1 /* 2 * Copyright (c) 1999, 2020, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/macroAssembler.hpp" 27 #include "ci/ciUtilities.inline.hpp" 28 #include "classfile/systemDictionary.hpp" 29 #include "classfile/vmSymbols.hpp" 30 #include "compiler/compileBroker.hpp" 31 #include "compiler/compileLog.hpp" 32 #include "gc/shared/barrierSet.hpp" 33 #include "jfr/support/jfrIntrinsics.hpp" 34 #include "memory/resourceArea.hpp" 35 #include "oops/klass.inline.hpp" 36 #include "oops/objArrayKlass.hpp" 37 #include "opto/addnode.hpp" 38 #include "opto/arraycopynode.hpp" 39 #include "opto/c2compiler.hpp" 40 #include "opto/callGenerator.hpp" 41 #include "opto/castnode.hpp" 42 #include "opto/cfgnode.hpp" 43 #include "opto/convertnode.hpp" 44 #include "opto/countbitsnode.hpp" 45 #include "opto/intrinsicnode.hpp" 46 #include "opto/idealKit.hpp" 47 #include "opto/mathexactnode.hpp" 48 #include "opto/movenode.hpp" 49 #include "opto/mulnode.hpp" 50 #include "opto/narrowptrnode.hpp" 51 #include "opto/opaquenode.hpp" 52 #include "opto/parse.hpp" 53 #include "opto/runtime.hpp" 54 #include "opto/rootnode.hpp" 55 #include "opto/subnode.hpp" 56 #include "prims/nativeLookup.hpp" 57 #include "prims/unsafe.hpp" 58 #include "runtime/objectMonitor.hpp" 59 #include "runtime/sharedRuntime.hpp" 60 #include "utilities/macros.hpp" 61 #include "utilities/powerOfTwo.hpp" 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_mirror_from_klass(Node* klass); 168 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null, 169 RegionNode* region, int null_path, 170 int offset); 171 Node* load_klass_from_mirror(Node* mirror, bool never_see_null, 172 RegionNode* region, int null_path) { 173 int offset = java_lang_Class::klass_offset(); 174 return load_klass_from_mirror_common(mirror, never_see_null, 175 region, null_path, 176 offset); 177 } 178 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null, 179 RegionNode* region, int null_path) { 180 int offset = java_lang_Class::array_klass_offset(); 181 return load_klass_from_mirror_common(mirror, never_see_null, 182 region, null_path, 183 offset); 184 } 185 Node* generate_access_flags_guard(Node* kls, 186 int modifier_mask, int modifier_bits, 187 RegionNode* region); 188 Node* generate_interface_guard(Node* kls, RegionNode* region); 189 Node* generate_hidden_class_guard(Node* kls, RegionNode* region); 190 Node* generate_array_guard(Node* kls, RegionNode* region) { 191 return generate_array_guard_common(kls, region, false, false); 192 } 193 Node* generate_non_array_guard(Node* kls, RegionNode* region) { 194 return generate_array_guard_common(kls, region, false, true); 195 } 196 Node* generate_objArray_guard(Node* kls, RegionNode* region) { 197 return generate_array_guard_common(kls, region, true, false); 198 } 199 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) { 200 return generate_array_guard_common(kls, region, true, true); 201 } 202 Node* generate_array_guard_common(Node* kls, RegionNode* region, 203 bool obj_array, bool not_array); 204 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region); 205 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id, 206 bool is_virtual = false, bool is_static = false); 207 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) { 208 return generate_method_call(method_id, false, true); 209 } 210 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) { 211 return generate_method_call(method_id, true, false); 212 } 213 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 214 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls); 215 216 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae); 217 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae); 218 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae); 219 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae); 220 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 221 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae); 222 bool inline_string_indexOfChar(); 223 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae); 224 bool inline_string_toBytesU(); 225 bool inline_string_getCharsU(); 226 bool inline_string_copy(bool compress); 227 bool inline_string_char_access(bool is_store); 228 Node* round_double_node(Node* n); 229 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName); 230 bool inline_math_native(vmIntrinsics::ID id); 231 bool inline_math(vmIntrinsics::ID id); 232 bool inline_double_math(vmIntrinsics::ID id); 233 template <typename OverflowOp> 234 bool inline_math_overflow(Node* arg1, Node* arg2); 235 void inline_math_mathExact(Node* math, Node* test); 236 bool inline_math_addExactI(bool is_increment); 237 bool inline_math_addExactL(bool is_increment); 238 bool inline_math_multiplyExactI(); 239 bool inline_math_multiplyExactL(); 240 bool inline_math_multiplyHigh(); 241 bool inline_math_negateExactI(); 242 bool inline_math_negateExactL(); 243 bool inline_math_subtractExactI(bool is_decrement); 244 bool inline_math_subtractExactL(bool is_decrement); 245 bool inline_min_max(vmIntrinsics::ID id); 246 bool inline_notify(vmIntrinsics::ID id); 247 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y); 248 // This returns Type::AnyPtr, RawPtr, or OopPtr. 249 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type); 250 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false); 251 252 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind; 253 DecoratorSet mo_decorator_for_access_kind(AccessKind kind); 254 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned); 255 static bool klass_needs_init_guard(Node* kls); 256 bool inline_unsafe_allocate(); 257 bool inline_unsafe_newArray(bool uninitialized); 258 bool inline_unsafe_writeback0(); 259 bool inline_unsafe_writebackSync0(bool is_pre); 260 bool inline_unsafe_copyMemory(); 261 bool inline_native_currentThread(); 262 263 bool inline_native_time_funcs(address method, const char* funcName); 264 #ifdef JFR_HAVE_INTRINSICS 265 bool inline_native_classID(); 266 bool inline_native_getEventWriter(); 267 #endif 268 bool inline_native_Class_query(vmIntrinsics::ID id); 269 bool inline_native_subtype_check(); 270 bool inline_native_getLength(); 271 bool inline_array_copyOf(bool is_copyOfRange); 272 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae); 273 bool inline_preconditions_checkIndex(); 274 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array); 275 bool inline_native_clone(bool is_virtual); 276 bool inline_native_Reflection_getCallerClass(); 277 // Helper function for inlining native object hash method 278 bool inline_native_hashcode(bool is_virtual, bool is_static); 279 bool inline_native_getClass(); 280 281 // Helper functions for inlining arraycopy 282 bool inline_arraycopy(); 283 AllocateArrayNode* tightly_coupled_allocation(Node* ptr, 284 RegionNode* slow_region); 285 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp); 286 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp, 287 uint new_idx); 288 289 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind; 290 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind); 291 bool inline_unsafe_fence(vmIntrinsics::ID id); 292 bool inline_onspinwait(); 293 bool inline_fp_conversions(vmIntrinsics::ID id); 294 bool inline_number_methods(vmIntrinsics::ID id); 295 bool inline_reference_get(); 296 bool inline_Class_cast(); 297 bool inline_aescrypt_Block(vmIntrinsics::ID id); 298 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id); 299 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id); 300 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id); 301 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting); 302 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting); 303 Node* inline_counterMode_AESCrypt_predicate(); 304 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object); 305 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object); 306 bool inline_ghash_processBlocks(); 307 bool inline_base64_encodeBlock(); 308 bool inline_digestBase_implCompress(vmIntrinsics::ID id); 309 bool inline_digestBase_implCompressMB(int predicate); 310 bool inline_digestBase_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass, 311 bool long_state, address stubAddr, const char *stubName, 312 Node* src_start, Node* ofs, Node* limit); 313 Node* get_state_from_digest_object(Node *digestBase_object); 314 Node* get_long_state_from_digest_object(Node *digestBase_object); 315 Node* inline_digestBase_implCompressMB_predicate(int predicate); 316 bool inline_encodeISOArray(); 317 bool inline_updateCRC32(); 318 bool inline_updateBytesCRC32(); 319 bool inline_updateByteBufferCRC32(); 320 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class); 321 bool inline_updateBytesCRC32C(); 322 bool inline_updateDirectByteBufferCRC32C(); 323 bool inline_updateBytesAdler32(); 324 bool inline_updateByteBufferAdler32(); 325 bool inline_multiplyToLen(); 326 bool inline_hasNegatives(); 327 bool inline_squareToLen(); 328 bool inline_mulAdd(); 329 bool inline_montgomeryMultiply(); 330 bool inline_montgomerySquare(); 331 bool inline_bigIntegerShift(bool isRightShift); 332 bool inline_vectorizedMismatch(); 333 bool inline_fma(vmIntrinsics::ID id); 334 bool inline_character_compare(vmIntrinsics::ID id); 335 bool inline_fp_min_max(vmIntrinsics::ID id); 336 337 bool inline_profileBoolean(); 338 bool inline_isCompileConstant(); 339 void clear_upper_avx() { 340 #ifdef X86 341 if (UseAVX >= 2) { 342 C->set_clear_upper_avx(true); 343 } 344 #endif 345 } 346 }; 347 348 //---------------------------make_vm_intrinsic---------------------------- 349 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) { 350 vmIntrinsics::ID id = m->intrinsic_id(); 351 assert(id != vmIntrinsics::_none, "must be a VM intrinsic"); 352 353 if (!m->is_loaded()) { 354 // Do not attempt to inline unloaded methods. 355 return NULL; 356 } 357 358 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization); 359 bool is_available = false; 360 361 { 362 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag 363 // the compiler must transition to '_thread_in_vm' state because both 364 // methods access VM-internal data. 365 VM_ENTRY_MARK; 366 methodHandle mh(THREAD, m->get_Method()); 367 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) && 368 !C->directive()->is_intrinsic_disabled(mh) && 369 !vmIntrinsics::is_disabled_by_flags(mh); 370 371 } 372 373 if (is_available) { 374 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility"); 375 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?"); 376 return new LibraryIntrinsic(m, is_virtual, 377 vmIntrinsics::predicates_needed(id), 378 vmIntrinsics::does_virtual_dispatch(id), 379 (vmIntrinsics::ID) id); 380 } else { 381 return NULL; 382 } 383 } 384 385 //----------------------register_library_intrinsics----------------------- 386 // Initialize this file's data structures, for each Compile instance. 387 void Compile::register_library_intrinsics() { 388 // Nothing to do here. 389 } 390 391 JVMState* LibraryIntrinsic::generate(JVMState* jvms) { 392 LibraryCallKit kit(jvms, this); 393 Compile* C = kit.C; 394 int nodes = C->unique(); 395 #ifndef PRODUCT 396 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 397 char buf[1000]; 398 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 399 tty->print_cr("Intrinsic %s", str); 400 } 401 #endif 402 ciMethod* callee = kit.callee(); 403 const int bci = kit.bci(); 404 405 // Try to inline the intrinsic. 406 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) && 407 kit.try_to_inline(_last_predicate)) { 408 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)" 409 : "(intrinsic)"; 410 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 411 if (C->print_intrinsics() || C->print_inlining()) { 412 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 413 } 414 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 415 if (C->log()) { 416 C->log()->elem("intrinsic id='%s'%s nodes='%d'", 417 vmIntrinsics::name_at(intrinsic_id()), 418 (is_virtual() ? " virtual='1'" : ""), 419 C->unique() - nodes); 420 } 421 // Push the result from the inlined method onto the stack. 422 kit.push_result(); 423 C->print_inlining_update(this); 424 return kit.transfer_exceptions_into_jvms(); 425 } 426 427 // The intrinsic bailed out 428 if (jvms->has_method()) { 429 // Not a root compile. 430 const char* msg; 431 if (callee->intrinsic_candidate()) { 432 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)"; 433 } else { 434 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated" 435 : "failed to inline (intrinsic), method not annotated"; 436 } 437 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg); 438 if (C->print_intrinsics() || C->print_inlining()) { 439 C->print_inlining(callee, jvms->depth() - 1, bci, msg); 440 } 441 } else { 442 // Root compile 443 ResourceMark rm; 444 stringStream msg_stream; 445 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 446 vmIntrinsics::name_at(intrinsic_id()), 447 is_virtual() ? " (virtual)" : "", bci); 448 const char *msg = msg_stream.as_string(); 449 log_debug(jit, inlining)("%s", msg); 450 if (C->print_intrinsics() || C->print_inlining()) { 451 tty->print("%s", msg); 452 } 453 } 454 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 455 C->print_inlining_update(this); 456 return NULL; 457 } 458 459 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) { 460 LibraryCallKit kit(jvms, this); 461 Compile* C = kit.C; 462 int nodes = C->unique(); 463 _last_predicate = predicate; 464 #ifndef PRODUCT 465 assert(is_predicated() && predicate < predicates_count(), "sanity"); 466 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 467 char buf[1000]; 468 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf)); 469 tty->print_cr("Predicate for intrinsic %s", str); 470 } 471 #endif 472 ciMethod* callee = kit.callee(); 473 const int bci = kit.bci(); 474 475 Node* slow_ctl = kit.try_to_predicate(predicate); 476 if (!kit.failing()) { 477 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)" 478 : "(intrinsic, predicate)"; 479 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg); 480 if (C->print_intrinsics() || C->print_inlining()) { 481 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg); 482 } 483 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked); 484 if (C->log()) { 485 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'", 486 vmIntrinsics::name_at(intrinsic_id()), 487 (is_virtual() ? " virtual='1'" : ""), 488 C->unique() - nodes); 489 } 490 return slow_ctl; // Could be NULL if the check folds. 491 } 492 493 // The intrinsic bailed out 494 if (jvms->has_method()) { 495 // Not a root compile. 496 const char* msg = "failed to generate predicate for intrinsic"; 497 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg); 498 if (C->print_intrinsics() || C->print_inlining()) { 499 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg); 500 } 501 } else { 502 // Root compile 503 ResourceMark rm; 504 stringStream msg_stream; 505 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in", 506 vmIntrinsics::name_at(intrinsic_id()), 507 is_virtual() ? " (virtual)" : "", bci); 508 const char *msg = msg_stream.as_string(); 509 log_debug(jit, inlining)("%s", msg); 510 if (C->print_intrinsics() || C->print_inlining()) { 511 C->print_inlining_stream()->print("%s", msg); 512 } 513 } 514 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); 515 return NULL; 516 } 517 518 bool LibraryCallKit::try_to_inline(int predicate) { 519 // Handle symbolic names for otherwise undistinguished boolean switches: 520 const bool is_store = true; 521 const bool is_compress = true; 522 const bool is_static = true; 523 const bool is_volatile = true; 524 525 if (!jvms()->has_method()) { 526 // Root JVMState has a null method. 527 assert(map()->memory()->Opcode() == Op_Parm, ""); 528 // Insert the memory aliasing node 529 set_all_memory(reset_memory()); 530 } 531 assert(merged_memory(), ""); 532 533 534 switch (intrinsic_id()) { 535 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static); 536 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static); 537 case vmIntrinsics::_getClass: return inline_native_getClass(); 538 539 case vmIntrinsics::_ceil: 540 case vmIntrinsics::_floor: 541 case vmIntrinsics::_rint: 542 case vmIntrinsics::_dsin: 543 case vmIntrinsics::_dcos: 544 case vmIntrinsics::_dtan: 545 case vmIntrinsics::_dabs: 546 case vmIntrinsics::_fabs: 547 case vmIntrinsics::_iabs: 548 case vmIntrinsics::_labs: 549 case vmIntrinsics::_datan2: 550 case vmIntrinsics::_dsqrt: 551 case vmIntrinsics::_dexp: 552 case vmIntrinsics::_dlog: 553 case vmIntrinsics::_dlog10: 554 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id()); 555 556 case vmIntrinsics::_min: 557 case vmIntrinsics::_max: return inline_min_max(intrinsic_id()); 558 559 case vmIntrinsics::_notify: 560 case vmIntrinsics::_notifyAll: 561 return inline_notify(intrinsic_id()); 562 563 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */); 564 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */); 565 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */); 566 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */); 567 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */); 568 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */); 569 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI(); 570 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL(); 571 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh(); 572 case vmIntrinsics::_negateExactI: return inline_math_negateExactI(); 573 case vmIntrinsics::_negateExactL: return inline_math_negateExactL(); 574 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */); 575 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */); 576 577 case vmIntrinsics::_arraycopy: return inline_arraycopy(); 578 579 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL); 580 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU); 581 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU); 582 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL); 583 584 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL); 585 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU); 586 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL); 587 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL); 588 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU); 589 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL); 590 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(); 591 592 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL); 593 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU); 594 595 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU(); 596 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU(); 597 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store); 598 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store); 599 600 case vmIntrinsics::_compressStringC: 601 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress); 602 case vmIntrinsics::_inflateStringC: 603 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress); 604 605 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false); 606 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false); 607 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false); 608 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false); 609 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false); 610 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false); 611 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false); 612 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false); 613 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false); 614 615 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false); 616 