1 /* 2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/macroAssembler.hpp" 27 #include "ci/ciUtilities.inline.hpp" 28 #include "classfile/systemDictionary.hpp" 29 #include "classfile/vmSymbols.hpp" 30 #include "compiler/compileBroker.hpp" 31 #include "compiler/compileLog.hpp" 32 #include "gc/shared/barrierSet.hpp" 33 #include "jfr/support/jfrIntrinsics.hpp" 34 #include "memory/resourceArea.hpp" 35 #include "oops/objArrayKlass.hpp" 36 #include "opto/addnode.hpp" 37 #include "opto/arraycopynode.hpp" 38 #include "opto/c2compiler.hpp" 39 #include "opto/callGenerator.hpp" 40 #include "opto/castnode.hpp" 41 #include "opto/cfgnode.hpp" 42 #include "opto/convertnode.hpp" 43 #include "opto/countbitsnode.hpp" 44 #include "opto/intrinsicnode.hpp" 45 #include "opto/idealKit.hpp" 46 #include "opto/mathexactnode.hpp" 47 #include "opto/movenode.hpp" 48 #include "opto/mulnode.hpp" 49 #include "opto/narrowptrnode.hpp" 50 #include "opto/opaquenode.hpp" 51 #include "opto/parse.hpp" 52 #include "opto/runtime.hpp" 53 #include "opto/rootnode.hpp" 54 #include "opto/subnode.hpp" 55 #include "opto/valuetypenode.hpp" 56 #include "prims/nativeLookup.hpp" 57 #include "prims/unsafe.hpp" 58 #include "runtime/objectMonitor.hpp" 59 #include "runtime/sharedRuntime.hpp" 60 #include "utilities/macros.hpp" 61 62 63 class LibraryIntrinsic : public InlineCallGenerator { 64 // Extend the set of intrinsics known to the runtime: 65 public: 66 private: 67 bool _is_virtual; 68 bool _does_virtual_dispatch; 69 int8_t _predicates_count; // Intrinsic is predicated by several conditions 70 int8_t _last_predicate; // Last generated predicate 71 vmIntrinsics::ID _intrinsic_id; 72 73 public: 74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id) 75 : InlineCallGenerator(m), 76 _is_virtual(is_virtual), 77 _does_virtual_dispatch(does_virtual_dispatch), 78 _predicates_count((int8_t)predicates_count), 79 _last_predicate((int8_t)-1), 80 _intrinsic_id(id) 81 { 82 } 83 virtual bool is_intrinsic() const { return true; } 84 virtual bool is_virtual() const { return _is_virtual; } 85 virtual bool is_predicated() const { return _predicates_count > 0; } 86 virtual int predicates_count() const { return _predicates_count; } 87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; } 88 virtual JVMState* generate(JVMState* jvms); 89 virtual Node* generate_predicate(JVMState* jvms, int predicate); 90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; } 91 }; 92 93 94 // Local helper class for LibraryIntrinsic: 95 class LibraryCallKit : public GraphKit { 96 private: 97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called 98 Node* _result; // the result node, if any 99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted 100 101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type); 102 103 public: 104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic) 105 : GraphKit(jvms), 106 _intrinsic(intrinsic), 107 _result(NULL) 108 { 109 // Check if this is a root compile. In that case we don't have a caller. 110 if (!jvms->has_method()) { 111 _reexecute_sp = sp(); 112 } else { 113 // Find out how many arguments the interpreter needs when deoptimizing 114 // and save the stack pointer value so it can used by uncommon_trap. 115 // We find the argument count by looking at the declared signature. 116 bool ignored_will_link; 117 ciSignature* declared_signature = NULL; 118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature); 119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci())); 120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack 121 } 122 } 123 124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; } 125 126 ciMethod* caller() const { return jvms()->method(); } 127 int bci() const { return jvms()->bci(); } 128 LibraryIntrinsic* intrinsic() const { return _intrinsic; } 129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); } 130 ciMethod* callee() const { return _intrinsic->method(); } 131 132 bool try_to_inline(int predicate); 133 Node* try_to_predicate(int predicate); 134 135 void push_result() { 136 // Push the result onto the stack. 137 if (!stopped() && result() != NULL) { 138 BasicType bt = result()->bottom_type()->basic_type(); 139 push_node(bt, result()); 140 } 141 } 142 143 private: 144 void fatal_unexpected_iid(vmIntrinsics::ID iid) { 145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid)); 146 } 147 148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; } 149 void set_result(RegionNode* region, PhiNode* value); 150 Node* result() { return _result; } 151 152 virtual int reexecute_sp() { return _reexecute_sp; } 153 154 // Helper functions to inline natives 155 Node* generate_guard(Node* test, RegionNode* region, float true_prob); 156 Node* generate_slow_guard(Node* test, RegionNode* region); 157 Node* generate_fair_guard(Node* test, RegionNode* region); 158 Node* generate_negative_guard(Node* index, RegionNode* region, 159 // resulting CastII of index: 160 Node* *pos_index = NULL); 161 Node* generate_limit_guard(Node* offset, Node* subseq_length, 162 Node* array_length, 163 RegionNode* region); 164 void generate_string_range_check(Node* array, Node* offset, 165 Node* length, bool char_count); 166 Node* generate_current_thread(Node* &tls_output); 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_value_guard(Node* kls, RegionNode* region); 189 190 enum ArrayKind { 191 AnyArray, 192 NonArray, 193 ObjectArray, 194 NonObjectArray, 195 TypeArray, 196 ValueArray 197 }; 198 199 Node* generate_array_guard(Node* kls, RegionNode* region) { 200 return generate_array_guard_common(kls, region, AnyArray); 201 } 202 Node* generate_non_array_guard(Node* kls, RegionNode* region) { 203 return generate_array_guard_common(kls, region, NonArray); 204 } 205 Node* generate_objArray_guard(Node* kls, RegionNode* region) { 206 return generate_array_guard_common(kls, region, ObjectArray); 207 } 208 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) { 209 return generate_array_guard_common(kls, region, NonObjectArray); 210 } 211 Node* generate_typeArray_guard(Node* kls, RegionNode* region) { 212 return generate_array_guard_common(kls, region, TypeArray); 213 } 214 Node* generate_valueArray_guard(Node* kls, RegionNode* region) { 215 assert(ValueArrayFlatten, "can never be flattened"); 216 return generate_array_guard_common(kls, region, ValueArray); 217 } 218 Node* generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind); 219 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region); 220 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id, 221 bool is_virtual = false, bool is_static = false); 222 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) { 223 return generate_method_call(method_id, false, true); 224 } 225 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) { 226 return generate_method_call(method_id, true, false); 227 } 228 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 229 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 230 231 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae); 232 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae); 233 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae); 234 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae); 235 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 236 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae); 237 bool inline_string_indexOfChar(); 238 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae); 239 bool inline_string_toBytesU(); 240 bool inline_string_getCharsU(); 241 bool inline_string_copy(bool compress); 242 bool inline_string_char_access(bool is_store); 243 Node* round_double_node(Node* n); 244 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName); 245 bool inline_math_native(vmIntrinsics::ID id); 246 bool inline_math(vmIntrinsics::ID id); 247 bool inline_double_math(vmIntrinsics::ID id); 248 template <typename OverflowOp> 249 bool inline_math_overflow(Node* arg1, Node* arg2); 250 void inline_math_mathExact(Node* math, Node* test); 251 bool inline_math_addExactI(bool is_increment); 252 bool inline_math_addExactL(bool is_increment); 253 bool inline_math_multiplyExactI(); 254 bool inline_math_multiplyExactL(); 255 bool inline_math_multiplyHigh(); 256 bool inline_math_negateExactI(); 257 bool inline_math_negateExactL(); 258 bool inline_math_subtractExactI(bool is_decrement); 259 bool inline_math_subtractExactL(bool is_decrement); 260 bool inline_min_max(vmIntrinsics::ID id); 261 bool inline_notify(vmIntrinsics::ID id); 262 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y); 263 // This returns Type::AnyPtr, RawPtr, or OopPtr. 264 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type); 265 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false); 266 267 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind; 268 DecoratorSet mo_decorator_for_access_kind(AccessKind kind); 269 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned); 270 static bool klass_needs_init_guard(Node* kls); 271 bool inline_unsafe_allocate(); 272 bool inline_unsafe_newArray(bool uninitialized); 273 bool inline_unsafe_writeback0(); 274 bool inline_unsafe_writebackSync0(bool is_pre); 275 bool inline_unsafe_copyMemory(); 276 bool inline_unsafe_make_private_buffer(); 277 bool inline_unsafe_finish_private_buffer(); 278 bool inline_native_currentThread(); 279 280 bool inline_native_time_funcs(address method, const char* funcName); 281 #ifdef JFR_HAVE_INTRINSICS 282 bool inline_native_classID(); 283 bool inline_native_getEventWriter(); 284 #endif 285 bool inline_native_isInterrupted(); 286 bool inline_native_Class_query(vmIntrinsics::ID id); 287 bool inline_value_Class_conversion(vmIntrinsics::ID id); 288 bool inline_native_subtype_check(); 289 bool inline_native_getLength(); 290 bool inline_array_copyOf(bool is_copyOfRange); 291 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae); 292 bool inline_preconditions_checkIndex(); 293 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array); 294 bool inline_native_clone(bool is_virtual); 295 bool inline_native_Reflection_getCallerClass(); 296 // Helper function for inlining native object hash method 297 bool inline_native_hashcode(bool is_virtual, bool is_static); 298 bool inline_native_getClass(); 299 300 // Helper functions for inlining arraycopy 301 bool inline_arraycopy(); 302 AllocateArrayNode* tightly_coupled_allocation(Node* ptr, 303 RegionNode* slow_region); 304 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp); 305 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp, 306 uint new_idx); 307 308 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind; 309 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind); 310 bool inline_unsafe_fence(vmIntrinsics::ID id); 311 bool inline_onspinwait(); 312 bool inline_fp_conversions(vmIntrinsics::ID id); 313 bool inline_number_methods(vmIntrinsics::ID id); 314 bool inline_reference_get(); 315 bool inline_Class_cast(); 316 bool inline_aescrypt_Block(vmIntrinsics::ID id); 317 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id); 318 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id); 319 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id); 320 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting); 321 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting); 322 Node* inline_counterMode_AESCrypt_predicate(); 323 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object); 324 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object); 325 bool inline_ghash_processBlocks(); 326 bool inline_base64_encodeBlock(); 327 bool inline_sha_implCompress(vmIntrinsics::ID id); 328 bool inline_digestBase_implCompressMB(int predicate); 329 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA, 330 bool long_state, address stubAddr, const char *stubName, 331 Node* src_start, Node* ofs, Node* limit); 332 Node* get_state_from_sha_object(Node *sha_object); 333 Node* get_state_from_sha5_object(Node *sha_object); 334 Node* inline_digestBase_implCompressMB_predicate(int predicate); 335 bool inline_encodeISOArray(); 336 bool inline_updateCRC32(); 337 bool inline_updateBytesCRC32(); 338 bool inline_updateByteBufferCRC32(); 339 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class); 340 bool inline_updateBytesCRC32C(); 341 bool inline_updateDirectByteBufferCRC32C(); 342 bool inline_updateBytesAdler32(); 343 bool inline_updateByteBufferAdler32(); 344 bool inline_multiplyToLen(); 345 bool inline_hasNegatives(); 346 bool inline_squareToLen(); 347 bool inline_mulAdd(); 348 bool inline_montgomeryMultiply(); 349 bool inline_montgomerySquare(); 350 bool inline_vectorizedMismatch(); 351 bool inline_fma(vmIntrinsics::ID id); 352 bool inline_character_compare(vmIntrinsics::ID id); 353 bool inline_fp_min_max(vmIntrinsics::ID id); 354 355 bool inline_profileBoolean(); 356 bool inline_isCompileConstant(); 357 void clear_upper_avx() { 358 #ifdef X86 359 if (UseAVX >= 2) { 360 C->set_clear_upper_avx(true); 361 } 362 #endif 363 } 364 }; 365 366 //---------------------------make_vm_intrinsic---------------------------- 367 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) { 368 vmIntrinsics::ID id = m->intrinsic_id(); 369 assert(id != vmIntrinsics::_none, "must be a VM intrinsic"); 370 371 if (!m->is_loaded()) { 372 // Do not attempt to inline unloaded methods. 373 return NULL; 374 } 375 376 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization); 377 bool is_available = false; 378 379 { 380 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag 381 // the compiler must transition to '_thread_in_vm' state because both 382 // methods access VM-internal data. 383 VM_ENTRY_MARK; 384 methodHandle mh(THREAD, m->get_Method()); 385 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) && 386 !C->directive()->is_intrinsic_disabled(mh) && 387 !vmIntrinsics::is_disabled_by_flags(mh); 388 389 } 390 391 if (is_available) { 392 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility"); 393 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?"); 394 return new LibraryIntrinsic(m, is_virtual, 395 vmIntrinsics::predicates_needed(id), 396 vmIntrinsics::does_virtual_dispatch(id), 397 (vmIntrinsics::ID) id); 398 } else { 399 return NULL; 400 } 401 } 402 403 //----------------------register_library_intrinsics----------------------- 404 // Initialize this file's data structures, for each Compile instance. 405 void Compile::register_library_intrinsics() { 406 // Nothing to do here. 407 } 408 409 JVMState* LibraryIntrinsic::generate(JVMState* jvms) { 410 LibraryCallKit kit(jvms, this); 411 Compile* C = kit.C; 412 int nodes = C->unique(); 413 #ifndef PRODUCT 414 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 415 char buf[1000]; 416 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 417 tty->print_cr("Intrinsic %s", str); 418 } 419 #endif 420 ciMethod* callee = kit.callee(); 421 const int bci = kit.bci(); 422 423 // Try to inline the intrinsic. 424 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) && 425 kit.try_to_inline(_last_predicate)) { 426 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)" 427 : "(intrinsic)"; 428 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 429 if (C->print_intrinsics() || C->print_inlining()) { 430 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 431 } 432 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 433 if (C->log()) { 434 C->log()->elem("intrinsic id='%s'%s nodes='%d'", 435 vmIntrinsics::name_at(intrinsic_id()), 436 (is_virtual() ? " virtual='1'" : ""), 437 C->unique() - nodes); 438 } 439 // Push the result from the inlined method onto the stack. 440 kit.push_result(); 441 C->print_inlining_update(this); 442 return kit.transfer_exceptions_into_jvms(); 443 } 444 445 // The intrinsic bailed out 446 if (jvms->has_method()) { 447 // Not a root compile. 448 const char* msg; 449 if (callee->intrinsic_candidate()) { 450 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)"; 451 } else { 452 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated" 453 : "failed to inline (intrinsic), method not annotated"; 454 } 455 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg); 456 if (C->print_intrinsics() || C->print_inlining()) { 457 C->print_inlining(callee, jvms->depth() - 1, bci, msg); 458 } 459 } else { 460 // Root compile 461 ResourceMark rm; 462 stringStream msg_stream; 463 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 464 vmIntrinsics::name_at(intrinsic_id()), 465 is_virtual() ? " (virtual)" : "", bci); 466 const char *msg = msg_stream.as_string(); 467 log_debug(jit, inlining)("%s", msg); 468 if (C->print_intrinsics() || C->print_inlining()) { 469 tty->print("%s", msg); 470 } 471 } 472 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 473 C->print_inlining_update(this); 474 return NULL; 475 } 476 477 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) { 478 LibraryCallKit kit(jvms, this); 479 Compile* C = kit.C; 480 int nodes = C->unique(); 481 _last_predicate = predicate; 482 #ifndef PRODUCT 483 assert(is_predicated() && predicate < predicates_count(), "sanity"); 484 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 485 char buf[1000]; 486 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 487 tty->print_cr("Predicate for intrinsic %s", str); 488 } 489 #endif 490 ciMethod* callee = kit.callee(); 491 const int bci = kit.bci(); 492 493 Node* slow_ctl = kit.try_to_predicate(predicate); 494 if (!kit.failing()) { 495 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)" 496 : "(intrinsic, predicate)"; 497 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 498 if (C->print_intrinsics() || C->print_inlining()) { 499 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 500 } 501 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 502 if (C->log()) { 503 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'", 504 vmIntrinsics::name_at(intrinsic_id()), 505 (is_virtual() ? " virtual='1'" : ""), 506 C->unique() - nodes); 507 } 508 return slow_ctl; // Could be NULL if the check folds. 509 } 510 511 // The intrinsic bailed out 512 if (jvms->has_method()) { 513 // Not a root compile. 514 const char* msg = "failed to generate predicate for intrinsic"; 515 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg); 516 if (C->print_intrinsics() || C->print_inlining()) { 517 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg); 518 } 519 } else { 520 // Root compile 521 ResourceMark rm; 522 stringStream msg_stream; 523 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 524 vmIntrinsics::name_at(intrinsic_id()), 525 is_virtual() ? " (virtual)" : "", bci); 526 const char *msg = msg_stream.as_string(); 527 log_debug(jit, inlining)("%s", msg); 528 if (C->print_intrinsics() || C->print_inlining()) { 529 C->print_inlining_stream()->print("%s", msg); 530 } 531 } 532 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 533 return NULL; 534 } 535 536 bool LibraryCallKit::try_to_inline(int predicate) { 537 // Handle symbolic names for otherwise undistinguished boolean switches: 538 const bool is_store = true; 539 const bool is_compress = true; 540 const bool is_static = true; 541 const bool is_volatile = true; 542 543 if (!jvms()->has_method()) { 544 // Root JVMState has a null method. 545 assert(map()->memory()->Opcode() == Op_Parm, ""); 546 // Insert the memory aliasing node 547 set_all_memory(reset_memory()); 548 } 549 assert(merged_memory(), ""); 550 551 552 switch (intrinsic_id()) { 553 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static); 554 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static); 555 case vmIntrinsics::_getClass: return inline_native_getClass(); 556 557 case vmIntrinsics::_dsin: 558 case vmIntrinsics::_dcos: 559 case vmIntrinsics::_dtan: 560 case vmIntrinsics::_dabs: 561 case vmIntrinsics::_fabs: 562 case vmIntrinsics::_iabs: 563 case vmIntrinsics::_labs: 564 case vmIntrinsics::_datan2: 565 case vmIntrinsics::_dsqrt: 566 case vmIntrinsics::_dexp: 567 case vmIntrinsics::_dlog: 568 case vmIntrinsics::_dlog10: 569 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id()); 570 571 case vmIntrinsics::_min: 572 case vmIntrinsics::_max: return inline_min_max(intrinsic_id()); 573 574 case vmIntrinsics::_notify: 575 case vmIntrinsics::_notifyAll: 576 return inline_notify(intrinsic_id()); 577 578 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 579 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 580 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 581 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 582 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 583 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 584 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 585 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 586 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh(); 587 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 588 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 589 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 590 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 591 592 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 593 594 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 595 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 596 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 597 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 598 599 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 600 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 601 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 602 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 603 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 604 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 605 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(); 606 607 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 608 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU); 609 610 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 611 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 612 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 613 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 614 615 case vmIntrinsics::_compressStringC: 616 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 617 case vmIntrinsics::_inflateStringC: 618 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 619 620 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer(); 621 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer(); 622 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false); 623 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false); 624 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false); 625 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false); 626 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false); 627 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false); 628 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false); 629 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false); 630 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false); 631 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_VALUETYPE,Relaxed, false); 632 633 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false); 634 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false); 635 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false); 636 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false); 637 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false); 638 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false); 639 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false); 640 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false); 641 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false); 642 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_VALUETYPE,Relaxed, false); 643 644 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false); 645 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false); 646 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false); 647 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false); 648 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false); 649 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false); 650 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false); 651 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false); 652 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false); 653 654 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false); 655 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false); 656 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false); 657 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false); 658 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false); 659 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false); 660 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false); 661 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false); 662 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false); 663 664 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true); 665 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true); 666 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true); 667 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true); 668 669 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true); 670 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true); 671 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true); 672 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true); 673 674 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false); 675 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false); 676 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false); 677 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false); 678 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false); 679 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false); 680 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false); 681 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false); 682 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false); 683 684 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false); 685 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false); 686 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false); 687 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false); 688 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false); 689 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false); 690 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false); 691 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false); 692 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false); 693 694 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false); 695 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false); 696 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false); 697 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false); 698 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false); 699 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false); 700 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false); 701 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false); 702 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false); 703 704 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false); 705 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false); 706 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false); 707 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false); 708 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false); 709 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false); 710 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false); 711 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false); 712 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false); 713 714 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 715 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile); 716 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile); 717 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 718 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 719 720 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 721 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 722 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 723 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile); 724 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed); 725 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire); 726 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release); 727 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile); 728 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed); 729 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire); 730 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release); 731 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile); 732 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 733 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 734 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 735 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile); 736 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 737 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 738 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 739 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile); 740 741 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 742 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 743 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 744 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile); 745 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire); 746 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release); 747 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile); 748 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire); 749 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release); 750 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 751 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 752 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 753 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 754 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 755 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 756 757 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile); 758 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile); 759 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 760 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 761 762 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile); 763 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile); 764 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 765 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 766 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 767 768 case vmIntrinsics::_loadFence: 769 case vmIntrinsics::_storeFence: 770 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 771 772 case vmIntrinsics::_onSpinWait: return inline_onspinwait(); 773 774 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 775 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted(); 776 777 #ifdef JFR_HAVE_INTRINSICS 778 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime"); 779 case vmIntrinsics::_getClassId: return inline_native_classID(); 780 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter(); 781 #endif 782 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 783 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 784 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0(); 785 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true); 786 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false); 787 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 788 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 789 case vmIntrinsics::_getLength: return inline_native_getLength(); 790 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 791 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 792 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 793 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 794 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(); 795 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 796 797 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true); 798 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false); 799 800 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 801 802 case vmIntrinsics::_isInstance: 803 case vmIntrinsics::_getModifiers: 804 case vmIntrinsics::_isInterface: 805 case vmIntrinsics::_isArray: 806 case vmIntrinsics::_isPrimitive: 807 case vmIntrinsics::_getSuperclass: 808 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id()); 809 810 case vmIntrinsics::_asPrimaryType: 811 case vmIntrinsics::_asIndirectType: return inline_value_Class_conversion(intrinsic_id()); 812 813 case vmIntrinsics::_floatToRawIntBits: 814 case vmIntrinsics::_floatToIntBits: 815 case vmIntrinsics::_intBitsToFloat: 816 case vmIntrinsics::_doubleToRawLongBits: 817 case vmIntrinsics::_doubleToLongBits: 818 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id()); 819 820 case vmIntrinsics::_numberOfLeadingZeros_i: 821 case vmIntrinsics::_numberOfLeadingZeros_l: 822 case vmIntrinsics::_numberOfTrailingZeros_i: 823 case vmIntrinsics::_numberOfTrailingZeros_l: 824 case vmIntrinsics::_bitCount_i: 825 case vmIntrinsics::_bitCount_l: 826 case vmIntrinsics::_reverseBytes_i: 827 case vmIntrinsics::_reverseBytes_l: 828 case vmIntrinsics::_reverseBytes_s: 829 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id()); 830 831 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass(); 832 833 case vmIntrinsics::_Reference_get: return inline_reference_get(); 834 835 case vmIntrinsics::_Class_cast: return inline_Class_cast(); 836 837 case vmIntrinsics::_aescrypt_encryptBlock: 838 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id()); 839 840 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 841 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 842 return inline_cipherBlockChaining_AESCrypt(intrinsic_id()); 843 844 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 845 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 846 return inline_electronicCodeBook_AESCrypt(intrinsic_id()); 847 848 case vmIntrinsics::_counterMode_AESCrypt: 849 return inline_counterMode_AESCrypt(intrinsic_id()); 850 851 case vmIntrinsics::_sha_implCompress: 852 case vmIntrinsics::_sha2_implCompress: 853 case vmIntrinsics::_sha5_implCompress: 854 return inline_sha_implCompress(intrinsic_id()); 855 856 case vmIntrinsics::_digestBase_implCompressMB: 857 return inline_digestBase_implCompressMB(predicate); 858 859 case vmIntrinsics::_multiplyToLen: 860 return inline_multiplyToLen(); 861 862 case vmIntrinsics::_squareToLen: 863 return inline_squareToLen(); 864 865 case vmIntrinsics::_mulAdd: 866 return inline_mulAdd(); 867 868 case vmIntrinsics::_montgomeryMultiply: 869 return inline_montgomeryMultiply(); 870 case vmIntrinsics::_montgomerySquare: 871 return inline_montgomerySquare(); 872 873 case vmIntrinsics::_vectorizedMismatch: 874 return inline_vectorizedMismatch(); 875 876 case vmIntrinsics::_ghash_processBlocks: 877 return inline_ghash_processBlocks(); 878 case vmIntrinsics::_base64_encodeBlock: 879 return inline_base64_encodeBlock(); 880 881 case vmIntrinsics::_encodeISOArray: 882 case vmIntrinsics::_encodeByteISOArray: 883 return inline_encodeISOArray(); 884 885 case vmIntrinsics::_updateCRC32: 886 return inline_updateCRC32(); 887 case vmIntrinsics::_updateBytesCRC32: 888 return inline_updateBytesCRC32(); 889 case vmIntrinsics::_updateByteBufferCRC32: 890 return inline_updateByteBufferCRC32(); 891 892 case vmIntrinsics::_updateBytesCRC32C: 893 return inline_updateBytesCRC32C(); 894 case vmIntrinsics::_updateDirectByteBufferCRC32C: 895 return inline_updateDirectByteBufferCRC32C(); 896 897 case vmIntrinsics::_updateBytesAdler32: 898 return inline_updateBytesAdler32(); 899 case vmIntrinsics::_updateByteBufferAdler32: 900 return inline_updateByteBufferAdler32(); 901 902 case vmIntrinsics::_profileBoolean: 903 return inline_profileBoolean(); 904 case vmIntrinsics::_isCompileConstant: 905 return inline_isCompileConstant(); 906 907 case vmIntrinsics::_hasNegatives: 908 return inline_hasNegatives(); 909 910 case vmIntrinsics::_fmaD: 911 case vmIntrinsics::_fmaF: 912 return inline_fma(intrinsic_id()); 913 914 case vmIntrinsics::_isDigit: 915 case vmIntrinsics::_isLowerCase: 916 case vmIntrinsics::_isUpperCase: 917 case vmIntrinsics::_isWhitespace: 918 return inline_character_compare(intrinsic_id()); 919 920 case vmIntrinsics::_maxF: 921 case vmIntrinsics::_minF: 922 case vmIntrinsics::_maxD: 923 case vmIntrinsics::_minD: 924 return inline_fp_min_max(intrinsic_id()); 925 926 default: 927 // If you get here, it may be that someone has added a new intrinsic 928 // to the list in vmSymbols.hpp without implementing it here. 929 #ifndef PRODUCT 930 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 931 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 932 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 933 } 934 #endif 935 return false; 936 } 937 } 938 939 Node* LibraryCallKit::try_to_predicate(int predicate) { 940 if (!jvms()->has_method()) { 941 // Root JVMState has a null method. 