt"; 6150 break; 6151 } 6152 if (stubAddr == NULL) return false; 6153 6154 Node* cipherBlockChaining_object = argument(0); 6155 Node* src = argument(1); 6156 Node* src_offset = argument(2); 6157 Node* len = argument(3); 6158 Node* dest = argument(4); 6159 Node* dest_offset = argument(5); 6160 6161 // (1) src and dest are arrays. 6162 const Type* src_type = src->Value(&_gvn); 6163 const Type* dest_type = dest->Value(&_gvn); 6164 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6165 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6166 assert (top_src != NULL && top_src->klass() != NULL 6167 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6168 6169 // checks are the responsibility of the caller 6170 Node* src_start = src; 6171 Node* dest_start = dest; 6172 if (src_offset != NULL || dest_offset != NULL) { 6173 assert(src_offset != NULL && dest_offset != NULL, ""); 6174 src_start = array_element_address(src, src_offset, T_BYTE); 6175 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6176 } 6177 6178 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6179 // (because of the predicated logic executed earlier). 6180 // so we cast it here safely. 6181 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6182 6183 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6184 if (embeddedCipherObj == NULL) return false; 6185 6186 // cast it to what we know it will be at runtime 6187 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 6188 assert(tinst != NULL, "CBC obj is null"); 6189 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded"); 6190 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6191 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6192 6193 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6194 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6195 const TypeOopPtr* xtype = aklass->as_instance_type(); 6196 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6197 aescrypt_object = _gvn.transform(aescrypt_object); 6198 6199 // we need to get the start of the aescrypt_object's expanded key array 6200 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6201 if (k_start == NULL) return false; 6202 6203 // similarly, get the start address of the r vector 6204 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false); 6205 if (objRvec == NULL) return false; 6206 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 6207 6208 Node* cbcCrypt; 6209 if (Matcher::pass_original_key_for_aes()) { 6210 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 6211 // compatibility issues between Java key expansion and SPARC crypto instructions 6212 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 6213 if (original_k_start == NULL) return false; 6214 6215 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start 6216 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6217 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 6218 stubAddr, stubName, TypePtr::BOTTOM, 6219 src_start, dest_start, k_start, r_start, len, original_k_start); 6220 } else { 6221 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6222 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6223 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 6224 stubAddr, stubName, TypePtr::BOTTOM, 6225 src_start, dest_start, k_start, r_start, len); 6226 } 6227 6228 // return cipher length (int) 6229 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 6230 set_result(retvalue); 6231 return true; 6232 } 6233 6234 //------------------------------inline_counterMode_AESCrypt----------------------- 6235 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 6236 assert(UseAES, "need AES instruction support"); 6237 if (!UseAESCTRIntrinsics) return false; 6238 6239 address stubAddr = NULL; 6240 const char *stubName = NULL; 6241 if (id == vmIntrinsics::_counterMode_AESCrypt) { 6242 stubAddr = StubRoutines::counterMode_AESCrypt(); 6243 stubName = "counterMode_AESCrypt"; 6244 } 6245 if (stubAddr == NULL) return false; 6246 6247 Node* counterMode_object = argument(0); 6248 Node* src = argument(1); 6249 Node* src_offset = argument(2); 6250 Node* len = argument(3); 6251 Node* dest = argument(4); 6252 Node* dest_offset = argument(5); 6253 6254 // (1) src and dest are arrays. 6255 const Type* src_type = src->Value(&_gvn); 6256 const Type* dest_type = dest->Value(&_gvn); 6257 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6258 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6259 assert(top_src != NULL && top_src->klass() != NULL && 6260 top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6261 6262 // checks are the responsibility of the caller 6263 Node* src_start = src; 6264 Node* dest_start = dest; 6265 if (src_offset != NULL || dest_offset != NULL) { 6266 assert(src_offset != NULL && dest_offset != NULL, ""); 6267 src_start = array_element_address(src, src_offset, T_BYTE); 6268 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6269 } 6270 6271 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6272 // (because of the predicated logic executed earlier). 6273 // so we cast it here safely. 