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