942 assert(map()->memory()->Opcode() == Op_Parm, ""); 943 // Insert the memory aliasing node 944 set_all_memory(reset_memory()); 945 } 946 assert(merged_memory(), ""); 947 948 switch (intrinsic_id()) { 949 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 950 return inline_cipherBlockChaining_AESCrypt_predicate(false); 951 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 952 return inline_cipherBlockChaining_AESCrypt_predicate(true); 953 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 954 return inline_electronicCodeBook_AESCrypt_predicate(false); 955 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 956 return inline_electronicCodeBook_AESCrypt_predicate(true); 957 case vmIntrinsics::_counterMode_AESCrypt: 958 return inline_counterMode_AESCrypt_predicate(); 959 case vmIntrinsics::_digestBase_implCompressMB: 960 return inline_digestBase_implCompressMB_predicate(predicate); 961 962 default: 963 // If you get here, it may be that someone has added a new intrinsic 964 // to the list in vmSymbols.hpp without implementing it here. 965 #ifndef PRODUCT 966 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 967 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 968 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 969 } 970 #endif 971 Node* slow_ctl = control(); 972 set_control(top()); // No fast path instrinsic 973 return slow_ctl; 974 } 975 } 976 977 //------------------------------set_result------------------------------- 978 // Helper function for finishing intrinsics. 979 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 980 record_for_igvn(region); 981 set_control(_gvn.transform(region)); 982 set_result( _gvn.transform(value)); 983 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 984 } 985 986 //------------------------------generate_guard--------------------------- 987 // Helper function for generating guarded fast-slow graph structures. 988 // The given 'test', if true, guards a slow path. If the test fails 989 // then a fast path can be taken. (We generally hope it fails.) 990 // In all cases, GraphKit::control() is updated to the fast path. 991 // The returned value represents the control for the slow path. 992 // The return value is never 'top'; it is either a valid control 993 // or NULL if it is obvious that the slow path can never be taken. 994 // Also, if region and the slow control are not NULL, the slow edge 995 // is appended to the region. 996 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 997 if (stopped()) { 998 // Already short circuited. 999 return NULL; 1000 } 1001 1002 // Build an if node and its projections. 1003 // If test is true we take the slow path, which we assume is uncommon. 1004 if (_gvn.type(test) == TypeInt::ZERO) { 1005 // The slow branch is never taken. No need to build this guard. 1006 return NULL; 1007 } 1008 1009 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 1010 1011 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 1012 if (if_slow == top()) { 1013 // The slow branch is never taken. No need to build this guard. 1014 return NULL; 1015 } 1016 1017 if (region != NULL) 1018 region->add_req(if_slow); 1019 1020 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 1021 set_control(if_fast); 1022 1023 return if_slow; 1024 } 1025 1026 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 1027 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 1028 } 1029 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 1030 return generate_guard(test, region, PROB_FAIR); 1031 } 1032 1033 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 1034 Node* *pos_index) { 1035 if (stopped()) 1036 return NULL; // already stopped 1037 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 1038 return NULL; // index is already adequately typed 1039 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 1040 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1041 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 1042 if (is_neg != NULL && pos_index != NULL) { 1043 // Emulate effect of Parse::adjust_map_after_if. 1044 Node* ccast = new CastIINode(index, TypeInt::POS); 1045 ccast->set_req(0, control()); 1046 (*pos_index) = _gvn.transform(ccast); 1047 } 1048 return is_neg; 1049 } 1050 1051 // Make sure that 'position' is a valid limit index, in [0..length]. 1052 // There are two equivalent plans for checking this: 1053 // A. (offset + copyLength) unsigned<= arrayLength 1054 // B. offset <= (arrayLength - copyLength) 1055 // We require that all of the values above, except for the sum and 1056 // difference, are already known to be non-negative. 1057 // Plan A is robust in the face of overflow, if offset and copyLength 1058 // are both hugely positive. 1059 // 1060 // Plan B is less direct and intuitive, but it does not overflow at 1061 // all, since the difference of two non-negatives is always 1062 // representable. Whenever Java methods must perform the equivalent 1063 // check they generally use Plan B instead of Plan A. 1064 // For the moment we use Plan A. 1065 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 1066 Node* subseq_length, 1067 Node* array_length, 1068 RegionNode* region) { 1069 if (stopped()) 1070 return NULL; // already stopped 1071 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 1072 if (zero_offset && subseq_length->eqv_uncast(array_length)) 1073 return NULL; // common case of whole-array copy 1074 Node* last = subseq_length; 1075 if (!zero_offset) // last += offset 1076 last = _gvn.transform(new AddINode(last, offset)); 1077 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 1078 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1079 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 1080 return is_over; 1081 } 1082 1083 // Emit range checks for the given String.value byte array 1084 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) { 1085 if (stopped()) { 1086 return; // already stopped 1087 } 1088 RegionNode* bailout = new RegionNode(1); 1089 record_for_igvn(bailout); 1090 if (char_count) { 1091 // Convert char count to byte count 1092 count = _gvn.transform(new LShiftINode(count, intcon(1))); 1093 } 1094 1095 // Offset and count must not be negative 1096 generate_negative_guard(offset, bailout); 1097 generate_negative_guard(count, bailout); 1098 // Offset + count must not exceed length of array 1099 generate_limit_guard(offset, count, load_array_length(array), bailout); 1100 1101 if (bailout->req() > 1) { 1102 PreserveJVMState pjvms(this); 1103 set_control(_gvn.transform(bailout)); 1104 uncommon_trap(Deoptimization::Reason_intrinsic, 1105 Deoptimization::Action_maybe_recompile); 1106 } 1107 } 1108 1109 //--------------------------generate_current_thread-------------------- 1110 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 1111 ciKlass* thread_klass = env()->Thread_klass(); 1112 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 1113 Node* thread = _gvn.transform(new ThreadLocalNode()); 1114 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset())); 1115 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered); 1116 tls_output = thread; 1117 return threadObj; 1118 } 1119 1120 1121 //------------------------------make_string_method_node------------------------ 1122 // Helper method for String intrinsic functions. This version is called with 1123 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 1124 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 1125 // containing the lengths of str1 and str2. 1126 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 1127 Node* result = NULL; 1128 switch (opcode) { 1129 case Op_StrIndexOf: 1130 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 1131 str1_start, cnt1, str2_start, cnt2, ae); 1132 break; 1133 case Op_StrComp: 1134 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 1135 str1_start, cnt1, str2_start, cnt2, ae); 1136 break; 1137 case Op_StrEquals: 1138 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1139 // Use the constant length if there is one because optimized match rule may exist. 1140 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1141 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1142 break; 1143 default: 1144 ShouldNotReachHere(); 1145 return NULL; 1146 } 1147 1148 // All these intrinsics have checks. 1149 C->set_has_split_ifs(true); // Has chance for split-if optimization 1150 clear_upper_avx(); 1151 1152 return _gvn.transform(result); 1153 } 1154 1155 //------------------------------inline_string_compareTo------------------------ 1156 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1157 Node* arg1 = argument(0); 1158 Node* arg2 = argument(1); 1159 1160 arg1 = must_be_not_null(arg1, true); 1161 arg2 = must_be_not_null(arg2, true); 1162 1163 arg1 = access_resolve(arg1, ACCESS_READ); 1164 arg2 = access_resolve(arg2, ACCESS_READ); 1165 1166 // Get start addr and length of first argument 1167 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1168 Node* arg1_cnt = load_array_length(arg1); 1169 1170 // Get start addr and length of second argument 1171 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1172 Node* arg2_cnt = load_array_length(arg2); 1173 1174 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1175 set_result(result); 1176 return true; 1177 } 1178 1179 //------------------------------inline_string_equals------------------------ 1180 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1181 Node* arg1 = argument(0); 1182 Node* arg2 = argument(1); 1183 1184 // paths (plus control) merge 1185 RegionNode* region = new RegionNode(3); 1186 Node* phi = new PhiNode(region, TypeInt::BOOL); 1187 1188 if (!stopped()) { 1189 1190 arg1 = must_be_not_null(arg1, true); 1191 arg2 = must_be_not_null(arg2, true); 1192 1193 arg1 = access_resolve(arg1, ACCESS_READ); 1194 arg2 = access_resolve(arg2, ACCESS_READ); 1195 1196 // Get start addr and length of first argument 1197 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1198 Node* arg1_cnt = load_array_length(arg1); 1199 1200 // Get start addr and length of second argument 1201 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1202 Node* arg2_cnt = load_array_length(arg2); 1203 1204 // Check for arg1_cnt != arg2_cnt 1205 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1206 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1207 Node* if_ne = generate_slow_guard(bol, NULL); 1208 if (if_ne != NULL) { 1209 phi->init_req(2, intcon(0)); 1210 region->init_req(2, if_ne); 1211 } 1212 1213 // Check for count == 0 is done by assembler code for StrEquals. 1214 1215 if (!stopped()) { 1216 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1217 phi->init_req(1, equals); 1218 region->init_req(1, control()); 1219 } 1220 } 1221 1222 // post merge 1223 set_control(_gvn.transform(region)); 1224 record_for_igvn(region); 1225 1226 set_result(_gvn.transform(phi)); 1227 return true; 1228 } 1229 1230 //------------------------------inline_array_equals---------------------------- 1231 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1232 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1233 Node* arg1 = argument(0); 1234 Node* arg2 = argument(1); 1235 1236 arg1 = access_resolve(arg1, ACCESS_READ); 1237 arg2 = access_resolve(arg2, ACCESS_READ); 1238 1239 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1240 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1241 clear_upper_avx(); 1242 1243 return true; 1244 } 1245 1246 //------------------------------inline_hasNegatives------------------------------ 1247 bool LibraryCallKit::inline_hasNegatives() { 1248 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1249 return false; 1250 } 1251 1252 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters"); 1253 // no receiver since it is static method 1254 Node* ba = argument(0); 1255 Node* offset = argument(1); 1256 Node* len = argument(2); 1257 1258 ba = must_be_not_null(ba, true); 1259 1260 // Range checks 1261 generate_string_range_check(ba, offset, len, false); 1262 if (stopped()) { 1263 return true; 1264 } 1265 ba = access_resolve(ba, ACCESS_READ); 1266 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1267 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1268 set_result(_gvn.transform(result)); 1269 return true; 1270 } 1271 1272 bool LibraryCallKit::inline_preconditions_checkIndex() { 1273 Node* index = argument(0); 1274 Node* length = argument(1); 1275 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1276 return false; 1277 } 1278 1279 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0))); 1280 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1281 1282 { 1283 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1284 uncommon_trap(Deoptimization::Reason_intrinsic, 1285 Deoptimization::Action_make_not_entrant); 1286 } 1287 1288 if (stopped()) { 1289 return false; 1290 } 1291 1292 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length)); 1293 BoolTest::mask btest = BoolTest::lt; 1294 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1295 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1296 _gvn.set_type(rc, rc->Value(&_gvn)); 1297 if (!rc_bool->is_Con()) { 1298 record_for_igvn(rc); 1299 } 1300 set_control(_gvn.transform(new IfTrueNode(rc))); 1301 { 1302 PreserveJVMState pjvms(this); 1303 set_control(_gvn.transform(new IfFalseNode(rc))); 1304 uncommon_trap(Deoptimization::Reason_range_check, 1305 Deoptimization::Action_make_not_entrant); 1306 } 1307 1308 if (stopped()) { 1309 return false; 1310 } 1311 1312 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax)); 1313 result->set_req(0, control()); 1314 result = _gvn.transform(result); 1315 set_result(result); 1316 replace_in_map(index, result); 1317 clear_upper_avx(); 1318 return true; 1319 } 1320 1321 //------------------------------inline_string_indexOf------------------------ 1322 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1323 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1324 return false; 1325 } 1326 Node* src = argument(0); 1327 Node* tgt = argument(1); 1328 1329 // Make the merge point 1330 RegionNode* result_rgn = new RegionNode(4); 1331 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 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 // Get start addr and length of source string 1340 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1341 Node* src_count = load_array_length(src); 1342 1343 // Get start addr and length of substring 1344 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1345 Node* tgt_count = load_array_length(tgt); 1346 1347 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1348 // Divide src size by 2 if String is UTF16 encoded 1349 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1350 } 1351 if (ae == StrIntrinsicNode::UU) { 1352 // Divide substring size by 2 if String is UTF16 encoded 1353 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1354 } 1355 1356 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae); 1357 if (result != NULL) { 1358 result_phi->init_req(3, result); 1359 result_rgn->init_req(3, control()); 1360 } 1361 set_control(_gvn.transform(result_rgn)); 1362 record_for_igvn(result_rgn); 1363 set_result(_gvn.transform(result_phi)); 1364 1365 return true; 1366 } 1367 1368 //-----------------------------inline_string_indexOf----------------------- 1369 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1370 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1371 return false; 1372 } 1373 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1374 return false; 1375 } 1376 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1377 Node* src = argument(0); // byte[] 1378 Node* src_count = argument(1); // char count 1379 Node* tgt = argument(2); // byte[] 1380 Node* tgt_count = argument(3); // char count 1381 Node* from_index = argument(4); // char index 1382 1383 src = must_be_not_null(src, true); 1384 tgt = must_be_not_null(tgt, true); 1385 1386 src = access_resolve(src, ACCESS_READ); 1387 tgt = access_resolve(tgt, ACCESS_READ); 1388 1389 // Multiply byte array index by 2 if String is UTF16 encoded 1390 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1391 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1392 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1393 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1394 1395 // Range checks 1396 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1397 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1398 if (stopped()) { 1399 return true; 1400 } 1401 1402 RegionNode* region = new RegionNode(5); 1403 Node* phi = new PhiNode(region, TypeInt::INT); 1404 1405 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae); 1406 if (result != NULL) { 1407 // The result is index relative to from_index if substring was found, -1 otherwise. 1408 // Generate code which will fold into cmove. 1409 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1410 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1411 1412 Node* if_lt = generate_slow_guard(bol, NULL); 1413 if (if_lt != NULL) { 1414 // result == -1 1415 phi->init_req(3, result); 1416 region->init_req(3, if_lt); 1417 } 1418 if (!stopped()) { 1419 result = _gvn.transform(new AddINode(result, from_index)); 1420 phi->init_req(4, result); 1421 region->init_req(4, control()); 1422 } 1423 } 1424 1425 set_control(_gvn.transform(region)); 1426 record_for_igvn(region); 1427 set_result(_gvn.transform(phi)); 1428 clear_upper_avx(); 1429 1430 return true; 1431 } 1432 1433 // Create StrIndexOfNode with fast path checks 1434 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1435 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1436 // Check for substr count > string count 1437 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1438 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1439 Node* if_gt = generate_slow_guard(bol, NULL); 1440 if (if_gt != NULL) { 1441 phi->init_req(1, intcon(-1)); 1442 region->init_req(1, if_gt); 1443 } 1444 if (!stopped()) { 1445 // Check for substr count == 0 1446 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1447 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1448 Node* if_zero = generate_slow_guard(bol, NULL); 1449 if (if_zero != NULL) { 1450 phi->init_req(2, intcon(0)); 1451 region->init_req(2, if_zero); 1452 } 1453 } 1454 if (!stopped()) { 1455 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1456 } 1457 return NULL; 1458 } 1459 1460 //-----------------------------inline_string_indexOfChar----------------------- 1461 bool LibraryCallKit::inline_string_indexOfChar() { 1462 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1463 return false; 1464 } 1465 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1466 return false; 1467 } 1468 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1469 Node* src = argument(0); // byte[] 1470 Node* tgt = argument(1); // tgt is int ch 1471 Node* from_index = argument(2); 1472 Node* max = argument(3); 1473 1474 src = must_be_not_null(src, true); 1475 src = access_resolve(src, ACCESS_READ); 1476 1477 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1))); 1478 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1479 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1480 1481 // Range checks 1482 generate_string_range_check(src, src_offset, src_count, true); 1483 if (stopped()) { 1484 return true; 1485 } 1486 1487 RegionNode* region = new RegionNode(3); 1488 Node* phi = new PhiNode(region, TypeInt::INT); 1489 1490 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none); 1491 C->set_has_split_ifs(true); // Has chance for split-if optimization 1492 _gvn.transform(result); 1493 1494 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1495 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1496 1497 Node* if_lt = generate_slow_guard(bol, NULL); 1498 if (if_lt != NULL) { 1499 // result == -1 1500 phi->init_req(2, result); 1501 region->init_req(2, if_lt); 1502 } 1503 if (!stopped()) { 1504 result = _gvn.transform(new AddINode(result, from_index)); 1505 phi->init_req(1, result); 1506 region->init_req(1, control()); 1507 } 1508 set_control(_gvn.transform(region)); 1509 record_for_igvn(region); 1510 set_result(_gvn.transform(phi)); 1511 1512 return true; 1513 } 1514 //---------------------------inline_string_copy--------------------- 1515 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1516 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1517 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1518 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1519 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1520 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1521 bool LibraryCallKit::inline_string_copy(bool compress) { 1522 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1523 return false; 1524 } 1525 int nargs = 5; // 2 oops, 3 ints 1526 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1527 1528 Node* src = argument(0); 1529 Node* src_offset = argument(1); 1530 Node* dst = argument(2); 1531 Node* dst_offset = argument(3); 1532 Node* length = argument(4); 1533 1534 // Check for allocation before we add nodes that would confuse 1535 // tightly_coupled_allocation() 1536 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1537 1538 // Figure out the size and type of the elements we will be copying. 1539 const Type* src_type = src->Value(&_gvn); 1540 const Type* dst_type = dst->Value(&_gvn); 1541 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1542 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1543 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1544 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1545 "Unsupported array types for inline_string_copy"); 1546 1547 src = must_be_not_null(src, true); 1548 dst = must_be_not_null(dst, true); 1549 1550 // Convert char[] offsets to byte[] offsets 1551 bool convert_src = (compress && src_elem == T_BYTE); 1552 bool convert_dst = (!compress && dst_elem == T_BYTE); 1553 if (convert_src) { 1554 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1555 } else if (convert_dst) { 1556 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1557 } 1558 1559 // Range checks 1560 generate_string_range_check(src, src_offset, length, convert_src); 1561 generate_string_range_check(dst, dst_offset, length, convert_dst); 1562 if (stopped()) { 1563 return true; 1564 } 1565 1566 src = access_resolve(src, ACCESS_READ); 1567 dst = access_resolve(dst, ACCESS_WRITE); 1568 1569 Node* src_start = array_element_address(src, src_offset, src_elem); 1570 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1571 // 'src_start' points to src array + scaled offset 1572 // 'dst_start' points to dst array + scaled offset 1573 Node* count = NULL; 1574 if (compress) { 1575 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1576 } else { 1577 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1578 } 1579 1580 if (alloc != NULL) { 1581 if (alloc->maybe_set_complete(&_gvn)) { 1582 // "You break it, you buy it." 1583 InitializeNode* init = alloc->initialization(); 1584 assert(init->is_complete(), "we just did this"); 1585 init->set_complete_with_arraycopy(); 1586 assert(dst->is_CheckCastPP(), "sanity"); 1587 assert(dst->in(0)->in(0) == init, "dest pinned"); 1588 } 1589 // Do not let stores that initialize this object be reordered with 1590 // a subsequent store that would make this object accessible by 1591 // other threads. 1592 // Record what AllocateNode this StoreStore protects so that 1593 // escape analysis can go from the MemBarStoreStoreNode to the 1594 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1595 // based on the escape status of the AllocateNode. 1596 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1597 } 1598 if (compress) { 1599 set_result(_gvn.transform(count)); 1600 } 1601 clear_upper_avx(); 1602 1603 return true; 1604 } 1605 1606 #ifdef _LP64 1607 #define XTOP ,top() /*additional argument*/ 1608 #else //_LP64 1609 #define XTOP /*no additional argument*/ 1610 #endif //_LP64 1611 1612 //------------------------inline_string_toBytesU-------------------------- 1613 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1614 bool LibraryCallKit::inline_string_toBytesU() { 1615 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1616 return false; 1617 } 1618 // Get the arguments. 1619 Node* value = argument(0); 1620 Node* offset = argument(1); 1621 Node* length = argument(2); 1622 1623 Node* newcopy = NULL; 1624 1625 // Set the original stack and the reexecute bit for the interpreter to reexecute 1626 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1627 { PreserveReexecuteState preexecs(this); 1628 jvms()->set_should_reexecute(true); 1629 1630 // Check if a null path was taken unconditionally. 1631 value = null_check(value); 1632 1633 RegionNode* bailout = new RegionNode(1); 1634 record_for_igvn(bailout); 1635 1636 // Range checks 1637 generate_negative_guard(offset, bailout); 1638 generate_negative_guard(length, bailout); 1639 generate_limit_guard(offset, length, load_array_length(value), bailout); 1640 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1641 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1642 1643 if (bailout->req() > 1) { 1644 PreserveJVMState pjvms(this); 1645 set_control(_gvn.transform(bailout)); 1646 uncommon_trap(Deoptimization::Reason_intrinsic, 1647 Deoptimization::Action_maybe_recompile); 1648 } 1649 if (stopped()) { 1650 return true; 1651 } 1652 1653 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1654 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1655 newcopy = new_array(klass_node, size, 0); // no arguments to push 1656 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL); 1657 1658 // Calculate starting addresses. 1659 value = access_resolve(value, ACCESS_READ); 1660 Node* src_start = array_element_address(value, offset, T_CHAR); 1661 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1662 1663 // Check if src array address is aligned to HeapWordSize (dst is always aligned) 1664 const TypeInt* toffset = gvn().type(offset)->is_int(); 1665 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1666 1667 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1668 const char* copyfunc_name = "arraycopy"; 1669 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1670 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1671 OptoRuntime::fast_arraycopy_Type(), 1672 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1673 src_start, dst_start, ConvI2X(length) XTOP); 1674 // Do not let reads from the cloned object float above the arraycopy. 1675 if (alloc != NULL) { 1676 if (alloc->maybe_set_complete(&_gvn)) { 1677 // "You break it, you buy it." 1678 InitializeNode* init = alloc->initialization(); 1679 assert(init->is_complete(), "we just did this"); 1680 init->set_complete_with_arraycopy(); 1681 assert(newcopy->is_CheckCastPP(), "sanity"); 1682 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1683 } 1684 // Do not let stores that initialize this object be reordered with 1685 // a subsequent store that would make this object accessible by 1686 // other threads. 1687 // Record what AllocateNode this StoreStore protects so that 1688 // escape analysis can go from the MemBarStoreStoreNode to the 1689 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1690 // based on the escape status of the AllocateNode. 1691 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1692 } else { 1693 insert_mem_bar(Op_MemBarCPUOrder); 1694 } 1695 } // original reexecute is set back here 1696 1697 C->set_has_split_ifs(true); // Has chance for split-if optimization 1698 if (!stopped()) { 1699 set_result(newcopy); 1700 } 1701 clear_upper_avx(); 1702 1703 return true; 1704 } 1705 1706 //------------------------inline_string_getCharsU-------------------------- 1707 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1708 bool LibraryCallKit::inline_string_getCharsU() { 1709 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1710 return false; 1711 } 1712 1713 // Get the arguments. 1714 Node* src = argument(0); 1715 Node* src_begin = argument(1); 1716 Node* src_end = argument(2); // exclusive offset (i < src_end) 1717 Node* dst = argument(3); 1718 Node* dst_begin = argument(4); 1719 1720 // Check for allocation before we add nodes that would confuse 1721 // tightly_coupled_allocation() 1722 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1723 1724 // Check if a null path was taken unconditionally. 1725 src = null_check(src); 1726 dst = null_check(dst); 1727 if (stopped()) { 1728 return true; 1729 } 1730 1731 // Get length and convert char[] offset to byte[] offset 1732 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1733 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1734 1735 // Range checks 1736 generate_string_range_check(src, src_begin, length, true); 1737 generate_string_range_check(dst, dst_begin, length, false); 1738 if (stopped()) { 1739 return true; 1740 } 1741 1742 if (!stopped()) { 1743 src = access_resolve(src, ACCESS_READ); 1744 dst = access_resolve(dst, ACCESS_WRITE); 1745 1746 // Calculate starting addresses. 1747 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1748 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1749 1750 // Check if array addresses are aligned to HeapWordSize 1751 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1752 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1753 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1754 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1755 1756 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1757 const char* copyfunc_name = "arraycopy"; 1758 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1759 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1760 OptoRuntime::fast_arraycopy_Type(), 1761 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1762 src_start, dst_start, ConvI2X(length) XTOP); 1763 // Do not let reads from the cloned object float above the arraycopy. 1764 if (alloc != NULL) { 1765 if (alloc->maybe_set_complete(&_gvn)) { 1766 // "You break it, you buy it." 1767 InitializeNode* init = alloc->initialization(); 1768 assert(init->is_complete(), "we just did this"); 1769 init->set_complete_with_arraycopy(); 1770 assert(dst->is_CheckCastPP(), "sanity"); 1771 assert(dst->in(0)->in(0) == init, "dest pinned"); 1772 } 1773 // Do not let stores that initialize this object be reordered with 1774 // a subsequent store that would make this object accessible by 1775 // other threads. 1776 // Record what AllocateNode this StoreStore protects so that 1777 // escape analysis can go from the MemBarStoreStoreNode to the 1778 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1779 // based on the escape status of the AllocateNode. 1780 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1781 } else { 1782 insert_mem_bar(Op_MemBarCPUOrder); 1783 } 1784 } 1785 1786 C->set_has_split_ifs(true); // Has chance for split-if optimization 1787 return true; 1788 } 1789 1790 //----------------------inline_string_char_access---------------------------- 1791 // Store/Load char to/from byte[] array. 1792 // static void StringUTF16.putChar(byte[] val, int index, int c) 1793 // static char StringUTF16.getChar(byte[] val, int index) 1794 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1795 Node* value = argument(0); 1796 Node* index = argument(1); 1797 Node* ch = is_store ? argument(2) : NULL; 1798 1799 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1800 // correctly requires matched array shapes. 1801 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1802 "sanity: byte[] and char[] bases agree"); 1803 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1804 "sanity: byte[] and char[] scales agree"); 1805 1806 // Bail when getChar over constants is requested: constant folding would 1807 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1808 // Java method would constant fold nicely instead. 1809 if (!is_store && value->is_Con() && index->is_Con()) { 1810 return false; 1811 } 1812 1813 value = must_be_not_null(value, true); 1814 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ); 1815 1816 Node* adr = array_element_address(value, index, T_CHAR); 1817 if (adr->is_top()) { 1818 return false; 1819 } 1820 if (is_store) { 1821 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED); 1822 } else { 1823 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 1824 set_result(ch); 1825 } 1826 return true; 1827 } 1828 1829 //--------------------------round_double_node-------------------------------- 1830 // Round a double node if necessary. 