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false); 617 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false); 618 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false); 619 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false); 620 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false); 621 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false); 622 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false); 623 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false); 624 625 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false); 626 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false); 627 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false); 628 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false); 629 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false); 630 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false); 631 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false); 632 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false); 633 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false); 634 635 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false); 636 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false); 637 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false); 638 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false); 639 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false); 640 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false); 641 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false); 642 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false); 643 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false); 644 645 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true); 646 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true); 647 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true); 648 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true); 649 650 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true); 651 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true); 652 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true); 653 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true); 654 655 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false); 656 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false); 657 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false); 658 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false); 659 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false); 660 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false); 661 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false); 662 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false); 663 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false); 664 665 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false); 666 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false); 667 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false); 668 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false); 669 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false); 670 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false); 671 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false); 672 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false); 673 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false); 674 675 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false); 676 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false); 677 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false); 678 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false); 679 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false); 680 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false); 681 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false); 682 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false); 683 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false); 684 685 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false); 686 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false); 687 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false); 688 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false); 689 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false); 690 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false); 691 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false); 692 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false); 693 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false); 694 695 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile); 696 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile); 697 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile); 698 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile); 699 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile); 700 701 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed); 702 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire); 703 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release); 704 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile); 705 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed); 706 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire); 707 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release); 708 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile); 709 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed); 710 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire); 711 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release); 712 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile); 713 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed); 714 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire); 715 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release); 716 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile); 717 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed); 718 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire); 719 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release); 720 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile); 721 722 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile); 723 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire); 724 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release); 725 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile); 726 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire); 727 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release); 728 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile); 729 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire); 730 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release); 731 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile); 732 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire); 733 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release); 734 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile); 735 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire); 736 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release); 737 738 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile); 739 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile); 740 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile); 741 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile); 742 743 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile); 744 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile); 745 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile); 746 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile); 747 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile); 748 749 case vmIntrinsics::_loadFence: 750 case vmIntrinsics::_storeFence: 751 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id()); 752 753 case vmIntrinsics::_onSpinWait: return inline_onspinwait(); 754 755 case vmIntrinsics::_currentThread: return inline_native_currentThread(); 756 757 #ifdef JFR_HAVE_INTRINSICS 758 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime"); 759 case vmIntrinsics::_getClassId: return inline_native_classID(); 760 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter(); 761 #endif 762 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis"); 763 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime"); 764 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0(); 765 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true); 766 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false); 767 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate(); 768 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory(); 769 case vmIntrinsics::_getLength: return inline_native_getLength(); 770 case vmIntrinsics::_copyOf: return inline_array_copyOf(false); 771 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true); 772 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL); 773 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU); 774 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(); 775 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual()); 776 777 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true); 778 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false); 779 780 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check(); 781 782 case vmIntrinsics::_isInstance: 783 case vmIntrinsics::_getModifiers: 784 case vmIntrinsics::_isInterface: 785 case vmIntrinsics::_isArray: 786 case vmIntrinsics::_isPrimitive: 787 case vmIntrinsics::_isHidden: 788 case vmIntrinsics::_getSuperclass: 789 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id()); 790 791 case vmIntrinsics::_floatToRawIntBits: 792 case vmIntrinsics::_floatToIntBits: 793 case vmIntrinsics::_intBitsToFloat: 794 case vmIntrinsics::_doubleToRawLongBits: 795 case vmIntrinsics::_doubleToLongBits: 796 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id()); 797 798 case vmIntrinsics::_numberOfLeadingZeros_i: 799 case vmIntrinsics::_numberOfLeadingZeros_l: 800 case vmIntrinsics::_numberOfTrailingZeros_i: 801 case vmIntrinsics::_numberOfTrailingZeros_l: 802 case vmIntrinsics::_bitCount_i: 803 case vmIntrinsics::_bitCount_l: 804 case vmIntrinsics::_reverseBytes_i: 805 case vmIntrinsics::_reverseBytes_l: 806 case vmIntrinsics::_reverseBytes_s: 807 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id()); 808 809 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass(); 810 811 case vmIntrinsics::_Reference_get: return inline_reference_get(); 812 813 case vmIntrinsics::_Class_cast: return inline_Class_cast(); 814 815 case vmIntrinsics::_aescrypt_encryptBlock: 816 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id()); 817 818 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 819 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 820 return inline_cipherBlockChaining_AESCrypt(intrinsic_id()); 821 822 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 823 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 824 return inline_electronicCodeBook_AESCrypt(intrinsic_id()); 825 826 case vmIntrinsics::_counterMode_AESCrypt: 827 return inline_counterMode_AESCrypt(intrinsic_id()); 828 829 case vmIntrinsics::_md5_implCompress: 830 case vmIntrinsics::_sha_implCompress: 831 case vmIntrinsics::_sha2_implCompress: 832 case vmIntrinsics::_sha5_implCompress: 833 return inline_digestBase_implCompress(intrinsic_id()); 834 835 case vmIntrinsics::_digestBase_implCompressMB: 836 return inline_digestBase_implCompressMB(predicate); 837 838 case vmIntrinsics::_multiplyToLen: 839 return inline_multiplyToLen(); 840 841 case vmIntrinsics::_squareToLen: 842 return inline_squareToLen(); 843 844 case vmIntrinsics::_mulAdd: 845 return inline_mulAdd(); 846 847 case vmIntrinsics::_montgomeryMultiply: 848 return inline_montgomeryMultiply(); 849 case vmIntrinsics::_montgomerySquare: 850 return inline_montgomerySquare(); 851 852 case vmIntrinsics::_bigIntegerRightShiftWorker: 853 return inline_bigIntegerShift(true); 854 case vmIntrinsics::_bigIntegerLeftShiftWorker: 855 return inline_bigIntegerShift(false); 856 857 case vmIntrinsics::_vectorizedMismatch: 858 return inline_vectorizedMismatch(); 859 860 case vmIntrinsics::_ghash_processBlocks: 861 return inline_ghash_processBlocks(); 862 case vmIntrinsics::_base64_encodeBlock: 863 return inline_base64_encodeBlock(); 864 865 case vmIntrinsics::_encodeISOArray: 866 case vmIntrinsics::_encodeByteISOArray: 867 return inline_encodeISOArray(); 868 869 case vmIntrinsics::_updateCRC32: 870 return inline_updateCRC32(); 871 case vmIntrinsics::_updateBytesCRC32: 872 return inline_updateBytesCRC32(); 873 case vmIntrinsics::_updateByteBufferCRC32: 874 return inline_updateByteBufferCRC32(); 875 876 case vmIntrinsics::_updateBytesCRC32C: 877 return inline_updateBytesCRC32C(); 878 case vmIntrinsics::_updateDirectByteBufferCRC32C: 879 return inline_updateDirectByteBufferCRC32C(); 880 881 case vmIntrinsics::_updateBytesAdler32: 882 return inline_updateBytesAdler32(); 883 case vmIntrinsics::_updateByteBufferAdler32: 884 return inline_updateByteBufferAdler32(); 885 886 case vmIntrinsics::_profileBoolean: 887 return inline_profileBoolean(); 888 case vmIntrinsics::_isCompileConstant: 889 return inline_isCompileConstant(); 890 891 case vmIntrinsics::_hasNegatives: 892 return inline_hasNegatives(); 893 894 case vmIntrinsics::_fmaD: 895 case vmIntrinsics::_fmaF: 896 return inline_fma(intrinsic_id()); 897 898 case vmIntrinsics::_isDigit: 899 case vmIntrinsics::_isLowerCase: 900 case vmIntrinsics::_isUpperCase: 901 case vmIntrinsics::_isWhitespace: 902 return inline_character_compare(intrinsic_id()); 903 904 case vmIntrinsics::_maxF: 905 case vmIntrinsics::_minF: 906 case vmIntrinsics::_maxD: 907 case vmIntrinsics::_minD: 908 return inline_fp_min_max(intrinsic_id()); 909 910 default: 911 // If you get here, it may be that someone has added a new intrinsic 912 // to the list in vmSymbols.hpp without implementing it here. 913 #ifndef PRODUCT 914 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 915 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)", 916 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 917 } 918 #endif 919 return false; 920 } 921 } 922 923 Node* LibraryCallKit::try_to_predicate(int predicate) { 924 if (!jvms()->has_method()) { 925 // Root JVMState has a null method. 926 assert(map()->memory()->Opcode() == Op_Parm, ""); 927 // Insert the memory aliasing node 928 set_all_memory(reset_memory()); 929 } 930 assert(merged_memory(), ""); 931 932 switch (intrinsic_id()) { 933 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 934 return inline_cipherBlockChaining_AESCrypt_predicate(false); 935 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 936 return inline_cipherBlockChaining_AESCrypt_predicate(true); 937 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 938 return inline_electronicCodeBook_AESCrypt_predicate(false); 939 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 940 return inline_electronicCodeBook_AESCrypt_predicate(true); 941 case vmIntrinsics::_counterMode_AESCrypt: 942 return inline_counterMode_AESCrypt_predicate(); 943 case vmIntrinsics::_digestBase_implCompressMB: 944 return inline_digestBase_implCompressMB_predicate(predicate); 945 946 default: 947 // If you get here, it may be that someone has added a new intrinsic 948 // to the list in vmSymbols.hpp without implementing it here. 949 #ifndef PRODUCT 950 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) { 951 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)", 952 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id()); 953 } 954 #endif 955 Node* slow_ctl = control(); 956 set_control(top()); // No fast path instrinsic 957 return slow_ctl; 958 } 959 } 960 961 //------------------------------set_result------------------------------- 962 // Helper function for finishing intrinsics. 963 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) { 964 record_for_igvn(region); 965 set_control(_gvn.transform(region)); 966 set_result( _gvn.transform(value)); 967 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity"); 968 } 969 970 //------------------------------generate_guard--------------------------- 971 // Helper function for generating guarded fast-slow graph structures. 972 // The given 'test', if true, guards a slow path. If the test fails 973 // then a fast path can be taken. (We generally hope it fails.) 974 // In all cases, GraphKit::control() is updated to the fast path. 975 // The returned value represents the control for the slow path. 976 // The return value is never 'top'; it is either a valid control 977 // or NULL if it is obvious that the slow path can never be taken. 978 // Also, if region and the slow control are not NULL, the slow edge 979 // is appended to the region. 980 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) { 981 if (stopped()) { 982 // Already short circuited. 983 return NULL; 984 } 985 986 // Build an if node and its projections. 987 // If test is true we take the slow path, which we assume is uncommon. 988 if (_gvn.type(test) == TypeInt::ZERO) { 989 // The slow branch is never taken. No need to build this guard. 990 return NULL; 991 } 992 993 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN); 994 995 Node* if_slow = _gvn.transform(new IfTrueNode(iff)); 996 if (if_slow == top()) { 997 // The slow branch is never taken. No need to build this guard. 998 return NULL; 999 } 1000 1001 if (region != NULL) 1002 region->add_req(if_slow); 1003 1004 Node* if_fast = _gvn.transform(new IfFalseNode(iff)); 1005 set_control(if_fast); 1006 1007 return if_slow; 1008 } 1009 1010 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) { 1011 return generate_guard(test, region, PROB_UNLIKELY_MAG(3)); 1012 } 1013 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) { 1014 return generate_guard(test, region, PROB_FAIR); 1015 } 1016 1017 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region, 1018 Node* *pos_index) { 1019 if (stopped()) 1020 return NULL; // already stopped 1021 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint] 1022 return NULL; // index is already adequately typed 1023 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0))); 1024 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1025 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN); 1026 if (is_neg != NULL && pos_index != NULL) { 1027 // Emulate effect of Parse::adjust_map_after_if. 1028 Node* ccast = new CastIINode(index, TypeInt::POS); 1029 ccast->set_req(0, control()); 1030 (*pos_index) = _gvn.transform(ccast); 1031 } 1032 return is_neg; 1033 } 1034 1035 // Make sure that 'position' is a valid limit index, in [0..length]. 1036 // There are two equivalent plans for checking this: 1037 // A. (offset + copyLength) unsigned<= arrayLength 1038 // B. offset <= (arrayLength - copyLength) 1039 // We require that all of the values above, except for the sum and 1040 // difference, are already known to be non-negative. 1041 // Plan A is robust in the face of overflow, if offset and copyLength 1042 // are both hugely positive. 