6274 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6275 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6276 if (embeddedCipherObj == NULL) return false; 6277 // cast it to what we know it will be at runtime 6278 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 6279 assert(tinst != NULL, "CTR obj is null"); 6280 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded"); 6281 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6282 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6283 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6284 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6285 const TypeOopPtr* xtype = aklass->as_instance_type(); 6286 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6287 aescrypt_object = _gvn.transform(aescrypt_object); 6288 // we need to get the start of the aescrypt_object's expanded key array 6289 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6290 if (k_start == NULL) return false; 6291 // similarly, get the start address of the r vector 6292 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false); 6293 if (obj_counter == NULL) return false; 6294 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 6295 6296 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false); 6297 if (saved_encCounter == NULL) return false; 6298 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 6299 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 6300 6301 Node* ctrCrypt; 6302 if (Matcher::pass_original_key_for_aes()) { 6303 // no SPARC version for AES/CTR intrinsics now. 6304 return false; 6305 } 6306 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6307 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6308 OptoRuntime::counterMode_aescrypt_Type(), 6309 stubAddr, stubName, TypePtr::BOTTOM, 6310 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 6311 6312 // return cipher length (int) 6313 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 6314 set_result(retvalue); 6315 return true; 6316 } 6317 6318 //------------------------------get_key_start_from_aescrypt_object----------------------- 6319 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 6320 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false); 6321 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6322 if (objAESCryptKey == NULL) return (Node *) NULL; 6323 6324 // now have the array, need to get the start address of the K array 6325 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 6326 return k_start; 6327 } 6328 6329 //------------------------------get_original_key_start_from_aescrypt_object----------------------- 6330 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) { 6331 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false); 6332 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6333 if (objAESCryptKey == NULL) return (Node *) NULL; 6334 6335 // now have the array, need to get the start address of the lastKey array 6336 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE); 6337 return original_k_start; 6338 } 6339 6340 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 6341 // Return node representing slow path of predicate check. 6342 // the pseudo code we want to emulate with this predicate is: 6343 // for encryption: 6344 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6345 // for decryption: 6346 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6347 // note cipher==plain is more conservative than the original java code but that's OK 6348 // 6349 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 6350 // The receiver was checked for NULL already. 6351 Node* objCBC = argument(0); 6352 6353 // Load embeddedCipher field of CipherBlockChaining object. 6354 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6355 6356 // get AESCrypt klass for instanceOf check 6357 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6358 // will have same classloader as CipherBlockChaining object 6359 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 6360 assert(tinst != NULL, "CBCobj is null"); 6361 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded"); 6362 6363 // we want to do an instanceof comparison against the AESCrypt class 6364 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6365 if (!klass_AESCrypt->is_loaded()) { 6366 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6367 Node* ctrl = control(); 6368 set_control(top()); // no regular fast path 6369 return ctrl; 6370 } 6371 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6372 6373 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6374 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6375 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6376 6377 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6378 6379 // for encryption, we are done 6380 if (!decrypting) 6381 return instof_false; // even if it is NULL 6382 6383 // for decryption, we need to add a further check to avoid 6384 // taking the intrinsic path when cipher and plain are the same 6385 // see the original java code for why. 6386 RegionNode* region = new RegionNode(3); 6387 region->init_req(1, instof_false); 6388 Node* src = argument(1); 6389 Node* dest = argument(4); 6390 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6391 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6392 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6393 region->init_req(2, src_dest_conjoint); 6394 6395 record_for_igvn(region); 6396 return _gvn.transform(region); 6397 } 6398 6399 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 6400 // Return node representing slow path of predicate check. 