1831 Node* LibraryCallKit::round_double_node(Node* n) { 1832 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1) 1833 n = _gvn.transform(new RoundDoubleNode(0, n)); 1834 return n; 1835 } 1836 1837 //------------------------------inline_math----------------------------------- 1838 // public static double Math.abs(double) 1839 // public static double Math.sqrt(double) 1840 // public static double Math.log(double) 1841 // public static double Math.log10(double) 1842 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) { 1843 Node* arg = round_double_node(argument(0)); 1844 Node* n = NULL; 1845 switch (id) { 1846 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1847 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break; 1848 default: fatal_unexpected_iid(id); break; 1849 } 1850 set_result(_gvn.transform(n)); 1851 return true; 1852 } 1853 1854 //------------------------------inline_math----------------------------------- 1855 // public static float Math.abs(float) 1856 // public static int Math.abs(int) 1857 // public static long Math.abs(long) 1858 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1859 Node* arg = argument(0); 1860 Node* n = NULL; 1861 switch (id) { 1862 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break; 1863 case vmIntrinsics::_iabs: n = new AbsINode( arg); break; 1864 case vmIntrinsics::_labs: n = new AbsLNode( arg); break; 1865 default: fatal_unexpected_iid(id); break; 1866 } 1867 set_result(_gvn.transform(n)); 1868 return true; 1869 } 1870 1871 //------------------------------runtime_math----------------------------- 1872 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1873 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1874 "must be (DD)D or (D)D type"); 1875 1876 // Inputs 1877 Node* a = round_double_node(argument(0)); 1878 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL; 1879 1880 const TypePtr* no_memory_effects = NULL; 1881 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName, 1882 no_memory_effects, 1883 a, top(), b, b ? top() : NULL); 1884 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1885 #ifdef ASSERT 1886 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1887 assert(value_top == top(), "second value must be top"); 1888 #endif 1889 1890 set_result(value); 1891 return true; 1892 } 1893 1894 //------------------------------inline_math_native----------------------------- 1895 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1896 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f) 1897 switch (id) { 1898 // These intrinsics are not properly supported on all hardware 1899 case vmIntrinsics::_dsin: 1900 return StubRoutines::dsin() != NULL ? 1901 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1902 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN"); 1903 case vmIntrinsics::_dcos: 1904 return StubRoutines::dcos() != NULL ? 1905 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1906 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS"); 1907 case vmIntrinsics::_dtan: 1908 return StubRoutines::dtan() != NULL ? 1909 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") : 1910 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN"); 1911 case vmIntrinsics::_dlog: 1912 return StubRoutines::dlog() != NULL ? 1913 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1914 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG"); 1915 case vmIntrinsics::_dlog10: 1916 return StubRoutines::dlog10() != NULL ? 1917 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") : 1918 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10"); 1919 1920 // These intrinsics are supported on all hardware 1921 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false; 1922 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false; 1923 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false; 1924 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false; 1925 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false; 1926 1927 case vmIntrinsics::_dexp: 1928 return StubRoutines::dexp() != NULL ? 1929 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1930 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP"); 1931 case vmIntrinsics::_dpow: { 1932 Node* exp = round_double_node(argument(2)); 1933 const TypeD* d = _gvn.type(exp)->isa_double_constant(); 1934 if (d != NULL && d->getd() == 2.0) { 1935 // Special case: pow(x, 2.0) => x * x 1936 Node* base = round_double_node(argument(0)); 1937 set_result(_gvn.transform(new MulDNode(base, base))); 1938 return true; 1939 } 1940 return StubRoutines::dpow() != NULL ? 1941 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1942 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW"); 1943 } 1944 #undef FN_PTR 1945 1946 // These intrinsics are not yet correctly implemented 1947 case vmIntrinsics::_datan2: 1948 return false; 1949 1950 default: 1951 fatal_unexpected_iid(id); 1952 return false; 1953 } 1954 } 1955 1956 static bool is_simple_name(Node* n) { 1957 return (n->req() == 1 // constant 1958 || (n->is_Type() && n->as_Type()->type()->singleton()) 1959 || n->is_Proj() // parameter or return value 1960 || n->is_Phi() // local of some sort 1961 ); 1962 } 1963 1964 //----------------------------inline_notify-----------------------------------* 1965 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1966 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1967 address func; 1968 if (id == vmIntrinsics::_notify) { 1969 func = OptoRuntime::monitor_notify_Java(); 1970 } else { 1971 func = OptoRuntime::monitor_notifyAll_Java(); 1972 } 1973 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0)); 1974 make_slow_call_ex(call, env()->Throwable_klass(), false); 1975 return true; 1976 } 1977 1978 1979 //----------------------------inline_min_max----------------------------------- 1980 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1981 set_result(generate_min_max(id, argument(0), argument(1))); 1982 return true; 1983 } 1984 1985 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) { 1986 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 1987 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 1988 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 1989 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 1990 1991 { 1992 PreserveJVMState pjvms(this); 1993 PreserveReexecuteState preexecs(this); 1994 jvms()->set_should_reexecute(true); 1995 1996 set_control(slow_path); 1997 set_i_o(i_o()); 1998 1999 uncommon_trap(Deoptimization::Reason_intrinsic, 2000 Deoptimization::Action_none); 2001 } 2002 2003 set_control(fast_path); 2004 set_result(math); 2005 } 2006 2007 template <typename OverflowOp> 2008 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 2009 typedef typename OverflowOp::MathOp MathOp; 2010 2011 MathOp* mathOp = new MathOp(arg1, arg2); 2012 Node* operation = _gvn.transform( mathOp ); 2013 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 2014 inline_math_mathExact(operation, ofcheck); 2015 return true; 2016 } 2017 2018 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 2019 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 2020 } 2021 2022 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 2023 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 2024 } 2025 2026 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 2027 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 2028 } 2029 2030 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 2031 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 2032 } 2033 2034 bool LibraryCallKit::inline_math_negateExactI() { 2035 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 2036 } 2037 2038 bool LibraryCallKit::inline_math_negateExactL() { 2039 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 2040 } 2041 2042 bool LibraryCallKit::inline_math_multiplyExactI() { 2043 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 2044 } 2045 2046 bool LibraryCallKit::inline_math_multiplyExactL() { 2047 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 2048 } 2049 2050 bool LibraryCallKit::inline_math_multiplyHigh() { 2051 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2)))); 2052 return true; 2053 } 2054 2055 Node* 2056 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) { 2057 // These are the candidate return value: 2058 Node* xvalue = x0; 2059 Node* yvalue = y0; 2060 2061 if (xvalue == yvalue) { 2062 return xvalue; 2063 } 2064 2065 bool want_max = (id == vmIntrinsics::_max); 2066 2067 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int(); 2068 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int(); 2069 if (txvalue == NULL || tyvalue == NULL) return top(); 2070 // This is not really necessary, but it is consistent with a 2071 // hypothetical MaxINode::Value method: 2072 int widen = MAX2(txvalue->_widen, tyvalue->_widen); 2073 2074 // %%% This folding logic should (ideally) be in a different place. 2075 // Some should be inside IfNode, and there to be a more reliable 2076 // transformation of ?: style patterns into cmoves. We also want 2077 // more powerful optimizations around cmove and min/max. 2078 2079 // Try to find a dominating comparison of these guys. 2080 // It can simplify the index computation for Arrays.copyOf 2081 // and similar uses of System.arraycopy. 2082 // First, compute the normalized version of CmpI(x, y). 2083 int cmp_op = Op_CmpI; 2084 Node* xkey = xvalue; 2085 Node* ykey = yvalue; 2086 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey)); 2087 if (ideal_cmpxy->is_Cmp()) { 2088 // E.g., if we have CmpI(length - offset, count), 2089 // it might idealize to CmpI(length, count + offset) 2090 cmp_op = ideal_cmpxy->Opcode(); 2091 xkey = ideal_cmpxy->in(1); 2092 ykey = ideal_cmpxy->in(2); 2093 } 2094 2095 // Start by locating any relevant comparisons. 2096 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey; 2097 Node* cmpxy = NULL; 2098 Node* cmpyx = NULL; 2099 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) { 2100 Node* cmp = start_from->fast_out(k); 2101 if (cmp->outcnt() > 0 && // must have prior uses 2102 cmp->in(0) == NULL && // must be context-independent 2103 cmp->Opcode() == cmp_op) { // right kind of compare 2104 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp; 2105 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp; 2106 } 2107 } 2108 2109 const int NCMPS = 2; 2110 Node* cmps[NCMPS] = { cmpxy, cmpyx }; 2111 int cmpn; 2112 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2113 if (cmps[cmpn] != NULL) break; // find a result 2114 } 2115 if (cmpn < NCMPS) { 2116 // Look for a dominating test that tells us the min and max. 2117 int depth = 0; // Limit search depth for speed 2118 Node* dom = control(); 2119 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) { 2120 if (++depth >= 100) break; 2121 Node* ifproj = dom; 2122 if (!ifproj->is_Proj()) continue; 2123 Node* iff = ifproj->in(0); 2124 if (!iff->is_If()) continue; 2125 Node* bol = iff->in(1); 2126 if (!bol->is_Bool()) continue; 2127 Node* cmp = bol->in(1); 2128 if (cmp == NULL) continue; 2129 for (cmpn = 0; cmpn < NCMPS; cmpn++) 2130 if (cmps[cmpn] == cmp) break; 2131 if (cmpn == NCMPS) continue; 2132 BoolTest::mask btest = bol->as_Bool()->_test._test; 2133 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate(); 2134 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2135 // At this point, we know that 'x btest y' is true. 2136 switch (btest) { 2137 case BoolTest::eq: 2138 // They are proven equal, so we can collapse the min/max. 2139 // Either value is the answer. Choose the simpler. 2140 if (is_simple_name(yvalue) && !is_simple_name(xvalue)) 2141 return yvalue; 2142 return xvalue; 2143 case BoolTest::lt: // x < y 2144 case BoolTest::le: // x <= y 2145 return (want_max ? yvalue : xvalue); 2146 case BoolTest::gt: // x > y 2147 case BoolTest::ge: // x >= y 2148 return (want_max ? xvalue : yvalue); 2149 default: 2150 break; 2151 } 2152 } 2153 } 2154 2155 // We failed to find a dominating test. 2156 // Let's pick a test that might GVN with prior tests. 2157 Node* best_bol = NULL; 2158 BoolTest::mask best_btest = BoolTest::illegal; 2159 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2160 Node* cmp = cmps[cmpn]; 2161 if (cmp == NULL) continue; 2162 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) { 2163 Node* bol = cmp->fast_out(j); 2164 if (!bol->is_Bool()) continue; 2165 BoolTest::mask btest = bol->as_Bool()->_test._test; 2166 if (btest == BoolTest::eq || btest == BoolTest::ne) continue; 2167 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2168 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) { 2169 best_bol = bol->as_Bool(); 2170 best_btest = btest; 2171 } 2172 } 2173 } 2174 2175 Node* answer_if_true = NULL; 2176 Node* answer_if_false = NULL; 2177 switch (best_btest) { 2178 default: 2179 if (cmpxy == NULL) 2180 cmpxy = ideal_cmpxy; 2181 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt)); 2182 // and fall through: 2183 case BoolTest::lt: // x < y 2184 case BoolTest::le: // x <= y 2185 answer_if_true = (want_max ? yvalue : xvalue); 2186 answer_if_false = (want_max ? xvalue : yvalue); 2187 break; 2188 case BoolTest::gt: // x > y 2189 case BoolTest::ge: // x >= y 2190 answer_if_true = (want_max ? xvalue : yvalue); 2191 answer_if_false = (want_max ? yvalue : xvalue); 2192 break; 2193 } 2194 2195 jint hi, lo; 2196 if (want_max) { 2197 // We can sharpen the minimum. 2198 hi = MAX2(txvalue->_hi, tyvalue->_hi); 2199 lo = MAX2(txvalue->_lo, tyvalue->_lo); 2200 } else { 2201 // We can sharpen the maximum. 2202 hi = MIN2(txvalue->_hi, tyvalue->_hi); 2203 lo = MIN2(txvalue->_lo, tyvalue->_lo); 2204 } 2205 2206 // Use a flow-free graph structure, to avoid creating excess control edges 2207 // which could hinder other optimizations. 2208 // Since Math.min/max is often used with arraycopy, we want 2209 // tightly_coupled_allocation to be able to see beyond min/max expressions. 2210 Node* cmov = CMoveNode::make(NULL, best_bol, 2211 answer_if_false, answer_if_true, 2212 TypeInt::make(lo, hi, widen)); 2213 2214 return _gvn.transform(cmov); 2215 2216 /* 2217 // This is not as desirable as it may seem, since Min and Max 2218 // nodes do not have a full set of optimizations. 2219 // And they would interfere, anyway, with 'if' optimizations 2220 // and with CMoveI canonical forms. 2221 switch (id) { 2222 case vmIntrinsics::_min: 2223 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break; 2224 case vmIntrinsics::_max: 2225 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break; 2226 default: 2227 ShouldNotReachHere(); 2228 } 2229 */ 2230 } 2231 2232 inline int 2233 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) { 2234 const TypePtr* base_type = TypePtr::NULL_PTR; 2235 if (base != NULL) base_type = _gvn.type(base)->isa_ptr(); 2236 if (base_type == NULL) { 2237 // Unknown type. 2238 return Type::AnyPtr; 2239 } else if (base_type == TypePtr::NULL_PTR) { 2240 // Since this is a NULL+long form, we have to switch to a rawptr. 2241 base = _gvn.transform(new CastX2PNode(offset)); 2242 offset = MakeConX(0); 2243 return Type::RawPtr; 2244 } else if (base_type->base() == Type::RawPtr) { 2245 return Type::RawPtr; 2246 } else if (base_type->isa_oopptr()) { 2247 // Base is never null => always a heap address. 2248 if (!TypePtr::NULL_PTR->higher_equal(base_type)) { 2249 return Type::OopPtr; 2250 } 2251 // Offset is small => always a heap address. 2252 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2253 if (offset_type != NULL && 2254 base_type->offset() == 0 && // (should always be?) 2255 offset_type->_lo >= 0 && 2256 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2257 return Type::OopPtr; 2258 } else if (type == T_OBJECT) { 2259 // off heap access to an oop doesn't make any sense. Has to be on 2260 // heap. 2261 return Type::OopPtr; 2262 } 2263 // Otherwise, it might either be oop+off or NULL+addr. 2264 return Type::AnyPtr; 2265 } else { 2266 // No information: 2267 return Type::AnyPtr; 2268 } 2269 } 2270 2271 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) { 2272 Node* uncasted_base = base; 2273 int kind = classify_unsafe_addr(uncasted_base, offset, type); 2274 if (kind == Type::RawPtr) { 2275 return basic_plus_adr(top(), uncasted_base, offset); 2276 } else if (kind == Type::AnyPtr) { 2277 assert(base == uncasted_base, "unexpected base change"); 2278 if (can_cast) { 2279 if (!_gvn.type(base)->speculative_maybe_null() && 2280 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 2281 // According to profiling, this access is always on 2282 // heap. Casting the base to not null and thus avoiding membars 2283 // around the access should allow better optimizations 2284 Node* null_ctl = top(); 2285 base = null_check_oop(base, &null_ctl, true, true, true); 2286 assert(null_ctl->is_top(), "no null control here"); 2287 return basic_plus_adr(base, offset); 2288 } else if (_gvn.type(base)->speculative_always_null() && 2289 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 2290 // According to profiling, this access is always off 2291 // heap. 2292 base = null_assert(base); 2293 Node* raw_base = _gvn.transform(new CastX2PNode(offset)); 2294 offset = MakeConX(0); 2295 return basic_plus_adr(top(), raw_base, offset); 2296 } 2297 } 2298 // We don't know if it's an on heap or off heap access. Fall back 2299 // to raw memory access. 2300 base = access_resolve(base, decorators); 2301 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM)); 2302 return basic_plus_adr(top(), raw, offset); 2303 } else { 2304 assert(base == uncasted_base, "unexpected base change"); 2305 // We know it's an on heap access so base can't be null 2306 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) { 2307 base = must_be_not_null(base, true); 2308 } 2309 return basic_plus_adr(base, offset); 2310 } 2311 } 2312 2313 //--------------------------inline_number_methods----------------------------- 2314 // inline int Integer.numberOfLeadingZeros(int) 2315 // inline int Long.numberOfLeadingZeros(long) 2316 // 2317 // inline int Integer.numberOfTrailingZeros(int) 2318 // inline int Long.numberOfTrailingZeros(long) 2319 // 2320 // inline int Integer.bitCount(int) 2321 // inline int Long.bitCount(long) 2322 // 2323 // inline char Character.reverseBytes(char) 2324 // inline short Short.reverseBytes(short) 2325 // inline int Integer.reverseBytes(int) 2326 // inline long Long.reverseBytes(long) 2327 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2328 Node* arg = argument(0); 2329 Node* n = NULL; 2330 switch (id) { 2331 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2332 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2333 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2334 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2335 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2336 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2337 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break; 2338 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break; 2339 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break; 2340 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break; 2341 default: fatal_unexpected_iid(id); break; 2342 } 2343 set_result(_gvn.transform(n)); 2344 return true; 2345 } 2346 2347 //----------------------------inline_unsafe_access---------------------------- 2348 2349 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) { 2350 // Attempt to infer a sharper value type from the offset and base type. 2351 ciKlass* sharpened_klass = NULL; 2352 2353 // See if it is an instance field, with an object type. 2354 if (alias_type->field() != NULL) { 2355 if (alias_type->field()->type()->is_klass()) { 2356 sharpened_klass = alias_type->field()->type()->as_klass(); 2357 } 2358 } 2359 2360 // See if it is a narrow oop array. 2361 if (adr_type->isa_aryptr()) { 2362 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2363 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr(); 2364 if (elem_type != NULL) { 2365 sharpened_klass = elem_type->klass(); 2366 } 2367 } 2368 } 2369 2370 // The sharpened class might be unloaded if there is no class loader 2371 // contraint in place. 2372 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) { 2373 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass); 2374 2375 #ifndef PRODUCT 2376 if (C->print_intrinsics() || C->print_inlining()) { 2377 tty->print(" from base type: "); adr_type->dump(); tty->cr(); 2378 tty->print(" sharpened value: "); tjp->dump(); tty->cr(); 2379 } 2380 #endif 2381 // Sharpen the value type. 2382 return tjp; 2383 } 2384 return NULL; 2385 } 2386 2387 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) { 2388 switch (kind) { 2389 case Relaxed: 2390 return MO_UNORDERED; 2391 case Opaque: 2392 return MO_RELAXED; 2393 case Acquire: 2394 return MO_ACQUIRE; 2395 case Release: 2396 return MO_RELEASE; 2397 case Volatile: 2398 return MO_SEQ_CST; 2399 default: 2400 ShouldNotReachHere(); 2401 return 0; 2402 } 2403 } 2404 2405 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) { 2406 if (callee()->is_static()) return false; // caller must have the capability! 2407 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2408 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2409 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2410 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2411 2412 if (type == T_OBJECT || type == T_ARRAY) { 2413 decorators |= ON_UNKNOWN_OOP_REF; 2414 } 2415 2416 if (unaligned) { 2417 decorators |= C2_UNALIGNED; 2418 } 2419 2420 #ifndef PRODUCT 2421 { 2422 ResourceMark rm; 2423 // Check the signatures. 2424 ciSignature* sig = callee()->signature(); 2425 #ifdef ASSERT 2426 if (!is_store) { 2427 // Object getReference(Object base, int/long offset), etc. 2428 BasicType rtype = sig->return_type()->basic_type(); 2429 assert(rtype == type || (rtype == T_OBJECT && type == T_VALUETYPE), "getter must return the expected value"); 2430 assert(sig->count() == 2 || (type == T_VALUETYPE && sig->count() == 3), "oop getter has 2 or 3 arguments"); 2431 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2432 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2433 } else { 2434 // void putReference(Object base, int/long offset, Object x), etc. 2435 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2436 assert(sig->count() == 3 || (type == T_VALUETYPE && sig->count() == 4), "oop putter has 3 arguments"); 2437 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2438 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2439 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2440 assert(vtype == type || (type == T_VALUETYPE && vtype == T_OBJECT), "putter must accept the expected value"); 2441 } 2442 #endif // ASSERT 2443 } 2444 #endif //PRODUCT 2445 2446 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2447 2448 Node* receiver = argument(0); // type: oop 2449 2450 // Build address expression. 2451 Node* adr; 2452 Node* heap_base_oop = top(); 2453 Node* offset = top(); 2454 Node* val; 2455 2456 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2457 Node* base = argument(1); // type: oop 2458 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2459 offset = argument(2); // type: long 2460 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2461 // to be plain byte offsets, which are also the same as those accepted 2462 // by oopDesc::field_addr. 2463 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2464 "fieldOffset must be byte-scaled"); 2465 2466 ciValueKlass* value_klass = NULL; 2467 if (type == T_VALUETYPE) { 2468 Node* cls = null_check(argument(4)); 2469 if (stopped()) { 2470 return true; 2471 } 2472 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 2473 const TypeKlassPtr* kls_t = _gvn.type(kls)->isa_klassptr(); 2474 if (!kls_t->klass_is_exact()) { 2475 return false; 2476 } 2477 ciKlass* klass = kls_t->klass(); 2478 if (!klass->is_valuetype()) { 2479 return false; 2480 } 2481 value_klass = klass->as_value_klass(); 2482 } 2483 2484 receiver = null_check(receiver); 2485 if (stopped()) { 2486 return true; 2487 } 2488 2489 if (base->is_ValueType()) { 2490 ValueTypeNode* vt = base->as_ValueType(); 2491 2492 if (is_store) { 2493 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) { 2494 return false; 2495 } 2496 base = vt->get_oop(); 2497 } else { 2498 if (offset->is_Con()) { 2499 long off = find_long_con(offset, 0); 2500 ciValueKlass* vk = vt->type()->value_klass(); 2501 if ((long)(int)off != off || !vk->contains_field_offset(off)) { 2502 return false; 2503 } 2504 2505 ciField* f = vk->get_non_flattened_field_by_offset((int)off); 2506 2507 if (f != NULL) { 2508 BasicType bt = f->layout_type(); 2509 if (bt == T_ARRAY || bt == T_NARROWOOP) { 2510 bt = T_OBJECT; 2511 } 2512 if (bt == type) { 2513 if (bt != T_VALUETYPE || f->type() == value_klass) { 2514 set_result(vt->field_value_by_offset((int)off, false)); 2515 return true; 2516 } 2517 } 2518 } 2519 } 2520 vt = vt->allocate(this)->as_ValueType(); 2521 base = vt->get_oop(); 2522 } 2523 } 2524 2525 // 32-bit machines ignore the high half! 2526 offset = ConvL2X(offset); 2527 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed); 2528 2529 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) { 2530 heap_base_oop = base; 2531 } else if (type == T_OBJECT || (value_klass != NULL && value_klass->has_object_fields())) { 2532 return false; // off-heap oop accesses are not supported 2533 } 2534 2535 // Can base be NULL? Otherwise, always on-heap access. 2536 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base)); 2537 2538 if (!can_access_non_heap) { 2539 decorators |= IN_HEAP; 2540 } 2541 2542 val = is_store ? argument(4 + (type == T_VALUETYPE ? 1 : 0)) : NULL; 2543 2544 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr(); 2545 if (adr_type == TypePtr::NULL_PTR) { 2546 return false; // off-heap access with zero address 2547 } 2548 2549 // Try to categorize the address. 2550 Compile::AliasType* alias_type = C->alias_type(adr_type); 2551 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2552 2553 if (alias_type->adr_type() == TypeInstPtr::KLASS || 2554 alias_type->adr_type() == TypeAryPtr::RANGE) { 2555 return false; // not supported 2556 } 2557 2558 bool mismatched = false; 2559 BasicType bt = T_ILLEGAL; 2560 ciField* field = NULL; 2561 if (adr_type->isa_instptr()) { 2562 const TypeInstPtr* instptr = adr_type->is_instptr(); 2563 ciInstanceKlass* k = instptr->klass()->as_instance_klass(); 2564 int off = instptr->offset(); 2565 if (instptr->const_oop() != NULL && 2566 instptr->klass() == ciEnv::current()->Class_klass() && 2567 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) { 2568 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass(); 2569 field = k->get_field_by_offset(off, true); 2570 } else { 2571 field = k->get_non_flattened_field_by_offset(off); 2572 } 2573 if (field != NULL) { 2574 bt = field->layout_type(); 2575 } 2576 assert(bt == alias_type->basic_type() || bt == T_VALUETYPE, "should match"); 2577 if (field != NULL && bt == T_VALUETYPE && !field->is_flattened()) { 2578 bt = T_OBJECT; 2579 } 2580 } else { 2581 bt = alias_type->basic_type(); 2582 } 2583 2584 if (bt != T_ILLEGAL) { 2585 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access"); 2586 if (bt == T_BYTE && adr_type->isa_aryptr()) { 2587 // Alias type doesn't differentiate between byte[] and boolean[]). 2588 // Use address type to get the element type. 2589 bt = adr_type->is_aryptr()->elem()->array_element_basic_type(); 2590 } 2591 if (bt == T_ARRAY || bt == T_NARROWOOP) { 2592 // accessing an array field with getReference is not a mismatch 2593 bt = T_OBJECT; 2594 } 2595 if ((bt == T_OBJECT) != (type == T_OBJECT)) { 2596 // Don't intrinsify mismatched object accesses 2597 return false; 2598 } 2599 mismatched = (bt != type); 2600 } else if (alias_type->adr_type()->isa_oopptr()) { 2601 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched 2602 } 2603 2604 if (type == T_VALUETYPE) { 2605 if (adr_type->isa_instptr()) { 2606 if (field == NULL || field->type() != value_klass) { 2607 mismatched = true; 2608 } 2609 } else if (adr_type->isa_aryptr()) { 2610 const Type* elem = adr_type->is_aryptr()->elem(); 2611 if (!elem->isa_valuetype()) { 2612 mismatched = true; 2613 } else if (elem->value_klass() != value_klass) { 2614 mismatched = true; 2615 } 2616 } 2617 if (is_store) { 2618 const Type* val_t = _gvn.type(val); 2619 if (!val_t->isa_valuetype() || val_t->value_klass() != value_klass) { 2620 return false; 2621 } 2622 } 2623 } 2624 2625 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched"); 2626 2627 if (mismatched) { 2628 decorators |= C2_MISMATCHED; 2629 } 2630 2631 // First guess at the value type. 2632 const Type *value_type = Type::get_const_basic_type(type); 2633 2634 // Figure out the memory ordering. 2635 decorators |= mo_decorator_for_access_kind(kind); 2636 2637 if (!is_store) { 2638 if (type == T_OBJECT) { 2639 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2640 if (tjp != NULL) { 2641 value_type = tjp; 2642 } 2643 } else if (type == T_VALUETYPE) { 2644 value_type = NULL; 2645 } 2646 } 2647 2648 // Heap pointers get a null-check from the interpreter, 2649 // as a courtesy. However, this is not guaranteed by Unsafe, 2650 // and it is not possible to fully distinguish unintended nulls 2651 // from intended ones in this API. 2652 2653 if (!is_store) { 2654 Node* p = NULL; 2655 // Try to constant fold a load from a constant field 2656 2657 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) { 2658 // final or stable field 2659 p = make_constant_from_field(field, heap_base_oop); 2660 } 2661 2662 if (p == NULL) { // Could not constant fold the load 2663 if (type == T_VALUETYPE) { 2664 if (adr_type->isa_instptr() && !mismatched) { 2665 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass(); 2666 int offset = adr_type->is_instptr()->offset(); 2667 p = ValueTypeNode::make_from_flattened(this, value_klass, base, base, holder, offset, decorators); 2668 } else { 2669 p = ValueTypeNode::make_from_flattened(this, value_klass, base, adr, NULL, 0, decorators); 2670 } 2671 } else { 2672 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators); 2673 } 2674 // Normalize the value returned by getBoolean in the following cases 2675 if (type == T_BOOLEAN && 2676 (mismatched || 2677 heap_base_oop == top() || // - heap_base_oop is NULL or 2678 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL 2679 // and the unsafe access is made to large offset 2680 // (i.e., larger than the maximum offset necessary for any 2681 // field access) 2682 ) { 2683 IdealKit ideal = IdealKit(this); 2684 #define __ ideal. 2685 IdealVariable normalized_result(ideal); 2686 __ declarations_done(); 2687 __ set(normalized_result, p); 2688 __ if_then(p, BoolTest::ne, ideal.ConI(0)); 2689 __ set(normalized_result, ideal.ConI(1)); 2690 ideal.end_if(); 2691 final_sync(ideal); 2692 p = __ value(normalized_result); 2693 #undef __ 2694 } 2695 } 2696 if (type == T_ADDRESS) { 2697 p = gvn().transform(new CastP2XNode(NULL, p)); 2698 p = ConvX2UL(p); 2699 } 2700 if (field != NULL && field->is_flattenable() && !field->is_flattened()) { 2701 // Load a non-flattened but flattenable value type from memory 2702 if (value_type->value_klass()->is_scalarizable()) { 2703 p = ValueTypeNode::make_from_oop(this, p, value_type->value_klass()); 2704 } else { 2705 p = null2default(p, value_type->value_klass()); 2706 } 2707 } 2708 // The load node has the control of the preceding MemBarCPUOrder. All 2709 // following nodes will have the control of the MemBarCPUOrder inserted at 2710 // the end of this method. So, pushing the load onto the stack at a later 2711 // point is fine. 2712 set_result(p); 2713 } else { 2714 if (bt == T_ADDRESS) { 2715 // Repackage the long as a pointer. 2716 val = ConvL2X(val); 2717 val = gvn().transform(new CastX2PNode(val)); 2718 } 2719 if (type == T_VALUETYPE) { 2720 if (adr_type->isa_instptr() && !mismatched) { 2721 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass(); 2722 int offset = adr_type->is_instptr()->offset(); 2723 val->as_ValueType()->store_flattened(this, base, base, holder, offset, decorators); 2724 } else { 2725 val->as_ValueType()->store_flattened(this, base, adr, NULL, 0, decorators); 2726 } 2727 } else { 2728 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators); 2729 } 2730 } 2731 2732 if (argument(1)->is_ValueType() && is_store) { 2733 Node* value = ValueTypeNode::make_from_oop(this, base, _gvn.type(base)->value_klass()); 2734 value = value->as_ValueType()->make_larval(this, false); 2735 replace_in_map(argument(1), value); 2736 } 2737 2738 return true; 2739 } 2740 2741 bool LibraryCallKit::inline_unsafe_make_private_buffer() { 2742 Node* receiver = argument(0); 2743 Node* value = argument(1); 2744 2745 receiver = null_check(receiver); 2746 if (stopped()) { 2747 return true; 2748 } 2749 2750 if (!value->is_ValueType()) { 2751 return false; 2752 } 2753 2754 set_result(value->as_ValueType()->make_larval(this, true)); 2755 2756 return true; 2757 } 2758 2759 bool LibraryCallKit::inline_unsafe_finish_private_buffer() { 2760 Node* receiver = argument(0); 2761 Node* buffer = argument(1); 2762 2763 receiver = null_check(receiver); 2764 if (stopped()) { 2765 return true; 2766 } 2767 2768 if (!buffer->is_ValueType()) { 2769 return false; 2770 } 2771 2772 ValueTypeNode* vt = buffer->as_ValueType(); 2773 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) { 2774 return false; 2775 } 2776 2777 set_result(vt->finish_larval(this)); 2778 2779 return true; 2780 } 2781 2782 //----------------------------inline_unsafe_load_store---------------------------- 2783 // This method serves a couple of different customers (depending on LoadStoreKind): 2784 // 2785 // LS_cmp_swap: 2786 // 2787 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x); 2788 // boolean compareAndSetInt( Object o, long offset, int expected, int x); 2789 // boolean compareAndSetLong( Object o, long offset, long expected, long x); 2790 // 2791 // LS_cmp_swap_weak: 2792 // 2793 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x); 2794 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x); 2795 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x); 2796 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x); 2797 // 2798 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x); 2799 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x); 2800 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x); 2801 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x); 2802 // 2803 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x); 2804 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x); 2805 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x); 2806 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x); 2807 // 2808 // LS_cmp_exchange: 2809 // 2810 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x); 2811 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x); 2812 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x); 2813 // 2814 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2815 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2816 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2817 // 2818 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2819 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2820 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2821 // 2822 // LS_get_add: 2823 // 2824 // int getAndAddInt( Object o, long offset, int delta) 2825 // long getAndAddLong(Object o, long offset, long delta) 2826 // 2827 // LS_get_set: 2828 // 2829 // int getAndSet(Object o, long offset, int newValue) 2830 // long getAndSet(Object o, long offset, long newValue) 2831 // Object getAndSet(Object o, long offset, Object newValue) 2832 // 2833 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2834 // This basic scheme here is the same as inline_unsafe_access, but 2835 // differs in enough details that combining them would make the code 2836 // overly confusing. (This is a true fact! I originally combined 2837 // them, but even I was confused by it!) As much code/comments as 2838 // possible are retained from inline_unsafe_access though to make 2839 // the correspondences clearer. - dl 2840 2841 if (callee()->is_static()) return false; // caller must have the capability! 