1043 // 1044 // Plan B is less direct and intuitive, but it does not overflow at 1045 // all, since the difference of two non-negatives is always 1046 // representable. Whenever Java methods must perform the equivalent 1047 // check they generally use Plan B instead of Plan A. 1048 // For the moment we use Plan A. 1049 inline Node* LibraryCallKit::generate_limit_guard(Node* offset, 1050 Node* subseq_length, 1051 Node* array_length, 1052 RegionNode* region) { 1053 if (stopped()) 1054 return NULL; // already stopped 1055 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO; 1056 if (zero_offset && subseq_length->eqv_uncast(array_length)) 1057 return NULL; // common case of whole-array copy 1058 Node* last = subseq_length; 1059 if (!zero_offset) // last += offset 1060 last = _gvn.transform(new AddINode(last, offset)); 1061 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last)); 1062 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt)); 1063 Node* is_over = generate_guard(bol_lt, region, PROB_MIN); 1064 return is_over; 1065 } 1066 1067 // Emit range checks for the given String.value byte array 1068 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) { 1069 if (stopped()) { 1070 return; // already stopped 1071 } 1072 RegionNode* bailout = new RegionNode(1); 1073 record_for_igvn(bailout); 1074 if (char_count) { 1075 // Convert char count to byte count 1076 count = _gvn.transform(new LShiftINode(count, intcon(1))); 1077 } 1078 1079 // Offset and count must not be negative 1080 generate_negative_guard(offset, bailout); 1081 generate_negative_guard(count, bailout); 1082 // Offset + count must not exceed length of array 1083 generate_limit_guard(offset, count, load_array_length(array), bailout); 1084 1085 if (bailout->req() > 1) { 1086 PreserveJVMState pjvms(this); 1087 set_control(_gvn.transform(bailout)); 1088 uncommon_trap(Deoptimization::Reason_intrinsic, 1089 Deoptimization::Action_maybe_recompile); 1090 } 1091 } 1092 1093 //--------------------------generate_current_thread-------------------- 1094 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) { 1095 ciKlass* thread_klass = env()->Thread_klass(); 1096 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull); 1097 Node* thread = _gvn.transform(new ThreadLocalNode()); 1098 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset())); 1099 Node* threadObj = _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), thread_type, T_OBJECT, MemNode::unordered)); 1100 tls_output = thread; 1101 return threadObj; 1102 } 1103 1104 1105 //------------------------------make_string_method_node------------------------ 1106 // Helper method for String intrinsic functions. This version is called with 1107 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded 1108 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes 1109 // containing the lengths of str1 and str2. 1110 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) { 1111 Node* result = NULL; 1112 switch (opcode) { 1113 case Op_StrIndexOf: 1114 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES), 1115 str1_start, cnt1, str2_start, cnt2, ae); 1116 break; 1117 case Op_StrComp: 1118 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES), 1119 str1_start, cnt1, str2_start, cnt2, ae); 1120 break; 1121 case Op_StrEquals: 1122 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals'). 1123 // Use the constant length if there is one because optimized match rule may exist. 1124 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES), 1125 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae); 1126 break; 1127 default: 1128 ShouldNotReachHere(); 1129 return NULL; 1130 } 1131 1132 // All these intrinsics have checks. 1133 C->set_has_split_ifs(true); // Has chance for split-if optimization 1134 clear_upper_avx(); 1135 1136 return _gvn.transform(result); 1137 } 1138 1139 //------------------------------inline_string_compareTo------------------------ 1140 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) { 1141 Node* arg1 = argument(0); 1142 Node* arg2 = argument(1); 1143 1144 arg1 = must_be_not_null(arg1, true); 1145 arg2 = must_be_not_null(arg2, true); 1146 1147 // Get start addr and length of first argument 1148 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1149 Node* arg1_cnt = load_array_length(arg1); 1150 1151 // Get start addr and length of second argument 1152 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1153 Node* arg2_cnt = load_array_length(arg2); 1154 1155 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1156 set_result(result); 1157 return true; 1158 } 1159 1160 //------------------------------inline_string_equals------------------------ 1161 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) { 1162 Node* arg1 = argument(0); 1163 Node* arg2 = argument(1); 1164 1165 // paths (plus control) merge 1166 RegionNode* region = new RegionNode(3); 1167 Node* phi = new PhiNode(region, TypeInt::BOOL); 1168 1169 if (!stopped()) { 1170 1171 arg1 = must_be_not_null(arg1, true); 1172 arg2 = must_be_not_null(arg2, true); 1173 1174 // Get start addr and length of first argument 1175 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE); 1176 Node* arg1_cnt = load_array_length(arg1); 1177 1178 // Get start addr and length of second argument 1179 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE); 1180 Node* arg2_cnt = load_array_length(arg2); 1181 1182 // Check for arg1_cnt != arg2_cnt 1183 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt)); 1184 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 1185 Node* if_ne = generate_slow_guard(bol, NULL); 1186 if (if_ne != NULL) { 1187 phi->init_req(2, intcon(0)); 1188 region->init_req(2, if_ne); 1189 } 1190 1191 // Check for count == 0 is done by assembler code for StrEquals. 1192 1193 if (!stopped()) { 1194 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae); 1195 phi->init_req(1, equals); 1196 region->init_req(1, control()); 1197 } 1198 } 1199 1200 // post merge 1201 set_control(_gvn.transform(region)); 1202 record_for_igvn(region); 1203 1204 set_result(_gvn.transform(phi)); 1205 return true; 1206 } 1207 1208 //------------------------------inline_array_equals---------------------------- 1209 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) { 1210 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types"); 1211 Node* arg1 = argument(0); 1212 Node* arg2 = argument(1); 1213 1214 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES; 1215 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae))); 1216 clear_upper_avx(); 1217 1218 return true; 1219 } 1220 1221 //------------------------------inline_hasNegatives------------------------------ 1222 bool LibraryCallKit::inline_hasNegatives() { 1223 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1224 return false; 1225 } 1226 1227 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters"); 1228 // no receiver since it is static method 1229 Node* ba = argument(0); 1230 Node* offset = argument(1); 1231 Node* len = argument(2); 1232 1233 ba = must_be_not_null(ba, true); 1234 1235 // Range checks 1236 generate_string_range_check(ba, offset, len, false); 1237 if (stopped()) { 1238 return true; 1239 } 1240 Node* ba_start = array_element_address(ba, offset, T_BYTE); 1241 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len); 1242 set_result(_gvn.transform(result)); 1243 return true; 1244 } 1245 1246 bool LibraryCallKit::inline_preconditions_checkIndex() { 1247 Node* index = argument(0); 1248 Node* length = argument(1); 1249 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) { 1250 return false; 1251 } 1252 1253 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0))); 1254 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge)); 1255 1256 { 1257 BuildCutout unless(this, len_pos_bol, PROB_MAX); 1258 uncommon_trap(Deoptimization::Reason_intrinsic, 1259 Deoptimization::Action_make_not_entrant); 1260 } 1261 1262 if (stopped()) { 1263 return false; 1264 } 1265 1266 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length)); 1267 BoolTest::mask btest = BoolTest::lt; 1268 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest)); 1269 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN); 1270 _gvn.set_type(rc, rc->Value(&_gvn)); 1271 if (!rc_bool->is_Con()) { 1272 record_for_igvn(rc); 1273 } 1274 set_control(_gvn.transform(new IfTrueNode(rc))); 1275 { 1276 PreserveJVMState pjvms(this); 1277 set_control(_gvn.transform(new IfFalseNode(rc))); 1278 uncommon_trap(Deoptimization::Reason_range_check, 1279 Deoptimization::Action_make_not_entrant); 1280 } 1281 1282 if (stopped()) { 1283 return false; 1284 } 1285 1286 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax)); 1287 result->set_req(0, control()); 1288 result = _gvn.transform(result); 1289 set_result(result); 1290 replace_in_map(index, result); 1291 clear_upper_avx(); 1292 return true; 1293 } 1294 1295 //------------------------------inline_string_indexOf------------------------ 1296 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) { 1297 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1298 return false; 1299 } 1300 Node* src = argument(0); 1301 Node* tgt = argument(1); 1302 1303 // Make the merge point 1304 RegionNode* result_rgn = new RegionNode(4); 1305 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT); 1306 1307 src = must_be_not_null(src, true); 1308 tgt = must_be_not_null(tgt, true); 1309 1310 // Get start addr and length of source string 1311 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 1312 Node* src_count = load_array_length(src); 1313 1314 // Get start addr and length of substring 1315 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1316 Node* tgt_count = load_array_length(tgt); 1317 1318 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) { 1319 // Divide src size by 2 if String is UTF16 encoded 1320 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1))); 1321 } 1322 if (ae == StrIntrinsicNode::UU) { 1323 // Divide substring size by 2 if String is UTF16 encoded 1324 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1))); 1325 } 1326 1327 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae); 1328 if (result != NULL) { 1329 result_phi->init_req(3, result); 1330 result_rgn->init_req(3, control()); 1331 } 1332 set_control(_gvn.transform(result_rgn)); 1333 record_for_igvn(result_rgn); 1334 set_result(_gvn.transform(result_phi)); 1335 1336 return true; 1337 } 1338 1339 //-----------------------------inline_string_indexOf----------------------- 1340 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) { 1341 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1342 return false; 1343 } 1344 if (!Matcher::match_rule_supported(Op_StrIndexOf)) { 1345 return false; 1346 } 1347 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments"); 1348 Node* src = argument(0); // byte[] 1349 Node* src_count = argument(1); // char count 1350 Node* tgt = argument(2); // byte[] 1351 Node* tgt_count = argument(3); // char count 1352 Node* from_index = argument(4); // char index 1353 1354 src = must_be_not_null(src, true); 1355 tgt = must_be_not_null(tgt, true); 1356 1357 // Multiply byte array index by 2 if String is UTF16 encoded 1358 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1))); 1359 src_count = _gvn.transform(new SubINode(src_count, from_index)); 1360 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1361 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE); 1362 1363 // Range checks 1364 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL); 1365 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU); 1366 if (stopped()) { 1367 return true; 1368 } 1369 1370 RegionNode* region = new RegionNode(5); 1371 Node* phi = new PhiNode(region, TypeInt::INT); 1372 1373 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae); 1374 if (result != NULL) { 1375 // The result is index relative to from_index if substring was found, -1 otherwise. 1376 // Generate code which will fold into cmove. 1377 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1378 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1379 1380 Node* if_lt = generate_slow_guard(bol, NULL); 1381 if (if_lt != NULL) { 1382 // result == -1 1383 phi->init_req(3, result); 1384 region->init_req(3, if_lt); 1385 } 1386 if (!stopped()) { 1387 result = _gvn.transform(new AddINode(result, from_index)); 1388 phi->init_req(4, result); 1389 region->init_req(4, control()); 1390 } 1391 } 1392 1393 set_control(_gvn.transform(region)); 1394 record_for_igvn(region); 1395 set_result(_gvn.transform(phi)); 1396 clear_upper_avx(); 1397 1398 return true; 1399 } 1400 1401 // Create StrIndexOfNode with fast path checks 1402 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count, 1403 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) { 1404 // Check for substr count > string count 1405 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count)); 1406 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt)); 1407 Node* if_gt = generate_slow_guard(bol, NULL); 1408 if (if_gt != NULL) { 1409 phi->init_req(1, intcon(-1)); 1410 region->init_req(1, if_gt); 1411 } 1412 if (!stopped()) { 1413 // Check for substr count == 0 1414 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0))); 1415 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 1416 Node* if_zero = generate_slow_guard(bol, NULL); 1417 if (if_zero != NULL) { 1418 phi->init_req(2, intcon(0)); 1419 region->init_req(2, if_zero); 1420 } 1421 } 1422 if (!stopped()) { 1423 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae); 1424 } 1425 return NULL; 1426 } 1427 1428 //-----------------------------inline_string_indexOfChar----------------------- 1429 bool LibraryCallKit::inline_string_indexOfChar() { 1430 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1431 return false; 1432 } 1433 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) { 1434 return false; 1435 } 1436 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments"); 1437 Node* src = argument(0); // byte[] 1438 Node* tgt = argument(1); // tgt is int ch 1439 Node* from_index = argument(2); 1440 Node* max = argument(3); 1441 1442 src = must_be_not_null(src, true); 1443 1444 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1))); 1445 Node* src_start = array_element_address(src, src_offset, T_BYTE); 1446 Node* src_count = _gvn.transform(new SubINode(max, from_index)); 1447 1448 // Range checks 1449 generate_string_range_check(src, src_offset, src_count, true); 1450 if (stopped()) { 1451 return true; 1452 } 1453 1454 RegionNode* region = new RegionNode(3); 1455 Node* phi = new PhiNode(region, TypeInt::INT); 1456 1457 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none); 1458 C->set_has_split_ifs(true); // Has chance for split-if optimization 1459 _gvn.transform(result); 1460 1461 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0))); 1462 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt)); 1463 1464 Node* if_lt = generate_slow_guard(bol, NULL); 1465 if (if_lt != NULL) { 1466 // result == -1 1467 phi->init_req(2, result); 1468 region->init_req(2, if_lt); 1469 } 1470 if (!stopped()) { 1471 result = _gvn.transform(new AddINode(result, from_index)); 1472 phi->init_req(1, result); 1473 region->init_req(1, control()); 1474 } 1475 set_control(_gvn.transform(region)); 1476 record_for_igvn(region); 1477 set_result(_gvn.transform(phi)); 1478 1479 return true; 1480 } 1481 //---------------------------inline_string_copy--------------------- 1482 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[]) 1483 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) 1484 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1485 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[]) 1486 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) 1487 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len) 1488 bool LibraryCallKit::inline_string_copy(bool compress) { 1489 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1490 return false; 1491 } 1492 int nargs = 5; // 2 oops, 3 ints 1493 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments"); 1494 1495 Node* src = argument(0); 1496 Node* src_offset = argument(1); 1497 Node* dst = argument(2); 1498 Node* dst_offset = argument(3); 1499 Node* length = argument(4); 1500 1501 // Check for allocation before we add nodes that would confuse 1502 // tightly_coupled_allocation() 1503 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1504 1505 // Figure out the size and type of the elements we will be copying. 1506 const Type* src_type = src->Value(&_gvn); 1507 const Type* dst_type = dst->Value(&_gvn); 1508 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1509 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 1510 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) || 1511 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)), 1512 "Unsupported array types for inline_string_copy"); 1513 1514 src = must_be_not_null(src, true); 1515 dst = must_be_not_null(dst, true); 1516 1517 // Convert char[] offsets to byte[] offsets 1518 bool convert_src = (compress && src_elem == T_BYTE); 1519 bool convert_dst = (!compress && dst_elem == T_BYTE); 1520 if (convert_src) { 1521 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1))); 1522 } else if (convert_dst) { 1523 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1))); 1524 } 1525 1526 // Range checks 1527 generate_string_range_check(src, src_offset, length, convert_src); 1528 generate_string_range_check(dst, dst_offset, length, convert_dst); 1529 if (stopped()) { 1530 return true; 1531 } 1532 1533 Node* src_start = array_element_address(src, src_offset, src_elem); 1534 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 1535 // 'src_start' points to src array + scaled offset 1536 // 'dst_start' points to dst array + scaled offset 1537 Node* count = NULL; 1538 if (compress) { 1539 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length); 1540 } else { 1541 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length); 1542 } 1543 1544 if (alloc != NULL) { 1545 if (alloc->maybe_set_complete(&_gvn)) { 1546 // "You break it, you buy it." 1547 InitializeNode* init = alloc->initialization(); 1548 assert(init->is_complete(), "we just did this"); 1549 init->set_complete_with_arraycopy(); 1550 assert(dst->is_CheckCastPP(), "sanity"); 1551 assert(dst->in(0)->in(0) == init, "dest pinned"); 1552 } 1553 // Do not let stores that initialize this object be reordered with 1554 // a subsequent store that would make this object accessible by 1555 // other threads. 1556 // Record what AllocateNode this StoreStore protects so that 1557 // escape analysis can go from the MemBarStoreStoreNode to the 1558 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1559 // based on the escape status of the AllocateNode. 1560 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1561 } 1562 if (compress) { 1563 set_result(_gvn.transform(count)); 1564 } 1565 clear_upper_avx(); 1566 1567 return true; 1568 } 1569 1570 #ifdef _LP64 1571 #define XTOP ,top() /*additional argument*/ 1572 #else //_LP64 1573 #define XTOP /*no additional argument*/ 1574 #endif //_LP64 1575 1576 //------------------------inline_string_toBytesU-------------------------- 1577 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len) 1578 bool LibraryCallKit::inline_string_toBytesU() { 1579 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1580 return false; 1581 } 1582 // Get the arguments. 1583 Node* value = argument(0); 1584 Node* offset = argument(1); 1585 Node* length = argument(2); 1586 1587 Node* newcopy = NULL; 1588 1589 // Set the original stack and the reexecute bit for the interpreter to reexecute 1590 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens. 1591 { PreserveReexecuteState preexecs(this); 1592 jvms()->set_should_reexecute(true); 1593 1594 // Check if a null path was taken unconditionally. 1595 value = null_check(value); 1596 1597 RegionNode* bailout = new RegionNode(1); 1598 record_for_igvn(bailout); 1599 1600 // Range checks 1601 generate_negative_guard(offset, bailout); 1602 generate_negative_guard(length, bailout); 1603 generate_limit_guard(offset, length, load_array_length(value), bailout); 1604 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE 1605 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout); 1606 1607 if (bailout->req() > 1) { 1608 PreserveJVMState pjvms(this); 1609 set_control(_gvn.transform(bailout)); 1610 uncommon_trap(Deoptimization::Reason_intrinsic, 1611 Deoptimization::Action_maybe_recompile); 1612 } 1613 if (stopped()) { 1614 return true; 1615 } 1616 1617 Node* size = _gvn.transform(new LShiftINode(length, intcon(1))); 1618 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE))); 1619 newcopy = new_array(klass_node, size, 0); // no arguments to push 1620 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL); 1621 1622 // Calculate starting addresses. 