6401 // the pseudo code we want to emulate with this predicate is: 6402 // for encryption: 6403 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6404 // for decryption: 6405 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6406 // note cipher==plain is more conservative than the original java code but that's OK 6407 // 6408 6409 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 6410 // The receiver was checked for NULL already. 6411 Node* objCTR = argument(0); 6412 6413 // Load embeddedCipher field of CipherBlockChaining object. 6414 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6415 6416 // get AESCrypt klass for instanceOf check 6417 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6418 // will have same classloader as CipherBlockChaining object 6419 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 6420 assert(tinst != NULL, "CTRobj is null"); 6421 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded"); 6422 6423 // we want to do an instanceof comparison against the AESCrypt class 6424 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6425 if (!klass_AESCrypt->is_loaded()) { 6426 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6427 Node* ctrl = control(); 6428 set_control(top()); // no regular fast path 6429 return ctrl; 6430 } 6431 6432 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6433 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6434 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6435 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6436 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6437 6438 return instof_false; // even if it is NULL 6439 } 6440 6441 //------------------------------inline_ghash_processBlocks 6442 bool LibraryCallKit::inline_ghash_processBlocks() { 6443 address stubAddr; 6444 const char *stubName; 6445 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 6446 6447 stubAddr = StubRoutines::ghash_processBlocks(); 6448 stubName = "ghash_processBlocks"; 6449 6450 Node* data = argument(0); 6451 Node* offset = argument(1); 6452 Node* len = argument(2); 6453 Node* state = argument(3); 6454 Node* subkeyH = argument(4); 6455 6456 Node* state_start = array_element_address(state, intcon(0), T_LONG); 6457 assert(state_start, "state is NULL"); 6458 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 6459 assert(subkeyH_start, "subkeyH is NULL"); 6460 Node* data_start = array_element_address(data, offset, T_BYTE); 6461 assert(data_start, "data is NULL"); 6462 6463 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 6464 OptoRuntime::ghash_processBlocks_Type(), 6465 stubAddr, stubName, TypePtr::BOTTOM, 6466 state_start, subkeyH_start, data_start, len); 6467 return true; 6468 } 6469 6470 //------------------------------inline_sha_implCompress----------------------- 6471 // 6472 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 6473 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 6474 // 6475 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 6476 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 6477 // 6478 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 6479 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 6480 // 6481 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) { 6482 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 6483 6484 Node* sha_obj = argument(0); 6485 Node* src = argument(1); // type oop 6486 Node* ofs = argument(2); // type int 6487 6488 const Type* src_type = src->Value(&_gvn); 6489 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6490 if (top_src == NULL || top_src->klass() == NULL) { 6491 // failed array check 6492 return false; 6493 } 6494 // Figure out the size and type of the elements we will be copying. 6495 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6496 if (src_elem != T_BYTE) { 6497 return false; 6498 } 6499 // 'src_start' points to src array + offset 6500 Node* src_start = array_element_address(src, ofs, src_elem); 6501 Node* state = NULL; 6502 address stubAddr; 6503 const char *stubName; 6504 6505 switch(id) { 6506 case vmIntrinsics::_sha_implCompress: 6507 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 6508 state = get_state_from_sha_object(sha_obj); 6509 stubAddr = StubRoutines::sha1_implCompress(); 6510 stubName = "sha1_implCompress"; 6511 break; 6512 case vmIntrinsics::_sha2_implCompress: 6513 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 6514 state = get_state_from_sha_object(sha_obj); 6515 stubAddr = StubRoutines::sha256_implCompress(); 6516 stubName = "sha256_implCompress"; 6517 break; 6518 case vmIntrinsics::_sha5_implCompress: 6519 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 6520 state = get_state_from_sha5_object(sha_obj); 6521 stubAddr = StubRoutines::sha512_implCompress(); 6522 stubName = "sha512_implCompress"; 6523 break; 6524 default: 6525 fatal_unexpected_iid(id); 6526 return false; 6527 } 6528 if (state == NULL) return false; 6529 6530 // Call the stub. 6531 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(), 6532 stubAddr, stubName, TypePtr::BOTTOM, 6533 src_start, state); 6534 6535 return true; 6536 } 6537 6538 //------------------------------inline_digestBase_implCompressMB----------------------- 6539 // 6540 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array. 