2842 2843 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2844 decorators |= mo_decorator_for_access_kind(access_kind); 2845 2846 #ifndef PRODUCT 2847 BasicType rtype; 2848 { 2849 ResourceMark rm; 2850 // Check the signatures. 2851 ciSignature* sig = callee()->signature(); 2852 rtype = sig->return_type()->basic_type(); 2853 switch(kind) { 2854 case LS_get_add: 2855 case LS_get_set: { 2856 // Check the signatures. 2857 #ifdef ASSERT 2858 assert(rtype == type, "get and set must return the expected type"); 2859 assert(sig->count() == 3, "get and set has 3 arguments"); 2860 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2861 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2862 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2863 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation"); 2864 #endif // ASSERT 2865 break; 2866 } 2867 case LS_cmp_swap: 2868 case LS_cmp_swap_weak: { 2869 // Check the signatures. 2870 #ifdef ASSERT 2871 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2872 assert(sig->count() == 4, "CAS has 4 arguments"); 2873 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2874 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2875 #endif // ASSERT 2876 break; 2877 } 2878 case LS_cmp_exchange: { 2879 // Check the signatures. 2880 #ifdef ASSERT 2881 assert(rtype == type, "CAS must return the expected type"); 2882 assert(sig->count() == 4, "CAS has 4 arguments"); 2883 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2884 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2885 #endif // ASSERT 2886 break; 2887 } 2888 default: 2889 ShouldNotReachHere(); 2890 } 2891 } 2892 #endif //PRODUCT 2893 2894 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2895 2896 // Get arguments: 2897 Node* receiver = NULL; 2898 Node* base = NULL; 2899 Node* offset = NULL; 2900 Node* oldval = NULL; 2901 Node* newval = NULL; 2902 switch(kind) { 2903 case LS_cmp_swap: 2904 case LS_cmp_swap_weak: 2905 case LS_cmp_exchange: { 2906 const bool two_slot_type = type2size[type] == 2; 2907 receiver = argument(0); // type: oop 2908 base = argument(1); // type: oop 2909 offset = argument(2); // type: long 2910 oldval = argument(4); // type: oop, int, or long 2911 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2912 break; 2913 } 2914 case LS_get_add: 2915 case LS_get_set: { 2916 receiver = argument(0); // type: oop 2917 base = argument(1); // type: oop 2918 offset = argument(2); // type: long 2919 oldval = NULL; 2920 newval = argument(4); // type: oop, int, or long 2921 break; 2922 } 2923 default: 2924 ShouldNotReachHere(); 2925 } 2926 2927 // Build field offset expression. 2928 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2929 // to be plain byte offsets, which are also the same as those accepted 2930 // by oopDesc::field_addr. 2931 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2932 // 32-bit machines ignore the high half of long offsets 2933 offset = ConvL2X(offset); 2934 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false); 2935 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2936 2937 Compile::AliasType* alias_type = C->alias_type(adr_type); 2938 BasicType bt = alias_type->basic_type(); 2939 if (bt != T_ILLEGAL && 2940 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) { 2941 // Don't intrinsify mismatched object accesses. 2942 return false; 2943 } 2944 2945 // For CAS, unlike inline_unsafe_access, there seems no point in 2946 // trying to refine types. Just use the coarse types here. 2947 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2948 const Type *value_type = Type::get_const_basic_type(type); 2949 2950 switch (kind) { 2951 case LS_get_set: 2952 case LS_cmp_exchange: { 2953 if (type == T_OBJECT) { 2954 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2955 if (tjp != NULL) { 2956 value_type = tjp; 2957 } 2958 } 2959 break; 2960 } 2961 case LS_cmp_swap: 2962 case LS_cmp_swap_weak: 2963 case LS_get_add: 2964 break; 2965 default: 2966 ShouldNotReachHere(); 2967 } 2968 2969 // Null check receiver. 2970 receiver = null_check(receiver); 2971 if (stopped()) { 2972 return true; 2973 } 2974 2975 int alias_idx = C->get_alias_index(adr_type); 2976 2977 if (type == T_OBJECT || type == T_ARRAY) { 2978 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF; 2979 2980 // Transformation of a value which could be NULL pointer (CastPP #NULL) 2981 // could be delayed during Parse (for example, in adjust_map_after_if()). 2982 // Execute transformation here to avoid barrier generation in such case. 2983 if (_gvn.type(newval) == TypePtr::NULL_PTR) 2984 newval = _gvn.makecon(TypePtr::NULL_PTR); 2985 2986 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) { 2987 // Refine the value to a null constant, when it is known to be null 2988 oldval = _gvn.makecon(TypePtr::NULL_PTR); 2989 } 2990 } 2991 2992 Node* result = NULL; 2993 switch (kind) { 2994 case LS_cmp_exchange: { 2995 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx, 2996 oldval, newval, value_type, type, decorators); 2997 break; 2998 } 2999 case LS_cmp_swap_weak: 3000 decorators |= C2_WEAK_CMPXCHG; 3001 case LS_cmp_swap: { 3002 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx, 3003 oldval, newval, value_type, type, decorators); 3004 break; 3005 } 3006 case LS_get_set: { 3007 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx, 3008 newval, value_type, type, decorators); 3009 break; 3010 } 3011 case LS_get_add: { 3012 result = access_atomic_add_at(base, adr, adr_type, alias_idx, 3013 newval, value_type, type, decorators); 3014 break; 3015 } 3016 default: 3017 ShouldNotReachHere(); 3018 } 3019 3020 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 3021 set_result(result); 3022 return true; 3023 } 3024 3025 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 3026 // Regardless of form, don't allow previous ld/st to move down, 3027 // then issue acquire, release, or volatile mem_bar. 3028 insert_mem_bar(Op_MemBarCPUOrder); 3029 switch(id) { 3030 case vmIntrinsics::_loadFence: 3031 insert_mem_bar(Op_LoadFence); 3032 return true; 3033 case vmIntrinsics::_storeFence: 3034 insert_mem_bar(Op_StoreFence); 3035 return true; 3036 case vmIntrinsics::_fullFence: 3037 insert_mem_bar(Op_MemBarVolatile); 3038 return true; 3039 default: 3040 fatal_unexpected_iid(id); 3041 return false; 3042 } 3043 } 3044 3045 bool LibraryCallKit::inline_onspinwait() { 3046 insert_mem_bar(Op_OnSpinWait); 3047 return true; 3048 } 3049 3050 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 3051 if (!kls->is_Con()) { 3052 return true; 3053 } 3054 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr(); 3055 if (klsptr == NULL) { 3056 return true; 3057 } 3058 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass(); 3059 // don't need a guard for a klass that is already initialized 3060 return !ik->is_initialized(); 3061 } 3062 3063 //----------------------------inline_unsafe_writeback0------------------------- 3064 // public native void Unsafe.writeback0(long address) 3065 bool LibraryCallKit::inline_unsafe_writeback0() { 3066 if (!Matcher::has_match_rule(Op_CacheWB)) { 3067 return false; 3068 } 3069 #ifndef PRODUCT 3070 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync"); 3071 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync"); 3072 ciSignature* sig = callee()->signature(); 3073 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!"); 3074 #endif 3075 null_check_receiver(); // null-check, then ignore 3076 Node *addr = argument(1); 3077 addr = new CastX2PNode(addr); 3078 addr = _gvn.transform(addr); 3079 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr); 3080 flush = _gvn.transform(flush); 3081 set_memory(flush, TypeRawPtr::BOTTOM); 3082 return true; 3083 } 3084 3085 //----------------------------inline_unsafe_writeback0------------------------- 3086 // public native void Unsafe.writeback0(long address) 3087 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) { 3088 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) { 3089 return false; 3090 } 3091 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) { 3092 return false; 3093 } 3094 #ifndef PRODUCT 3095 assert(Matcher::has_match_rule(Op_CacheWB), 3096 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB" 3097 : "found match rule for CacheWBPostSync but not CacheWB")); 3098 3099 #endif 3100 null_check_receiver(); // null-check, then ignore 3101 Node *sync; 3102 if (is_pre) { 3103 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 3104 } else { 3105 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 3106 } 3107 sync = _gvn.transform(sync); 3108 set_memory(sync, TypeRawPtr::BOTTOM); 3109 return true; 3110 } 3111 3112 //----------------------------inline_unsafe_allocate--------------------------- 3113 // public native Object Unsafe.allocateInstance(Class<?> cls); 3114 bool LibraryCallKit::inline_unsafe_allocate() { 3115 if (callee()->is_static()) return false; // caller must have the capability! 3116 3117 null_check_receiver(); // null-check, then ignore 3118 Node* cls = null_check(argument(1)); 3119 if (stopped()) return true; 3120 3121 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 3122 kls = null_check(kls); 3123 if (stopped()) return true; // argument was like int.class 3124 3125 Node* test = NULL; 3126 if (LibraryCallKit::klass_needs_init_guard(kls)) { 3127 // Note: The argument might still be an illegal value like 3128 // Serializable.class or Object[].class. The runtime will handle it. 3129 // But we must make an explicit check for initialization. 3130 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 3131 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 3132 // can generate code to load it as unsigned byte. 3133 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered); 3134 Node* bits = intcon(InstanceKlass::fully_initialized); 3135 test = _gvn.transform(new SubINode(inst, bits)); 3136 // The 'test' is non-zero if we need to take a slow path. 3137 } 3138 3139 Node* obj = new_instance(kls, test); 3140 set_result(obj); 3141 return true; 3142 } 3143 3144 //------------------------inline_native_time_funcs-------------- 3145 // inline code for System.currentTimeMillis() and System.nanoTime() 3146 // these have the same type and signature 3147 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 3148 const TypeFunc* tf = OptoRuntime::void_long_Type(); 3149 const TypePtr* no_memory_effects = NULL; 3150 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 3151 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 3152 #ifdef ASSERT 3153 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 3154 assert(value_top == top(), "second value must be top"); 3155 #endif 3156 set_result(value); 3157 return true; 3158 } 3159 3160 #ifdef JFR_HAVE_INTRINSICS 3161 3162 /* 3163 * oop -> myklass 3164 * myklass->trace_id |= USED 3165 * return myklass->trace_id & ~0x3 3166 */ 3167 bool LibraryCallKit::inline_native_classID() { 3168 Node* cls = null_check(argument(0), T_OBJECT); 3169 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 3170 kls = null_check(kls, T_OBJECT); 3171 3172 ByteSize offset = KLASS_TRACE_ID_OFFSET; 3173 Node* insp = basic_plus_adr(kls, in_bytes(offset)); 3174 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered); 3175 3176 Node* clsused = longcon(0x01l); // set the class bit 3177 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused)); 3178 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr(); 3179 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered); 3180 3181 #ifdef TRACE_ID_META_BITS 3182 Node* mbits = longcon(~TRACE_ID_META_BITS); 3183 tvalue = _gvn.transform(new AndLNode(tvalue, mbits)); 3184 #endif 3185 #ifdef TRACE_ID_SHIFT 3186 Node* cbits = intcon(TRACE_ID_SHIFT); 3187 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits)); 3188 #endif 3189 3190 set_result(tvalue); 3191 return true; 3192 3193 } 3194 3195 bool LibraryCallKit::inline_native_getEventWriter() { 3196 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 3197 3198 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, 3199 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)); 3200 3201 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered); 3202 3203 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) ); 3204 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) ); 3205 3206 IfNode* iff_jobj_null = 3207 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN); 3208 3209 enum { _normal_path = 1, 3210 _null_path = 2, 3211 PATH_LIMIT }; 3212 3213 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3214 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM); 3215 3216 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null)); 3217 result_rgn->init_req(_null_path, jobj_is_null); 3218 result_val->init_req(_null_path, null()); 3219 3220 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null)); 3221 set_control(jobj_is_not_null); 3222 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT, 3223 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD); 3224 result_rgn->init_req(_normal_path, control()); 3225 result_val->init_req(_normal_path, res); 3226 3227 set_result(result_rgn, result_val); 3228 3229 return true; 3230 } 3231 3232 #endif // JFR_HAVE_INTRINSICS 3233 3234 //------------------------inline_native_currentThread------------------ 3235 bool LibraryCallKit::inline_native_currentThread() { 3236 Node* junk = NULL; 3237 set_result(generate_current_thread(junk)); 3238 return true; 3239 } 3240 3241 //------------------------inline_native_isInterrupted------------------ 3242 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted); 3243 bool LibraryCallKit::inline_native_isInterrupted() { 3244 // Add a fast path to t.isInterrupted(clear_int): 3245 // (t == Thread.current() && 3246 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int))) 3247 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int) 3248 // So, in the common case that the interrupt bit is false, 3249 // we avoid making a call into the VM. Even if the interrupt bit 3250 // is true, if the clear_int argument is false, we avoid the VM call. 3251 // However, if the receiver is not currentThread, we must call the VM, 3252 // because there must be some locking done around the operation. 3253 3254 // We only go to the fast case code if we pass two guards. 3255 // Paths which do not pass are accumulated in the slow_region. 3256 3257 enum { 3258 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted 3259 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int 3260 slow_result_path = 3, // slow path: t.isInterrupted(clear_int) 3261 PATH_LIMIT 3262 }; 3263 3264 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag 3265 // out of the function. 3266 insert_mem_bar(Op_MemBarCPUOrder); 3267 3268 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3269 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL); 3270 3271 RegionNode* slow_region = new RegionNode(1); 3272 record_for_igvn(slow_region); 3273 3274 // (a) Receiving thread must be the current thread. 3275 Node* rec_thr = argument(0); 3276 Node* tls_ptr = NULL; 3277 Node* cur_thr = generate_current_thread(tls_ptr); 3278 3279 // Resolve oops to stable for CmpP below. 3280 cur_thr = access_resolve(cur_thr, 0); 3281 rec_thr = access_resolve(rec_thr, 0); 3282 3283 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr)); 3284 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne)); 3285 3286 generate_slow_guard(bol_thr, slow_region); 3287 3288 // (b) Interrupt bit on TLS must be false. 3289 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset())); 3290 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3291 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset())); 3292 3293 // Set the control input on the field _interrupted read to prevent it floating up. 3294 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered); 3295 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0))); 3296 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne)); 3297 3298 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 3299 3300 // First fast path: if (!TLS._interrupted) return false; 3301 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit)); 3302 result_rgn->init_req(no_int_result_path, false_bit); 3303 result_val->init_req(no_int_result_path, intcon(0)); 3304 3305 // drop through to next case 3306 set_control( _gvn.transform(new IfTrueNode(iff_bit))); 3307 3308 #ifndef _WINDOWS 3309 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path. 3310 Node* clr_arg = argument(1); 3311 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0))); 3312 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne)); 3313 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN); 3314 3315 // Second fast path: ... else if (!clear_int) return true; 3316 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg)); 3317 result_rgn->init_req(no_clear_result_path, false_arg); 3318 result_val->init_req(no_clear_result_path, intcon(1)); 3319 3320 // drop through to next case 3321 set_control( _gvn.transform(new IfTrueNode(iff_arg))); 3322 #else 3323 // To return true on Windows you must read the _interrupted field 3324 // and check the event state i.e. take the slow path. 3325 #endif // _WINDOWS 3326 3327 // (d) Otherwise, go to the slow path. 3328 slow_region->add_req(control()); 3329 set_control( _gvn.transform(slow_region)); 3330 3331 if (stopped()) { 3332 // There is no slow path. 3333 result_rgn->init_req(slow_result_path, top()); 3334 result_val->init_req(slow_result_path, top()); 3335 } else { 3336 // non-virtual because it is a private non-static 3337 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted); 3338 3339 Node* slow_val = set_results_for_java_call(slow_call); 3340 // this->control() comes from set_results_for_java_call 3341 3342 Node* fast_io = slow_call->in(TypeFunc::I_O); 3343 Node* fast_mem = slow_call->in(TypeFunc::Memory); 3344 3345 // These two phis are pre-filled with copies of of the fast IO and Memory 3346 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM); 3347 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO); 3348 3349 result_rgn->init_req(slow_result_path, control()); 3350 result_io ->init_req(slow_result_path, i_o()); 3351 result_mem->init_req(slow_result_path, reset_memory()); 3352 result_val->init_req(slow_result_path, slow_val); 3353 3354 set_all_memory(_gvn.transform(result_mem)); 3355 set_i_o( _gvn.transform(result_io)); 3356 } 3357 3358 C->set_has_split_ifs(true); // Has chance for split-if optimization 3359 set_result(result_rgn, result_val); 3360 return true; 3361 } 3362 3363 //-----------------------load_klass_from_mirror_common------------------------- 3364 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3365 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3366 // and branch to the given path on the region. 3367 // If never_see_null, take an uncommon trap on null, so we can optimistically 3368 // compile for the non-null case. 3369 // If the region is NULL, force never_see_null = true. 3370 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3371 bool never_see_null, 3372 RegionNode* region, 3373 int null_path, 3374 int offset) { 3375 if (region == NULL) never_see_null = true; 3376 Node* p = basic_plus_adr(mirror, offset); 3377 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3378 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3379 Node* null_ctl = top(); 3380 kls = null_check_oop(kls, &null_ctl, never_see_null); 3381 if (region != NULL) { 3382 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3383 region->init_req(null_path, null_ctl); 3384 } else { 3385 assert(null_ctl == top(), "no loose ends"); 3386 } 3387 return kls; 3388 } 3389 3390 //--------------------(inline_native_Class_query helpers)--------------------- 3391 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER. 3392 // Fall through if (mods & mask) == bits, take the guard otherwise. 3393 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 3394 // Branch around if the given klass has the given modifier bit set. 3395 // Like generate_guard, adds a new path onto the region. 3396 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3397 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered); 3398 Node* mask = intcon(modifier_mask); 3399 Node* bits = intcon(modifier_bits); 3400 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3401 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3402 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3403 return generate_fair_guard(bol, region); 3404 } 3405 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3406 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region); 3407 } 3408 3409 Node* LibraryCallKit::generate_value_guard(Node* kls, RegionNode* region) { 3410 return generate_access_flags_guard(kls, JVM_ACC_VALUE, 0, region); 3411 } 3412 3413 //-------------------------inline_native_Class_query------------------- 3414 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 3415 const Type* return_type = TypeInt::BOOL; 3416 Node* prim_return_value = top(); // what happens if it's a primitive class? 3417 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3418 bool expect_prim = false; // most of these guys expect to work on refs 3419 3420 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 3421 3422 Node* mirror = argument(0); 3423 Node* obj = top(); 3424 3425 switch (id) { 3426 case vmIntrinsics::_isInstance: 3427 // nothing is an instance of a primitive type 3428 prim_return_value = intcon(0); 3429 obj = argument(1); 3430 break; 3431 case vmIntrinsics::_getModifiers: 3432 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3433 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line"); 3434 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin); 3435 break; 3436 case vmIntrinsics::_isInterface: 3437 prim_return_value = intcon(0); 3438 break; 3439 case vmIntrinsics::_isArray: 3440 prim_return_value = intcon(0); 3441 expect_prim = true; // cf. ObjectStreamClass.getClassSignature 3442 break; 3443 case vmIntrinsics::_isPrimitive: 3444 prim_return_value = intcon(1); 3445 expect_prim = true; // obviously 3446 break; 3447 case vmIntrinsics::_getSuperclass: 3448 prim_return_value = null(); 3449 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 3450 break; 3451 case vmIntrinsics::_getClassAccessFlags: 3452 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3453 return_type = TypeInt::INT; // not bool! 6297094 3454 break; 3455 default: 3456 fatal_unexpected_iid(id); 3457 break; 3458 } 3459 3460 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3461 if (mirror_con == NULL) return false; // cannot happen? 3462 3463 #ifndef PRODUCT 3464 if (C->print_intrinsics() || C->print_inlining()) { 3465 ciType* k = mirror_con->java_mirror_type(); 3466 if (k) { 3467 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 3468 k->print_name(); 3469 tty->cr(); 3470 } 3471 } 3472 #endif 3473 3474 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 3475 RegionNode* region = new RegionNode(PATH_LIMIT); 3476 record_for_igvn(region); 3477 PhiNode* phi = new PhiNode(region, return_type); 3478 3479 // The mirror will never be null of Reflection.getClassAccessFlags, however 3480 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 3481 // if it is. See bug 4774291. 3482 3483 // For Reflection.getClassAccessFlags(), the null check occurs in 3484 // the wrong place; see inline_unsafe_access(), above, for a similar 3485 // situation. 3486 mirror = null_check(mirror); 3487 // If mirror or obj is dead, only null-path is taken. 3488 if (stopped()) return true; 3489 3490 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 3491 3492 // Now load the mirror's klass metaobject, and null-check it. 3493 // Side-effects region with the control path if the klass is null. 3494 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 3495 // If kls is null, we have a primitive mirror. 3496 phi->init_req(_prim_path, prim_return_value); 3497 if (stopped()) { set_result(region, phi); return true; } 3498 bool safe_for_replace = (region->in(_prim_path) == top()); 3499 3500 Node* p; // handy temp 3501 Node* null_ctl; 3502 3503 // Now that we have the non-null klass, we can perform the real query. 3504 // For constant classes, the query will constant-fold in LoadNode::Value. 3505 Node* query_value = top(); 3506 switch (id) { 3507 case vmIntrinsics::_isInstance: 3508 // nothing is an instance of a primitive type 3509 query_value = gen_instanceof(obj, kls, safe_for_replace); 3510 break; 3511 3512 case vmIntrinsics::_getModifiers: 3513 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset())); 3514 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3515 break; 3516 3517 case vmIntrinsics::_isInterface: 3518 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3519 if (generate_interface_guard(kls, region) != NULL) 3520 // A guard was added. If the guard is taken, it was an interface. 3521 phi->add_req(intcon(1)); 3522 // If we fall through, it's a plain class. 3523 query_value = intcon(0); 3524 break; 3525 3526 case vmIntrinsics::_isArray: 3527 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.) 3528 if (generate_array_guard(kls, region) != NULL) 3529 // A guard was added. If the guard is taken, it was an array. 3530 phi->add_req(intcon(1)); 3531 // If we fall through, it's a plain class. 3532 query_value = intcon(0); 3533 break; 3534 3535 case vmIntrinsics::_isPrimitive: 3536 query_value = intcon(0); // "normal" path produces false 3537 break; 3538 3539 case vmIntrinsics::_getSuperclass: 3540 // The rules here are somewhat unfortunate, but we can still do better 3541 // with random logic than with a JNI call. 3542 // Interfaces store null or Object as _super, but must report null. 3543 // Arrays store an intermediate super as _super, but must report Object. 3544 // Other types can report the actual _super. 3545 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3546 if (generate_interface_guard(kls, region) != NULL) 3547 // A guard was added. If the guard is taken, it was an interface. 3548 phi->add_req(null()); 3549 if (generate_array_guard(kls, region) != NULL) 3550 // A guard was added. If the guard is taken, it was an array. 3551 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 3552 // If we fall through, it's a plain class. Get its _super. 3553 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 3554 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL)); 3555 null_ctl = top(); 3556 kls = null_check_oop(kls, &null_ctl); 3557 if (null_ctl != top()) { 3558 // If the guard is taken, Object.superClass is null (both klass and mirror). 3559 region->add_req(null_ctl); 3560 phi ->add_req(null()); 3561 } 3562 if (!stopped()) { 3563 query_value = load_mirror_from_klass(kls); 3564 } 3565 break; 3566 3567 case vmIntrinsics::_getClassAccessFlags: 3568 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3569 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3570 break; 3571 3572 default: 3573 fatal_unexpected_iid(id); 3574 break; 3575 } 3576 3577 // Fall-through is the normal case of a query to a real class. 3578 phi->init_req(1, query_value); 3579 region->init_req(1, control()); 3580 3581 C->set_has_split_ifs(true); // Has chance for split-if optimization 3582 set_result(region, phi); 3583 return true; 3584 } 3585 3586 //-------------------------inline_value_Class_conversion------------------- 3587 // public Class<T> java.lang.Class.asPrimaryType(); 3588 // public Class<T> java.lang.Class.asIndirectType() 3589 bool LibraryCallKit::inline_value_Class_conversion(vmIntrinsics::ID id) { 3590 Node* mirror = argument(0); // Receiver Class 3591 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3592 if (mirror_con == NULL) { 3593 return false; 3594 } 3595 3596 bool is_indirect_type = true; 3597 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type); 3598 if (tm != NULL) { 3599 Node* result = mirror; 3600 if (tm->is_valuetype()) { 3601 if (id == vmIntrinsics::_asPrimaryType && is_indirect_type) { 3602 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->inline_mirror_instance())); 3603 } else if (id == vmIntrinsics::_asIndirectType && !is_indirect_type) { 3604 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->indirect_mirror_instance())); 3605 } 3606 } 3607 set_result(result); 3608 return true; 3609 } 3610 return false; 3611 } 3612 3613 //-------------------------inline_Class_cast------------------- 3614 bool LibraryCallKit::inline_Class_cast() { 3615 Node* mirror = argument(0); // Class 3616 Node* obj = argument(1); 3617 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3618 if (mirror_con == NULL) { 3619 return false; // dead path (mirror->is_top()). 3620 } 3621 if (obj == NULL || obj->is_top()) { 3622 return false; // dead path 3623 } 3624 3625 ciKlass* obj_klass = NULL; 3626 if (obj->is_ValueType()) { 3627 obj_klass = _gvn.type(obj)->value_klass(); 3628 } else { 3629 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr(); 3630 if (tp != NULL) { 3631 obj_klass = tp->klass(); 3632 } 3633 } 3634 3635 // First, see if Class.cast() can be folded statically. 3636 // java_mirror_type() returns non-null for compile-time Class constants. 3637 bool is_indirect_type = true; 3638 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type); 3639 if (!obj->is_ValueType() && !is_indirect_type) { 3640 obj = null_check(obj); 3641 if (stopped()) { 3642 return true; 3643 } 3644 } 3645 if (tm != NULL && tm->is_klass() && obj_klass != NULL) { 3646 if (!obj_klass->is_loaded()) { 3647 // Don't use intrinsic when class is not loaded. 3648 return false; 3649 } else { 3650 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass); 3651 if (static_res == Compile::SSC_always_true) { 3652 // isInstance() is true - fold the code. 3653 set_result(obj); 3654 return true; 3655 } else if (static_res == Compile::SSC_always_false) { 3656 // Don't use intrinsic, have to throw ClassCastException. 3657 // If the reference is null, the non-intrinsic bytecode will 3658 // be optimized appropriately. 3659 return false; 3660 } 3661 } 3662 } 3663 3664 // Bailout intrinsic and do normal inlining if exception path is frequent. 3665 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 3666 return false; 3667 } 3668 3669 // Generate dynamic checks. 3670 // Class.cast() is java implementation of _checkcast bytecode. 3671 // Do checkcast (Parse::do_checkcast()) optimizations here. 3672 3673 mirror = null_check(mirror); 3674 // If mirror is dead, only null-path is taken. 3675 if (stopped()) { 3676 return true; 3677 } 3678 3679 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive). 3680 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT }; 3681 RegionNode* region = new RegionNode(PATH_LIMIT); 3682 record_for_igvn(region); 3683 3684 // Now load the mirror's klass metaobject, and null-check it. 3685 // If kls is null, we have a primitive mirror and 3686 // nothing is an instance of a primitive type. 3687 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 3688 3689 Node* res = top(); 3690 if (!stopped()) { 3691 if (EnableValhalla && !obj->is_ValueType() && is_indirect_type) { 3692 // Check if (mirror == inline_mirror && obj == null) 3693 Node* is_val_mirror = generate_fair_guard(is_value_mirror(mirror), NULL); 3694 if (is_val_mirror != NULL) { 3695 RegionNode* r = new RegionNode(3); 3696 record_for_igvn(r); 3697 r->init_req(1, control()); 3698 3699 // Casting to .val, check for null 3700 set_control(is_val_mirror); 3701 Node *null_ctr = top(); 3702 null_check_oop(obj, &null_ctr); 3703 region->init_req(_npe_path, null_ctr); 3704 r->init_req(2, control()); 3705 3706 set_control(_gvn.transform(r)); 3707 } 3708 } 3709 3710 Node* bad_type_ctrl = top(); 3711 // Do checkcast optimizations. 3712 res = gen_checkcast(obj, kls, &bad_type_ctrl); 3713 region->init_req(_bad_type_path, bad_type_ctrl); 3714 } 3715 if (region->in(_prim_path) != top() || 3716 region->in(_bad_type_path) != top() || 3717 region->in(_npe_path) != top()) { 3718 // Let Interpreter throw ClassCastException. 3719 PreserveJVMState pjvms(this); 3720 set_control(_gvn.transform(region)); 3721 uncommon_trap(Deoptimization::Reason_intrinsic, 3722 Deoptimization::Action_maybe_recompile); 3723 } 3724 if (!stopped()) { 3725 set_result(res); 3726 } 3727 return true; 3728 } 3729 3730 3731 //--------------------------inline_native_subtype_check------------------------ 3732 // This intrinsic takes the JNI calls out of the heart of 3733 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 3734 bool LibraryCallKit::inline_native_subtype_check() { 3735 // Pull both arguments off the stack. 3736 Node* args[2]; // two java.lang.Class mirrors: superc, subc 3737 args[0] = argument(0); 3738 args[1] = argument(1); 3739 Node* klasses[2]; // corresponding Klasses: superk, subk 3740 klasses[0] = klasses[1] = top(); 3741 3742 enum { 3743 // A full decision tree on {superc is prim, subc is prim}: 3744 _prim_0_path = 1, // {P,N} => false 3745 // {P,P} & superc!=subc => false 3746 _prim_same_path, // {P,P} & superc==subc => true 3747 _prim_1_path, // {N,P} => false 3748 _ref_subtype_path, // {N,N} & subtype check wins => true 3749 _both_ref_path, // {N,N} & subtype check loses => false 3750 PATH_LIMIT 3751 }; 3752 3753 RegionNode* region = new RegionNode(PATH_LIMIT); 3754 RegionNode* prim_region = new RegionNode(2); 3755 Node* phi = new PhiNode(region, TypeInt::BOOL); 3756 record_for_igvn(region); 3757 record_for_igvn(prim_region); 3758 3759 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 3760 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3761 int class_klass_offset = java_lang_Class::klass_offset_in_bytes(); 3762 3763 // First null-check both mirrors and load each mirror's klass metaobject. 