1623 Node* src_start = array_element_address(value, offset, T_CHAR); 1624 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE)); 1625 1626 // Check if src array address is aligned to HeapWordSize (dst is always aligned) 1627 const TypeInt* toffset = gvn().type(offset)->is_int(); 1628 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1629 1630 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1631 const char* copyfunc_name = "arraycopy"; 1632 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1633 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1634 OptoRuntime::fast_arraycopy_Type(), 1635 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1636 src_start, dst_start, ConvI2X(length) XTOP); 1637 // Do not let reads from the cloned object float above the arraycopy. 1638 if (alloc != NULL) { 1639 if (alloc->maybe_set_complete(&_gvn)) { 1640 // "You break it, you buy it." 1641 InitializeNode* init = alloc->initialization(); 1642 assert(init->is_complete(), "we just did this"); 1643 init->set_complete_with_arraycopy(); 1644 assert(newcopy->is_CheckCastPP(), "sanity"); 1645 assert(newcopy->in(0)->in(0) == init, "dest pinned"); 1646 } 1647 // Do not let stores that initialize this object be reordered with 1648 // a subsequent store that would make this object accessible by 1649 // other threads. 1650 // Record what AllocateNode this StoreStore protects so that 1651 // escape analysis can go from the MemBarStoreStoreNode to the 1652 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1653 // based on the escape status of the AllocateNode. 1654 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1655 } else { 1656 insert_mem_bar(Op_MemBarCPUOrder); 1657 } 1658 } // original reexecute is set back here 1659 1660 C->set_has_split_ifs(true); // Has chance for split-if optimization 1661 if (!stopped()) { 1662 set_result(newcopy); 1663 } 1664 clear_upper_avx(); 1665 1666 return true; 1667 } 1668 1669 //------------------------inline_string_getCharsU-------------------------- 1670 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin) 1671 bool LibraryCallKit::inline_string_getCharsU() { 1672 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 1673 return false; 1674 } 1675 1676 // Get the arguments. 1677 Node* src = argument(0); 1678 Node* src_begin = argument(1); 1679 Node* src_end = argument(2); // exclusive offset (i < src_end) 1680 Node* dst = argument(3); 1681 Node* dst_begin = argument(4); 1682 1683 // Check for allocation before we add nodes that would confuse 1684 // tightly_coupled_allocation() 1685 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL); 1686 1687 // Check if a null path was taken unconditionally. 1688 src = null_check(src); 1689 dst = null_check(dst); 1690 if (stopped()) { 1691 return true; 1692 } 1693 1694 // Get length and convert char[] offset to byte[] offset 1695 Node* length = _gvn.transform(new SubINode(src_end, src_begin)); 1696 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1))); 1697 1698 // Range checks 1699 generate_string_range_check(src, src_begin, length, true); 1700 generate_string_range_check(dst, dst_begin, length, false); 1701 if (stopped()) { 1702 return true; 1703 } 1704 1705 if (!stopped()) { 1706 // Calculate starting addresses. 1707 Node* src_start = array_element_address(src, src_begin, T_BYTE); 1708 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR); 1709 1710 // Check if array addresses are aligned to HeapWordSize 1711 const TypeInt* tsrc = gvn().type(src_begin)->is_int(); 1712 const TypeInt* tdst = gvn().type(dst_begin)->is_int(); 1713 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) && 1714 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0); 1715 1716 // Figure out which arraycopy runtime method to call (disjoint, uninitialized). 1717 const char* copyfunc_name = "arraycopy"; 1718 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true); 1719 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 1720 OptoRuntime::fast_arraycopy_Type(), 1721 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM, 1722 src_start, dst_start, ConvI2X(length) XTOP); 1723 // Do not let reads from the cloned object float above the arraycopy. 1724 if (alloc != NULL) { 1725 if (alloc->maybe_set_complete(&_gvn)) { 1726 // "You break it, you buy it." 1727 InitializeNode* init = alloc->initialization(); 1728 assert(init->is_complete(), "we just did this"); 1729 init->set_complete_with_arraycopy(); 1730 assert(dst->is_CheckCastPP(), "sanity"); 1731 assert(dst->in(0)->in(0) == init, "dest pinned"); 1732 } 1733 // Do not let stores that initialize this object be reordered with 1734 // a subsequent store that would make this object accessible by 1735 // other threads. 1736 // Record what AllocateNode this StoreStore protects so that 1737 // escape analysis can go from the MemBarStoreStoreNode to the 1738 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 1739 // based on the escape status of the AllocateNode. 1740 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 1741 } else { 1742 insert_mem_bar(Op_MemBarCPUOrder); 1743 } 1744 } 1745 1746 C->set_has_split_ifs(true); // Has chance for split-if optimization 1747 return true; 1748 } 1749 1750 //----------------------inline_string_char_access---------------------------- 1751 // Store/Load char to/from byte[] array. 1752 // static void StringUTF16.putChar(byte[] val, int index, int c) 1753 // static char StringUTF16.getChar(byte[] val, int index) 1754 bool LibraryCallKit::inline_string_char_access(bool is_store) { 1755 Node* value = argument(0); 1756 Node* index = argument(1); 1757 Node* ch = is_store ? argument(2) : NULL; 1758 1759 // This intrinsic accesses byte[] array as char[] array. Computing the offsets 1760 // correctly requires matched array shapes. 1761 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE), 1762 "sanity: byte[] and char[] bases agree"); 1763 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2, 1764 "sanity: byte[] and char[] scales agree"); 1765 1766 // Bail when getChar over constants is requested: constant folding would 1767 // reject folding mismatched char access over byte[]. A normal inlining for getChar 1768 // Java method would constant fold nicely instead. 1769 if (!is_store && value->is_Con() && index->is_Con()) { 1770 return false; 1771 } 1772 1773 value = must_be_not_null(value, true); 1774 1775 Node* adr = array_element_address(value, index, T_CHAR); 1776 if (adr->is_top()) { 1777 return false; 1778 } 1779 if (is_store) { 1780 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED); 1781 } else { 1782 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD); 1783 set_result(ch); 1784 } 1785 return true; 1786 } 1787 1788 //--------------------------round_double_node-------------------------------- 1789 // Round a double node if necessary. 1790 Node* LibraryCallKit::round_double_node(Node* n) { 1791 if (Matcher::strict_fp_requires_explicit_rounding) { 1792 #ifdef IA32 1793 if (UseSSE < 2) { 1794 n = _gvn.transform(new RoundDoubleNode(NULL, n)); 1795 } 1796 #else 1797 Unimplemented(); 1798 #endif // IA32 1799 } 1800 return n; 1801 } 1802 1803 //------------------------------inline_math----------------------------------- 1804 // public static double Math.abs(double) 1805 // public static double Math.sqrt(double) 1806 // public static double Math.log(double) 1807 // public static double Math.log10(double) 1808 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) { 1809 Node* arg = round_double_node(argument(0)); 1810 Node* n = NULL; 1811 switch (id) { 1812 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break; 1813 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break; 1814 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break; 1815 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break; 1816 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break; 1817 default: fatal_unexpected_iid(id); break; 1818 } 1819 set_result(_gvn.transform(n)); 1820 return true; 1821 } 1822 1823 //------------------------------inline_math----------------------------------- 1824 // public static float Math.abs(float) 1825 // public static int Math.abs(int) 1826 // public static long Math.abs(long) 1827 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) { 1828 Node* arg = argument(0); 1829 Node* n = NULL; 1830 switch (id) { 1831 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break; 1832 case vmIntrinsics::_iabs: n = new AbsINode( arg); break; 1833 case vmIntrinsics::_labs: n = new AbsLNode( arg); break; 1834 default: fatal_unexpected_iid(id); break; 1835 } 1836 set_result(_gvn.transform(n)); 1837 return true; 1838 } 1839 1840 //------------------------------runtime_math----------------------------- 1841 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) { 1842 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(), 1843 "must be (DD)D or (D)D type"); 1844 1845 // Inputs 1846 Node* a = round_double_node(argument(0)); 1847 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL; 1848 1849 const TypePtr* no_memory_effects = NULL; 1850 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName, 1851 no_memory_effects, 1852 a, top(), b, b ? top() : NULL); 1853 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0)); 1854 #ifdef ASSERT 1855 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1)); 1856 assert(value_top == top(), "second value must be top"); 1857 #endif 1858 1859 set_result(value); 1860 return true; 1861 } 1862 1863 //------------------------------inline_math_native----------------------------- 1864 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) { 1865 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f) 1866 switch (id) { 1867 // These intrinsics are not properly supported on all hardware 1868 case vmIntrinsics::_dsin: 1869 return StubRoutines::dsin() != NULL ? 1870 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") : 1871 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN"); 1872 case vmIntrinsics::_dcos: 1873 return StubRoutines::dcos() != NULL ? 1874 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") : 1875 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS"); 1876 case vmIntrinsics::_dtan: 1877 return StubRoutines::dtan() != NULL ? 1878 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") : 1879 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN"); 1880 case vmIntrinsics::_dlog: 1881 return StubRoutines::dlog() != NULL ? 1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") : 1883 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG"); 1884 case vmIntrinsics::_dlog10: 1885 return StubRoutines::dlog10() != NULL ? 1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") : 1887 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10"); 1888 1889 // These intrinsics are supported on all hardware 1890 case vmIntrinsics::_ceil: 1891 case vmIntrinsics::_floor: 1892 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false; 1893 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false; 1894 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false; 1895 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false; 1896 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false; 1897 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false; 1898 1899 case vmIntrinsics::_dexp: 1900 return StubRoutines::dexp() != NULL ? 1901 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") : 1902 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP"); 1903 case vmIntrinsics::_dpow: { 1904 Node* exp = round_double_node(argument(2)); 1905 const TypeD* d = _gvn.type(exp)->isa_double_constant(); 1906 if (d != NULL && d->getd() == 2.0) { 1907 // Special case: pow(x, 2.0) => x * x 1908 Node* base = round_double_node(argument(0)); 1909 set_result(_gvn.transform(new MulDNode(base, base))); 1910 return true; 1911 } 1912 return StubRoutines::dpow() != NULL ? 1913 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") : 1914 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW"); 1915 } 1916 #undef FN_PTR 1917 1918 // These intrinsics are not yet correctly implemented 1919 case vmIntrinsics::_datan2: 1920 return false; 1921 1922 default: 1923 fatal_unexpected_iid(id); 1924 return false; 1925 } 1926 } 1927 1928 static bool is_simple_name(Node* n) { 1929 return (n->req() == 1 // constant 1930 || (n->is_Type() && n->as_Type()->type()->singleton()) 1931 || n->is_Proj() // parameter or return value 1932 || n->is_Phi() // local of some sort 1933 ); 1934 } 1935 1936 //----------------------------inline_notify-----------------------------------* 1937 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) { 1938 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type(); 1939 address func; 1940 if (id == vmIntrinsics::_notify) { 1941 func = OptoRuntime::monitor_notify_Java(); 1942 } else { 1943 func = OptoRuntime::monitor_notifyAll_Java(); 1944 } 1945 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0)); 1946 make_slow_call_ex(call, env()->Throwable_klass(), false); 1947 return true; 1948 } 1949 1950 1951 //----------------------------inline_min_max----------------------------------- 1952 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) { 1953 set_result(generate_min_max(id, argument(0), argument(1))); 1954 return true; 1955 } 1956 1957 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) { 1958 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) ); 1959 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN); 1960 Node* fast_path = _gvn.transform( new IfFalseNode(check)); 1961 Node* slow_path = _gvn.transform( new IfTrueNode(check) ); 1962 1963 { 1964 PreserveJVMState pjvms(this); 1965 PreserveReexecuteState preexecs(this); 1966 jvms()->set_should_reexecute(true); 1967 1968 set_control(slow_path); 1969 set_i_o(i_o()); 1970 1971 uncommon_trap(Deoptimization::Reason_intrinsic, 1972 Deoptimization::Action_none); 1973 } 1974 1975 set_control(fast_path); 1976 set_result(math); 1977 } 1978 1979 template <typename OverflowOp> 1980 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) { 1981 typedef typename OverflowOp::MathOp MathOp; 1982 1983 MathOp* mathOp = new MathOp(arg1, arg2); 1984 Node* operation = _gvn.transform( mathOp ); 1985 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) ); 1986 inline_math_mathExact(operation, ofcheck); 1987 return true; 1988 } 1989 1990 bool LibraryCallKit::inline_math_addExactI(bool is_increment) { 1991 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1)); 1992 } 1993 1994 bool LibraryCallKit::inline_math_addExactL(bool is_increment) { 1995 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2)); 1996 } 1997 1998 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) { 1999 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1)); 2000 } 2001 2002 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) { 2003 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2)); 2004 } 2005 2006 bool LibraryCallKit::inline_math_negateExactI() { 2007 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0)); 2008 } 2009 2010 bool LibraryCallKit::inline_math_negateExactL() { 2011 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0)); 2012 } 2013 2014 bool LibraryCallKit::inline_math_multiplyExactI() { 2015 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1)); 2016 } 2017 2018 bool LibraryCallKit::inline_math_multiplyExactL() { 2019 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2)); 2020 } 2021 2022 bool LibraryCallKit::inline_math_multiplyHigh() { 2023 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2)))); 2024 return true; 2025 } 2026 2027 Node* 2028 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) { 2029 // These are the candidate return value: 2030 Node* xvalue = x0; 2031 Node* yvalue = y0; 2032 2033 if (xvalue == yvalue) { 2034 return xvalue; 2035 } 2036 2037 bool want_max = (id == vmIntrinsics::_max); 2038 2039 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int(); 2040 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int(); 2041 if (txvalue == NULL || tyvalue == NULL) return top(); 2042 // This is not really necessary, but it is consistent with a 2043 // hypothetical MaxINode::Value method: 2044 int widen = MAX2(txvalue->_widen, tyvalue->_widen); 2045 2046 // %%% This folding logic should (ideally) be in a different place. 2047 // Some should be inside IfNode, and there to be a more reliable 2048 // transformation of ?: style patterns into cmoves. We also want 2049 // more powerful optimizations around cmove and min/max. 2050 2051 // Try to find a dominating comparison of these guys. 2052 // It can simplify the index computation for Arrays.copyOf 2053 // and similar uses of System.arraycopy. 2054 // First, compute the normalized version of CmpI(x, y). 2055 int cmp_op = Op_CmpI; 2056 Node* xkey = xvalue; 2057 Node* ykey = yvalue; 2058 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey)); 2059 if (ideal_cmpxy->is_Cmp()) { 2060 // E.g., if we have CmpI(length - offset, count), 2061 // it might idealize to CmpI(length, count + offset) 2062 cmp_op = ideal_cmpxy->Opcode(); 2063 xkey = ideal_cmpxy->in(1); 2064 ykey = ideal_cmpxy->in(2); 2065 } 2066 2067 // Start by locating any relevant comparisons. 2068 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey; 2069 Node* cmpxy = NULL; 2070 Node* cmpyx = NULL; 2071 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) { 2072 Node* cmp = start_from->fast_out(k); 2073 if (cmp->outcnt() > 0 && // must have prior uses 2074 cmp->in(0) == NULL && // must be context-independent 2075 cmp->Opcode() == cmp_op) { // right kind of compare 2076 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp; 2077 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp; 2078 } 2079 } 2080 2081 const int NCMPS = 2; 2082 Node* cmps[NCMPS] = { cmpxy, cmpyx }; 2083 int cmpn; 2084 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2085 if (cmps[cmpn] != NULL) break; // find a result 2086 } 2087 if (cmpn < NCMPS) { 2088 // Look for a dominating test that tells us the min and max. 2089 int depth = 0; // Limit search depth for speed 2090 Node* dom = control(); 2091 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) { 2092 if (++depth >= 100) break; 2093 Node* ifproj = dom; 2094 if (!ifproj->is_Proj()) continue; 2095 Node* iff = ifproj->in(0); 2096 if (!iff->is_If()) continue; 2097 Node* bol = iff->in(1); 2098 if (!bol->is_Bool()) continue; 2099 Node* cmp = bol->in(1); 2100 if (cmp == NULL) continue; 2101 for (cmpn = 0; cmpn < NCMPS; cmpn++) 2102 if (cmps[cmpn] == cmp) break; 2103 if (cmpn == NCMPS) continue; 2104 BoolTest::mask btest = bol->as_Bool()->_test._test; 2105 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate(); 2106 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2107 // At this point, we know that 'x btest y' is true. 2108 switch (btest) { 2109 case BoolTest::eq: 2110 // They are proven equal, so we can collapse the min/max. 2111 // Either value is the answer. Choose the simpler. 2112 if (is_simple_name(yvalue) && !is_simple_name(xvalue)) 2113 return yvalue; 2114 return xvalue; 2115 case BoolTest::lt: // x < y 2116 case BoolTest::le: // x <= y 2117 return (want_max ? yvalue : xvalue); 2118 case BoolTest::gt: // x > y 2119 case BoolTest::ge: // x >= y 2120 return (want_max ? xvalue : yvalue); 2121 default: 2122 break; 2123 } 2124 } 2125 } 2126 2127 // We failed to find a dominating test. 2128 // Let's pick a test that might GVN with prior tests. 