6541 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 6542 // 6543 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 6544 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 6545 "need SHA1/SHA256/SHA512 instruction support"); 6546 assert((uint)predicate < 3, "sanity"); 6547 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 6548 6549 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already. 6550 Node* src = argument(1); // byte[] array 6551 Node* ofs = argument(2); // type int 6552 Node* limit = argument(3); // type int 6553 6554 const Type* src_type = src->Value(&_gvn); 6555 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6556 if (top_src == NULL || top_src->klass() == NULL) { 6557 // failed array check 6558 return false; 6559 } 6560 // Figure out the size and type of the elements we will be copying. 6561 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6562 if (src_elem != T_BYTE) { 6563 return false; 6564 } 6565 // 'src_start' points to src array + offset 6566 Node* src_start = array_element_address(src, ofs, src_elem); 6567 6568 const char* klass_SHA_name = NULL; 6569 const char* stub_name = NULL; 6570 address stub_addr = NULL; 6571 bool long_state = false; 6572 6573 switch (predicate) { 6574 case 0: 6575 if (UseSHA1Intrinsics) { 6576 klass_SHA_name = "sun/security/provider/SHA"; 6577 stub_name = "sha1_implCompressMB"; 6578 stub_addr = StubRoutines::sha1_implCompressMB(); 6579 } 6580 break; 6581 case 1: 6582 if (UseSHA256Intrinsics) { 6583 klass_SHA_name = "sun/security/provider/SHA2"; 6584 stub_name = "sha256_implCompressMB"; 6585 stub_addr = StubRoutines::sha256_implCompressMB(); 6586 } 6587 break; 6588 case 2: 6589 if (UseSHA512Intrinsics) { 6590 klass_SHA_name = "sun/security/provider/SHA5"; 6591 stub_name = "sha512_implCompressMB"; 6592 stub_addr = StubRoutines::sha512_implCompressMB(); 6593 long_state = true; 6594 } 6595 break; 6596 default: 6597 fatal("unknown SHA intrinsic predicate: %d", predicate); 6598 } 6599 if (klass_SHA_name != NULL) { 6600 // get DigestBase klass to lookup for SHA klass 6601 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 6602 assert(tinst != NULL, "digestBase_obj is not instance???"); 6603 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 6604 6605 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 6606 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded"); 6607 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 6608 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit); 6609 } 6610 return false; 6611 } 6612 //------------------------------inline_sha_implCompressMB----------------------- 6613 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA, 6614 bool long_state, address stubAddr, const char *stubName, 6615 Node* src_start, Node* ofs, Node* limit) { 6616 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA); 6617 const TypeOopPtr* xtype = aklass->as_instance_type(); 6618 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 6619 sha_obj = _gvn.transform(sha_obj); 6620 6621 Node* state; 6622 if (long_state) { 6623 state = get_state_from_sha5_object(sha_obj); 6624 } else { 6625 state = get_state_from_sha_object(sha_obj); 6626 } 6627 if (state == NULL) return false; 6628 6629 // Call the stub. 6630 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6631 OptoRuntime::digestBase_implCompressMB_Type(), 6632 stubAddr, stubName, TypePtr::BOTTOM, 6633 src_start, state, ofs, limit); 6634 // return ofs (int) 6635 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6636 set_result(result); 6637 6638 return true; 6639 } 6640 6641 //------------------------------get_state_from_sha_object----------------------- 6642 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) { 6643 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false); 6644 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2"); 6645 if (sha_state == NULL) return (Node *) NULL; 6646 6647 // now have the array, need to get the start address of the state array 6648 Node* state = array_element_address(sha_state, intcon(0), T_INT); 6649 return state; 6650 } 6651 6652 //------------------------------get_state_from_sha5_object----------------------- 6653 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) { 6654 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false); 6655 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5"); 6656 if (sha_state == NULL) return (Node *) NULL; 6657 6658 // now have the array, need to get the start address of the state array 6659 Node* state = array_element_address(sha_state, intcon(0), T_LONG); 6660 return state; 6661 } 6662 6663 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 6664 // Return node representing slow path of predicate check. 6665 // the pseudo code we want to emulate with this predicate is: 6666 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath 6667 // 6668 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 6669 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 6670 "need SHA1/SHA256/SHA512 instruction support"); 6671 assert((uint)predicate < 3, "sanity"); 6672 6673 // The receiver was checked for NULL already. 