3764 int which_arg; 3765 for (which_arg = 0; which_arg <= 1; which_arg++) { 3766 Node* arg = args[which_arg]; 3767 arg = null_check(arg); 3768 if (stopped()) break; 3769 args[which_arg] = arg; 3770 3771 Node* p = basic_plus_adr(arg, class_klass_offset); 3772 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type); 3773 klasses[which_arg] = _gvn.transform(kls); 3774 } 3775 3776 // Resolve oops to stable for CmpP below. 3777 args[0] = access_resolve(args[0], 0); 3778 args[1] = access_resolve(args[1], 0); 3779 3780 // Having loaded both klasses, test each for null. 3781 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3782 for (which_arg = 0; which_arg <= 1; which_arg++) { 3783 Node* kls = klasses[which_arg]; 3784 Node* null_ctl = top(); 3785 kls = null_check_oop(kls, &null_ctl, never_see_null); 3786 if (which_arg == 0) { 3787 prim_region->init_req(1, null_ctl); 3788 } else { 3789 region->init_req(_prim_1_path, null_ctl); 3790 } 3791 if (stopped()) break; 3792 klasses[which_arg] = kls; 3793 } 3794 3795 if (!stopped()) { 3796 // now we have two reference types, in klasses[0..1] 3797 Node* subk = klasses[1]; // the argument to isAssignableFrom 3798 Node* superk = klasses[0]; // the receiver 3799 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 3800 // If superc is a value mirror, we also need to check if superc == subc because 3801 // V? is not a subtype of V but due to subk == superk the subtype check will pass. 3802 generate_fair_guard(is_value_mirror(args[0]), prim_region); 3803 // now we have a successful reference subtype check 3804 region->set_req(_ref_subtype_path, control()); 3805 } 3806 3807 // If both operands are primitive (both klasses null), then 3808 // we must return true when they are identical primitives. 3809 // It is convenient to test this after the first null klass check. 3810 // This path is also used if superc is a value mirror. 3811 set_control(_gvn.transform(prim_region)); 3812 if (!stopped()) { 3813 // Since superc is primitive, make a guard for the superc==subc case. 3814 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 3815 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 3816 generate_fair_guard(bol_eq, region); 3817 if (region->req() == PATH_LIMIT+1) { 3818 // A guard was added. If the added guard is taken, superc==subc. 3819 region->swap_edges(PATH_LIMIT, _prim_same_path); 3820 region->del_req(PATH_LIMIT); 3821 } 3822 region->set_req(_prim_0_path, control()); // Not equal after all. 3823 } 3824 3825 // these are the only paths that produce 'true': 3826 phi->set_req(_prim_same_path, intcon(1)); 3827 phi->set_req(_ref_subtype_path, intcon(1)); 3828 3829 // pull together the cases: 3830 assert(region->req() == PATH_LIMIT, "sane region"); 3831 for (uint i = 1; i < region->req(); i++) { 3832 Node* ctl = region->in(i); 3833 if (ctl == NULL || ctl == top()) { 3834 region->set_req(i, top()); 3835 phi ->set_req(i, top()); 3836 } else if (phi->in(i) == NULL) { 3837 phi->set_req(i, intcon(0)); // all other paths produce 'false' 3838 } 3839 } 3840 3841 set_control(_gvn.transform(region)); 3842 set_result(_gvn.transform(phi)); 3843 return true; 3844 } 3845 3846 //---------------------generate_array_guard_common------------------------ 3847 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) { 3848 3849 if (stopped()) { 3850 return NULL; 3851 } 3852 3853 // Like generate_guard, adds a new path onto the region. 3854 jint layout_con = 0; 3855 Node* layout_val = get_layout_helper(kls, layout_con); 3856 if (layout_val == NULL) { 3857 bool query = 0; 3858 switch(kind) { 3859 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break; 3860 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break; 3861 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break; 3862 case ValueArray: query = Klass::layout_helper_is_valueArray(layout_con); break; 3863 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break; 3864 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break; 3865 default: 3866 ShouldNotReachHere(); 3867 } 3868 if (!query) { 3869 return NULL; // never a branch 3870 } else { // always a branch 3871 Node* always_branch = control(); 3872 if (region != NULL) 3873 region->add_req(always_branch); 3874 set_control(top()); 3875 return always_branch; 3876 } 3877 } 3878 unsigned int value = 0; 3879 BoolTest::mask btest = BoolTest::illegal; 3880 switch(kind) { 3881 case ObjectArray: 3882 case NonObjectArray: { 3883 value = Klass::_lh_array_tag_obj_value; 3884 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 3885 btest = kind == ObjectArray ? BoolTest::eq : BoolTest::ne; 3886 break; 3887 } 3888 case TypeArray: { 3889 value = Klass::_lh_array_tag_type_value; 3890 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 3891 btest = BoolTest::eq; 3892 break; 3893 } 3894 case ValueArray: { 3895 value = Klass::_lh_array_tag_vt_value; 3896 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift))); 3897 btest = BoolTest::eq; 3898 break; 3899 } 3900 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break; 3901 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break; 3902 default: 3903 ShouldNotReachHere(); 3904 } 3905 // Now test the correct condition. 3906 jint nval = (jint)value; 3907 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 3908 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 3909 return generate_fair_guard(bol, region); 3910 } 3911 3912 3913 //-----------------------inline_native_newArray-------------------------- 3914 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length); 3915 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size); 3916 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) { 3917 Node* mirror; 3918 Node* count_val; 3919 if (uninitialized) { 3920 mirror = argument(1); 3921 count_val = argument(2); 3922 } else { 3923 mirror = argument(0); 3924 count_val = argument(1); 3925 } 3926 3927 mirror = null_check(mirror); 3928 // If mirror or obj is dead, only null-path is taken. 3929 if (stopped()) return true; 3930 3931 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 3932 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 3933 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 3934 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 3935 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 3936 3937 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3938 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 3939 result_reg, _slow_path); 3940 Node* normal_ctl = control(); 3941 Node* no_array_ctl = result_reg->in(_slow_path); 3942 3943 // Generate code for the slow case. We make a call to newArray(). 3944 set_control(no_array_ctl); 3945 if (!stopped()) { 3946 // Either the input type is void.class, or else the 3947 // array klass has not yet been cached. Either the 3948 // ensuing call will throw an exception, or else it 3949 // will cache the array klass for next time. 3950 PreserveJVMState pjvms(this); 3951 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray); 3952 Node* slow_result = set_results_for_java_call(slow_call); 3953 // this->control() comes from set_results_for_java_call 3954 result_reg->set_req(_slow_path, control()); 3955 result_val->set_req(_slow_path, slow_result); 3956 result_io ->set_req(_slow_path, i_o()); 3957 result_mem->set_req(_slow_path, reset_memory()); 3958 } 3959 3960 set_control(normal_ctl); 3961 if (!stopped()) { 3962 // Normal case: The array type has been cached in the java.lang.Class. 3963 // The following call works fine even if the array type is polymorphic. 3964 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3965 Node* obj = new_array(klass_node, count_val, 0, NULL, false, mirror); // no arguments to push 3966 result_reg->init_req(_normal_path, control()); 3967 result_val->init_req(_normal_path, obj); 3968 result_io ->init_req(_normal_path, i_o()); 3969 result_mem->init_req(_normal_path, reset_memory()); 3970 3971 if (uninitialized) { 3972 // Mark the allocation so that zeroing is skipped 3973 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn); 3974 alloc->maybe_set_complete(&_gvn); 3975 } 3976 } 3977 3978 // Return the combined state. 3979 set_i_o( _gvn.transform(result_io) ); 3980 set_all_memory( _gvn.transform(result_mem)); 3981 3982 C->set_has_split_ifs(true); // Has chance for split-if optimization 3983 set_result(result_reg, result_val); 3984 return true; 3985 } 3986 3987 //----------------------inline_native_getLength-------------------------- 3988 // public static native int java.lang.reflect.Array.getLength(Object array); 3989 bool LibraryCallKit::inline_native_getLength() { 3990 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3991 3992 Node* array = null_check(argument(0)); 3993 // If array is dead, only null-path is taken. 3994 if (stopped()) return true; 3995 3996 // Deoptimize if it is a non-array. 3997 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL); 3998 3999 if (non_array != NULL) { 4000 PreserveJVMState pjvms(this); 4001 set_control(non_array); 4002 uncommon_trap(Deoptimization::Reason_intrinsic, 4003 Deoptimization::Action_maybe_recompile); 4004 } 4005 4006 // If control is dead, only non-array-path is taken. 4007 if (stopped()) return true; 4008 4009 // The works fine even if the array type is polymorphic. 4010 // It could be a dynamic mix of int[], boolean[], Object[], etc. 4011 Node* result = load_array_length(array); 4012 4013 C->set_has_split_ifs(true); // Has chance for split-if optimization 4014 set_result(result); 4015 return true; 4016 } 4017 4018 //------------------------inline_array_copyOf---------------------------- 4019 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 4020 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 4021 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 4022 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 4023 4024 // Get the arguments. 4025 Node* original = argument(0); 4026 Node* start = is_copyOfRange? argument(1): intcon(0); 4027 Node* end = is_copyOfRange? argument(2): argument(1); 4028 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 4029 4030 const TypeAryPtr* original_t = _gvn.type(original)->isa_aryptr(); 4031 const TypeInstPtr* mirror_t = _gvn.type(array_type_mirror)->isa_instptr(); 4032 if (EnableValhalla && ValueArrayFlatten && 4033 (original_t == NULL || mirror_t == NULL || 4034 (mirror_t->java_mirror_type() == NULL && 4035 (original_t->elem()->isa_valuetype() || 4036 (original_t->elem()->make_oopptr() != NULL && 4037 original_t->elem()->make_oopptr()->can_be_value_type()))))) { 4038 // We need to know statically if the copy is to a flattened array 4039 // or not but can't tell. 4040 return false; 4041 } 4042 4043 Node* newcopy = NULL; 4044 4045 // Set the original stack and the reexecute bit for the interpreter to reexecute 4046 // the bytecode that invokes Arrays.copyOf if deoptimization happens. 4047 { PreserveReexecuteState preexecs(this); 4048 jvms()->set_should_reexecute(true); 4049 4050 array_type_mirror = null_check(array_type_mirror); 4051 original = null_check(original); 4052 4053 // Check if a null path was taken unconditionally. 4054 if (stopped()) return true; 4055 4056 Node* orig_length = load_array_length(original); 4057 4058 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0); 4059 klass_node = null_check(klass_node); 4060 4061 RegionNode* bailout = new RegionNode(1); 4062 record_for_igvn(bailout); 4063 4064 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc. 4065 // Bail out if that is so. 4066 // Value type array may have object field that would require a 4067 // write barrier. Conservatively, go to slow path. 4068 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4069 Node* not_objArray = !bs->array_copy_requires_gc_barriers(false, T_OBJECT, false, BarrierSetC2::Parsing) ? 4070 generate_typeArray_guard(klass_node, bailout) : generate_non_objArray_guard(klass_node, bailout); 4071 if (not_objArray != NULL) { 4072 // Improve the klass node's type from the new optimistic assumption: 4073 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 4074 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), false); 4075 Node* cast = new CastPPNode(klass_node, akls); 4076 cast->init_req(0, control()); 4077 klass_node = _gvn.transform(cast); 4078 } 4079 4080 Node* original_kls = load_object_klass(original); 4081 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 4082 // loads/stores but it is legal only if we're sure the 4083 // Arrays.copyOf would succeed. So we need all input arguments 4084 // to the copyOf to be validated, including that the copy to the 4085 // new array won't trigger an ArrayStoreException. That subtype 4086 // check can be optimized if we know something on the type of 4087 // the input array from type speculation. 4088 if (_gvn.type(klass_node)->singleton() && !stopped()) { 4089 ciKlass* subk = _gvn.type(original_kls)->is_klassptr()->klass(); 4090 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass(); 4091 4092 int test = C->static_subtype_check(superk, subk); 4093 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 4094 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 4095 if (t_original->speculative_type() != NULL) { 4096 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 4097 original_kls = load_object_klass(original); 4098 } 4099 } 4100 } 4101 4102 if (ValueArrayFlatten) { 4103 // Either both or neither new array klass and original array 4104 // klass must be flattened 4105 Node* is_flat = generate_valueArray_guard(klass_node, NULL); 4106 if (!original_t->is_not_flat()) { 4107 generate_valueArray_guard(original_kls, bailout); 4108 } 4109 if (is_flat != NULL) { 4110 RegionNode* r = new RegionNode(2); 4111 record_for_igvn(r); 4112 r->init_req(1, control()); 4113 set_control(is_flat); 4114 if (!original_t->is_not_flat()) { 4115 generate_valueArray_guard(original_kls, r); 4116 } 4117 bailout->add_req(control()); 4118 set_control(_gvn.transform(r)); 4119 } 4120 } 4121 4122 // Bail out if either start or end is negative. 4123 generate_negative_guard(start, bailout, &start); 4124 generate_negative_guard(end, bailout, &end); 4125 4126 Node* length = end; 4127 if (_gvn.type(start) != TypeInt::ZERO) { 4128 length = _gvn.transform(new SubINode(end, start)); 4129 } 4130 4131 // Bail out if length is negative. 4132 // Without this the new_array would throw 4133 // NegativeArraySizeException but IllegalArgumentException is what 4134 // should be thrown 4135 generate_negative_guard(length, bailout, &length); 4136 4137 if (bailout->req() > 1) { 4138 PreserveJVMState pjvms(this); 4139 set_control(_gvn.transform(bailout)); 4140 uncommon_trap(Deoptimization::Reason_intrinsic, 4141 Deoptimization::Action_maybe_recompile); 4142 } 4143 4144 if (!stopped()) { 4145 // How many elements will we copy from the original? 4146 // The answer is MinI(orig_length - start, length). 4147 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 4148 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length); 4149 4150 original = access_resolve(original, ACCESS_READ); 4151 4152 // Generate a direct call to the right arraycopy function(s). 4153 // We know the copy is disjoint but we might not know if the 4154 // oop stores need checking. 4155 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 4156 // This will fail a store-check if x contains any non-nulls. 4157 4158 bool validated = false; 4159 // Reason_class_check rather than Reason_intrinsic because we 4160 // want to intrinsify even if this traps. 4161 if (!too_many_traps(Deoptimization::Reason_class_check)) { 4162 Node* not_subtype_ctrl = gen_subtype_check(original_kls, 4163 klass_node); 4164 4165 if (not_subtype_ctrl != top()) { 4166 PreserveJVMState pjvms(this); 4167 set_control(not_subtype_ctrl); 4168 uncommon_trap(Deoptimization::Reason_class_check, 4169 Deoptimization::Action_make_not_entrant); 4170 assert(stopped(), "Should be stopped"); 4171 } 4172 validated = true; 4173 } 4174 4175 if (!stopped()) { 4176 // Load element mirror 4177 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes()); 4178 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP); 4179 4180 newcopy = new_array(klass_node, length, 0, NULL, false, elem_mirror); 4181 4182 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false, 4183 original_kls, klass_node); 4184 if (!is_copyOfRange) { 4185 ac->set_copyof(validated); 4186 } else { 4187 ac->set_copyofrange(validated); 4188 } 4189 Node* n = _gvn.transform(ac); 4190 if (n == ac) { 4191 ac->connect_outputs(this); 4192 } else { 4193 assert(validated, "shouldn't transform if all arguments not validated"); 4194 set_all_memory(n); 4195 } 4196 } 4197 } 4198 } // original reexecute is set back here 4199 4200 C->set_has_split_ifs(true); // Has chance for split-if optimization 4201 if (!stopped()) { 4202 set_result(newcopy); 4203 } 4204 return true; 4205 } 4206 4207 4208 //----------------------generate_virtual_guard--------------------------- 4209 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 4210 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 4211 RegionNode* slow_region) { 4212 ciMethod* method = callee(); 4213 int vtable_index = method->vtable_index(); 4214 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4215 "bad index %d", vtable_index); 4216 // Get the Method* out of the appropriate vtable entry. 4217 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 4218 vtable_index*vtableEntry::size_in_bytes() + 4219 vtableEntry::method_offset_in_bytes(); 4220 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 4221 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 4222 4223 // Compare the target method with the expected method (e.g., Object.hashCode). 4224 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 4225 4226 Node* native_call = makecon(native_call_addr); 4227 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 4228 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 4229 4230 return generate_slow_guard(test_native, slow_region); 4231 } 4232 4233 //-----------------------generate_method_call---------------------------- 4234 // Use generate_method_call to make a slow-call to the real 4235 // method if the fast path fails. An alternative would be to 4236 // use a stub like OptoRuntime::slow_arraycopy_Java. 4237 // This only works for expanding the current library call, 4238 // not another intrinsic. (E.g., don't use this for making an 4239 // arraycopy call inside of the copyOf intrinsic.) 4240 CallJavaNode* 4241 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) { 4242 // When compiling the intrinsic method itself, do not use this technique. 4243 guarantee(callee() != C->method(), "cannot make slow-call to self"); 4244 4245 ciMethod* method = callee(); 4246 // ensure the JVMS we have will be correct for this call 4247 guarantee(method_id == method->intrinsic_id(), "must match"); 4248 4249 const TypeFunc* tf = TypeFunc::make(method); 4250 CallJavaNode* slow_call; 4251 if (is_static) { 4252 assert(!is_virtual, ""); 4253 slow_call = new CallStaticJavaNode(C, tf, 4254 SharedRuntime::get_resolve_static_call_stub(), 4255 method, bci()); 4256 } else if (is_virtual) { 4257 null_check_receiver(); 4258 int vtable_index = Method::invalid_vtable_index; 4259 if (UseInlineCaches) { 4260 // Suppress the vtable call 4261 } else { 4262 // hashCode and clone are not a miranda methods, 4263 // so the vtable index is fixed. 4264 // No need to use the linkResolver to get it. 4265 vtable_index = method->vtable_index(); 4266 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 4267 "bad index %d", vtable_index); 4268 } 4269 slow_call = new CallDynamicJavaNode(tf, 4270 SharedRuntime::get_resolve_virtual_call_stub(), 4271 method, vtable_index, bci()); 4272 } else { // neither virtual nor static: opt_virtual 4273 null_check_receiver(); 4274 slow_call = new CallStaticJavaNode(C, tf, 4275 SharedRuntime::get_resolve_opt_virtual_call_stub(), 4276 method, bci()); 4277 slow_call->set_optimized_virtual(true); 4278 } 4279 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) { 4280 // To be able to issue a direct call (optimized virtual or virtual) 4281 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information 4282 // about the method being invoked should be attached to the call site to 4283 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C). 4284 slow_call->set_override_symbolic_info(true); 4285 } 4286 set_arguments_for_java_call(slow_call); 4287 set_edges_for_java_call(slow_call); 4288 return slow_call; 4289 } 4290 4291 4292 /** 4293 * Build special case code for calls to hashCode on an object. This call may 4294 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 4295 * slightly different code. 4296 */ 4297 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 4298 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 4299 assert(!(is_virtual && is_static), "either virtual, special, or static"); 4300 4301 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 4302 4303 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4304 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 4305 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 4306 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4307 Node* obj = argument(0); 4308 4309 if (obj->is_ValueType() || gvn().type(obj)->is_valuetypeptr()) { 4310 return false; 4311 } 4312 4313 if (!is_static) { 4314 // Check for hashing null object 4315 obj = null_check_receiver(); 4316 if (stopped()) return true; // unconditionally null 4317 result_reg->init_req(_null_path, top()); 4318 result_val->init_req(_null_path, top()); 4319 } else { 4320 // Do a null check, and return zero if null. 4321 // System.identityHashCode(null) == 0 4322 Node* null_ctl = top(); 4323 obj = null_check_oop(obj, &null_ctl); 4324 result_reg->init_req(_null_path, null_ctl); 4325 result_val->init_req(_null_path, _gvn.intcon(0)); 4326 } 4327 4328 // Unconditionally null? Then return right away. 4329 if (stopped()) { 4330 set_control( result_reg->in(_null_path)); 4331 if (!stopped()) 4332 set_result(result_val->in(_null_path)); 4333 return true; 4334 } 4335 4336 // We only go to the fast case code if we pass a number of guards. The 4337 // paths which do not pass are accumulated in the slow_region. 4338 RegionNode* slow_region = new RegionNode(1); 4339 record_for_igvn(slow_region); 4340 4341 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 4342 assert(!obj_type->isa_valuetype() || !obj_type->is_valuetypeptr(), "no value type here"); 4343 if (is_static && obj_type->can_be_value_type()) { 4344 Node* obj_klass = load_object_klass(obj); 4345 generate_value_guard(obj_klass, slow_region); 4346 } 4347 4348 // If this is a virtual call, we generate a funny guard. We pull out 4349 // the vtable entry corresponding to hashCode() from the target object. 4350 // If the target method which we are calling happens to be the native 4351 // Object hashCode() method, we pass the guard. We do not need this 4352 // guard for non-virtual calls -- the caller is known to be the native 4353 // Object hashCode(). 4354 if (is_virtual) { 4355 // After null check, get the object's klass. 4356 Node* obj_klass = load_object_klass(obj); 4357 generate_virtual_guard(obj_klass, slow_region); 4358 } 4359 4360 // Get the header out of the object, use LoadMarkNode when available 4361 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 4362 // The control of the load must be NULL. Otherwise, the load can move before 4363 // the null check after castPP removal. 4364 Node* no_ctrl = NULL; 4365 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 4366 4367 // Test the header to see if it is unlocked. 4368 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place); 4369 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 4370 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value); 4371 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val)); 4372 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne)); 4373 4374 generate_slow_guard(test_unlocked, slow_region); 4375 4376 // Get the hash value and check to see that it has been properly assigned. 4377 // We depend on hash_mask being at most 32 bits and avoid the use of 4378 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 4379 // vm: see markWord.hpp. 4380 Node *hash_mask = _gvn.intcon(markWord::hash_mask); 4381 Node *hash_shift = _gvn.intcon(markWord::hash_shift); 4382 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 4383 // This hack lets the hash bits live anywhere in the mark object now, as long 4384 // as the shift drops the relevant bits into the low 32 bits. Note that 4385 // Java spec says that HashCode is an int so there's no point in capturing 4386 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 4387 hshifted_header = ConvX2I(hshifted_header); 4388 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 4389 4390 Node *no_hash_val = _gvn.intcon(markWord::no_hash); 4391 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 4392 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 4393 4394 generate_slow_guard(test_assigned, slow_region); 4395 4396 Node* init_mem = reset_memory(); 4397 // fill in the rest of the null path: 4398 result_io ->init_req(_null_path, i_o()); 4399 result_mem->init_req(_null_path, init_mem); 4400 4401 result_val->init_req(_fast_path, hash_val); 4402 result_reg->init_req(_fast_path, control()); 4403 result_io ->init_req(_fast_path, i_o()); 4404 result_mem->init_req(_fast_path, init_mem); 4405 4406 // Generate code for the slow case. We make a call to hashCode(). 4407 set_control(_gvn.transform(slow_region)); 4408 if (!stopped()) { 4409 // No need for PreserveJVMState, because we're using up the present state. 4410 set_all_memory(init_mem); 4411 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 4412 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static); 4413 Node* slow_result = set_results_for_java_call(slow_call); 4414 // this->control() comes from set_results_for_java_call 4415 result_reg->init_req(_slow_path, control()); 4416 result_val->init_req(_slow_path, slow_result); 4417 result_io ->set_req(_slow_path, i_o()); 4418 result_mem ->set_req(_slow_path, reset_memory()); 4419 } 4420 4421 // Return the combined state. 4422 set_i_o( _gvn.transform(result_io) ); 4423 set_all_memory( _gvn.transform(result_mem)); 4424 4425 set_result(result_reg, result_val); 4426 return true; 4427 } 4428 4429 //---------------------------inline_native_getClass---------------------------- 4430 // public final native Class<?> java.lang.Object.getClass(); 4431 // 4432 // Build special case code for calls to getClass on an object. 4433 bool LibraryCallKit::inline_native_getClass() { 4434 Node* obj = argument(0); 4435 if (obj->is_ValueType()) { 4436 ciKlass* vk = _gvn.type(obj)->value_klass(); 4437 set_result(makecon(TypeInstPtr::make(vk->java_mirror()))); 4438 return true; 4439 } 4440 obj = null_check_receiver(); 4441 if (stopped()) return true; 4442 set_result(load_mirror_from_klass(load_object_klass(obj))); 4443 return true; 4444 } 4445 4446 //-----------------inline_native_Reflection_getCallerClass--------------------- 4447 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 4448 // 4449 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 4450 // 4451 // NOTE: This code must perform the same logic as JVM_GetCallerClass 4452 // in that it must skip particular security frames and checks for 4453 // caller sensitive methods. 4454 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 4455 #ifndef PRODUCT 4456 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4457 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 4458 } 4459 #endif 4460 4461 if (!jvms()->has_method()) { 4462 #ifndef PRODUCT 4463 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4464 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 4465 } 4466 #endif 4467 return false; 4468 } 4469 4470 // Walk back up the JVM state to find the caller at the required 4471 // depth. 4472 JVMState* caller_jvms = jvms(); 4473 4474 // Cf. JVM_GetCallerClass 4475 // NOTE: Start the loop at depth 1 because the current JVM state does 4476 // not include the Reflection.getCallerClass() frame. 4477 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) { 4478 ciMethod* m = caller_jvms->method(); 4479 switch (n) { 4480 case 0: 4481 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 4482 break; 4483 case 1: 4484 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 4485 if (!m->caller_sensitive()) { 4486 #ifndef PRODUCT 4487 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4488 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 4489 } 4490 #endif 4491 return false; // bail-out; let JVM_GetCallerClass do the work 4492 } 4493 break; 4494 default: 4495 if (!m->is_ignored_by_security_stack_walk()) { 4496 // We have reached the desired frame; return the holder class. 4497 // Acquire method holder as java.lang.Class and push as constant. 4498 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 4499 ciInstance* caller_mirror = caller_klass->java_mirror(); 4500 set_result(makecon(TypeInstPtr::make(caller_mirror))); 4501 4502 #ifndef PRODUCT 4503 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4504 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()); 4505 tty->print_cr(" JVM state at this point:"); 4506 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4507 ciMethod* m = jvms()->of_depth(i)->method(); 4508 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4509 } 4510 } 4511 #endif 4512 return true; 4513 } 4514 break; 4515 } 4516 } 4517 4518 #ifndef PRODUCT 4519 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4520 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 4521 tty->print_cr(" JVM state at this point:"); 4522 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4523 ciMethod* m = jvms()->of_depth(i)->method(); 4524 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4525 } 4526 } 4527 #endif 4528 4529 return false; // bail-out; let JVM_GetCallerClass do the work 4530 } 4531 4532 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 4533 Node* arg = argument(0); 4534 Node* result = NULL; 4535 4536 switch (id) { 4537 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 4538 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 4539 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 4540 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 4541 4542 case vmIntrinsics::_doubleToLongBits: { 4543 // two paths (plus control) merge in a wood 4544 RegionNode *r = new RegionNode(3); 4545 Node *phi = new PhiNode(r, TypeLong::LONG); 4546 4547 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 4548 // Build the boolean node 4549 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4550 4551 // Branch either way. 4552 // NaN case is less traveled, which makes all the difference. 4553 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4554 Node *opt_isnan = _gvn.transform(ifisnan); 4555 assert( opt_isnan->is_If(), "Expect an IfNode"); 4556 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4557 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4558 4559 set_control(iftrue); 4560 4561 static const jlong nan_bits = CONST64(0x7ff8000000000000); 4562 Node *slow_result = longcon(nan_bits); // return NaN 4563 phi->init_req(1, _gvn.transform( slow_result )); 4564 r->init_req(1, iftrue); 4565 4566 // Else fall through 4567 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4568 set_control(iffalse); 4569 4570 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 4571 r->init_req(2, iffalse); 4572 4573 // Post merge 4574 set_control(_gvn.transform(r)); 4575 record_for_igvn(r); 4576 4577 C->set_has_split_ifs(true); // Has chance for split-if optimization 4578 result = phi; 4579 assert(result->bottom_type()->isa_long(), "must be"); 4580 break; 4581 } 4582 4583 case vmIntrinsics::_floatToIntBits: { 4584 // two paths (plus control) merge in a wood 4585 RegionNode *r = new RegionNode(3); 4586 Node *phi = new PhiNode(r, TypeInt::INT); 4587 4588 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 4589 // Build the boolean node 4590 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4591 4592 // Branch either way. 4593 // NaN case is less traveled, which makes all the difference. 4594 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4595 Node *opt_isnan = _gvn.transform(ifisnan); 4596 assert( opt_isnan->is_If(), "Expect an IfNode"); 4597 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4598 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4599 4600 set_control(iftrue); 4601 4602 static const jint nan_bits = 0x7fc00000; 4603 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 4604 phi->init_req(1, _gvn.transform( slow_result )); 4605 r->init_req(1, iftrue); 4606 4607 // Else fall through 4608 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4609 set_control(iffalse); 4610 4611 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 4612 r->init_req(2, iffalse); 4613 4614 // Post merge 4615 set_control(_gvn.transform(r)); 4616 record_for_igvn(r); 4617 4618 C->set_has_split_ifs(true); // Has chance for split-if optimization 4619 result = phi; 4620 assert(result->bottom_type()->isa_int(), "must be"); 4621 break; 4622 } 4623 4624 default: 4625 fatal_unexpected_iid(id); 4626 break; 4627 } 4628 set_result(_gvn.transform(result)); 4629 return true; 4630 } 4631 4632 //----------------------inline_unsafe_copyMemory------------------------- 4633 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 4634 bool LibraryCallKit::inline_unsafe_copyMemory() { 4635 if (callee()->is_static()) return false; // caller must have the capability! 4636 null_check_receiver(); // null-check receiver 4637 if (stopped()) return true; 4638 4639 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 4640 4641 Node* src_ptr = argument(1); // type: oop 4642 Node* src_off = ConvL2X(argument(2)); // type: long 4643 Node* dst_ptr = argument(4); // type: oop 4644 Node* dst_off = ConvL2X(argument(5)); // type: long 4645 Node* size = ConvL2X(argument(7)); // type: long 4646 4647 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 4648 "fieldOffset must be byte-scaled"); 4649 4650 src_ptr = access_resolve(src_ptr, ACCESS_READ); 4651 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE); 4652 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ); 4653 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE); 4654 4655 // Conservatively insert a memory barrier on all memory slices. 4656 // Do not let writes of the copy source or destination float below the copy. 4657 insert_mem_bar(Op_MemBarCPUOrder); 4658 4659 Node* thread = _gvn.transform(new ThreadLocalNode()); 4660 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 4661 BasicType doing_unsafe_access_bt = T_BYTE; 4662 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 4663 4664 // update volatile field 4665 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4666 4667 // Call it. Note that the length argument is not scaled. 