2129 Node* best_bol = NULL; 2130 BoolTest::mask best_btest = BoolTest::illegal; 2131 for (cmpn = 0; cmpn < NCMPS; cmpn++) { 2132 Node* cmp = cmps[cmpn]; 2133 if (cmp == NULL) continue; 2134 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) { 2135 Node* bol = cmp->fast_out(j); 2136 if (!bol->is_Bool()) continue; 2137 BoolTest::mask btest = bol->as_Bool()->_test._test; 2138 if (btest == BoolTest::eq || btest == BoolTest::ne) continue; 2139 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute(); 2140 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) { 2141 best_bol = bol->as_Bool(); 2142 best_btest = btest; 2143 } 2144 } 2145 } 2146 2147 Node* answer_if_true = NULL; 2148 Node* answer_if_false = NULL; 2149 switch (best_btest) { 2150 default: 2151 if (cmpxy == NULL) 2152 cmpxy = ideal_cmpxy; 2153 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt)); 2154 // and fall through: 2155 case BoolTest::lt: // x < y 2156 case BoolTest::le: // x <= y 2157 answer_if_true = (want_max ? yvalue : xvalue); 2158 answer_if_false = (want_max ? xvalue : yvalue); 2159 break; 2160 case BoolTest::gt: // x > y 2161 case BoolTest::ge: // x >= y 2162 answer_if_true = (want_max ? xvalue : yvalue); 2163 answer_if_false = (want_max ? yvalue : xvalue); 2164 break; 2165 } 2166 2167 jint hi, lo; 2168 if (want_max) { 2169 // We can sharpen the minimum. 2170 hi = MAX2(txvalue->_hi, tyvalue->_hi); 2171 lo = MAX2(txvalue->_lo, tyvalue->_lo); 2172 } else { 2173 // We can sharpen the maximum. 2174 hi = MIN2(txvalue->_hi, tyvalue->_hi); 2175 lo = MIN2(txvalue->_lo, tyvalue->_lo); 2176 } 2177 2178 // Use a flow-free graph structure, to avoid creating excess control edges 2179 // which could hinder other optimizations. 2180 // Since Math.min/max is often used with arraycopy, we want 2181 // tightly_coupled_allocation to be able to see beyond min/max expressions. 2182 Node* cmov = CMoveNode::make(NULL, best_bol, 2183 answer_if_false, answer_if_true, 2184 TypeInt::make(lo, hi, widen)); 2185 2186 return _gvn.transform(cmov); 2187 2188 /* 2189 // This is not as desirable as it may seem, since Min and Max 2190 // nodes do not have a full set of optimizations. 2191 // And they would interfere, anyway, with 'if' optimizations 2192 // and with CMoveI canonical forms. 2193 switch (id) { 2194 case vmIntrinsics::_min: 2195 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break; 2196 case vmIntrinsics::_max: 2197 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break; 2198 default: 2199 ShouldNotReachHere(); 2200 } 2201 */ 2202 } 2203 2204 inline int 2205 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) { 2206 const TypePtr* base_type = TypePtr::NULL_PTR; 2207 if (base != NULL) base_type = _gvn.type(base)->isa_ptr(); 2208 if (base_type == NULL) { 2209 // Unknown type. 2210 return Type::AnyPtr; 2211 } else if (base_type == TypePtr::NULL_PTR) { 2212 // Since this is a NULL+long form, we have to switch to a rawptr. 2213 base = _gvn.transform(new CastX2PNode(offset)); 2214 offset = MakeConX(0); 2215 return Type::RawPtr; 2216 } else if (base_type->base() == Type::RawPtr) { 2217 return Type::RawPtr; 2218 } else if (base_type->isa_oopptr()) { 2219 // Base is never null => always a heap address. 2220 if (!TypePtr::NULL_PTR->higher_equal(base_type)) { 2221 return Type::OopPtr; 2222 } 2223 // Offset is small => always a heap address. 2224 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t(); 2225 if (offset_type != NULL && 2226 base_type->offset() == 0 && // (should always be?) 2227 offset_type->_lo >= 0 && 2228 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) { 2229 return Type::OopPtr; 2230 } else if (type == T_OBJECT) { 2231 // off heap access to an oop doesn't make any sense. Has to be on 2232 // heap. 2233 return Type::OopPtr; 2234 } 2235 // Otherwise, it might either be oop+off or NULL+addr. 2236 return Type::AnyPtr; 2237 } else { 2238 // No information: 2239 return Type::AnyPtr; 2240 } 2241 } 2242 2243 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) { 2244 Node* uncasted_base = base; 2245 int kind = classify_unsafe_addr(uncasted_base, offset, type); 2246 if (kind == Type::RawPtr) { 2247 return basic_plus_adr(top(), uncasted_base, offset); 2248 } else if (kind == Type::AnyPtr) { 2249 assert(base == uncasted_base, "unexpected base change"); 2250 if (can_cast) { 2251 if (!_gvn.type(base)->speculative_maybe_null() && 2252 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 2253 // According to profiling, this access is always on 2254 // heap. Casting the base to not null and thus avoiding membars 2255 // around the access should allow better optimizations 2256 Node* null_ctl = top(); 2257 base = null_check_oop(base, &null_ctl, true, true, true); 2258 assert(null_ctl->is_top(), "no null control here"); 2259 return basic_plus_adr(base, offset); 2260 } else if (_gvn.type(base)->speculative_always_null() && 2261 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 2262 // According to profiling, this access is always off 2263 // heap. 2264 base = null_assert(base); 2265 Node* raw_base = _gvn.transform(new CastX2PNode(offset)); 2266 offset = MakeConX(0); 2267 return basic_plus_adr(top(), raw_base, offset); 2268 } 2269 } 2270 // We don't know if it's an on heap or off heap access. Fall back 2271 // to raw memory access. 2272 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM)); 2273 return basic_plus_adr(top(), raw, offset); 2274 } else { 2275 assert(base == uncasted_base, "unexpected base change"); 2276 // We know it's an on heap access so base can't be null 2277 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) { 2278 base = must_be_not_null(base, true); 2279 } 2280 return basic_plus_adr(base, offset); 2281 } 2282 } 2283 2284 //--------------------------inline_number_methods----------------------------- 2285 // inline int Integer.numberOfLeadingZeros(int) 2286 // inline int Long.numberOfLeadingZeros(long) 2287 // 2288 // inline int Integer.numberOfTrailingZeros(int) 2289 // inline int Long.numberOfTrailingZeros(long) 2290 // 2291 // inline int Integer.bitCount(int) 2292 // inline int Long.bitCount(long) 2293 // 2294 // inline char Character.reverseBytes(char) 2295 // inline short Short.reverseBytes(short) 2296 // inline int Integer.reverseBytes(int) 2297 // inline long Long.reverseBytes(long) 2298 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) { 2299 Node* arg = argument(0); 2300 Node* n = NULL; 2301 switch (id) { 2302 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break; 2303 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break; 2304 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break; 2305 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break; 2306 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break; 2307 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break; 2308 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break; 2309 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break; 2310 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break; 2311 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break; 2312 default: fatal_unexpected_iid(id); break; 2313 } 2314 set_result(_gvn.transform(n)); 2315 return true; 2316 } 2317 2318 //----------------------------inline_unsafe_access---------------------------- 2319 2320 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) { 2321 // Attempt to infer a sharper value type from the offset and base type. 2322 ciKlass* sharpened_klass = NULL; 2323 2324 // See if it is an instance field, with an object type. 2325 if (alias_type->field() != NULL) { 2326 if (alias_type->field()->type()->is_klass()) { 2327 sharpened_klass = alias_type->field()->type()->as_klass(); 2328 } 2329 } 2330 2331 // See if it is a narrow oop array. 2332 if (adr_type->isa_aryptr()) { 2333 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) { 2334 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr(); 2335 if (elem_type != NULL) { 2336 sharpened_klass = elem_type->klass(); 2337 } 2338 } 2339 } 2340 2341 // The sharpened class might be unloaded if there is no class loader 2342 // contraint in place. 2343 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) { 2344 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass); 2345 2346 #ifndef PRODUCT 2347 if (C->print_intrinsics() || C->print_inlining()) { 2348 tty->print(" from base type: "); adr_type->dump(); tty->cr(); 2349 tty->print(" sharpened value: "); tjp->dump(); tty->cr(); 2350 } 2351 #endif 2352 // Sharpen the value type. 2353 return tjp; 2354 } 2355 return NULL; 2356 } 2357 2358 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) { 2359 switch (kind) { 2360 case Relaxed: 2361 return MO_UNORDERED; 2362 case Opaque: 2363 return MO_RELAXED; 2364 case Acquire: 2365 return MO_ACQUIRE; 2366 case Release: 2367 return MO_RELEASE; 2368 case Volatile: 2369 return MO_SEQ_CST; 2370 default: 2371 ShouldNotReachHere(); 2372 return 0; 2373 } 2374 } 2375 2376 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) { 2377 if (callee()->is_static()) return false; // caller must have the capability! 2378 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2379 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads"); 2380 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores"); 2381 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type"); 2382 2383 if (is_reference_type(type)) { 2384 decorators |= ON_UNKNOWN_OOP_REF; 2385 } 2386 2387 if (unaligned) { 2388 decorators |= C2_UNALIGNED; 2389 } 2390 2391 #ifndef PRODUCT 2392 { 2393 ResourceMark rm; 2394 // Check the signatures. 2395 ciSignature* sig = callee()->signature(); 2396 #ifdef ASSERT 2397 if (!is_store) { 2398 // Object getReference(Object base, int/long offset), etc. 2399 BasicType rtype = sig->return_type()->basic_type(); 2400 assert(rtype == type, "getter must return the expected value"); 2401 assert(sig->count() == 2, "oop getter has 2 arguments"); 2402 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object"); 2403 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct"); 2404 } else { 2405 // void putReference(Object base, int/long offset, Object x), etc. 2406 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value"); 2407 assert(sig->count() == 3, "oop putter has 3 arguments"); 2408 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object"); 2409 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct"); 2410 BasicType vtype = sig->type_at(sig->count()-1)->basic_type(); 2411 assert(vtype == type, "putter must accept the expected value"); 2412 } 2413 #endif // ASSERT 2414 } 2415 #endif //PRODUCT 2416 2417 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2418 2419 Node* receiver = argument(0); // type: oop 2420 2421 // Build address expression. 2422 Node* adr; 2423 Node* heap_base_oop = top(); 2424 Node* offset = top(); 2425 Node* val; 2426 2427 // The base is either a Java object or a value produced by Unsafe.staticFieldBase 2428 Node* base = argument(1); // type: oop 2429 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset 2430 offset = argument(2); // type: long 2431 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2432 // to be plain byte offsets, which are also the same as those accepted 2433 // by oopDesc::field_addr. 2434 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 2435 "fieldOffset must be byte-scaled"); 2436 // 32-bit machines ignore the high half! 2437 offset = ConvL2X(offset); 2438 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed); 2439 2440 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) { 2441 if (type != T_OBJECT) { 2442 decorators |= IN_NATIVE; // off-heap primitive access 2443 } else { 2444 return false; // off-heap oop accesses are not supported 2445 } 2446 } else { 2447 heap_base_oop = base; // on-heap or mixed access 2448 } 2449 2450 // Can base be NULL? Otherwise, always on-heap access. 2451 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base)); 2452 2453 if (!can_access_non_heap) { 2454 decorators |= IN_HEAP; 2455 } 2456 2457 val = is_store ? argument(4) : NULL; 2458 2459 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr(); 2460 if (adr_type == TypePtr::NULL_PTR) { 2461 return false; // off-heap access with zero address 2462 } 2463 2464 // Try to categorize the address. 2465 Compile::AliasType* alias_type = C->alias_type(adr_type); 2466 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2467 2468 if (alias_type->adr_type() == TypeInstPtr::KLASS || 2469 alias_type->adr_type() == TypeAryPtr::RANGE) { 2470 return false; // not supported 2471 } 2472 2473 bool mismatched = false; 2474 BasicType bt = alias_type->basic_type(); 2475 if (bt != T_ILLEGAL) { 2476 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access"); 2477 if (bt == T_BYTE && adr_type->isa_aryptr()) { 2478 // Alias type doesn't differentiate between byte[] and boolean[]). 2479 // Use address type to get the element type. 2480 bt = adr_type->is_aryptr()->elem()->array_element_basic_type(); 2481 } 2482 if (bt == T_ARRAY || bt == T_NARROWOOP) { 2483 // accessing an array field with getReference is not a mismatch 2484 bt = T_OBJECT; 2485 } 2486 if ((bt == T_OBJECT) != (type == T_OBJECT)) { 2487 // Don't intrinsify mismatched object accesses 2488 return false; 2489 } 2490 mismatched = (bt != type); 2491 } else if (alias_type->adr_type()->isa_oopptr()) { 2492 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched 2493 } 2494 2495 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched"); 2496 2497 if (mismatched) { 2498 decorators |= C2_MISMATCHED; 2499 } 2500 2501 // First guess at the value type. 2502 const Type *value_type = Type::get_const_basic_type(type); 2503 2504 // Figure out the memory ordering. 2505 decorators |= mo_decorator_for_access_kind(kind); 2506 2507 if (!is_store && type == T_OBJECT) { 2508 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2509 if (tjp != NULL) { 2510 value_type = tjp; 2511 } 2512 } 2513 2514 receiver = null_check(receiver); 2515 if (stopped()) { 2516 return true; 2517 } 2518 // Heap pointers get a null-check from the interpreter, 2519 // as a courtesy. However, this is not guaranteed by Unsafe, 2520 // and it is not possible to fully distinguish unintended nulls 2521 // from intended ones in this API. 2522 2523 if (!is_store) { 2524 Node* p = NULL; 2525 // Try to constant fold a load from a constant field 2526 ciField* field = alias_type->field(); 2527 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) { 2528 // final or stable field 2529 p = make_constant_from_field(field, heap_base_oop); 2530 } 2531 2532 if (p == NULL) { // Could not constant fold the load 2533 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators); 2534 // Normalize the value returned by getBoolean in the following cases 2535 if (type == T_BOOLEAN && 2536 (mismatched || 2537 heap_base_oop == top() || // - heap_base_oop is NULL or 2538 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL 2539 // and the unsafe access is made to large offset 2540 // (i.e., larger than the maximum offset necessary for any 2541 // field access) 2542 ) { 2543 IdealKit ideal = IdealKit(this); 2544 #define __ ideal. 2545 IdealVariable normalized_result(ideal); 2546 __ declarations_done(); 2547 __ set(normalized_result, p); 2548 __ if_then(p, BoolTest::ne, ideal.ConI(0)); 2549 __ set(normalized_result, ideal.ConI(1)); 2550 ideal.end_if(); 2551 final_sync(ideal); 2552 p = __ value(normalized_result); 2553 #undef __ 2554 } 2555 } 2556 if (type == T_ADDRESS) { 2557 p = gvn().transform(new CastP2XNode(NULL, p)); 2558 p = ConvX2UL(p); 2559 } 2560 // The load node has the control of the preceding MemBarCPUOrder. All 2561 // following nodes will have the control of the MemBarCPUOrder inserted at 2562 // the end of this method. So, pushing the load onto the stack at a later 2563 // point is fine. 2564 set_result(p); 2565 } else { 2566 if (bt == T_ADDRESS) { 2567 // Repackage the long as a pointer. 2568 val = ConvL2X(val); 2569 val = gvn().transform(new CastX2PNode(val)); 2570 } 2571 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators); 2572 } 2573 2574 return true; 2575 } 2576 2577 //----------------------------inline_unsafe_load_store---------------------------- 2578 // This method serves a couple of different customers (depending on LoadStoreKind): 2579 // 2580 // LS_cmp_swap: 2581 // 2582 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x); 2583 // boolean compareAndSetInt( Object o, long offset, int expected, int x); 2584 // boolean compareAndSetLong( Object o, long offset, long expected, long x); 2585 // 2586 // LS_cmp_swap_weak: 2587 // 2588 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x); 2589 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x); 2590 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x); 2591 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x); 2592 // 2593 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x); 2594 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x); 2595 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x); 2596 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x); 2597 // 2598 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x); 2599 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x); 2600 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x); 2601 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x); 2602 // 2603 // LS_cmp_exchange: 2604 // 2605 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x); 2606 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x); 2607 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x); 2608 // 2609 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x); 2610 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x); 2611 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x); 2612 // 2613 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x); 2614 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x); 2615 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x); 2616 // 2617 // LS_get_add: 2618 // 2619 // int getAndAddInt( Object o, long offset, int delta) 2620 // long getAndAddLong(Object o, long offset, long delta) 2621 // 2622 // LS_get_set: 2623 // 2624 // int getAndSet(Object o, long offset, int newValue) 2625 // long getAndSet(Object o, long offset, long newValue) 2626 // Object getAndSet(Object o, long offset, Object newValue) 2627 // 2628 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) { 2629 // This basic scheme here is the same as inline_unsafe_access, but 2630 // differs in enough details that combining them would make the code 2631 // overly confusing. (This is a true fact! I originally combined 2632 // them, but even I was confused by it!) As much code/comments as 2633 // possible are retained from inline_unsafe_access though to make 2634 // the correspondences clearer. - dl 2635 2636 if (callee()->is_static()) return false; // caller must have the capability! 2637 2638 DecoratorSet decorators = C2_UNSAFE_ACCESS; 2639 decorators |= mo_decorator_for_access_kind(access_kind); 2640 2641 #ifndef PRODUCT 2642 BasicType rtype; 2643 { 2644 ResourceMark rm; 2645 // Check the signatures. 2646 ciSignature* sig = callee()->signature(); 2647 rtype = sig->return_type()->basic_type(); 2648 switch(kind) { 2649 case LS_get_add: 2650 case LS_get_set: { 2651 // Check the signatures. 2652 #ifdef ASSERT 2653 assert(rtype == type, "get and set must return the expected type"); 2654 assert(sig->count() == 3, "get and set has 3 arguments"); 2655 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object"); 2656 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long"); 2657 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta"); 2658 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation"); 2659 #endif // ASSERT 2660 break; 2661 } 2662 case LS_cmp_swap: 2663 case LS_cmp_swap_weak: { 2664 // Check the signatures. 