6674 Node* digestBaseObj = argument(0); 6675 6676 // get DigestBase klass for instanceOf check 6677 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 6678 assert(tinst != NULL, "digestBaseObj is null"); 6679 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 6680 6681 const char* klass_SHA_name = NULL; 6682 switch (predicate) { 6683 case 0: 6684 if (UseSHA1Intrinsics) { 6685 // we want to do an instanceof comparison against the SHA class 6686 klass_SHA_name = "sun/security/provider/SHA"; 6687 } 6688 break; 6689 case 1: 6690 if (UseSHA256Intrinsics) { 6691 // we want to do an instanceof comparison against the SHA2 class 6692 klass_SHA_name = "sun/security/provider/SHA2"; 6693 } 6694 break; 6695 case 2: 6696 if (UseSHA512Intrinsics) { 6697 // we want to do an instanceof comparison against the SHA5 class 6698 klass_SHA_name = "sun/security/provider/SHA5"; 6699 } 6700 break; 6701 default: 6702 fatal("unknown SHA intrinsic predicate: %d", predicate); 6703 } 6704 6705 ciKlass* klass_SHA = NULL; 6706 if (klass_SHA_name != NULL) { 6707 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 6708 } 6709 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) { 6710 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 6711 Node* ctrl = control(); 6712 set_control(top()); // no intrinsic path 6713 return ctrl; 6714 } 6715 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 6716 6717 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA))); 6718 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1))); 6719 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6720 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6721 6722 return instof_false; // even if it is NULL 6723 } 6724 6725 bool LibraryCallKit::inline_profileBoolean() { 6726 Node* counts = argument(1); 6727 const TypeAryPtr* ary = NULL; 6728 ciArray* aobj = NULL; 6729 if (counts->is_Con() 6730 && (ary = counts->bottom_type()->isa_aryptr()) != NULL 6731 && (aobj = ary->const_oop()->as_array()) != NULL 6732 && (aobj->length() == 2)) { 6733 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 6734 jint false_cnt = aobj->element_value(0).as_int(); 6735 jint true_cnt = aobj->element_value(1).as_int(); 6736 6737 if (C->log() != NULL) { 6738 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 6739 false_cnt, true_cnt); 6740 } 6741 6742 if (false_cnt + true_cnt == 0) { 6743 // According to profile, never executed. 6744 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 6745 Deoptimization::Action_reinterpret); 6746 return true; 6747 } 6748 6749 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 6750 // is a number of each value occurrences. 6751 Node* result = argument(0); 6752 if (false_cnt == 0 || true_cnt == 0) { 6753 // According to profile, one value has been never seen. 6754 int expected_val = (false_cnt == 0) ? 1 : 0; 6755 6756 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 6757 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 6758 6759 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 6760 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 6761 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 6762 6763 { // Slow path: uncommon trap for never seen value and then reexecute 6764 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 6765 // the value has been seen at least once. 6766 PreserveJVMState pjvms(this); 6767 PreserveReexecuteState preexecs(this); 6768 jvms()->set_should_reexecute(true); 6769 6770 set_control(slow_path); 6771 set_i_o(i_o()); 6772 6773 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 6774 Deoptimization::Action_reinterpret); 6775 } 6776 // The guard for never seen value enables sharpening of the result and 6777 // returning a constant. It allows to eliminate branches on the same value 6778 // later on. 6779 set_control(fast_path); 6780 result = intcon(expected_val); 6781 } 6782 // Stop profiling. 6783 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 6784 // By replacing method body with profile data (represented as ProfileBooleanNode 6785 // on IR level) we effectively disable profiling. 6786 // It enables full speed execution once optimized code is generated. 6787 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 6788 C->record_for_igvn(profile); 6789 set_result(profile); 6790 return true; 6791 } else { 6792 // Continue profiling. 6793 // Profile data isn't available at the moment. So, execute method's bytecode version. 6794 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 6795 // is compiled and counters aren't available since corresponding MethodHandle 6796 // isn't a compile-time constant. 6797 return false; 6798 } 6799 } 6800 6801 bool LibraryCallKit::inline_isCompileConstant() { 6802 Node* n = argument(0); 6803 set_result(n->is_Con() ? intcon(1) : intcon(0)); 6804 return true; 6805 } 6806 6807 bool LibraryCallKit::inline_deoptimize() { 6808 assert(WhiteBoxAPI, ""); 6809 PreserveReexecuteState preexecs(this); 6810 jvms()->set_should_reexecute(false); 6811 uncommon_trap(Deoptimization::Reason_intrinsic, 6812 Deoptimization::Action_none); 6813 return true; 6814 }