4668 make_runtime_call(RC_LEAF|RC_NO_FP, 4669 OptoRuntime::fast_arraycopy_Type(), 4670 StubRoutines::unsafe_arraycopy(), 4671 "unsafe_arraycopy", 4672 TypeRawPtr::BOTTOM, 4673 src, dst, size XTOP); 4674 4675 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4676 4677 // Do not let reads of the copy destination float above the copy. 4678 insert_mem_bar(Op_MemBarCPUOrder); 4679 4680 return true; 4681 } 4682 4683 //------------------------clone_coping----------------------------------- 4684 // Helper function for inline_native_clone. 4685 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) { 4686 assert(obj_size != NULL, ""); 4687 Node* raw_obj = alloc_obj->in(1); 4688 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 4689 4690 AllocateNode* alloc = NULL; 4691 if (ReduceBulkZeroing) { 4692 // We will be completely responsible for initializing this object - 4693 // mark Initialize node as complete. 4694 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn); 4695 // The object was just allocated - there should be no any stores! 4696 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), ""); 4697 // Mark as complete_with_arraycopy so that on AllocateNode 4698 // expansion, we know this AllocateNode is initialized by an array 4699 // copy and a StoreStore barrier exists after the array copy. 4700 alloc->initialization()->set_complete_with_arraycopy(); 4701 } 4702 4703 // Copy the fastest available way. 4704 // TODO: generate fields copies for small objects instead. 4705 Node* size = _gvn.transform(obj_size); 4706 4707 // Exclude the header but include array length to copy by 8 bytes words. 4708 // Can't use base_offset_in_bytes(bt) since basic type is unknown. 4709 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() : 4710 instanceOopDesc::base_offset_in_bytes(); 4711 // base_off: 4712 // 8 - 32-bit VM 4713 // 12 - 64-bit VM, compressed klass 4714 // 16 - 64-bit VM, normal klass 4715 if (base_off % BytesPerLong != 0) { 4716 assert(UseCompressedClassPointers, ""); 4717 if (is_array) { 4718 // Exclude length to copy by 8 bytes words. 4719 base_off += sizeof(int); 4720 } else { 4721 // Include klass to copy by 8 bytes words. 4722 base_off = instanceOopDesc::klass_offset_in_bytes(); 4723 } 4724 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment"); 4725 } 4726 Node* src_base = basic_plus_adr(obj, base_off); 4727 Node* dst_base = basic_plus_adr(alloc_obj, base_off); 4728 4729 // Compute the length also, if needed: 4730 Node* countx = size; 4731 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off))); 4732 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong))); 4733 4734 access_clone(src_base, dst_base, countx, is_array); 4735 4736 // Do not let reads from the cloned object float above the arraycopy. 4737 if (alloc != NULL) { 4738 // Do not let stores that initialize this object be reordered with 4739 // a subsequent store that would make this object accessible by 4740 // other threads. 4741 // Record what AllocateNode this StoreStore protects so that 4742 // escape analysis can go from the MemBarStoreStoreNode to the 4743 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 4744 // based on the escape status of the AllocateNode. 4745 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 4746 } else { 4747 insert_mem_bar(Op_MemBarCPUOrder); 4748 } 4749 } 4750 4751 //------------------------inline_native_clone---------------------------- 4752 // protected native Object java.lang.Object.clone(); 4753 // 4754 // Here are the simple edge cases: 4755 // null receiver => normal trap 4756 // virtual and clone was overridden => slow path to out-of-line clone 4757 // not cloneable or finalizer => slow path to out-of-line Object.clone 4758 // 4759 // The general case has two steps, allocation and copying. 4760 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 4761 // 4762 // Copying also has two cases, oop arrays and everything else. 4763 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 4764 // Everything else uses the tight inline loop supplied by CopyArrayNode. 4765 // 4766 // These steps fold up nicely if and when the cloned object's klass 4767 // can be sharply typed as an object array, a type array, or an instance. 4768 // 4769 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 4770 PhiNode* result_val; 4771 4772 // Set the reexecute bit for the interpreter to reexecute 4773 // the bytecode that invokes Object.clone if deoptimization happens. 4774 { PreserveReexecuteState preexecs(this); 4775 jvms()->set_should_reexecute(true); 4776 4777 Node* obj = argument(0); 4778 if (obj->is_ValueType()) { 4779 return false; 4780 } 4781 4782 obj = null_check_receiver(); 4783 if (stopped()) return true; 4784 4785 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 4786 4787 // If we are going to clone an instance, we need its exact type to 4788 // know the number and types of fields to convert the clone to 4789 // loads/stores. Maybe a speculative type can help us. 4790 if (!obj_type->klass_is_exact() && 4791 obj_type->speculative_type() != NULL && 4792 obj_type->speculative_type()->is_instance_klass() && 4793 !obj_type->speculative_type()->is_valuetype()) { 4794 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 4795 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 4796 !spec_ik->has_injected_fields()) { 4797 ciKlass* k = obj_type->klass(); 4798 if (!k->is_instance_klass() || 4799 k->as_instance_klass()->is_interface() || 4800 k->as_instance_klass()->has_subklass()) { 4801 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 4802 } 4803 } 4804 } 4805 4806 Node* obj_klass = load_object_klass(obj); 4807 4808 // Conservatively insert a memory barrier on all memory slices. 4809 // Do not let writes into the original float below the clone. 4810 insert_mem_bar(Op_MemBarCPUOrder); 4811 4812 // paths into result_reg: 4813 enum { 4814 _slow_path = 1, // out-of-line call to clone method (virtual or not) 4815 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 4816 _array_path, // plain array allocation, plus arrayof_long_arraycopy 4817 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 4818 PATH_LIMIT 4819 }; 4820 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4821 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4822 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 4823 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4824 record_for_igvn(result_reg); 4825 4826 // We only go to the fast case code if we pass a number of guards. 4827 // The paths which do not pass are accumulated in the slow_region. 4828 RegionNode* slow_region = new RegionNode(1); 4829 record_for_igvn(slow_region); 4830 4831 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL); 4832 if (array_ctl != NULL) { 4833 // It's an array. 4834 PreserveJVMState pjvms(this); 4835 set_control(array_ctl); 4836 4837 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4838 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing) && 4839 (!obj_type->isa_aryptr() || !obj_type->is_aryptr()->is_not_flat())) { 4840 // Flattened value type array may have object field that would require a 4841 // write barrier. Conservatively, go to slow path. 4842 generate_valueArray_guard(obj_klass, slow_region); 4843 } 4844 4845 if (!stopped()) { 4846 Node* obj_length = load_array_length(obj); 4847 Node* obj_size = NULL; 4848 // Load element mirror 4849 Node* array_type_mirror = load_mirror_from_klass(obj_klass); 4850 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes()); 4851 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP); 4852 4853 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, false, elem_mirror); 4854 4855 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4856 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) { 4857 // If it is an oop array, it requires very special treatment, 4858 // because gc barriers are required when accessing the array. 4859 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL); 4860 if (is_obja != NULL) { 4861 PreserveJVMState pjvms2(this); 4862 set_control(is_obja); 4863 // Generate a direct call to the right arraycopy function(s). 4864 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL); 4865 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false); 4866 ac->set_cloneoop(); 4867 Node* n = _gvn.transform(ac); 4868 assert(n == ac, "cannot disappear"); 4869 ac->connect_outputs(this); 4870 4871 result_reg->init_req(_objArray_path, control()); 4872 result_val->init_req(_objArray_path, alloc_obj); 4873 result_i_o ->set_req(_objArray_path, i_o()); 4874 result_mem ->set_req(_objArray_path, reset_memory()); 4875 } 4876 } 4877 4878 // Otherwise, there are no barriers to worry about. 4879 // (We can dispense with card marks if we know the allocation 4880 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 4881 // causes the non-eden paths to take compensating steps to 4882 // simulate a fresh allocation, so that no further 4883 // card marks are required in compiled code to initialize 4884 // the object.) 4885 4886 if (!stopped()) { 4887 copy_to_clone(obj, alloc_obj, obj_size, true); 4888 4889 // Present the results of the copy. 4890 result_reg->init_req(_array_path, control()); 4891 result_val->init_req(_array_path, alloc_obj); 4892 result_i_o ->set_req(_array_path, i_o()); 4893 result_mem ->set_req(_array_path, reset_memory()); 4894 } 4895 } 4896 } 4897 4898 if (!stopped()) { 4899 // It's an instance (we did array above). Make the slow-path tests. 4900 // If this is a virtual call, we generate a funny guard. We grab 4901 // the vtable entry corresponding to clone() from the target object. 4902 // If the target method which we are calling happens to be the 4903 // Object clone() method, we pass the guard. We do not need this 4904 // guard for non-virtual calls; the caller is known to be the native 4905 // Object clone(). 4906 if (is_virtual) { 4907 generate_virtual_guard(obj_klass, slow_region); 4908 } 4909 4910 // The object must be easily cloneable and must not have a finalizer. 4911 // Both of these conditions may be checked in a single test. 4912 // We could optimize the test further, but we don't care. 4913 generate_access_flags_guard(obj_klass, 4914 // Test both conditions: 4915 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER, 4916 // Must be cloneable but not finalizer: 4917 JVM_ACC_IS_CLONEABLE_FAST, 4918 slow_region); 4919 } 4920 4921 if (!stopped()) { 4922 // It's an instance, and it passed the slow-path tests. 4923 PreserveJVMState pjvms(this); 4924 Node* obj_size = NULL; 4925 // Need to deoptimize on exception from allocation since Object.clone intrinsic 4926 // is reexecuted if deoptimization occurs and there could be problems when merging 4927 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 4928 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true); 4929 4930 copy_to_clone(obj, alloc_obj, obj_size, false); 4931 4932 // Present the results of the slow call. 4933 result_reg->init_req(_instance_path, control()); 4934 result_val->init_req(_instance_path, alloc_obj); 4935 result_i_o ->set_req(_instance_path, i_o()); 4936 result_mem ->set_req(_instance_path, reset_memory()); 4937 } 4938 4939 // Generate code for the slow case. We make a call to clone(). 4940 set_control(_gvn.transform(slow_region)); 4941 if (!stopped()) { 4942 PreserveJVMState pjvms(this); 4943 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual); 4944 // We need to deoptimize on exception (see comment above) 4945 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true); 4946 // this->control() comes from set_results_for_java_call 4947 result_reg->init_req(_slow_path, control()); 4948 result_val->init_req(_slow_path, slow_result); 4949 result_i_o ->set_req(_slow_path, i_o()); 4950 result_mem ->set_req(_slow_path, reset_memory()); 4951 } 4952 4953 // Return the combined state. 4954 set_control( _gvn.transform(result_reg)); 4955 set_i_o( _gvn.transform(result_i_o)); 4956 set_all_memory( _gvn.transform(result_mem)); 4957 } // original reexecute is set back here 4958 4959 set_result(_gvn.transform(result_val)); 4960 return true; 4961 } 4962 4963 // If we have a tightly coupled allocation, the arraycopy may take care 4964 // of the array initialization. If one of the guards we insert between 4965 // the allocation and the arraycopy causes a deoptimization, an 4966 // unitialized array will escape the compiled method. To prevent that 4967 // we set the JVM state for uncommon traps between the allocation and 4968 // the arraycopy to the state before the allocation so, in case of 4969 // deoptimization, we'll reexecute the allocation and the 4970 // initialization. 4971 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 4972 if (alloc != NULL) { 4973 ciMethod* trap_method = alloc->jvms()->method(); 4974 int trap_bci = alloc->jvms()->bci(); 4975 4976 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 4977 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 4978 // Make sure there's no store between the allocation and the 4979 // arraycopy otherwise visible side effects could be rexecuted 4980 // in case of deoptimization and cause incorrect execution. 4981 bool no_interfering_store = true; 4982 Node* mem = alloc->in(TypeFunc::Memory); 4983 if (mem->is_MergeMem()) { 4984 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 4985 Node* n = mms.memory(); 4986 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4987 assert(n->is_Store(), "what else?"); 4988 no_interfering_store = false; 4989 break; 4990 } 4991 } 4992 } else { 4993 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 4994 Node* n = mms.memory(); 4995 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4996 assert(n->is_Store(), "what else?"); 4997 no_interfering_store = false; 4998 break; 4999 } 5000 } 5001 } 5002 5003 if (no_interfering_store) { 5004 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 5005 uint size = alloc->req(); 5006 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 5007 old_jvms->set_map(sfpt); 5008 for (uint i = 0; i < size; i++) { 5009 sfpt->init_req(i, alloc->in(i)); 5010 } 5011 // re-push array length for deoptimization 5012 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength)); 5013 old_jvms->set_sp(old_jvms->sp()+1); 5014 old_jvms->set_monoff(old_jvms->monoff()+1); 5015 old_jvms->set_scloff(old_jvms->scloff()+1); 5016 old_jvms->set_endoff(old_jvms->endoff()+1); 5017 old_jvms->set_should_reexecute(true); 5018 5019 sfpt->set_i_o(map()->i_o()); 5020 sfpt->set_memory(map()->memory()); 5021 sfpt->set_control(map()->control()); 5022 5023 JVMState* saved_jvms = jvms(); 5024 saved_reexecute_sp = _reexecute_sp; 5025 5026 set_jvms(sfpt->jvms()); 5027 _reexecute_sp = jvms()->sp(); 5028 5029 return saved_jvms; 5030 } 5031 } 5032 } 5033 return NULL; 5034 } 5035 5036 // In case of a deoptimization, we restart execution at the 5037 // allocation, allocating a new array. We would leave an uninitialized 5038 // array in the heap that GCs wouldn't expect. Move the allocation 5039 // after the traps so we don't allocate the array if we 5040 // deoptimize. This is possible because tightly_coupled_allocation() 5041 // guarantees there's no observer of the allocated array at this point 5042 // and the control flow is simple enough. 5043 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, 5044 int saved_reexecute_sp, uint new_idx) { 5045 if (saved_jvms != NULL && !stopped()) { 5046 assert(alloc != NULL, "only with a tightly coupled allocation"); 5047 // restore JVM state to the state at the arraycopy 5048 saved_jvms->map()->set_control(map()->control()); 5049 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?"); 5050 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?"); 5051 // If we've improved the types of some nodes (null check) while 5052 // emitting the guards, propagate them to the current state 5053 map()->replaced_nodes().apply(saved_jvms->map(), new_idx); 5054 set_jvms(saved_jvms); 5055 _reexecute_sp = saved_reexecute_sp; 5056 5057 // Remove the allocation from above the guards 5058 CallProjections* callprojs = alloc->extract_projections(true); 5059 InitializeNode* init = alloc->initialization(); 5060 Node* alloc_mem = alloc->in(TypeFunc::Memory); 5061 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 5062 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 5063 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 5064 5065 // move the allocation here (after the guards) 5066 _gvn.hash_delete(alloc); 5067 alloc->set_req(TypeFunc::Control, control()); 5068 alloc->set_req(TypeFunc::I_O, i_o()); 5069 Node *mem = reset_memory(); 5070 set_all_memory(mem); 5071 alloc->set_req(TypeFunc::Memory, mem); 5072 set_control(init->proj_out_or_null(TypeFunc::Control)); 5073 set_i_o(callprojs->fallthrough_ioproj); 5074 5075 // Update memory as done in GraphKit::set_output_for_allocation() 5076 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 5077 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 5078 if (ary_type->isa_aryptr() && length_type != NULL) { 5079 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 5080 } 5081 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 5082 int elemidx = C->get_alias_index(telemref); 5083 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw); 5084 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx); 5085 5086 Node* allocx = _gvn.transform(alloc); 5087 assert(allocx == alloc, "where has the allocation gone?"); 5088 assert(dest->is_CheckCastPP(), "not an allocation result?"); 5089 5090 _gvn.hash_delete(dest); 5091 dest->set_req(0, control()); 5092 Node* destx = _gvn.transform(dest); 5093 assert(destx == dest, "where has the allocation result gone?"); 5094 } 5095 } 5096 5097 5098 //------------------------------inline_arraycopy----------------------- 5099 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 5100 // Object dest, int destPos, 5101 // int length); 5102 bool LibraryCallKit::inline_arraycopy() { 5103 // Get the arguments. 5104 Node* src = argument(0); // type: oop 5105 Node* src_offset = argument(1); // type: int 5106 Node* dest = argument(2); // type: oop 5107 Node* dest_offset = argument(3); // type: int 5108 Node* length = argument(4); // type: int 5109 5110 uint new_idx = C->unique(); 5111 5112 // Check for allocation before we add nodes that would confuse 5113 // tightly_coupled_allocation() 5114 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL); 5115 5116 int saved_reexecute_sp = -1; 5117 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 5118 // See arraycopy_restore_alloc_state() comment 5119 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards 5120 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation 5121 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards 5122 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL); 5123 5124 // The following tests must be performed 5125 // (1) src and dest are arrays. 5126 // (2) src and dest arrays must have elements of the same BasicType 5127 // (3) src and dest must not be null. 5128 // (4) src_offset must not be negative. 5129 // (5) dest_offset must not be negative. 5130 // (6) length must not be negative. 5131 // (7) src_offset + length must not exceed length of src. 5132 // (8) dest_offset + length must not exceed length of dest. 5133 // (9) each element of an oop array must be assignable 5134 5135 // (3) src and dest must not be null. 5136 // always do this here because we need the JVM state for uncommon traps 5137 Node* null_ctl = top(); 5138 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 5139 assert(null_ctl->is_top(), "no null control here"); 5140 dest = null_check(dest, T_ARRAY); 5141 5142 if (!can_emit_guards) { 5143 // if saved_jvms == NULL and alloc != NULL, we don't emit any 5144 // guards but the arraycopy node could still take advantage of a 5145 // tightly allocated allocation. tightly_coupled_allocation() is 5146 // called again to make sure it takes the null check above into 5147 // account: the null check is mandatory and if it caused an 5148 // uncommon trap to be emitted then the allocation can't be 5149 // considered tightly coupled in this context. 5150 alloc = tightly_coupled_allocation(dest, NULL); 5151 } 5152 5153 bool validated = false; 5154 5155 const Type* src_type = _gvn.type(src); 5156 const Type* dest_type = _gvn.type(dest); 5157 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5158 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5159 5160 // Do we have the type of src? 5161 bool has_src = (top_src != NULL && top_src->klass() != NULL); 5162 // Do we have the type of dest? 5163 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL); 5164 // Is the type for src from speculation? 5165 bool src_spec = false; 5166 // Is the type for dest from speculation? 5167 bool dest_spec = false; 5168 5169 if ((!has_src || !has_dest) && can_emit_guards) { 5170 // We don't have sufficient type information, let's see if 5171 // speculative types can help. We need to have types for both src 5172 // and dest so that it pays off. 5173 5174 // Do we already have or could we have type information for src 5175 bool could_have_src = has_src; 5176 // Do we already have or could we have type information for dest 5177 bool could_have_dest = has_dest; 5178 5179 ciKlass* src_k = NULL; 5180 if (!has_src) { 5181 src_k = src_type->speculative_type_not_null(); 5182 if (src_k != NULL && src_k->is_array_klass()) { 5183 could_have_src = true; 5184 } 5185 } 5186 5187 ciKlass* dest_k = NULL; 5188 if (!has_dest) { 5189 dest_k = dest_type->speculative_type_not_null(); 5190 if (dest_k != NULL && dest_k->is_array_klass()) { 5191 could_have_dest = true; 5192 } 5193 } 5194 5195 if (could_have_src && could_have_dest) { 5196 // This is going to pay off so emit the required guards 5197 if (!has_src) { 5198 src = maybe_cast_profiled_obj(src, src_k, true); 5199 src_type = _gvn.type(src); 5200 top_src = src_type->isa_aryptr(); 5201 has_src = (top_src != NULL && top_src->klass() != NULL); 5202 src_spec = true; 5203 } 5204 if (!has_dest) { 5205 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5206 dest_type = _gvn.type(dest); 5207 top_dest = dest_type->isa_aryptr(); 5208 has_dest = (top_dest != NULL && top_dest->klass() != NULL); 5209 dest_spec = true; 5210 } 5211 } 5212 } 5213 5214 if (has_src && has_dest && can_emit_guards) { 5215 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type(); 5216 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type(); 5217 if (src_elem == T_ARRAY) src_elem = T_OBJECT; 5218 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT; 5219 5220 if (src_elem == dest_elem && src_elem == T_OBJECT) { 5221 // If both arrays are object arrays then having the exact types 5222 // for both will remove the need for a subtype check at runtime 5223 // before the call and may make it possible to pick a faster copy 5224 // routine (without a subtype check on every element) 5225 // Do we have the exact type of src? 5226 bool could_have_src = src_spec; 5227 // Do we have the exact type of dest? 5228 bool could_have_dest = dest_spec; 5229 ciKlass* src_k = top_src->klass(); 5230 ciKlass* dest_k = top_dest->klass(); 5231 if (!src_spec) { 5232 src_k = src_type->speculative_type_not_null(); 5233 if (src_k != NULL && src_k->is_array_klass()) { 5234 could_have_src = true; 5235 } 5236 } 5237 if (!dest_spec) { 5238 dest_k = dest_type->speculative_type_not_null(); 5239 if (dest_k != NULL && dest_k->is_array_klass()) { 5240 could_have_dest = true; 5241 } 5242 } 5243 if (could_have_src && could_have_dest) { 5244 // If we can have both exact types, emit the missing guards 5245 if (could_have_src && !src_spec) { 5246 src = maybe_cast_profiled_obj(src, src_k, true); 5247 } 5248 if (could_have_dest && !dest_spec) { 5249 dest = maybe_cast_profiled_obj(dest, dest_k, true); 5250 } 5251 } 5252 } 5253 } 5254 5255 ciMethod* trap_method = method(); 5256 int trap_bci = bci(); 5257 if (saved_jvms != NULL) { 5258 trap_method = alloc->jvms()->method(); 5259 trap_bci = alloc->jvms()->bci(); 5260 } 5261 5262 bool negative_length_guard_generated = false; 5263 5264 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 5265 can_emit_guards && 5266 !src->is_top() && !dest->is_top()) { 5267 // validate arguments: enables transformation the ArrayCopyNode 5268 validated = true; 5269 5270 RegionNode* slow_region = new RegionNode(1); 5271 record_for_igvn(slow_region); 5272 5273 // (1) src and dest are arrays. 5274 generate_non_array_guard(load_object_klass(src), slow_region); 5275 generate_non_array_guard(load_object_klass(dest), slow_region); 5276 5277 // (2) src and dest arrays must have elements of the same BasicType 5278 // done at macro expansion or at Ideal transformation time 5279 5280 // (4) src_offset must not be negative. 5281 generate_negative_guard(src_offset, slow_region); 5282 5283 // (5) dest_offset must not be negative. 5284 generate_negative_guard(dest_offset, slow_region); 5285 5286 // (7) src_offset + length must not exceed length of src. 5287 generate_limit_guard(src_offset, length, 5288 load_array_length(src), 5289 slow_region); 5290 5291 // (8) dest_offset + length must not exceed length of dest. 5292 generate_limit_guard(dest_offset, length, 5293 load_array_length(dest), 5294 slow_region); 5295 5296 // (6) length must not be negative. 5297 // This is also checked in generate_arraycopy() during macro expansion, but 5298 // we also have to check it here for the case where the ArrayCopyNode will 5299 // be eliminated by Escape Analysis. 5300 if (EliminateAllocations) { 5301 generate_negative_guard(length, slow_region); 5302 negative_length_guard_generated = true; 5303 } 5304 5305 // (9) each element of an oop array must be assignable 5306 Node* src_klass = load_object_klass(src); 5307 Node* dest_klass = load_object_klass(dest); 5308 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass); 5309 5310 if (not_subtype_ctrl != top()) { 5311 PreserveJVMState pjvms(this); 5312 set_control(not_subtype_ctrl); 5313 uncommon_trap(Deoptimization::Reason_intrinsic, 5314 Deoptimization::Action_make_not_entrant); 5315 assert(stopped(), "Should be stopped"); 5316 } 5317 5318 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr(); 5319 const Type* toop = TypeOopPtr::make_from_klass(dest_klass_t->klass()); 5320 src = _gvn.transform(new CheckCastPPNode(control(), src, toop)); 5321 src_type = _gvn.type(src); 5322 top_src = src_type->isa_aryptr(); 5323 5324 if (top_dest != NULL && !top_dest->elem()->isa_valuetype() && !top_dest->is_not_flat()) { 5325 generate_valueArray_guard(dest_klass, slow_region); 5326 } 5327 5328 if (top_src != NULL && !top_src->elem()->isa_valuetype() && !top_src->is_not_flat()) { 5329 generate_valueArray_guard(src_klass, slow_region); 5330 } 5331 5332 { 5333 PreserveJVMState pjvms(this); 5334 set_control(_gvn.transform(slow_region)); 5335 uncommon_trap(Deoptimization::Reason_intrinsic, 5336 Deoptimization::Action_make_not_entrant); 5337 assert(stopped(), "Should be stopped"); 5338 } 5339 } 5340 5341 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx); 5342 5343 if (stopped()) { 5344 return true; 5345 } 5346 5347 Node* new_src = access_resolve(src, ACCESS_READ); 5348 Node* new_dest = access_resolve(dest, ACCESS_WRITE); 5349 5350 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated, 5351 // Create LoadRange and LoadKlass nodes for use during macro expansion here 5352 // so the compiler has a chance to eliminate them: during macro expansion, 5353 // we have to set their control (CastPP nodes are eliminated). 5354 load_object_klass(src), load_object_klass(dest), 5355 load_array_length(src), load_array_length(dest)); 5356 5357 ac->set_arraycopy(validated); 5358 5359 Node* n = _gvn.transform(ac); 5360 if (n == ac) { 5361 ac->connect_outputs(this); 5362 } else { 5363 assert(validated, "shouldn't transform if all arguments not validated"); 5364 set_all_memory(n); 5365 } 5366 clear_upper_avx(); 5367 5368 5369 return true; 5370 } 5371 5372 5373 // Helper function which determines if an arraycopy immediately follows 5374 // an allocation, with no intervening tests or other escapes for the object. 5375 AllocateArrayNode* 5376 LibraryCallKit::tightly_coupled_allocation(Node* ptr, 5377 RegionNode* slow_region) { 5378 if (stopped()) return NULL; // no fast path 5379 if (C->AliasLevel() == 0) return NULL; // no MergeMems around 5380 5381 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn); 5382 if (alloc == NULL) return NULL; 5383 5384 Node* rawmem = memory(Compile::AliasIdxRaw); 5385 // Is the allocation's memory state untouched? 5386 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 5387 // Bail out if there have been raw-memory effects since the allocation. 5388 // (Example: There might have been a call or safepoint.) 5389 return NULL; 5390 } 5391 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 5392 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 5393 return NULL; 5394 } 5395 5396 // There must be no unexpected observers of this allocation. 5397 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 5398 Node* obs = ptr->fast_out(i); 5399 if (obs != this->map()) { 5400 return NULL; 5401 } 5402 } 5403 5404 // This arraycopy must unconditionally follow the allocation of the ptr. 5405 Node* alloc_ctl = ptr->in(0); 5406 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo"); 5407 5408 Node* ctl = control(); 5409 while (ctl != alloc_ctl) { 5410 // There may be guards which feed into the slow_region. 5411 // Any other control flow means that we might not get a chance 5412 // to finish initializing the allocated object. 5413 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) { 5414 IfNode* iff = ctl->in(0)->as_If(); 5415 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con); 5416 assert(not_ctl != NULL && not_ctl != ctl, "found alternate"); 5417 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) { 5418 ctl = iff->in(0); // This test feeds the known slow_region. 5419 continue; 5420 } 5421 // One more try: Various low-level checks bottom out in 5422 // uncommon traps. If the debug-info of the trap omits 5423 // any reference to the allocation, as we've already 5424 // observed, then there can be no objection to the trap. 5425 bool found_trap = false; 5426 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) { 5427 Node* obs = not_ctl->fast_out(j); 5428 if (obs->in(0) == not_ctl && obs->is_Call() && 5429 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) { 5430 found_trap = true; break; 5431 } 5432 } 5433 if (found_trap) { 5434 ctl = iff->in(0); // This test feeds a harmless uncommon trap. 5435 continue; 5436 } 5437 } 5438 return NULL; 5439 } 5440 5441 // If we get this far, we have an allocation which immediately 5442 // precedes the arraycopy, and we can take over zeroing the new object. 5443 // The arraycopy will finish the initialization, and provide 5444 // a new control state to which we will anchor the destination pointer. 5445 5446 return alloc; 5447 } 5448 5449 //-------------inline_encodeISOArray----------------------------------- 5450 // encode char[] to byte[] in ISO_8859_1 5451 bool LibraryCallKit::inline_encodeISOArray() { 5452 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 5453 // no receiver since it is static method 5454 Node *src = argument(0); 5455 Node *src_offset = argument(1); 5456 Node *dst = argument(2); 5457 Node *dst_offset = argument(3); 5458 Node *length = argument(4); 5459 5460 src = must_be_not_null(src, true); 5461 dst = must_be_not_null(dst, true); 5462 5463 src = access_resolve(src, ACCESS_READ); 5464 dst = access_resolve(dst, ACCESS_WRITE); 5465 5466 const Type* src_type = src->Value(&_gvn); 5467 const Type* dst_type = dst->Value(&_gvn); 5468 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5469 const TypeAryPtr* top_dest = dst_type->isa_aryptr(); 5470 if (top_src == NULL || top_src->klass() == NULL || 5471 top_dest == NULL || top_dest->klass() == NULL) { 5472 // failed array check 5473 return false; 5474 } 5475 5476 // Figure out the size and type of the elements we will be copying. 5477 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5478 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5479 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 5480 return false; 5481 } 5482 5483 Node* src_start = array_element_address(src, src_offset, T_CHAR); 5484 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 5485 // 'src_start' points to src array + scaled offset 5486 // 'dst_start' points to dst array + scaled offset 5487 5488 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 5489 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length); 5490 enc = _gvn.transform(enc); 5491 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 5492 set_memory(res_mem, mtype); 5493 set_result(enc); 5494 clear_upper_avx(); 5495 5496 return true; 5497 } 5498 5499 //-------------inline_multiplyToLen----------------------------------- 5500 bool LibraryCallKit::inline_multiplyToLen() { 5501 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 5502 5503 address stubAddr = StubRoutines::multiplyToLen(); 5504 if (stubAddr == NULL) { 5505 return false; // Intrinsic's stub is not implemented on this platform 5506 } 5507 const char* stubName = "multiplyToLen"; 5508 5509 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 5510 5511 // no receiver because it is a static method 5512 Node* x = argument(0); 5513 Node* xlen = argument(1); 5514 Node* y = argument(2); 5515 Node* ylen = argument(3); 5516 Node* z = argument(4); 5517 5518 x = must_be_not_null(x, true); 5519 y = must_be_not_null(y, true); 5520 5521 x = access_resolve(x, ACCESS_READ); 5522 y = access_resolve(y, ACCESS_READ); 5523 z = access_resolve(z, ACCESS_WRITE); 5524 5525 const Type* x_type = x->Value(&_gvn); 5526 const Type* y_type = y->Value(&_gvn); 5527 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5528 const TypeAryPtr* top_y = y_type->isa_aryptr(); 5529 if (top_x == NULL || top_x->klass() == NULL || 5530 top_y == NULL || top_y->klass() == NULL) { 5531 // failed array check 5532 return false; 5533 } 5534 5535 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5536 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5537 if (x_elem != T_INT || y_elem != T_INT) { 5538 return false; 5539 } 5540 5541 // Set the original stack and the reexecute bit for the interpreter to reexecute 5542 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens 5543 // on the return from z array allocation in runtime. 