2665 #ifdef ASSERT 2666 assert(rtype == T_BOOLEAN, "CAS must return boolean"); 2667 assert(sig->count() == 4, "CAS has 4 arguments"); 2668 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2669 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2670 #endif // ASSERT 2671 break; 2672 } 2673 case LS_cmp_exchange: { 2674 // Check the signatures. 2675 #ifdef ASSERT 2676 assert(rtype == type, "CAS must return the expected type"); 2677 assert(sig->count() == 4, "CAS has 4 arguments"); 2678 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object"); 2679 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long"); 2680 #endif // ASSERT 2681 break; 2682 } 2683 default: 2684 ShouldNotReachHere(); 2685 } 2686 } 2687 #endif //PRODUCT 2688 2689 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 2690 2691 // Get arguments: 2692 Node* receiver = NULL; 2693 Node* base = NULL; 2694 Node* offset = NULL; 2695 Node* oldval = NULL; 2696 Node* newval = NULL; 2697 switch(kind) { 2698 case LS_cmp_swap: 2699 case LS_cmp_swap_weak: 2700 case LS_cmp_exchange: { 2701 const bool two_slot_type = type2size[type] == 2; 2702 receiver = argument(0); // type: oop 2703 base = argument(1); // type: oop 2704 offset = argument(2); // type: long 2705 oldval = argument(4); // type: oop, int, or long 2706 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long 2707 break; 2708 } 2709 case LS_get_add: 2710 case LS_get_set: { 2711 receiver = argument(0); // type: oop 2712 base = argument(1); // type: oop 2713 offset = argument(2); // type: long 2714 oldval = NULL; 2715 newval = argument(4); // type: oop, int, or long 2716 break; 2717 } 2718 default: 2719 ShouldNotReachHere(); 2720 } 2721 2722 // Build field offset expression. 2723 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset 2724 // to be plain byte offsets, which are also the same as those accepted 2725 // by oopDesc::field_addr. 2726 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled"); 2727 // 32-bit machines ignore the high half of long offsets 2728 offset = ConvL2X(offset); 2729 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false); 2730 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr(); 2731 2732 Compile::AliasType* alias_type = C->alias_type(adr_type); 2733 BasicType bt = alias_type->basic_type(); 2734 if (bt != T_ILLEGAL && 2735 (is_reference_type(bt) != (type == T_OBJECT))) { 2736 // Don't intrinsify mismatched object accesses. 2737 return false; 2738 } 2739 2740 // For CAS, unlike inline_unsafe_access, there seems no point in 2741 // trying to refine types. Just use the coarse types here. 2742 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here"); 2743 const Type *value_type = Type::get_const_basic_type(type); 2744 2745 switch (kind) { 2746 case LS_get_set: 2747 case LS_cmp_exchange: { 2748 if (type == T_OBJECT) { 2749 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type); 2750 if (tjp != NULL) { 2751 value_type = tjp; 2752 } 2753 } 2754 break; 2755 } 2756 case LS_cmp_swap: 2757 case LS_cmp_swap_weak: 2758 case LS_get_add: 2759 break; 2760 default: 2761 ShouldNotReachHere(); 2762 } 2763 2764 // Null check receiver. 2765 receiver = null_check(receiver); 2766 if (stopped()) { 2767 return true; 2768 } 2769 2770 int alias_idx = C->get_alias_index(adr_type); 2771 2772 if (is_reference_type(type)) { 2773 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF; 2774 2775 // Transformation of a value which could be NULL pointer (CastPP #NULL) 2776 // could be delayed during Parse (for example, in adjust_map_after_if()). 2777 // Execute transformation here to avoid barrier generation in such case. 2778 if (_gvn.type(newval) == TypePtr::NULL_PTR) 2779 newval = _gvn.makecon(TypePtr::NULL_PTR); 2780 2781 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) { 2782 // Refine the value to a null constant, when it is known to be null 2783 oldval = _gvn.makecon(TypePtr::NULL_PTR); 2784 } 2785 } 2786 2787 Node* result = NULL; 2788 switch (kind) { 2789 case LS_cmp_exchange: { 2790 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx, 2791 oldval, newval, value_type, type, decorators); 2792 break; 2793 } 2794 case LS_cmp_swap_weak: 2795 decorators |= C2_WEAK_CMPXCHG; 2796 case LS_cmp_swap: { 2797 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx, 2798 oldval, newval, value_type, type, decorators); 2799 break; 2800 } 2801 case LS_get_set: { 2802 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx, 2803 newval, value_type, type, decorators); 2804 break; 2805 } 2806 case LS_get_add: { 2807 result = access_atomic_add_at(base, adr, adr_type, alias_idx, 2808 newval, value_type, type, decorators); 2809 break; 2810 } 2811 default: 2812 ShouldNotReachHere(); 2813 } 2814 2815 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match"); 2816 set_result(result); 2817 return true; 2818 } 2819 2820 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) { 2821 // Regardless of form, don't allow previous ld/st to move down, 2822 // then issue acquire, release, or volatile mem_bar. 2823 insert_mem_bar(Op_MemBarCPUOrder); 2824 switch(id) { 2825 case vmIntrinsics::_loadFence: 2826 insert_mem_bar(Op_LoadFence); 2827 return true; 2828 case vmIntrinsics::_storeFence: 2829 insert_mem_bar(Op_StoreFence); 2830 return true; 2831 case vmIntrinsics::_fullFence: 2832 insert_mem_bar(Op_MemBarVolatile); 2833 return true; 2834 default: 2835 fatal_unexpected_iid(id); 2836 return false; 2837 } 2838 } 2839 2840 bool LibraryCallKit::inline_onspinwait() { 2841 insert_mem_bar(Op_OnSpinWait); 2842 return true; 2843 } 2844 2845 bool LibraryCallKit::klass_needs_init_guard(Node* kls) { 2846 if (!kls->is_Con()) { 2847 return true; 2848 } 2849 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr(); 2850 if (klsptr == NULL) { 2851 return true; 2852 } 2853 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass(); 2854 // don't need a guard for a klass that is already initialized 2855 return !ik->is_initialized(); 2856 } 2857 2858 //----------------------------inline_unsafe_writeback0------------------------- 2859 // public native void Unsafe.writeback0(long address) 2860 bool LibraryCallKit::inline_unsafe_writeback0() { 2861 if (!Matcher::has_match_rule(Op_CacheWB)) { 2862 return false; 2863 } 2864 #ifndef PRODUCT 2865 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync"); 2866 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync"); 2867 ciSignature* sig = callee()->signature(); 2868 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!"); 2869 #endif 2870 null_check_receiver(); // null-check, then ignore 2871 Node *addr = argument(1); 2872 addr = new CastX2PNode(addr); 2873 addr = _gvn.transform(addr); 2874 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr); 2875 flush = _gvn.transform(flush); 2876 set_memory(flush, TypeRawPtr::BOTTOM); 2877 return true; 2878 } 2879 2880 //----------------------------inline_unsafe_writeback0------------------------- 2881 // public native void Unsafe.writeback0(long address) 2882 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) { 2883 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) { 2884 return false; 2885 } 2886 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) { 2887 return false; 2888 } 2889 #ifndef PRODUCT 2890 assert(Matcher::has_match_rule(Op_CacheWB), 2891 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB" 2892 : "found match rule for CacheWBPostSync but not CacheWB")); 2893 2894 #endif 2895 null_check_receiver(); // null-check, then ignore 2896 Node *sync; 2897 if (is_pre) { 2898 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2899 } else { 2900 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM)); 2901 } 2902 sync = _gvn.transform(sync); 2903 set_memory(sync, TypeRawPtr::BOTTOM); 2904 return true; 2905 } 2906 2907 //----------------------------inline_unsafe_allocate--------------------------- 2908 // public native Object Unsafe.allocateInstance(Class<?> cls); 2909 bool LibraryCallKit::inline_unsafe_allocate() { 2910 if (callee()->is_static()) return false; // caller must have the capability! 2911 2912 null_check_receiver(); // null-check, then ignore 2913 Node* cls = null_check(argument(1)); 2914 if (stopped()) return true; 2915 2916 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 2917 kls = null_check(kls); 2918 if (stopped()) return true; // argument was like int.class 2919 2920 Node* test = NULL; 2921 if (LibraryCallKit::klass_needs_init_guard(kls)) { 2922 // Note: The argument might still be an illegal value like 2923 // Serializable.class or Object[].class. The runtime will handle it. 2924 // But we must make an explicit check for initialization. 2925 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset())); 2926 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler 2927 // can generate code to load it as unsigned byte. 2928 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered); 2929 Node* bits = intcon(InstanceKlass::fully_initialized); 2930 test = _gvn.transform(new SubINode(inst, bits)); 2931 // The 'test' is non-zero if we need to take a slow path. 2932 } 2933 2934 Node* obj = new_instance(kls, test); 2935 set_result(obj); 2936 return true; 2937 } 2938 2939 //------------------------inline_native_time_funcs-------------- 2940 // inline code for System.currentTimeMillis() and System.nanoTime() 2941 // these have the same type and signature 2942 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) { 2943 const TypeFunc* tf = OptoRuntime::void_long_Type(); 2944 const TypePtr* no_memory_effects = NULL; 2945 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects); 2946 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0)); 2947 #ifdef ASSERT 2948 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1)); 2949 assert(value_top == top(), "second value must be top"); 2950 #endif 2951 set_result(value); 2952 return true; 2953 } 2954 2955 #ifdef JFR_HAVE_INTRINSICS 2956 2957 /* 2958 * oop -> myklass 2959 * myklass->trace_id |= USED 2960 * return myklass->trace_id & ~0x3 2961 */ 2962 bool LibraryCallKit::inline_native_classID() { 2963 Node* cls = null_check(argument(0), T_OBJECT); 2964 Node* kls = load_klass_from_mirror(cls, false, NULL, 0); 2965 kls = null_check(kls, T_OBJECT); 2966 2967 ByteSize offset = KLASS_TRACE_ID_OFFSET; 2968 Node* insp = basic_plus_adr(kls, in_bytes(offset)); 2969 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered); 2970 2971 Node* clsused = longcon(0x01l); // set the class bit 2972 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused)); 2973 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr(); 2974 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered); 2975 2976 #ifdef TRACE_ID_META_BITS 2977 Node* mbits = longcon(~TRACE_ID_META_BITS); 2978 tvalue = _gvn.transform(new AndLNode(tvalue, mbits)); 2979 #endif 2980 #ifdef TRACE_ID_SHIFT 2981 Node* cbits = intcon(TRACE_ID_SHIFT); 2982 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits)); 2983 #endif 2984 2985 set_result(tvalue); 2986 return true; 2987 2988 } 2989 2990 bool LibraryCallKit::inline_native_getEventWriter() { 2991 Node* tls_ptr = _gvn.transform(new ThreadLocalNode()); 2992 2993 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, 2994 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)); 2995 2996 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered); 2997 2998 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) ); 2999 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) ); 3000 3001 IfNode* iff_jobj_null = 3002 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN); 3003 3004 enum { _normal_path = 1, 3005 _null_path = 2, 3006 PATH_LIMIT }; 3007 3008 RegionNode* result_rgn = new RegionNode(PATH_LIMIT); 3009 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM); 3010 3011 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null)); 3012 result_rgn->init_req(_null_path, jobj_is_null); 3013 result_val->init_req(_null_path, null()); 3014 3015 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null)); 3016 set_control(jobj_is_not_null); 3017 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT, 3018 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD); 3019 result_rgn->init_req(_normal_path, control()); 3020 result_val->init_req(_normal_path, res); 3021 3022 set_result(result_rgn, result_val); 3023 3024 return true; 3025 } 3026 3027 #endif // JFR_HAVE_INTRINSICS 3028 3029 //------------------------inline_native_currentThread------------------ 3030 bool LibraryCallKit::inline_native_currentThread() { 3031 Node* junk = NULL; 3032 set_result(generate_current_thread(junk)); 3033 return true; 3034 } 3035 3036 //---------------------------load_mirror_from_klass---------------------------- 3037 // Given a klass oop, load its java mirror (a java.lang.Class oop). 3038 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) { 3039 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset())); 3040 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3041 // mirror = ((OopHandle)mirror)->resolve(); 3042 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE); 3043 } 3044 3045 //-----------------------load_klass_from_mirror_common------------------------- 3046 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop. 3047 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE), 3048 // and branch to the given path on the region. 3049 // If never_see_null, take an uncommon trap on null, so we can optimistically 3050 // compile for the non-null case. 3051 // If the region is NULL, force never_see_null = true. 3052 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror, 3053 bool never_see_null, 3054 RegionNode* region, 3055 int null_path, 3056 int offset) { 3057 if (region == NULL) never_see_null = true; 3058 Node* p = basic_plus_adr(mirror, offset); 3059 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3060 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type)); 3061 Node* null_ctl = top(); 3062 kls = null_check_oop(kls, &null_ctl, never_see_null); 3063 if (region != NULL) { 3064 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class). 3065 region->init_req(null_path, null_ctl); 3066 } else { 3067 assert(null_ctl == top(), "no loose ends"); 3068 } 3069 return kls; 3070 } 3071 3072 //--------------------(inline_native_Class_query helpers)--------------------- 3073 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER. 3074 // Fall through if (mods & mask) == bits, take the guard otherwise. 3075 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) { 3076 // Branch around if the given klass has the given modifier bit set. 3077 // Like generate_guard, adds a new path onto the region. 3078 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3079 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered); 3080 Node* mask = intcon(modifier_mask); 3081 Node* bits = intcon(modifier_bits); 3082 Node* mbit = _gvn.transform(new AndINode(mods, mask)); 3083 Node* cmp = _gvn.transform(new CmpINode(mbit, bits)); 3084 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne)); 3085 return generate_fair_guard(bol, region); 3086 } 3087 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) { 3088 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region); 3089 } 3090 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) { 3091 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region); 3092 } 3093 3094 //-------------------------inline_native_Class_query------------------- 3095 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) { 3096 const Type* return_type = TypeInt::BOOL; 3097 Node* prim_return_value = top(); // what happens if it's a primitive class? 3098 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3099 bool expect_prim = false; // most of these guys expect to work on refs 3100 3101 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT }; 3102 3103 Node* mirror = argument(0); 3104 Node* obj = top(); 3105 3106 switch (id) { 3107 case vmIntrinsics::_isInstance: 3108 // nothing is an instance of a primitive type 3109 prim_return_value = intcon(0); 3110 obj = argument(1); 3111 break; 3112 case vmIntrinsics::_getModifiers: 3113 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3114 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line"); 3115 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin); 3116 break; 3117 case vmIntrinsics::_isInterface: 3118 prim_return_value = intcon(0); 3119 break; 3120 case vmIntrinsics::_isArray: 3121 prim_return_value = intcon(0); 3122 expect_prim = true; // cf. ObjectStreamClass.getClassSignature 3123 break; 3124 case vmIntrinsics::_isPrimitive: 3125 prim_return_value = intcon(1); 3126 expect_prim = true; // obviously 3127 break; 3128 case vmIntrinsics::_isHidden: 3129 prim_return_value = intcon(0); 3130 break; 3131 case vmIntrinsics::_getSuperclass: 3132 prim_return_value = null(); 3133 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR); 3134 break; 3135 case vmIntrinsics::_getClassAccessFlags: 3136 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC); 3137 return_type = TypeInt::INT; // not bool! 6297094 3138 break; 3139 default: 3140 fatal_unexpected_iid(id); 3141 break; 3142 } 3143 3144 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3145 if (mirror_con == NULL) return false; // cannot happen? 3146 3147 #ifndef PRODUCT 3148 if (C->print_intrinsics() || C->print_inlining()) { 3149 ciType* k = mirror_con->java_mirror_type(); 3150 if (k) { 3151 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id())); 3152 k->print_name(); 3153 tty->cr(); 3154 } 3155 } 3156 #endif 3157 3158 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive). 3159 RegionNode* region = new RegionNode(PATH_LIMIT); 3160 record_for_igvn(region); 3161 PhiNode* phi = new PhiNode(region, return_type); 3162 3163 // The mirror will never be null of Reflection.getClassAccessFlags, however 3164 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE 3165 // if it is. See bug 4774291. 3166 3167 // For Reflection.getClassAccessFlags(), the null check occurs in 3168 // the wrong place; see inline_unsafe_access(), above, for a similar 3169 // situation. 3170 mirror = null_check(mirror); 3171 // If mirror or obj is dead, only null-path is taken. 3172 if (stopped()) return true; 3173 3174 if (expect_prim) never_see_null = false; // expect nulls (meaning prims) 3175 3176 // Now load the mirror's klass metaobject, and null-check it. 3177 // Side-effects region with the control path if the klass is null. 3178 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path); 3179 // If kls is null, we have a primitive mirror. 3180 phi->init_req(_prim_path, prim_return_value); 3181 if (stopped()) { set_result(region, phi); return true; } 3182 bool safe_for_replace = (region->in(_prim_path) == top()); 3183 3184 Node* p; // handy temp 3185 Node* null_ctl; 3186 3187 // Now that we have the non-null klass, we can perform the real query. 3188 // For constant classes, the query will constant-fold in LoadNode::Value. 3189 Node* query_value = top(); 3190 switch (id) { 3191 case vmIntrinsics::_isInstance: 3192 // nothing is an instance of a primitive type 3193 query_value = gen_instanceof(obj, kls, safe_for_replace); 3194 break; 3195 3196 case vmIntrinsics::_getModifiers: 3197 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset())); 3198 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3199 break; 3200 3201 case vmIntrinsics::_isInterface: 3202 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3203 if (generate_interface_guard(kls, region) != NULL) 3204 // A guard was added. If the guard is taken, it was an interface. 