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