5544 { PreserveReexecuteState preexecs(this); 5545 jvms()->set_should_reexecute(true); 5546 5547 Node* x_start = array_element_address(x, intcon(0), x_elem); 5548 Node* y_start = array_element_address(y, intcon(0), y_elem); 5549 // 'x_start' points to x array + scaled xlen 5550 // 'y_start' points to y array + scaled ylen 5551 5552 // Allocate the result array 5553 Node* zlen = _gvn.transform(new AddINode(xlen, ylen)); 5554 ciKlass* klass = ciTypeArrayKlass::make(T_INT); 5555 Node* klass_node = makecon(TypeKlassPtr::make(klass)); 5556 5557 IdealKit ideal(this); 5558 5559 #define __ ideal. 5560 Node* one = __ ConI(1); 5561 Node* zero = __ ConI(0); 5562 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done(); 5563 __ set(need_alloc, zero); 5564 __ set(z_alloc, z); 5565 __ if_then(z, BoolTest::eq, null()); { 5566 __ increment (need_alloc, one); 5567 } __ else_(); { 5568 // Update graphKit memory and control from IdealKit. 5569 sync_kit(ideal); 5570 Node *cast = new CastPPNode(z, TypePtr::NOTNULL); 5571 cast->init_req(0, control()); 5572 _gvn.set_type(cast, cast->bottom_type()); 5573 C->record_for_igvn(cast); 5574 5575 Node* zlen_arg = load_array_length(cast); 5576 // Update IdealKit memory and control from graphKit. 5577 __ sync_kit(this); 5578 __ if_then(zlen_arg, BoolTest::lt, zlen); { 5579 __ increment (need_alloc, one); 5580 } __ end_if(); 5581 } __ end_if(); 5582 5583 __ if_then(__ value(need_alloc), BoolTest::ne, zero); { 5584 // Update graphKit memory and control from IdealKit. 5585 sync_kit(ideal); 5586 Node * narr = new_array(klass_node, zlen, 1); 5587 // Update IdealKit memory and control from graphKit. 5588 __ sync_kit(this); 5589 __ set(z_alloc, narr); 5590 } __ end_if(); 5591 5592 sync_kit(ideal); 5593 z = __ value(z_alloc); 5594 // Can't use TypeAryPtr::INTS which uses Bottom offset. 5595 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass)); 5596 // Final sync IdealKit and GraphKit. 5597 final_sync(ideal); 5598 #undef __ 5599 5600 Node* z_start = array_element_address(z, intcon(0), T_INT); 5601 5602 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5603 OptoRuntime::multiplyToLen_Type(), 5604 stubAddr, stubName, TypePtr::BOTTOM, 5605 x_start, xlen, y_start, ylen, z_start, zlen); 5606 } // original reexecute is set back here 5607 5608 C->set_has_split_ifs(true); // Has chance for split-if optimization 5609 set_result(z); 5610 return true; 5611 } 5612 5613 //-------------inline_squareToLen------------------------------------ 5614 bool LibraryCallKit::inline_squareToLen() { 5615 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 5616 5617 address stubAddr = StubRoutines::squareToLen(); 5618 if (stubAddr == NULL) { 5619 return false; // Intrinsic's stub is not implemented on this platform 5620 } 5621 const char* stubName = "squareToLen"; 5622 5623 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 5624 5625 Node* x = argument(0); 5626 Node* len = argument(1); 5627 Node* z = argument(2); 5628 Node* zlen = argument(3); 5629 5630 x = must_be_not_null(x, true); 5631 z = must_be_not_null(z, true); 5632 5633 x = access_resolve(x, ACCESS_READ); 5634 z = access_resolve(z, ACCESS_WRITE); 5635 5636 const Type* x_type = x->Value(&_gvn); 5637 const Type* z_type = z->Value(&_gvn); 5638 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5639 const TypeAryPtr* top_z = z_type->isa_aryptr(); 5640 if (top_x == NULL || top_x->klass() == NULL || 5641 top_z == NULL || top_z->klass() == NULL) { 5642 // failed array check 5643 return false; 5644 } 5645 5646 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5647 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5648 if (x_elem != T_INT || z_elem != T_INT) { 5649 return false; 5650 } 5651 5652 5653 Node* x_start = array_element_address(x, intcon(0), x_elem); 5654 Node* z_start = array_element_address(z, intcon(0), z_elem); 5655 5656 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5657 OptoRuntime::squareToLen_Type(), 5658 stubAddr, stubName, TypePtr::BOTTOM, 5659 x_start, len, z_start, zlen); 5660 5661 set_result(z); 5662 return true; 5663 } 5664 5665 //-------------inline_mulAdd------------------------------------------ 5666 bool LibraryCallKit::inline_mulAdd() { 5667 assert(UseMulAddIntrinsic, "not implemented on this platform"); 5668 5669 address stubAddr = StubRoutines::mulAdd(); 5670 if (stubAddr == NULL) { 5671 return false; // Intrinsic's stub is not implemented on this platform 5672 } 5673 const char* stubName = "mulAdd"; 5674 5675 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 5676 5677 Node* out = argument(0); 5678 Node* in = argument(1); 5679 Node* offset = argument(2); 5680 Node* len = argument(3); 5681 Node* k = argument(4); 5682 5683 out = must_be_not_null(out, true); 5684 5685 in = access_resolve(in, ACCESS_READ); 5686 out = access_resolve(out, ACCESS_WRITE); 5687 5688 const Type* out_type = out->Value(&_gvn); 5689 const Type* in_type = in->Value(&_gvn); 5690 const TypeAryPtr* top_out = out_type->isa_aryptr(); 5691 const TypeAryPtr* top_in = in_type->isa_aryptr(); 5692 if (top_out == NULL || top_out->klass() == NULL || 5693 top_in == NULL || top_in->klass() == NULL) { 5694 // failed array check 5695 return false; 5696 } 5697 5698 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5699 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5700 if (out_elem != T_INT || in_elem != T_INT) { 5701 return false; 5702 } 5703 5704 Node* outlen = load_array_length(out); 5705 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 5706 Node* out_start = array_element_address(out, intcon(0), out_elem); 5707 Node* in_start = array_element_address(in, intcon(0), in_elem); 5708 5709 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5710 OptoRuntime::mulAdd_Type(), 5711 stubAddr, stubName, TypePtr::BOTTOM, 5712 out_start,in_start, new_offset, len, k); 5713 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5714 set_result(result); 5715 return true; 5716 } 5717 5718 //-------------inline_montgomeryMultiply----------------------------------- 5719 bool LibraryCallKit::inline_montgomeryMultiply() { 5720 address stubAddr = StubRoutines::montgomeryMultiply(); 5721 if (stubAddr == NULL) { 5722 return false; // Intrinsic's stub is not implemented on this platform 5723 } 5724 5725 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 5726 const char* stubName = "montgomery_multiply"; 5727 5728 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 5729 5730 Node* a = argument(0); 5731 Node* b = argument(1); 5732 Node* n = argument(2); 5733 Node* len = argument(3); 5734 Node* inv = argument(4); 5735 Node* m = argument(6); 5736 5737 a = access_resolve(a, ACCESS_READ); 5738 b = access_resolve(b, ACCESS_READ); 5739 n = access_resolve(n, ACCESS_READ); 5740 m = access_resolve(m, ACCESS_WRITE); 5741 5742 const Type* a_type = a->Value(&_gvn); 5743 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5744 const Type* b_type = b->Value(&_gvn); 5745 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5746 const Type* n_type = a->Value(&_gvn); 5747 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5748 const Type* m_type = a->Value(&_gvn); 5749 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5750 if (top_a == NULL || top_a->klass() == NULL || 5751 top_b == NULL || top_b->klass() == NULL || 5752 top_n == NULL || top_n->klass() == NULL || 5753 top_m == NULL || top_m->klass() == NULL) { 5754 // failed array check 5755 return false; 5756 } 5757 5758 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5759 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5760 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5761 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5762 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5763 return false; 5764 } 5765 5766 // Make the call 5767 { 5768 Node* a_start = array_element_address(a, intcon(0), a_elem); 5769 Node* b_start = array_element_address(b, intcon(0), b_elem); 5770 Node* n_start = array_element_address(n, intcon(0), n_elem); 5771 Node* m_start = array_element_address(m, intcon(0), m_elem); 5772 5773 Node* call = make_runtime_call(RC_LEAF, 5774 OptoRuntime::montgomeryMultiply_Type(), 5775 stubAddr, stubName, TypePtr::BOTTOM, 5776 a_start, b_start, n_start, len, inv, top(), 5777 m_start); 5778 set_result(m); 5779 } 5780 5781 return true; 5782 } 5783 5784 bool LibraryCallKit::inline_montgomerySquare() { 5785 address stubAddr = StubRoutines::montgomerySquare(); 5786 if (stubAddr == NULL) { 5787 return false; // Intrinsic's stub is not implemented on this platform 5788 } 5789 5790 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 5791 const char* stubName = "montgomery_square"; 5792 5793 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 5794 5795 Node* a = argument(0); 5796 Node* n = argument(1); 5797 Node* len = argument(2); 5798 Node* inv = argument(3); 5799 Node* m = argument(5); 5800 5801 a = access_resolve(a, ACCESS_READ); 5802 n = access_resolve(n, ACCESS_READ); 5803 m = access_resolve(m, ACCESS_WRITE); 5804 5805 const Type* a_type = a->Value(&_gvn); 5806 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5807 const Type* n_type = a->Value(&_gvn); 5808 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5809 const Type* m_type = a->Value(&_gvn); 5810 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5811 if (top_a == NULL || top_a->klass() == NULL || 5812 top_n == NULL || top_n->klass() == NULL || 5813 top_m == NULL || top_m->klass() == NULL) { 5814 // failed array check 5815 return false; 5816 } 5817 5818 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5819 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5820 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5821 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5822 return false; 5823 } 5824 5825 // Make the call 5826 { 5827 Node* a_start = array_element_address(a, intcon(0), a_elem); 5828 Node* n_start = array_element_address(n, intcon(0), n_elem); 5829 Node* m_start = array_element_address(m, intcon(0), m_elem); 5830 5831 Node* call = make_runtime_call(RC_LEAF, 5832 OptoRuntime::montgomerySquare_Type(), 5833 stubAddr, stubName, TypePtr::BOTTOM, 5834 a_start, n_start, len, inv, top(), 5835 m_start); 5836 set_result(m); 5837 } 5838 5839 return true; 5840 } 5841 5842 //-------------inline_vectorizedMismatch------------------------------ 5843 bool LibraryCallKit::inline_vectorizedMismatch() { 5844 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform"); 5845 5846 address stubAddr = StubRoutines::vectorizedMismatch(); 5847 if (stubAddr == NULL) { 5848 return false; // Intrinsic's stub is not implemented on this platform 5849 } 5850 const char* stubName = "vectorizedMismatch"; 5851 int size_l = callee()->signature()->size(); 5852 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 5853 5854 Node* obja = argument(0); 5855 Node* aoffset = argument(1); 5856 Node* objb = argument(3); 5857 Node* boffset = argument(4); 5858 Node* length = argument(6); 5859 Node* scale = argument(7); 5860 5861 const Type* a_type = obja->Value(&_gvn); 5862 const Type* b_type = objb->Value(&_gvn); 5863 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5864 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5865 if (top_a == NULL || top_a->klass() == NULL || 5866 top_b == NULL || top_b->klass() == NULL) { 5867 // failed array check 5868 return false; 5869 } 5870 5871 Node* call; 5872 jvms()->set_should_reexecute(true); 5873 5874 obja = access_resolve(obja, ACCESS_READ); 5875 objb = access_resolve(objb, ACCESS_READ); 5876 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ); 5877 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ); 5878 5879 call = make_runtime_call(RC_LEAF, 5880 OptoRuntime::vectorizedMismatch_Type(), 5881 stubAddr, stubName, TypePtr::BOTTOM, 5882 obja_adr, objb_adr, length, scale); 5883 5884 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5885 set_result(result); 5886 return true; 5887 } 5888 5889 /** 5890 * Calculate CRC32 for byte. 5891 * int java.util.zip.CRC32.update(int crc, int b) 5892 */ 5893 bool LibraryCallKit::inline_updateCRC32() { 5894 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5895 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 5896 // no receiver since it is static method 5897 Node* crc = argument(0); // type: int 5898 Node* b = argument(1); // type: int 5899 5900 /* 5901 * int c = ~ crc; 5902 * b = timesXtoThe32[(b ^ c) & 0xFF]; 5903 * b = b ^ (c >>> 8); 5904 * crc = ~b; 5905 */ 5906 5907 Node* M1 = intcon(-1); 5908 crc = _gvn.transform(new XorINode(crc, M1)); 5909 Node* result = _gvn.transform(new XorINode(crc, b)); 5910 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 5911 5912 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 5913 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 5914 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 5915 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 5916 5917 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 5918 result = _gvn.transform(new XorINode(crc, result)); 5919 result = _gvn.transform(new XorINode(result, M1)); 5920 set_result(result); 5921 return true; 5922 } 5923 5924 /** 5925 * Calculate CRC32 for byte[] array. 5926 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 5927 */ 5928 bool LibraryCallKit::inline_updateBytesCRC32() { 5929 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5930 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5931 // no receiver since it is static method 5932 Node* crc = argument(0); // type: int 5933 Node* src = argument(1); // type: oop 5934 Node* offset = argument(2); // type: int 5935 Node* length = argument(3); // type: int 5936 5937 const Type* src_type = src->Value(&_gvn); 5938 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5939 if (top_src == NULL || top_src->klass() == NULL) { 5940 // failed array check 5941 return false; 5942 } 5943 5944 // Figure out the size and type of the elements we will be copying. 5945 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5946 if (src_elem != T_BYTE) { 5947 return false; 5948 } 5949 5950 // 'src_start' points to src array + scaled offset 5951 src = must_be_not_null(src, true); 5952 src = access_resolve(src, ACCESS_READ); 5953 Node* src_start = array_element_address(src, offset, src_elem); 5954 5955 // We assume that range check is done by caller. 5956 // TODO: generate range check (offset+length < src.length) in debug VM. 5957 5958 // Call the stub. 5959 address stubAddr = StubRoutines::updateBytesCRC32(); 5960 const char *stubName = "updateBytesCRC32"; 5961 5962 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5963 stubAddr, stubName, TypePtr::BOTTOM, 5964 crc, src_start, length); 5965 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5966 set_result(result); 5967 return true; 5968 } 5969 5970 /** 5971 * Calculate CRC32 for ByteBuffer. 5972 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 5973 */ 5974 bool LibraryCallKit::inline_updateByteBufferCRC32() { 5975 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5976 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5977 // no receiver since it is static method 5978 Node* crc = argument(0); // type: int 5979 Node* src = argument(1); // type: long 5980 Node* offset = argument(3); // type: int 5981 Node* length = argument(4); // type: int 5982 5983 src = ConvL2X(src); // adjust Java long to machine word 5984 Node* base = _gvn.transform(new CastX2PNode(src)); 5985 offset = ConvI2X(offset); 5986 5987 // 'src_start' points to src array + scaled offset 5988 Node* src_start = basic_plus_adr(top(), base, offset); 5989 5990 // Call the stub. 5991 address stubAddr = StubRoutines::updateBytesCRC32(); 5992 const char *stubName = "updateBytesCRC32"; 5993 5994 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5995 stubAddr, stubName, TypePtr::BOTTOM, 5996 crc, src_start, length); 5997 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5998 set_result(result); 5999 return true; 6000 } 6001 6002 //------------------------------get_table_from_crc32c_class----------------------- 6003 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 6004 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class); 6005 assert (table != NULL, "wrong version of java.util.zip.CRC32C"); 6006 6007 return table; 6008 } 6009 6010 //------------------------------inline_updateBytesCRC32C----------------------- 6011 // 6012 // Calculate CRC32C for byte[] array. 6013 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 6014 // 6015 bool LibraryCallKit::inline_updateBytesCRC32C() { 6016 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6017 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6018 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6019 // no receiver since it is a static method 6020 Node* crc = argument(0); // type: int 6021 Node* src = argument(1); // type: oop 6022 Node* offset = argument(2); // type: int 6023 Node* end = argument(3); // type: int 6024 6025 Node* length = _gvn.transform(new SubINode(end, offset)); 6026 6027 const Type* src_type = src->Value(&_gvn); 6028 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6029 if (top_src == NULL || top_src->klass() == NULL) { 6030 // failed array check 6031 return false; 6032 } 6033 6034 // Figure out the size and type of the elements we will be copying. 6035 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6036 if (src_elem != T_BYTE) { 6037 return false; 6038 } 6039 6040 // 'src_start' points to src array + scaled offset 6041 src = must_be_not_null(src, true); 6042 src = access_resolve(src, ACCESS_READ); 6043 Node* src_start = array_element_address(src, offset, src_elem); 6044 6045 // static final int[] byteTable in class CRC32C 6046 Node* table = get_table_from_crc32c_class(callee()->holder()); 6047 table = must_be_not_null(table, true); 6048 table = access_resolve(table, ACCESS_READ); 6049 Node* table_start = array_element_address(table, intcon(0), T_INT); 6050 6051 // We assume that range check is done by caller. 6052 // TODO: generate range check (offset+length < src.length) in debug VM. 6053 6054 // Call the stub. 6055 address stubAddr = StubRoutines::updateBytesCRC32C(); 6056 const char *stubName = "updateBytesCRC32C"; 6057 6058 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 6059 stubAddr, stubName, TypePtr::BOTTOM, 6060 crc, src_start, length, table_start); 6061 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6062 set_result(result); 6063 return true; 6064 } 6065 6066 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 6067 // 6068 // Calculate CRC32C for DirectByteBuffer. 6069 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 6070 // 6071 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 6072 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 6073 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 6074 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 6075 // no receiver since it is a static method 6076 Node* crc = argument(0); // type: int 6077 Node* src = argument(1); // type: long 6078 Node* offset = argument(3); // type: int 6079 Node* end = argument(4); // type: int 6080 6081 Node* length = _gvn.transform(new SubINode(end, offset)); 6082 6083 src = ConvL2X(src); // adjust Java long to machine word 6084 Node* base = _gvn.transform(new CastX2PNode(src)); 6085 offset = ConvI2X(offset); 6086 6087 // 'src_start' points to src array + scaled offset 6088 Node* src_start = basic_plus_adr(top(), base, offset); 6089 6090 // static final int[] byteTable in class CRC32C 6091 Node* table = get_table_from_crc32c_class(callee()->holder()); 6092 table = must_be_not_null(table, true); 6093 table = access_resolve(table, ACCESS_READ); 6094 Node* table_start = array_element_address(table, intcon(0), T_INT); 6095 6096 // Call the stub. 6097 address stubAddr = StubRoutines::updateBytesCRC32C(); 6098 const char *stubName = "updateBytesCRC32C"; 6099 6100 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 6101 stubAddr, stubName, TypePtr::BOTTOM, 6102 crc, src_start, length, table_start); 6103 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6104 set_result(result); 6105 return true; 6106 } 6107 6108 //------------------------------inline_updateBytesAdler32---------------------- 6109 // 6110 // Calculate Adler32 checksum for byte[] array. 6111 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 6112 // 6113 bool LibraryCallKit::inline_updateBytesAdler32() { 6114 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 6115 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 6116 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 6117 // no receiver since it is static method 6118 Node* crc = argument(0); // type: int 6119 Node* src = argument(1); // type: oop 6120 Node* offset = argument(2); // type: int 6121 Node* length = argument(3); // type: int 6122 6123 const Type* src_type = src->Value(&_gvn); 6124 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6125 if (top_src == NULL || top_src->klass() == NULL) { 6126 // failed array check 6127 return false; 6128 } 6129 6130 // Figure out the size and type of the elements we will be copying. 6131 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6132 if (src_elem != T_BYTE) { 6133 return false; 6134 } 6135 6136 // 'src_start' points to src array + scaled offset 6137 src = access_resolve(src, ACCESS_READ); 6138 Node* src_start = array_element_address(src, offset, src_elem); 6139 6140 // We assume that range check is done by caller. 6141 // TODO: generate range check (offset+length < src.length) in debug VM. 6142 6143 // Call the stub. 6144 address stubAddr = StubRoutines::updateBytesAdler32(); 6145 const char *stubName = "updateBytesAdler32"; 6146 6147 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6148 stubAddr, stubName, TypePtr::BOTTOM, 6149 crc, src_start, length); 6150 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6151 set_result(result); 6152 return true; 6153 } 6154 6155 //------------------------------inline_updateByteBufferAdler32--------------- 6156 // 6157 // Calculate Adler32 checksum for DirectByteBuffer. 6158 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 6159 // 6160 bool LibraryCallKit::inline_updateByteBufferAdler32() { 6161 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 6162 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 6163 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 6164 // no receiver since it is static method 6165 Node* crc = argument(0); // type: int 6166 Node* src = argument(1); // type: long 6167 Node* offset = argument(3); // type: int 6168 Node* length = argument(4); // type: int 6169 6170 src = ConvL2X(src); // adjust Java long to machine word 6171 Node* base = _gvn.transform(new CastX2PNode(src)); 6172 offset = ConvI2X(offset); 6173 6174 // 'src_start' points to src array + scaled offset 6175 Node* src_start = basic_plus_adr(top(), base, offset); 6176 6177 // Call the stub. 6178 address stubAddr = StubRoutines::updateBytesAdler32(); 6179 const char *stubName = "updateBytesAdler32"; 6180 6181 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 6182 stubAddr, stubName, TypePtr::BOTTOM, 6183 crc, src_start, length); 6184 6185 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6186 set_result(result); 6187 return true; 6188 } 6189 6190 //----------------------------inline_reference_get---------------------------- 6191 // public T java.lang.ref.Reference.get(); 6192 bool LibraryCallKit::inline_reference_get() { 6193 const int referent_offset = java_lang_ref_Reference::referent_offset; 6194 guarantee(referent_offset > 0, "should have already been set"); 6195 6196 // Get the argument: 6197 Node* reference_obj = null_check_receiver(); 6198 if (stopped()) return true; 6199 6200 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr(); 6201 assert(tinst != NULL, "obj is null"); 6202 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 6203 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass(); 6204 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"), 6205 ciSymbol::make("Ljava/lang/Object;"), 6206 false); 6207 assert (field != NULL, "undefined field"); 6208 6209 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset); 6210 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 6211 6212 ciInstanceKlass* klass = env()->Object_klass(); 6213 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass); 6214 6215 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF; 6216 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators); 6217 // Add memory barrier to prevent commoning reads from this field 6218 // across safepoint since GC can change its value. 6219 insert_mem_bar(Op_MemBarCPUOrder); 6220 6221 set_result(result); 6222 return true; 6223 } 6224 6225 6226 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 6227 bool is_exact=true, bool is_static=false, 6228 ciInstanceKlass * fromKls=NULL) { 6229 if (fromKls == NULL) { 6230 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 6231 assert(tinst != NULL, "obj is null"); 6232 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 6233 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 6234 fromKls = tinst->klass()->as_instance_klass(); 6235 } else { 6236 assert(is_static, "only for static field access"); 6237 } 6238 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 6239 ciSymbol::make(fieldTypeString), 6240 is_static); 6241 6242 assert (field != NULL, "undefined field"); 6243 if (field == NULL) return (Node *) NULL; 6244 6245 if (is_static) { 6246 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 6247 fromObj = makecon(tip); 6248 } 6249 6250 // Next code copied from Parse::do_get_xxx(): 6251 6252 // Compute address and memory type. 6253 int offset = field->offset_in_bytes(); 6254 bool is_vol = field->is_volatile(); 6255 ciType* field_klass = field->type(); 6256 assert(field_klass->is_loaded(), "should be loaded"); 6257 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 6258 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 6259 BasicType bt = field->layout_type(); 6260 6261 // Build the resultant type of the load 6262 const Type *type; 6263 if (bt == T_OBJECT) { 6264 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 6265 } else { 6266 type = Type::get_const_basic_type(bt); 6267 } 6268 6269 DecoratorSet decorators = IN_HEAP; 6270 6271 if (is_vol) { 6272 decorators |= MO_SEQ_CST; 6273 } 6274 6275 return access_load_at(fromObj, adr, adr_type, type, bt, decorators); 6276 } 6277 6278 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 6279 bool is_exact = true, bool is_static = false, 6280 ciInstanceKlass * fromKls = NULL) { 6281 if (fromKls == NULL) { 6282 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 6283 assert(tinst != NULL, "obj is null"); 6284 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 6285 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 6286 fromKls = tinst->klass()->as_instance_klass(); 6287 } 6288 else { 6289 assert(is_static, "only for static field access"); 6290 } 6291 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 6292 ciSymbol::make(fieldTypeString), 6293 is_static); 6294 6295 assert(field != NULL, "undefined field"); 6296 assert(!field->is_volatile(), "not defined for volatile fields"); 6297 6298 if (is_static) { 6299 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 6300 fromObj = makecon(tip); 6301 } 6302 6303 // Next code copied from Parse::do_get_xxx(): 6304 6305 // Compute address and memory type. 6306 int offset = field->offset_in_bytes(); 6307 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 6308 6309 return adr; 6310 } 6311 6312 //------------------------------inline_aescrypt_Block----------------------- 6313 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 6314 address stubAddr = NULL; 6315 const char *stubName; 6316 assert(UseAES, "need AES instruction support"); 6317 6318 switch(id) { 6319 case vmIntrinsics::_aescrypt_encryptBlock: 6320 stubAddr = StubRoutines::aescrypt_encryptBlock(); 6321 stubName = "aescrypt_encryptBlock"; 6322 break; 6323 case vmIntrinsics::_aescrypt_decryptBlock: 6324 stubAddr = StubRoutines::aescrypt_decryptBlock(); 6325 stubName = "aescrypt_decryptBlock"; 6326 break; 6327 default: 6328 break; 6329 } 6330 if (stubAddr == NULL) return false; 6331 6332 Node* aescrypt_object = argument(0); 6333 Node* src = argument(1); 6334 Node* src_offset = argument(2); 6335 Node* dest = argument(3); 6336 Node* dest_offset = argument(4); 6337 6338 src = must_be_not_null(src, true); 6339 dest = must_be_not_null(dest, true); 6340 6341 src = access_resolve(src, ACCESS_READ); 6342 dest = access_resolve(dest, ACCESS_WRITE); 6343 6344 // (1) src and dest are arrays. 6345 const Type* src_type = src->Value(&_gvn); 6346 const Type* dest_type = dest->Value(&_gvn); 6347 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6348 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6349 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6350 6351 // for the quick and dirty code we will skip all the checks. 6352 // we are just trying to get the call to be generated. 6353 Node* src_start = src; 6354 Node* dest_start = dest; 6355 if (src_offset != NULL || dest_offset != NULL) { 6356 assert(src_offset != NULL && dest_offset != NULL, ""); 6357 src_start = array_element_address(src, src_offset, T_BYTE); 6358 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6359 } 6360 6361 // now need to get the start of its expanded key array 6362 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6363 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6364 if (k_start == NULL) return false; 6365 6366 if (Matcher::pass_original_key_for_aes()) { 6367 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 6368 // compatibility issues between Java key expansion and SPARC crypto instructions 6369 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 6370 if (original_k_start == NULL) return false; 6371 6372 // Call the stub. 6373 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 6374 stubAddr, stubName, TypePtr::BOTTOM, 6375 src_start, dest_start, k_start, original_k_start); 6376 } else { 6377 // Call the stub. 6378 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 6379 stubAddr, stubName, TypePtr::BOTTOM, 6380 src_start, dest_start, k_start); 6381 } 6382 6383 return true; 6384 } 6385 6386 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 6387 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 6388 address stubAddr = NULL; 6389 const char *stubName = NULL; 6390 6391 assert(UseAES, "need AES instruction support"); 6392 6393 switch(id) { 6394 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 6395 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 6396 stubName = "cipherBlockChaining_encryptAESCrypt"; 6397 break; 6398 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 6399 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 6400 stubName = "cipherBlockChaining_decryptAESCrypt"; 6401 break; 6402 default: 6403 break; 6404 } 6405 if (stubAddr == NULL) return false; 6406 6407 Node* cipherBlockChaining_object = argument(0); 6408 Node* src = argument(1); 6409 Node* src_offset = argument(2); 6410 Node* len = argument(3); 6411 Node* dest = argument(4); 6412 Node* dest_offset = argument(5); 6413 6414 src = must_be_not_null(src, false); 6415 dest = must_be_not_null(dest, false); 6416 6417 src = access_resolve(src, ACCESS_READ); 6418 dest = access_resolve(dest, ACCESS_WRITE); 6419 6420 // (1) src and dest are arrays. 6421 const Type* src_type = src->Value(&_gvn); 6422 const Type* dest_type = dest->Value(&_gvn); 6423 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6424 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6425 assert (top_src != NULL && top_src->klass() != NULL 6426 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6427 6428 // checks are the responsibility of the caller 6429 Node* src_start = src; 6430 Node* dest_start = dest; 6431 if (src_offset != NULL || dest_offset != NULL) { 6432 assert(src_offset != NULL && dest_offset != NULL, ""); 6433 src_start = array_element_address(src, src_offset, T_BYTE); 6434 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6435 } 6436 6437 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6438 // (because of the predicated logic executed earlier). 6439 // so we cast it here safely. 6440 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6441 6442 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6443 if (embeddedCipherObj == NULL) return false; 6444 6445 // cast it to what we know it will be at runtime 6446 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 6447 assert(tinst != NULL, "CBC obj is null"); 6448 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded"); 6449 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6450 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6451 6452 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6453 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6454 const TypeOopPtr* xtype = aklass->as_instance_type(); 6455 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6456 aescrypt_object = _gvn.transform(aescrypt_object); 6457 6458 // we need to get the start of the aescrypt_object's expanded key array 6459 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6460 if (k_start == NULL) return false; 6461 6462 // similarly, get the start address of the r vector 6463 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false); 6464 if (objRvec == NULL) return false; 6465 objRvec = access_resolve(objRvec, ACCESS_WRITE); 6466 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 6467 6468 Node* cbcCrypt; 6469 if (Matcher::pass_original_key_for_aes()) { 6470 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 6471 // compatibility issues between Java key expansion and SPARC crypto instructions 6472 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 6473 if (original_k_start == NULL) return false; 6474 6475 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start 6476 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6477 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 6478 stubAddr, stubName, TypePtr::BOTTOM, 6479 src_start, dest_start, k_start, r_start, len, original_k_start); 6480 } else { 6481 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6482 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6483 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 6484 stubAddr, stubName, TypePtr::BOTTOM, 6485 src_start, dest_start, k_start, r_start, len); 6486 } 6487 6488 // return cipher length (int) 6489 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 6490 set_result(retvalue); 6491 return true; 6492 } 6493 6494 //------------------------------inline_electronicCodeBook_AESCrypt----------------------- 6495 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) { 6496 address stubAddr = NULL; 6497 const char *stubName = NULL; 6498 6499 assert(UseAES, "need AES instruction support"); 6500 6501 switch (id) { 6502 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 6503 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt(); 6504 stubName = "electronicCodeBook_encryptAESCrypt"; 6505 break; 6506 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 6507 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt(); 6508 stubName = "electronicCodeBook_decryptAESCrypt"; 6509 break; 6510 default: 6511 break; 6512 } 6513 6514 if (stubAddr == NULL) return false; 6515 6516 Node* electronicCodeBook_object = argument(0); 6517 Node* src = argument(1); 6518 Node* src_offset = argument(2); 6519 Node* len = argument(3); 6520 Node* dest = argument(4); 6521 Node* dest_offset = argument(5); 6522 6523 // (1) src and dest are arrays. 