3205 phi->add_req(intcon(1)); 3206 // If we fall through, it's a plain class. 3207 query_value = intcon(0); 3208 break; 3209 3210 case vmIntrinsics::_isArray: 3211 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.) 3212 if (generate_array_guard(kls, region) != NULL) 3213 // A guard was added. If the guard is taken, it was an array. 3214 phi->add_req(intcon(1)); 3215 // If we fall through, it's a plain class. 3216 query_value = intcon(0); 3217 break; 3218 3219 case vmIntrinsics::_isPrimitive: 3220 query_value = intcon(0); // "normal" path produces false 3221 break; 3222 3223 case vmIntrinsics::_isHidden: 3224 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.) 3225 if (generate_hidden_class_guard(kls, region) != NULL) 3226 // A guard was added. If the guard is taken, it was an hidden class. 3227 phi->add_req(intcon(1)); 3228 // If we fall through, it's a plain class. 3229 query_value = intcon(0); 3230 break; 3231 3232 3233 case vmIntrinsics::_getSuperclass: 3234 // The rules here are somewhat unfortunate, but we can still do better 3235 // with random logic than with a JNI call. 3236 // Interfaces store null or Object as _super, but must report null. 3237 // Arrays store an intermediate super as _super, but must report Object. 3238 // Other types can report the actual _super. 3239 // (To verify this code sequence, check the asserts in JVM_IsInterface.) 3240 if (generate_interface_guard(kls, region) != NULL) 3241 // A guard was added. If the guard is taken, it was an interface. 3242 phi->add_req(null()); 3243 if (generate_array_guard(kls, region) != NULL) 3244 // A guard was added. If the guard is taken, it was an array. 3245 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror()))); 3246 // If we fall through, it's a plain class. Get its _super. 3247 p = basic_plus_adr(kls, in_bytes(Klass::super_offset())); 3248 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL)); 3249 null_ctl = top(); 3250 kls = null_check_oop(kls, &null_ctl); 3251 if (null_ctl != top()) { 3252 // If the guard is taken, Object.superClass is null (both klass and mirror). 3253 region->add_req(null_ctl); 3254 phi ->add_req(null()); 3255 } 3256 if (!stopped()) { 3257 query_value = load_mirror_from_klass(kls); 3258 } 3259 break; 3260 3261 case vmIntrinsics::_getClassAccessFlags: 3262 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset())); 3263 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered); 3264 break; 3265 3266 default: 3267 fatal_unexpected_iid(id); 3268 break; 3269 } 3270 3271 // Fall-through is the normal case of a query to a real class. 3272 phi->init_req(1, query_value); 3273 region->init_req(1, control()); 3274 3275 C->set_has_split_ifs(true); // Has chance for split-if optimization 3276 set_result(region, phi); 3277 return true; 3278 } 3279 3280 //-------------------------inline_Class_cast------------------- 3281 bool LibraryCallKit::inline_Class_cast() { 3282 Node* mirror = argument(0); // Class 3283 Node* obj = argument(1); 3284 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr(); 3285 if (mirror_con == NULL) { 3286 return false; // dead path (mirror->is_top()). 3287 } 3288 if (obj == NULL || obj->is_top()) { 3289 return false; // dead path 3290 } 3291 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr(); 3292 3293 // First, see if Class.cast() can be folded statically. 3294 // java_mirror_type() returns non-null for compile-time Class constants. 3295 ciType* tm = mirror_con->java_mirror_type(); 3296 if (tm != NULL && tm->is_klass() && 3297 tp != NULL && tp->klass() != NULL) { 3298 if (!tp->klass()->is_loaded()) { 3299 // Don't use intrinsic when class is not loaded. 3300 return false; 3301 } else { 3302 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass()); 3303 if (static_res == Compile::SSC_always_true) { 3304 // isInstance() is true - fold the code. 3305 set_result(obj); 3306 return true; 3307 } else if (static_res == Compile::SSC_always_false) { 3308 // Don't use intrinsic, have to throw ClassCastException. 3309 // If the reference is null, the non-intrinsic bytecode will 3310 // be optimized appropriately. 3311 return false; 3312 } 3313 } 3314 } 3315 3316 // Bailout intrinsic and do normal inlining if exception path is frequent. 3317 if (too_many_traps(Deoptimization::Reason_intrinsic)) { 3318 return false; 3319 } 3320 3321 // Generate dynamic checks. 3322 // Class.cast() is java implementation of _checkcast bytecode. 3323 // Do checkcast (Parse::do_checkcast()) optimizations here. 3324 3325 mirror = null_check(mirror); 3326 // If mirror is dead, only null-path is taken. 3327 if (stopped()) { 3328 return true; 3329 } 3330 3331 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive). 3332 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT }; 3333 RegionNode* region = new RegionNode(PATH_LIMIT); 3334 record_for_igvn(region); 3335 3336 // Now load the mirror's klass metaobject, and null-check it. 3337 // If kls is null, we have a primitive mirror and 3338 // nothing is an instance of a primitive type. 3339 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path); 3340 3341 Node* res = top(); 3342 if (!stopped()) { 3343 Node* bad_type_ctrl = top(); 3344 // Do checkcast optimizations. 3345 res = gen_checkcast(obj, kls, &bad_type_ctrl); 3346 region->init_req(_bad_type_path, bad_type_ctrl); 3347 } 3348 if (region->in(_prim_path) != top() || 3349 region->in(_bad_type_path) != top()) { 3350 // Let Interpreter throw ClassCastException. 3351 PreserveJVMState pjvms(this); 3352 set_control(_gvn.transform(region)); 3353 uncommon_trap(Deoptimization::Reason_intrinsic, 3354 Deoptimization::Action_maybe_recompile); 3355 } 3356 if (!stopped()) { 3357 set_result(res); 3358 } 3359 return true; 3360 } 3361 3362 3363 //--------------------------inline_native_subtype_check------------------------ 3364 // This intrinsic takes the JNI calls out of the heart of 3365 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc. 3366 bool LibraryCallKit::inline_native_subtype_check() { 3367 // Pull both arguments off the stack. 3368 Node* args[2]; // two java.lang.Class mirrors: superc, subc 3369 args[0] = argument(0); 3370 args[1] = argument(1); 3371 Node* klasses[2]; // corresponding Klasses: superk, subk 3372 klasses[0] = klasses[1] = top(); 3373 3374 enum { 3375 // A full decision tree on {superc is prim, subc is prim}: 3376 _prim_0_path = 1, // {P,N} => false 3377 // {P,P} & superc!=subc => false 3378 _prim_same_path, // {P,P} & superc==subc => true 3379 _prim_1_path, // {N,P} => false 3380 _ref_subtype_path, // {N,N} & subtype check wins => true 3381 _both_ref_path, // {N,N} & subtype check loses => false 3382 PATH_LIMIT 3383 }; 3384 3385 RegionNode* region = new RegionNode(PATH_LIMIT); 3386 Node* phi = new PhiNode(region, TypeInt::BOOL); 3387 record_for_igvn(region); 3388 3389 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads 3390 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL; 3391 int class_klass_offset = java_lang_Class::klass_offset(); 3392 3393 // First null-check both mirrors and load each mirror's klass metaobject. 3394 int which_arg; 3395 for (which_arg = 0; which_arg <= 1; which_arg++) { 3396 Node* arg = args[which_arg]; 3397 arg = null_check(arg); 3398 if (stopped()) break; 3399 args[which_arg] = arg; 3400 3401 Node* p = basic_plus_adr(arg, class_klass_offset); 3402 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type); 3403 klasses[which_arg] = _gvn.transform(kls); 3404 } 3405 3406 // Having loaded both klasses, test each for null. 3407 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3408 for (which_arg = 0; which_arg <= 1; which_arg++) { 3409 Node* kls = klasses[which_arg]; 3410 Node* null_ctl = top(); 3411 kls = null_check_oop(kls, &null_ctl, never_see_null); 3412 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path); 3413 region->init_req(prim_path, null_ctl); 3414 if (stopped()) break; 3415 klasses[which_arg] = kls; 3416 } 3417 3418 if (!stopped()) { 3419 // now we have two reference types, in klasses[0..1] 3420 Node* subk = klasses[1]; // the argument to isAssignableFrom 3421 Node* superk = klasses[0]; // the receiver 3422 region->set_req(_both_ref_path, gen_subtype_check(subk, superk)); 3423 // now we have a successful reference subtype check 3424 region->set_req(_ref_subtype_path, control()); 3425 } 3426 3427 // If both operands are primitive (both klasses null), then 3428 // we must return true when they are identical primitives. 3429 // It is convenient to test this after the first null klass check. 3430 set_control(region->in(_prim_0_path)); // go back to first null check 3431 if (!stopped()) { 3432 // Since superc is primitive, make a guard for the superc==subc case. 3433 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1])); 3434 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq)); 3435 generate_guard(bol_eq, region, PROB_FAIR); 3436 if (region->req() == PATH_LIMIT+1) { 3437 // A guard was added. If the added guard is taken, superc==subc. 3438 region->swap_edges(PATH_LIMIT, _prim_same_path); 3439 region->del_req(PATH_LIMIT); 3440 } 3441 region->set_req(_prim_0_path, control()); // Not equal after all. 3442 } 3443 3444 // these are the only paths that produce 'true': 3445 phi->set_req(_prim_same_path, intcon(1)); 3446 phi->set_req(_ref_subtype_path, intcon(1)); 3447 3448 // pull together the cases: 3449 assert(region->req() == PATH_LIMIT, "sane region"); 3450 for (uint i = 1; i < region->req(); i++) { 3451 Node* ctl = region->in(i); 3452 if (ctl == NULL || ctl == top()) { 3453 region->set_req(i, top()); 3454 phi ->set_req(i, top()); 3455 } else if (phi->in(i) == NULL) { 3456 phi->set_req(i, intcon(0)); // all other paths produce 'false' 3457 } 3458 } 3459 3460 set_control(_gvn.transform(region)); 3461 set_result(_gvn.transform(phi)); 3462 return true; 3463 } 3464 3465 //---------------------generate_array_guard_common------------------------ 3466 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, 3467 bool obj_array, bool not_array) { 3468 3469 if (stopped()) { 3470 return NULL; 3471 } 3472 3473 // If obj_array/non_array==false/false: 3474 // Branch around if the given klass is in fact an array (either obj or prim). 3475 // If obj_array/non_array==false/true: 3476 // Branch around if the given klass is not an array klass of any kind. 3477 // If obj_array/non_array==true/true: 3478 // Branch around if the kls is not an oop array (kls is int[], String, etc.) 3479 // If obj_array/non_array==true/false: 3480 // Branch around if the kls is an oop array (Object[] or subtype) 3481 // 3482 // Like generate_guard, adds a new path onto the region. 3483 jint layout_con = 0; 3484 Node* layout_val = get_layout_helper(kls, layout_con); 3485 if (layout_val == NULL) { 3486 bool query = (obj_array 3487 ? Klass::layout_helper_is_objArray(layout_con) 3488 : Klass::layout_helper_is_array(layout_con)); 3489 if (query == not_array) { 3490 return NULL; // never a branch 3491 } else { // always a branch 3492 Node* always_branch = control(); 3493 if (region != NULL) 3494 region->add_req(always_branch); 3495 set_control(top()); 3496 return always_branch; 3497 } 3498 } 3499 // Now test the correct condition. 3500 jint nval = (obj_array 3501 ? (jint)(Klass::_lh_array_tag_type_value 3502 << Klass::_lh_array_tag_shift) 3503 : Klass::_lh_neutral_value); 3504 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval))); 3505 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array 3506 // invert the test if we are looking for a non-array 3507 if (not_array) btest = BoolTest(btest).negate(); 3508 Node* bol = _gvn.transform(new BoolNode(cmp, btest)); 3509 return generate_fair_guard(bol, region); 3510 } 3511 3512 3513 //-----------------------inline_native_newArray-------------------------- 3514 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length); 3515 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size); 3516 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) { 3517 Node* mirror; 3518 Node* count_val; 3519 if (uninitialized) { 3520 mirror = argument(1); 3521 count_val = argument(2); 3522 } else { 3523 mirror = argument(0); 3524 count_val = argument(1); 3525 } 3526 3527 mirror = null_check(mirror); 3528 // If mirror or obj is dead, only null-path is taken. 3529 if (stopped()) return true; 3530 3531 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT }; 3532 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 3533 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 3534 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 3535 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 3536 3537 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check); 3538 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null, 3539 result_reg, _slow_path); 3540 Node* normal_ctl = control(); 3541 Node* no_array_ctl = result_reg->in(_slow_path); 3542 3543 // Generate code for the slow case. We make a call to newArray(). 3544 set_control(no_array_ctl); 3545 if (!stopped()) { 3546 // Either the input type is void.class, or else the 3547 // array klass has not yet been cached. Either the 3548 // ensuing call will throw an exception, or else it 3549 // will cache the array klass for next time. 3550 PreserveJVMState pjvms(this); 3551 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray); 3552 Node* slow_result = set_results_for_java_call(slow_call); 3553 // this->control() comes from set_results_for_java_call 3554 result_reg->set_req(_slow_path, control()); 3555 result_val->set_req(_slow_path, slow_result); 3556 result_io ->set_req(_slow_path, i_o()); 3557 result_mem->set_req(_slow_path, reset_memory()); 3558 } 3559 3560 set_control(normal_ctl); 3561 if (!stopped()) { 3562 // Normal case: The array type has been cached in the java.lang.Class. 3563 // The following call works fine even if the array type is polymorphic. 3564 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3565 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push 3566 result_reg->init_req(_normal_path, control()); 3567 result_val->init_req(_normal_path, obj); 3568 result_io ->init_req(_normal_path, i_o()); 3569 result_mem->init_req(_normal_path, reset_memory()); 3570 3571 if (uninitialized) { 3572 // Mark the allocation so that zeroing is skipped 3573 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn); 3574 alloc->maybe_set_complete(&_gvn); 3575 } 3576 } 3577 3578 // Return the combined state. 3579 set_i_o( _gvn.transform(result_io) ); 3580 set_all_memory( _gvn.transform(result_mem)); 3581 3582 C->set_has_split_ifs(true); // Has chance for split-if optimization 3583 set_result(result_reg, result_val); 3584 return true; 3585 } 3586 3587 //----------------------inline_native_getLength-------------------------- 3588 // public static native int java.lang.reflect.Array.getLength(Object array); 3589 bool LibraryCallKit::inline_native_getLength() { 3590 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3591 3592 Node* array = null_check(argument(0)); 3593 // If array is dead, only null-path is taken. 3594 if (stopped()) return true; 3595 3596 // Deoptimize if it is a non-array. 3597 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL); 3598 3599 if (non_array != NULL) { 3600 PreserveJVMState pjvms(this); 3601 set_control(non_array); 3602 uncommon_trap(Deoptimization::Reason_intrinsic, 3603 Deoptimization::Action_maybe_recompile); 3604 } 3605 3606 // If control is dead, only non-array-path is taken. 3607 if (stopped()) return true; 3608 3609 // The works fine even if the array type is polymorphic. 3610 // It could be a dynamic mix of int[], boolean[], Object[], etc. 3611 Node* result = load_array_length(array); 3612 3613 C->set_has_split_ifs(true); // Has chance for split-if optimization 3614 set_result(result); 3615 return true; 3616 } 3617 3618 //------------------------inline_array_copyOf---------------------------- 3619 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType); 3620 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType); 3621 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) { 3622 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false; 3623 3624 // Get the arguments. 3625 Node* original = argument(0); 3626 Node* start = is_copyOfRange? argument(1): intcon(0); 3627 Node* end = is_copyOfRange? argument(2): argument(1); 3628 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2); 3629 3630 Node* newcopy = NULL; 3631 3632 // Set the original stack and the reexecute bit for the interpreter to reexecute 3633 // the bytecode that invokes Arrays.copyOf if deoptimization happens. 3634 { PreserveReexecuteState preexecs(this); 3635 jvms()->set_should_reexecute(true); 3636 3637 array_type_mirror = null_check(array_type_mirror); 3638 original = null_check(original); 3639 3640 // Check if a null path was taken unconditionally. 3641 if (stopped()) return true; 3642 3643 Node* orig_length = load_array_length(original); 3644 3645 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0); 3646 klass_node = null_check(klass_node); 3647 3648 RegionNode* bailout = new RegionNode(1); 3649 record_for_igvn(bailout); 3650 3651 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc. 3652 // Bail out if that is so. 3653 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout); 3654 if (not_objArray != NULL) { 3655 // Improve the klass node's type from the new optimistic assumption: 3656 ciKlass* ak = ciArrayKlass::make(env()->Object_klass()); 3657 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/); 3658 Node* cast = new CastPPNode(klass_node, akls); 3659 cast->init_req(0, control()); 3660 klass_node = _gvn.transform(cast); 3661 } 3662 3663 // Bail out if either start or end is negative. 3664 generate_negative_guard(start, bailout, &start); 3665 generate_negative_guard(end, bailout, &end); 3666 3667 Node* length = end; 3668 if (_gvn.type(start) != TypeInt::ZERO) { 3669 length = _gvn.transform(new SubINode(end, start)); 3670 } 3671 3672 // Bail out if length is negative. 3673 // Without this the new_array would throw 3674 // NegativeArraySizeException but IllegalArgumentException is what 3675 // should be thrown 3676 generate_negative_guard(length, bailout, &length); 3677 3678 if (bailout->req() > 1) { 3679 PreserveJVMState pjvms(this); 3680 set_control(_gvn.transform(bailout)); 3681 uncommon_trap(Deoptimization::Reason_intrinsic, 3682 Deoptimization::Action_maybe_recompile); 3683 } 3684 3685 if (!stopped()) { 3686 // How many elements will we copy from the original? 3687 // The answer is MinI(orig_length - start, length). 3688 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start)); 3689 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length); 3690 3691 // Generate a direct call to the right arraycopy function(s). 3692 // We know the copy is disjoint but we might not know if the 3693 // oop stores need checking. 3694 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class). 3695 // This will fail a store-check if x contains any non-nulls. 3696 3697 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to 3698 // loads/stores but it is legal only if we're sure the 3699 // Arrays.copyOf would succeed. So we need all input arguments 3700 // to the copyOf to be validated, including that the copy to the 3701 // new array won't trigger an ArrayStoreException. That subtype 3702 // check can be optimized if we know something on the type of 3703 // the input array from type speculation. 