6524 const Type* src_type = src->Value(&_gvn); 6525 const Type* dest_type = dest->Value(&_gvn); 6526 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6527 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6528 assert(top_src != NULL && top_src->klass() != NULL 6529 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6530 6531 // checks are the responsibility of the caller 6532 Node* src_start = src; 6533 Node* dest_start = dest; 6534 if (src_offset != NULL || dest_offset != NULL) { 6535 assert(src_offset != NULL && dest_offset != NULL, ""); 6536 src_start = array_element_address(src, src_offset, T_BYTE); 6537 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6538 } 6539 6540 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6541 // (because of the predicated logic executed earlier). 6542 // so we cast it here safely. 6543 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6544 6545 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6546 if (embeddedCipherObj == NULL) return false; 6547 6548 // cast it to what we know it will be at runtime 6549 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr(); 6550 assert(tinst != NULL, "ECB obj is null"); 6551 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded"); 6552 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6553 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6554 6555 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6556 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6557 const TypeOopPtr* xtype = aklass->as_instance_type(); 6558 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6559 aescrypt_object = _gvn.transform(aescrypt_object); 6560 6561 // we need to get the start of the aescrypt_object's expanded key array 6562 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6563 if (k_start == NULL) return false; 6564 6565 Node* ecbCrypt; 6566 if (Matcher::pass_original_key_for_aes()) { 6567 // no SPARC version for AES/ECB intrinsics now. 6568 return false; 6569 } 6570 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6571 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP, 6572 OptoRuntime::electronicCodeBook_aescrypt_Type(), 6573 stubAddr, stubName, TypePtr::BOTTOM, 6574 src_start, dest_start, k_start, len); 6575 6576 // return cipher length (int) 6577 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms)); 6578 set_result(retvalue); 6579 return true; 6580 } 6581 6582 //------------------------------inline_counterMode_AESCrypt----------------------- 6583 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 6584 assert(UseAES, "need AES instruction support"); 6585 if (!UseAESCTRIntrinsics) return false; 6586 6587 address stubAddr = NULL; 6588 const char *stubName = NULL; 6589 if (id == vmIntrinsics::_counterMode_AESCrypt) { 6590 stubAddr = StubRoutines::counterMode_AESCrypt(); 6591 stubName = "counterMode_AESCrypt"; 6592 } 6593 if (stubAddr == NULL) return false; 6594 6595 Node* counterMode_object = argument(0); 6596 Node* src = argument(1); 6597 Node* src_offset = argument(2); 6598 Node* len = argument(3); 6599 Node* dest = argument(4); 6600 Node* dest_offset = argument(5); 6601 6602 src = access_resolve(src, ACCESS_READ); 6603 dest = access_resolve(dest, ACCESS_WRITE); 6604 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE); 6605 6606 // (1) src and dest are arrays. 6607 const Type* src_type = src->Value(&_gvn); 6608 const Type* dest_type = dest->Value(&_gvn); 6609 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6610 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6611 assert(top_src != NULL && top_src->klass() != NULL && 6612 top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6613 6614 // checks are the responsibility of the caller 6615 Node* src_start = src; 6616 Node* dest_start = dest; 6617 if (src_offset != NULL || dest_offset != NULL) { 6618 assert(src_offset != NULL && dest_offset != NULL, ""); 6619 src_start = array_element_address(src, src_offset, T_BYTE); 6620 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6621 } 6622 6623 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6624 // (because of the predicated logic executed earlier). 6625 // so we cast it here safely. 6626 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6627 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6628 if (embeddedCipherObj == NULL) return false; 6629 // cast it to what we know it will be at runtime 6630 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 6631 assert(tinst != NULL, "CTR obj is null"); 6632 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded"); 6633 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6634 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6635 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6636 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6637 const TypeOopPtr* xtype = aklass->as_instance_type(); 6638 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6639 aescrypt_object = _gvn.transform(aescrypt_object); 6640 // we need to get the start of the aescrypt_object's expanded key array 6641 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6642 if (k_start == NULL) return false; 6643 // similarly, get the start address of the r vector 6644 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false); 6645 if (obj_counter == NULL) return false; 6646 obj_counter = access_resolve(obj_counter, ACCESS_WRITE); 6647 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 6648 6649 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false); 6650 if (saved_encCounter == NULL) return false; 6651 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE); 6652 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 6653 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 6654 6655 Node* ctrCrypt; 6656 if (Matcher::pass_original_key_for_aes()) { 6657 // no SPARC version for AES/CTR intrinsics now. 6658 return false; 6659 } 6660 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6661 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6662 OptoRuntime::counterMode_aescrypt_Type(), 6663 stubAddr, stubName, TypePtr::BOTTOM, 6664 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 6665 6666 // return cipher length (int) 6667 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 6668 set_result(retvalue); 6669 return true; 6670 } 6671 6672 //------------------------------get_key_start_from_aescrypt_object----------------------- 6673 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 6674 #if defined(PPC64) || defined(S390) 6675 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys. 6676 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns. 6677 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption. 6678 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]). 6679 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false); 6680 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6681 if (objSessionK == NULL) { 6682 return (Node *) NULL; 6683 } 6684 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS); 6685 #else 6686 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false); 6687 #endif // PPC64 6688 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6689 if (objAESCryptKey == NULL) return (Node *) NULL; 6690 6691 // now have the array, need to get the start address of the K array 6692 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ); 6693 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 6694 return k_start; 6695 } 6696 6697 //------------------------------get_original_key_start_from_aescrypt_object----------------------- 6698 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) { 6699 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false); 6700 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6701 if (objAESCryptKey == NULL) return (Node *) NULL; 6702 6703 // now have the array, need to get the start address of the lastKey array 6704 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ); 6705 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE); 6706 return original_k_start; 6707 } 6708 6709 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 6710 // Return node representing slow path of predicate check. 6711 // the pseudo code we want to emulate with this predicate is: 6712 // for encryption: 6713 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6714 // for decryption: 6715 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6716 // note cipher==plain is more conservative than the original java code but that's OK 6717 // 6718 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 6719 // The receiver was checked for NULL already. 6720 Node* objCBC = argument(0); 6721 6722 Node* src = argument(1); 6723 Node* dest = argument(4); 6724 6725 // Load embeddedCipher field of CipherBlockChaining object. 6726 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6727 6728 // get AESCrypt klass for instanceOf check 6729 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6730 // will have same classloader as CipherBlockChaining object 6731 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 6732 assert(tinst != NULL, "CBCobj is null"); 6733 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded"); 6734 6735 // we want to do an instanceof comparison against the AESCrypt class 6736 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6737 if (!klass_AESCrypt->is_loaded()) { 6738 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6739 Node* ctrl = control(); 6740 set_control(top()); // no regular fast path 6741 return ctrl; 6742 } 6743 6744 src = must_be_not_null(src, true); 6745 dest = must_be_not_null(dest, true); 6746 6747 // Resolve oops to stable for CmpP below. 6748 src = access_resolve(src, 0); 6749 dest = access_resolve(dest, 0); 6750 6751 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6752 6753 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6754 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6755 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6756 6757 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6758 6759 // for encryption, we are done 6760 if (!decrypting) 6761 return instof_false; // even if it is NULL 6762 6763 // for decryption, we need to add a further check to avoid 6764 // taking the intrinsic path when cipher and plain are the same 6765 // see the original java code for why. 6766 RegionNode* region = new RegionNode(3); 6767 region->init_req(1, instof_false); 6768 6769 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6770 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6771 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6772 region->init_req(2, src_dest_conjoint); 6773 6774 record_for_igvn(region); 6775 return _gvn.transform(region); 6776 } 6777 6778 //----------------------------inline_electronicCodeBook_AESCrypt_predicate---------------------------- 6779 // Return node representing slow path of predicate check. 6780 // the pseudo code we want to emulate with this predicate is: 6781 // for encryption: 6782 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6783 // for decryption: 6784 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6785 // note cipher==plain is more conservative than the original java code but that's OK 6786 // 6787 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) { 6788 // The receiver was checked for NULL already. 6789 Node* objECB = argument(0); 6790 6791 // Load embeddedCipher field of ElectronicCodeBook object. 6792 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6793 6794 // get AESCrypt klass for instanceOf check 6795 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6796 // will have same classloader as ElectronicCodeBook object 6797 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr(); 6798 assert(tinst != NULL, "ECBobj is null"); 6799 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded"); 6800 6801 // we want to do an instanceof comparison against the AESCrypt class 6802 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6803 if (!klass_AESCrypt->is_loaded()) { 6804 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6805 Node* ctrl = control(); 6806 set_control(top()); // no regular fast path 6807 return ctrl; 6808 } 6809 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6810 6811 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6812 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6813 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6814 6815 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6816 6817 // for encryption, we are done 6818 if (!decrypting) 6819 return instof_false; // even if it is NULL 6820 6821 // for decryption, we need to add a further check to avoid 6822 // taking the intrinsic path when cipher and plain are the same 6823 // see the original java code for why. 6824 RegionNode* region = new RegionNode(3); 6825 region->init_req(1, instof_false); 6826 Node* src = argument(1); 6827 Node* dest = argument(4); 6828 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6829 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6830 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6831 region->init_req(2, src_dest_conjoint); 6832 6833 record_for_igvn(region); 6834 return _gvn.transform(region); 6835 } 6836 6837 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 6838 // Return node representing slow path of predicate check. 6839 // the pseudo code we want to emulate with this predicate is: 6840 // for encryption: 6841 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6842 // for decryption: 6843 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6844 // note cipher==plain is more conservative than the original java code but that's OK 6845 // 6846 6847 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 6848 // The receiver was checked for NULL already. 6849 Node* objCTR = argument(0); 6850 6851 // Load embeddedCipher field of CipherBlockChaining object. 6852 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6853 6854 // get AESCrypt klass for instanceOf check 6855 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6856 // will have same classloader as CipherBlockChaining object 6857 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 6858 assert(tinst != NULL, "CTRobj is null"); 6859 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded"); 6860 6861 // we want to do an instanceof comparison against the AESCrypt class 6862 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6863 if (!klass_AESCrypt->is_loaded()) { 6864 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6865 Node* ctrl = control(); 6866 set_control(top()); // no regular fast path 6867 return ctrl; 6868 } 6869 6870 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6871 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6872 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6873 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6874 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6875 6876 return instof_false; // even if it is NULL 6877 } 6878 6879 //------------------------------inline_ghash_processBlocks 6880 bool LibraryCallKit::inline_ghash_processBlocks() { 6881 address stubAddr; 6882 const char *stubName; 6883 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 6884 6885 stubAddr = StubRoutines::ghash_processBlocks(); 6886 stubName = "ghash_processBlocks"; 6887 6888 Node* data = argument(0); 6889 Node* offset = argument(1); 6890 Node* len = argument(2); 6891 Node* state = argument(3); 6892 Node* subkeyH = argument(4); 6893 6894 state = must_be_not_null(state, true); 6895 subkeyH = must_be_not_null(subkeyH, true); 6896 data = must_be_not_null(data, true); 6897 6898 state = access_resolve(state, ACCESS_WRITE); 6899 subkeyH = access_resolve(subkeyH, ACCESS_READ); 6900 data = access_resolve(data, ACCESS_READ); 6901 6902 Node* state_start = array_element_address(state, intcon(0), T_LONG); 6903 assert(state_start, "state is NULL"); 6904 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 6905 assert(subkeyH_start, "subkeyH is NULL"); 6906 Node* data_start = array_element_address(data, offset, T_BYTE); 6907 assert(data_start, "data is NULL"); 6908 6909 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 6910 OptoRuntime::ghash_processBlocks_Type(), 6911 stubAddr, stubName, TypePtr::BOTTOM, 6912 state_start, subkeyH_start, data_start, len); 6913 return true; 6914 } 6915 6916 bool LibraryCallKit::inline_base64_encodeBlock() { 6917 address stubAddr; 6918 const char *stubName; 6919 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 6920 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters"); 6921 stubAddr = StubRoutines::base64_encodeBlock(); 6922 stubName = "encodeBlock"; 6923 6924 if (!stubAddr) return false; 6925 Node* base64obj = argument(0); 6926 Node* src = argument(1); 6927 Node* offset = argument(2); 6928 Node* len = argument(3); 6929 Node* dest = argument(4); 6930 Node* dp = argument(5); 6931 Node* isURL = argument(6); 6932 6933 src = must_be_not_null(src, true); 6934 src = access_resolve(src, ACCESS_READ); 6935 dest = must_be_not_null(dest, true); 6936 dest = access_resolve(dest, ACCESS_WRITE); 6937 6938 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 6939 assert(src_start, "source array is NULL"); 6940 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 6941 assert(dest_start, "destination array is NULL"); 6942 6943 Node* base64 = make_runtime_call(RC_LEAF, 6944 OptoRuntime::base64_encodeBlock_Type(), 6945 stubAddr, stubName, TypePtr::BOTTOM, 6946 src_start, offset, len, dest_start, dp, isURL); 6947 return true; 6948 } 6949 6950 //------------------------------inline_sha_implCompress----------------------- 6951 // 6952 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 6953 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 6954 // 6955 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 6956 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 6957 // 6958 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 6959 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 6960 // 6961 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) { 6962 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 6963 6964 Node* sha_obj = argument(0); 6965 Node* src = argument(1); // type oop 6966 Node* ofs = argument(2); // type int 6967 6968 const Type* src_type = src->Value(&_gvn); 6969 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6970 if (top_src == NULL || top_src->klass() == NULL) { 6971 // failed array check 6972 return false; 6973 } 6974 // Figure out the size and type of the elements we will be copying. 6975 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6976 if (src_elem != T_BYTE) { 6977 return false; 6978 } 6979 // 'src_start' points to src array + offset 6980 src = must_be_not_null(src, true); 6981 src = access_resolve(src, ACCESS_READ); 6982 Node* src_start = array_element_address(src, ofs, src_elem); 6983 Node* state = NULL; 6984 address stubAddr; 6985 const char *stubName; 6986 6987 switch(id) { 6988 case vmIntrinsics::_sha_implCompress: 6989 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 6990 state = get_state_from_sha_object(sha_obj); 6991 stubAddr = StubRoutines::sha1_implCompress(); 6992 stubName = "sha1_implCompress"; 6993 break; 6994 case vmIntrinsics::_sha2_implCompress: 6995 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 6996 state = get_state_from_sha_object(sha_obj); 6997 stubAddr = StubRoutines::sha256_implCompress(); 6998 stubName = "sha256_implCompress"; 6999 break; 7000 case vmIntrinsics::_sha5_implCompress: 7001 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 7002 state = get_state_from_sha5_object(sha_obj); 7003 stubAddr = StubRoutines::sha512_implCompress(); 7004 stubName = "sha512_implCompress"; 7005 break; 7006 default: 7007 fatal_unexpected_iid(id); 7008 return false; 7009 } 7010 if (state == NULL) return false; 7011 7012 assert(stubAddr != NULL, "Stub is generated"); 7013 if (stubAddr == NULL) return false; 7014 7015 // Call the stub. 7016 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(), 7017 stubAddr, stubName, TypePtr::BOTTOM, 7018 src_start, state); 7019 7020 return true; 7021 } 7022 7023 //------------------------------inline_digestBase_implCompressMB----------------------- 7024 // 7025 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array. 7026 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 7027 // 7028 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 7029 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 7030 "need SHA1/SHA256/SHA512 instruction support"); 7031 assert((uint)predicate < 3, "sanity"); 7032 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 7033 7034 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already. 7035 Node* src = argument(1); // byte[] array 7036 Node* ofs = argument(2); // type int 7037 Node* limit = argument(3); // type int 7038 7039 const Type* src_type = src->Value(&_gvn); 7040 const TypeAryPtr* top_src = src_type->isa_aryptr(); 7041 if (top_src == NULL || top_src->klass() == NULL) { 7042 // failed array check 7043 return false; 7044 } 7045 // Figure out the size and type of the elements we will be copying. 7046 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 7047 if (src_elem != T_BYTE) { 7048 return false; 7049 } 7050 // 'src_start' points to src array + offset 7051 src = must_be_not_null(src, false); 7052 src = access_resolve(src, ACCESS_READ); 7053 Node* src_start = array_element_address(src, ofs, src_elem); 7054 7055 const char* klass_SHA_name = NULL; 7056 const char* stub_name = NULL; 7057 address stub_addr = NULL; 7058 bool long_state = false; 7059 7060 switch (predicate) { 7061 case 0: 7062 if (UseSHA1Intrinsics) { 7063 klass_SHA_name = "sun/security/provider/SHA"; 7064 stub_name = "sha1_implCompressMB"; 7065 stub_addr = StubRoutines::sha1_implCompressMB(); 7066 } 7067 break; 7068 case 1: 7069 if (UseSHA256Intrinsics) { 7070 klass_SHA_name = "sun/security/provider/SHA2"; 7071 stub_name = "sha256_implCompressMB"; 7072 stub_addr = StubRoutines::sha256_implCompressMB(); 7073 } 7074 break; 7075 case 2: 7076 if (UseSHA512Intrinsics) { 7077 klass_SHA_name = "sun/security/provider/SHA5"; 7078 stub_name = "sha512_implCompressMB"; 7079 stub_addr = StubRoutines::sha512_implCompressMB(); 7080 long_state = true; 7081 } 7082 break; 7083 default: 7084 fatal("unknown SHA intrinsic predicate: %d", predicate); 7085 } 7086 if (klass_SHA_name != NULL) { 7087 assert(stub_addr != NULL, "Stub is generated"); 7088 if (stub_addr == NULL) return false; 7089 7090 // get DigestBase klass to lookup for SHA klass 7091 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 7092 assert(tinst != NULL, "digestBase_obj is not instance???"); 7093 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 7094 7095 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 7096 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded"); 7097 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 7098 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit); 7099 } 7100 return false; 7101 } 7102 //------------------------------inline_sha_implCompressMB----------------------- 7103 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA, 7104 bool long_state, address stubAddr, const char *stubName, 7105 Node* src_start, Node* ofs, Node* limit) { 7106 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA); 7107 const TypeOopPtr* xtype = aklass->as_instance_type(); 7108 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 7109 sha_obj = _gvn.transform(sha_obj); 7110 7111 Node* state; 7112 if (long_state) { 7113 state = get_state_from_sha5_object(sha_obj); 7114 } else { 7115 state = get_state_from_sha_object(sha_obj); 7116 } 7117 if (state == NULL) return false; 7118 7119 // Call the stub. 7120 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 7121 OptoRuntime::digestBase_implCompressMB_Type(), 7122 stubAddr, stubName, TypePtr::BOTTOM, 7123 src_start, state, ofs, limit); 7124 // return ofs (int) 7125 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 7126 set_result(result); 7127 7128 return true; 7129 } 7130 7131 //------------------------------get_state_from_sha_object----------------------- 7132 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) { 7133 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false); 7134 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2"); 7135 if (sha_state == NULL) return (Node *) NULL; 7136 7137 // now have the array, need to get the start address of the state array 7138 sha_state = access_resolve(sha_state, ACCESS_WRITE); 7139 Node* state = array_element_address(sha_state, intcon(0), T_INT); 7140 return state; 7141 } 7142 7143 //------------------------------get_state_from_sha5_object----------------------- 7144 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) { 7145 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false); 7146 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5"); 7147 if (sha_state == NULL) return (Node *) NULL; 7148 7149 // now have the array, need to get the start address of the state array 7150 sha_state = access_resolve(sha_state, ACCESS_WRITE); 7151 Node* state = array_element_address(sha_state, intcon(0), T_LONG); 7152 return state; 7153 } 7154 7155 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 7156 // Return node representing slow path of predicate check. 7157 // the pseudo code we want to emulate with this predicate is: 7158 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath 7159 // 7160 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 7161 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 7162 "need SHA1/SHA256/SHA512 instruction support"); 7163 assert((uint)predicate < 3, "sanity"); 7164 7165 // The receiver was checked for NULL already. 7166 Node* digestBaseObj = argument(0); 7167 7168 // get DigestBase klass for instanceOf check 7169 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 7170 assert(tinst != NULL, "digestBaseObj is null"); 7171 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 7172 7173 const char* klass_SHA_name = NULL; 7174 switch (predicate) { 7175 case 0: 7176 if (UseSHA1Intrinsics) { 7177 // we want to do an instanceof comparison against the SHA class 7178 klass_SHA_name = "sun/security/provider/SHA"; 7179 } 7180 break; 7181 case 1: 7182 if (UseSHA256Intrinsics) { 7183 // we want to do an instanceof comparison against the SHA2 class 7184 klass_SHA_name = "sun/security/provider/SHA2"; 7185 } 7186 break; 7187 case 2: 7188 if (UseSHA512Intrinsics) { 7189 // we want to do an instanceof comparison against the SHA5 class 7190 klass_SHA_name = "sun/security/provider/SHA5"; 7191 } 7192 break; 7193 default: 7194 fatal("unknown SHA intrinsic predicate: %d", predicate); 7195 } 7196 7197 ciKlass* klass_SHA = NULL; 7198 if (klass_SHA_name != NULL) { 7199 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name)); 7200 } 7201 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) { 7202 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 7203 Node* ctrl = control(); 7204 set_control(top()); // no intrinsic path 7205 return ctrl; 7206 } 7207 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass(); 7208 7209 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA))); 7210 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1))); 7211 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 7212 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 7213 7214 return instof_false; // even if it is NULL 7215 } 7216 7217 //-------------inline_fma----------------------------------- 7218 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) { 7219 Node *a = NULL; 7220 Node *b = NULL; 7221 Node *c = NULL; 7222 Node* result = NULL; 7223 switch (id) { 7224 case vmIntrinsics::_fmaD: 7225 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each."); 7226 // no receiver since it is static method 7227 a = round_double_node(argument(0)); 7228 b = round_double_node(argument(2)); 7229 c = round_double_node(argument(4)); 7230 result = _gvn.transform(new FmaDNode(control(), a, b, c)); 7231 break; 7232 case vmIntrinsics::_fmaF: 7233 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each."); 7234 a = argument(0); 7235 b = argument(1); 7236 c = argument(2); 7237 result = _gvn.transform(new FmaFNode(control(), a, b, c)); 7238 break; 7239 default: 7240 fatal_unexpected_iid(id); break; 7241 } 7242 set_result(result); 7243 return true; 7244 } 7245 7246 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) { 7247 // argument(0) is receiver 7248 Node* codePoint = argument(1); 7249 Node* n = NULL; 7250 7251 switch (id) { 7252 case vmIntrinsics::_isDigit : 7253 n = new DigitNode(control(), codePoint); 7254 break; 7255 case vmIntrinsics::_isLowerCase : 7256 n = new LowerCaseNode(control(), codePoint); 7257 break; 7258 case vmIntrinsics::_isUpperCase : 7259 n = new UpperCaseNode(control(), codePoint); 7260 break; 7261 case vmIntrinsics::_isWhitespace : 7262 n = new WhitespaceNode(control(), codePoint); 7263 break; 7264 default: 7265 fatal_unexpected_iid(id); 7266 } 7267 7268 set_result(_gvn.transform(n)); 7269 return true; 7270 } 7271 7272 //------------------------------inline_fp_min_max------------------------------ 7273 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) { 7274 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416. 7275 7276 // The intrinsic should be used only when the API branches aren't predictable, 7277 // the last one performing the most important comparison. The following heuristic 7278 // uses the branch statistics to eventually bail out if necessary. 7279 7280 ciMethodData *md = callee()->method_data(); 7281 7282 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) { 7283 ciCallProfile cp = caller()->call_profile_at_bci(bci()); 7284 7285 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) { 7286 // Bail out if the call-site didn't contribute enough to the statistics. 7287 return false; 7288 } 7289 7290 uint taken = 0, not_taken = 0; 7291 7292 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) { 7293 if (p->is_BranchData()) { 7294 taken = ((ciBranchData*)p)->taken(); 7295 not_taken = ((ciBranchData*)p)->not_taken(); 7296 } 7297 } 7298 7299 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count()); 7300 balance = balance < 0 ? -balance : balance; 7301 if ( balance > 0.2 ) { 7302 // Bail out if the most important branch is predictable enough. 7303 return false; 7304 } 7305 } 7306 */ 7307 7308 Node *a = NULL; 7309 Node *b = NULL; 7310 Node *n = NULL; 7311 switch (id) { 7312 case vmIntrinsics::_maxF: 7313 case vmIntrinsics::_minF: 7314 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each."); 7315 a = argument(0); 7316 b = argument(1); 7317 break; 7318 case vmIntrinsics::_maxD: 7319 case vmIntrinsics::_minD: 7320 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each."); 7321 a = round_double_node(argument(0)); 7322 b = round_double_node(argument(2)); 7323 break; 7324 default: 7325 fatal_unexpected_iid(id); 7326 break; 7327 } 7328 switch (id) { 7329 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break; 7330 case vmIntrinsics::_minF: n = new MinFNode(a, b); break; 7331 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break; 7332 case vmIntrinsics::_minD: n = new MinDNode(a, b); break; 7333 default: fatal_unexpected_iid(id); break; 7334 } 7335 set_result(_gvn.transform(n)); 7336 return true; 7337 } 7338 7339 bool LibraryCallKit::inline_profileBoolean() { 7340 Node* counts = argument(1); 7341 const TypeAryPtr* ary = NULL; 7342 ciArray* aobj = NULL; 7343 if (counts->is_Con() 7344 && (ary = counts->bottom_type()->isa_aryptr()) != NULL 7345 && (aobj = ary->const_oop()->as_array()) != NULL 7346 && (aobj->length() == 2)) { 7347 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 7348 jint false_cnt = aobj->element_value(0).as_int(); 7349 jint true_cnt = aobj->element_value(1).as_int(); 7350 7351 if (C->log() != NULL) { 7352 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 7353 false_cnt, true_cnt); 7354 } 7355 7356 if (false_cnt + true_cnt == 0) { 7357 // According to profile, never executed. 7358 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 7359 Deoptimization::Action_reinterpret); 7360 return true; 7361 } 7362 7363 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 7364 // is a number of each value occurrences. 7365 Node* result = argument(0); 7366 if (false_cnt == 0 || true_cnt == 0) { 7367 // According to profile, one value has been never seen. 7368 int expected_val = (false_cnt == 0) ? 1 : 0; 7369 7370 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 7371 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 7372 7373 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 7374 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 7375 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 7376 7377 { // Slow path: uncommon trap for never seen value and then reexecute 7378 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 7379 // the value has been seen at least once. 7380 PreserveJVMState pjvms(this); 7381 PreserveReexecuteState preexecs(this); 7382 jvms()->set_should_reexecute(true); 7383 7384 set_control(slow_path); 7385 set_i_o(i_o()); 7386 7387 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 7388 Deoptimization::Action_reinterpret); 7389 } 7390 // The guard for never seen value enables sharpening of the result and 7391 // returning a constant. It allows to eliminate branches on the same value 7392 // later on. 7393 set_control(fast_path); 7394 result = intcon(expected_val); 7395 } 7396 // Stop profiling. 7397 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 7398 // By replacing method body with profile data (represented as ProfileBooleanNode 7399 // on IR level) we effectively disable profiling. 7400 // It enables full speed execution once optimized code is generated. 7401 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 7402 C->record_for_igvn(profile); 7403 set_result(profile); 7404 return true; 7405 } else { 7406 // Continue profiling. 7407 // Profile data isn't available at the moment. So, execute method's bytecode version. 7408 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 7409 // is compiled and counters aren't available since corresponding MethodHandle 7410 // isn't a compile-time constant. 7411 return false; 7412 } 7413 } 7414 7415 bool LibraryCallKit::inline_isCompileConstant() { 7416 Node* n = argument(0); 7417 set_result(n->is_Con() ? intcon(1) : intcon(0)); 7418 return true; 7419 }