3704 if (_gvn.type(klass_node)->singleton()) { 3705 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass(); 3706 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass(); 3707 3708 int test = C->static_subtype_check(superk, subk); 3709 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) { 3710 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr(); 3711 if (t_original->speculative_type() != NULL) { 3712 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true); 3713 } 3714 } 3715 } 3716 3717 bool validated = false; 3718 // Reason_class_check rather than Reason_intrinsic because we 3719 // want to intrinsify even if this traps. 3720 if (!too_many_traps(Deoptimization::Reason_class_check)) { 3721 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node); 3722 3723 if (not_subtype_ctrl != top()) { 3724 PreserveJVMState pjvms(this); 3725 set_control(not_subtype_ctrl); 3726 uncommon_trap(Deoptimization::Reason_class_check, 3727 Deoptimization::Action_make_not_entrant); 3728 assert(stopped(), "Should be stopped"); 3729 } 3730 validated = true; 3731 } 3732 3733 if (!stopped()) { 3734 newcopy = new_array(klass_node, length, 0); // no arguments to push 3735 3736 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false, 3737 load_object_klass(original), klass_node); 3738 if (!is_copyOfRange) { 3739 ac->set_copyof(validated); 3740 } else { 3741 ac->set_copyofrange(validated); 3742 } 3743 Node* n = _gvn.transform(ac); 3744 if (n == ac) { 3745 ac->connect_outputs(this); 3746 } else { 3747 assert(validated, "shouldn't transform if all arguments not validated"); 3748 set_all_memory(n); 3749 } 3750 } 3751 } 3752 } // original reexecute is set back here 3753 3754 C->set_has_split_ifs(true); // Has chance for split-if optimization 3755 if (!stopped()) { 3756 set_result(newcopy); 3757 } 3758 return true; 3759 } 3760 3761 3762 //----------------------generate_virtual_guard--------------------------- 3763 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call. 3764 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass, 3765 RegionNode* slow_region) { 3766 ciMethod* method = callee(); 3767 int vtable_index = method->vtable_index(); 3768 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 3769 "bad index %d", vtable_index); 3770 // Get the Method* out of the appropriate vtable entry. 3771 int entry_offset = in_bytes(Klass::vtable_start_offset()) + 3772 vtable_index*vtableEntry::size_in_bytes() + 3773 vtableEntry::method_offset_in_bytes(); 3774 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset); 3775 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered); 3776 3777 // Compare the target method with the expected method (e.g., Object.hashCode). 3778 const TypePtr* native_call_addr = TypeMetadataPtr::make(method); 3779 3780 Node* native_call = makecon(native_call_addr); 3781 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call)); 3782 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne)); 3783 3784 return generate_slow_guard(test_native, slow_region); 3785 } 3786 3787 //-----------------------generate_method_call---------------------------- 3788 // Use generate_method_call to make a slow-call to the real 3789 // method if the fast path fails. An alternative would be to 3790 // use a stub like OptoRuntime::slow_arraycopy_Java. 3791 // This only works for expanding the current library call, 3792 // not another intrinsic. (E.g., don't use this for making an 3793 // arraycopy call inside of the copyOf intrinsic.) 3794 CallJavaNode* 3795 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) { 3796 // When compiling the intrinsic method itself, do not use this technique. 3797 guarantee(callee() != C->method(), "cannot make slow-call to self"); 3798 3799 ciMethod* method = callee(); 3800 // ensure the JVMS we have will be correct for this call 3801 guarantee(method_id == method->intrinsic_id(), "must match"); 3802 3803 const TypeFunc* tf = TypeFunc::make(method); 3804 CallJavaNode* slow_call; 3805 if (is_static) { 3806 assert(!is_virtual, ""); 3807 slow_call = new CallStaticJavaNode(C, tf, 3808 SharedRuntime::get_resolve_static_call_stub(), 3809 method, bci()); 3810 } else if (is_virtual) { 3811 null_check_receiver(); 3812 int vtable_index = Method::invalid_vtable_index; 3813 if (UseInlineCaches) { 3814 // Suppress the vtable call 3815 } else { 3816 // hashCode and clone are not a miranda methods, 3817 // so the vtable index is fixed. 3818 // No need to use the linkResolver to get it. 3819 vtable_index = method->vtable_index(); 3820 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index, 3821 "bad index %d", vtable_index); 3822 } 3823 slow_call = new CallDynamicJavaNode(tf, 3824 SharedRuntime::get_resolve_virtual_call_stub(), 3825 method, vtable_index, bci()); 3826 } else { // neither virtual nor static: opt_virtual 3827 null_check_receiver(); 3828 slow_call = new CallStaticJavaNode(C, tf, 3829 SharedRuntime::get_resolve_opt_virtual_call_stub(), 3830 method, bci()); 3831 slow_call->set_optimized_virtual(true); 3832 } 3833 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) { 3834 // To be able to issue a direct call (optimized virtual or virtual) 3835 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information 3836 // about the method being invoked should be attached to the call site to 3837 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C). 3838 slow_call->set_override_symbolic_info(true); 3839 } 3840 set_arguments_for_java_call(slow_call); 3841 set_edges_for_java_call(slow_call); 3842 return slow_call; 3843 } 3844 3845 3846 /** 3847 * Build special case code for calls to hashCode on an object. This call may 3848 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate 3849 * slightly different code. 3850 */ 3851 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) { 3852 assert(is_static == callee()->is_static(), "correct intrinsic selection"); 3853 assert(!(is_virtual && is_static), "either virtual, special, or static"); 3854 3855 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT }; 3856 3857 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 3858 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT); 3859 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO); 3860 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 3861 Node* obj = NULL; 3862 if (!is_static) { 3863 // Check for hashing null object 3864 obj = null_check_receiver(); 3865 if (stopped()) return true; // unconditionally null 3866 result_reg->init_req(_null_path, top()); 3867 result_val->init_req(_null_path, top()); 3868 } else { 3869 // Do a null check, and return zero if null. 3870 // System.identityHashCode(null) == 0 3871 obj = argument(0); 3872 Node* null_ctl = top(); 3873 obj = null_check_oop(obj, &null_ctl); 3874 result_reg->init_req(_null_path, null_ctl); 3875 result_val->init_req(_null_path, _gvn.intcon(0)); 3876 } 3877 3878 // Unconditionally null? Then return right away. 3879 if (stopped()) { 3880 set_control( result_reg->in(_null_path)); 3881 if (!stopped()) 3882 set_result(result_val->in(_null_path)); 3883 return true; 3884 } 3885 3886 // We only go to the fast case code if we pass a number of guards. The 3887 // paths which do not pass are accumulated in the slow_region. 3888 RegionNode* slow_region = new RegionNode(1); 3889 record_for_igvn(slow_region); 3890 3891 // If this is a virtual call, we generate a funny guard. We pull out 3892 // the vtable entry corresponding to hashCode() from the target object. 3893 // If the target method which we are calling happens to be the native 3894 // Object hashCode() method, we pass the guard. We do not need this 3895 // guard for non-virtual calls -- the caller is known to be the native 3896 // Object hashCode(). 3897 if (is_virtual) { 3898 // After null check, get the object's klass. 3899 Node* obj_klass = load_object_klass(obj); 3900 generate_virtual_guard(obj_klass, slow_region); 3901 } 3902 3903 // Get the header out of the object, use LoadMarkNode when available 3904 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes()); 3905 // The control of the load must be NULL. Otherwise, the load can move before 3906 // the null check after castPP removal. 3907 Node* no_ctrl = NULL; 3908 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered); 3909 3910 // Test the header to see if it is unlocked. 3911 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place); 3912 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask)); 3913 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value); 3914 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val)); 3915 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne)); 3916 3917 generate_slow_guard(test_unlocked, slow_region); 3918 3919 // Get the hash value and check to see that it has been properly assigned. 3920 // We depend on hash_mask being at most 32 bits and avoid the use of 3921 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit 3922 // vm: see markWord.hpp. 3923 Node *hash_mask = _gvn.intcon(markWord::hash_mask); 3924 Node *hash_shift = _gvn.intcon(markWord::hash_shift); 3925 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift)); 3926 // This hack lets the hash bits live anywhere in the mark object now, as long 3927 // as the shift drops the relevant bits into the low 32 bits. Note that 3928 // Java spec says that HashCode is an int so there's no point in capturing 3929 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build). 3930 hshifted_header = ConvX2I(hshifted_header); 3931 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask)); 3932 3933 Node *no_hash_val = _gvn.intcon(markWord::no_hash); 3934 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val)); 3935 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq)); 3936 3937 generate_slow_guard(test_assigned, slow_region); 3938 3939 Node* init_mem = reset_memory(); 3940 // fill in the rest of the null path: 3941 result_io ->init_req(_null_path, i_o()); 3942 result_mem->init_req(_null_path, init_mem); 3943 3944 result_val->init_req(_fast_path, hash_val); 3945 result_reg->init_req(_fast_path, control()); 3946 result_io ->init_req(_fast_path, i_o()); 3947 result_mem->init_req(_fast_path, init_mem); 3948 3949 // Generate code for the slow case. We make a call to hashCode(). 3950 set_control(_gvn.transform(slow_region)); 3951 if (!stopped()) { 3952 // No need for PreserveJVMState, because we're using up the present state. 3953 set_all_memory(init_mem); 3954 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode; 3955 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static); 3956 Node* slow_result = set_results_for_java_call(slow_call); 3957 // this->control() comes from set_results_for_java_call 3958 result_reg->init_req(_slow_path, control()); 3959 result_val->init_req(_slow_path, slow_result); 3960 result_io ->set_req(_slow_path, i_o()); 3961 result_mem ->set_req(_slow_path, reset_memory()); 3962 } 3963 3964 // Return the combined state. 3965 set_i_o( _gvn.transform(result_io) ); 3966 set_all_memory( _gvn.transform(result_mem)); 3967 3968 set_result(result_reg, result_val); 3969 return true; 3970 } 3971 3972 //---------------------------inline_native_getClass---------------------------- 3973 // public final native Class<?> java.lang.Object.getClass(); 3974 // 3975 // Build special case code for calls to getClass on an object. 3976 bool LibraryCallKit::inline_native_getClass() { 3977 Node* obj = null_check_receiver(); 3978 if (stopped()) return true; 3979 set_result(load_mirror_from_klass(load_object_klass(obj))); 3980 return true; 3981 } 3982 3983 //-----------------inline_native_Reflection_getCallerClass--------------------- 3984 // public static native Class<?> sun.reflect.Reflection.getCallerClass(); 3985 // 3986 // In the presence of deep enough inlining, getCallerClass() becomes a no-op. 3987 // 3988 // NOTE: This code must perform the same logic as JVM_GetCallerClass 3989 // in that it must skip particular security frames and checks for 3990 // caller sensitive methods. 3991 bool LibraryCallKit::inline_native_Reflection_getCallerClass() { 3992 #ifndef PRODUCT 3993 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 3994 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass"); 3995 } 3996 #endif 3997 3998 if (!jvms()->has_method()) { 3999 #ifndef PRODUCT 4000 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4001 tty->print_cr(" Bailing out because intrinsic was inlined at top level"); 4002 } 4003 #endif 4004 return false; 4005 } 4006 4007 // Walk back up the JVM state to find the caller at the required 4008 // depth. 4009 JVMState* caller_jvms = jvms(); 4010 4011 // Cf. JVM_GetCallerClass 4012 // NOTE: Start the loop at depth 1 because the current JVM state does 4013 // not include the Reflection.getCallerClass() frame. 4014 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) { 4015 ciMethod* m = caller_jvms->method(); 4016 switch (n) { 4017 case 0: 4018 fatal("current JVM state does not include the Reflection.getCallerClass frame"); 4019 break; 4020 case 1: 4021 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). 4022 if (!m->caller_sensitive()) { 4023 #ifndef PRODUCT 4024 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4025 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); 4026 } 4027 #endif 4028 return false; // bail-out; let JVM_GetCallerClass do the work 4029 } 4030 break; 4031 default: 4032 if (!m->is_ignored_by_security_stack_walk()) { 4033 // We have reached the desired frame; return the holder class. 4034 // Acquire method holder as java.lang.Class and push as constant. 4035 ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); 4036 ciInstance* caller_mirror = caller_klass->java_mirror(); 4037 set_result(makecon(TypeInstPtr::make(caller_mirror))); 4038 4039 #ifndef PRODUCT 4040 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4041 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()); 4042 tty->print_cr(" JVM state at this point:"); 4043 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4044 ciMethod* m = jvms()->of_depth(i)->method(); 4045 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4046 } 4047 } 4048 #endif 4049 return true; 4050 } 4051 break; 4052 } 4053 } 4054 4055 #ifndef PRODUCT 4056 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) { 4057 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); 4058 tty->print_cr(" JVM state at this point:"); 4059 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { 4060 ciMethod* m = jvms()->of_depth(i)->method(); 4061 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); 4062 } 4063 } 4064 #endif 4065 4066 return false; // bail-out; let JVM_GetCallerClass do the work 4067 } 4068 4069 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { 4070 Node* arg = argument(0); 4071 Node* result = NULL; 4072 4073 switch (id) { 4074 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break; 4075 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break; 4076 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break; 4077 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break; 4078 4079 case vmIntrinsics::_doubleToLongBits: { 4080 // two paths (plus control) merge in a wood 4081 RegionNode *r = new RegionNode(3); 4082 Node *phi = new PhiNode(r, TypeLong::LONG); 4083 4084 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg)); 4085 // Build the boolean node 4086 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4087 4088 // Branch either way. 4089 // NaN case is less traveled, which makes all the difference. 4090 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4091 Node *opt_isnan = _gvn.transform(ifisnan); 4092 assert( opt_isnan->is_If(), "Expect an IfNode"); 4093 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4094 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4095 4096 set_control(iftrue); 4097 4098 static const jlong nan_bits = CONST64(0x7ff8000000000000); 4099 Node *slow_result = longcon(nan_bits); // return NaN 4100 phi->init_req(1, _gvn.transform( slow_result )); 4101 r->init_req(1, iftrue); 4102 4103 // Else fall through 4104 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4105 set_control(iffalse); 4106 4107 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg))); 4108 r->init_req(2, iffalse); 4109 4110 // Post merge 4111 set_control(_gvn.transform(r)); 4112 record_for_igvn(r); 4113 4114 C->set_has_split_ifs(true); // Has chance for split-if optimization 4115 result = phi; 4116 assert(result->bottom_type()->isa_long(), "must be"); 4117 break; 4118 } 4119 4120 case vmIntrinsics::_floatToIntBits: { 4121 // two paths (plus control) merge in a wood 4122 RegionNode *r = new RegionNode(3); 4123 Node *phi = new PhiNode(r, TypeInt::INT); 4124 4125 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg)); 4126 // Build the boolean node 4127 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne)); 4128 4129 // Branch either way. 4130 // NaN case is less traveled, which makes all the difference. 4131 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN); 4132 Node *opt_isnan = _gvn.transform(ifisnan); 4133 assert( opt_isnan->is_If(), "Expect an IfNode"); 4134 IfNode *opt_ifisnan = (IfNode*)opt_isnan; 4135 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan)); 4136 4137 set_control(iftrue); 4138 4139 static const jint nan_bits = 0x7fc00000; 4140 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN 4141 phi->init_req(1, _gvn.transform( slow_result )); 4142 r->init_req(1, iftrue); 4143 4144 // Else fall through 4145 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan)); 4146 set_control(iffalse); 4147 4148 phi->init_req(2, _gvn.transform(new MoveF2INode(arg))); 4149 r->init_req(2, iffalse); 4150 4151 // Post merge 4152 set_control(_gvn.transform(r)); 4153 record_for_igvn(r); 4154 4155 C->set_has_split_ifs(true); // Has chance for split-if optimization 4156 result = phi; 4157 assert(result->bottom_type()->isa_int(), "must be"); 4158 break; 4159 } 4160 4161 default: 4162 fatal_unexpected_iid(id); 4163 break; 4164 } 4165 set_result(_gvn.transform(result)); 4166 return true; 4167 } 4168 4169 //----------------------inline_unsafe_copyMemory------------------------- 4170 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes); 4171 bool LibraryCallKit::inline_unsafe_copyMemory() { 4172 if (callee()->is_static()) return false; // caller must have the capability! 4173 null_check_receiver(); // null-check receiver 4174 if (stopped()) return true; 4175 4176 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe". 4177 4178 Node* src_ptr = argument(1); // type: oop 4179 Node* src_off = ConvL2X(argument(2)); // type: long 4180 Node* dst_ptr = argument(4); // type: oop 4181 Node* dst_off = ConvL2X(argument(5)); // type: long 4182 Node* size = ConvL2X(argument(7)); // type: long 4183 4184 assert(Unsafe_field_offset_to_byte_offset(11) == 11, 4185 "fieldOffset must be byte-scaled"); 4186 4187 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ); 4188 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE); 4189 4190 // Conservatively insert a memory barrier on all memory slices. 4191 // Do not let writes of the copy source or destination float below the copy. 4192 insert_mem_bar(Op_MemBarCPUOrder); 4193 4194 Node* thread = _gvn.transform(new ThreadLocalNode()); 4195 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset())); 4196 BasicType doing_unsafe_access_bt = T_BYTE; 4197 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented"); 4198 4199 // update volatile field 4200 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4201 4202 // Call it. Note that the length argument is not scaled. 4203 make_runtime_call(RC_LEAF|RC_NO_FP, 4204 OptoRuntime::fast_arraycopy_Type(), 4205 StubRoutines::unsafe_arraycopy(), 4206 "unsafe_arraycopy", 4207 TypeRawPtr::BOTTOM, 4208 src, dst, size XTOP); 4209 4210 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered); 4211 4212 // Do not let reads of the copy destination float above the copy. 4213 insert_mem_bar(Op_MemBarCPUOrder); 4214 4215 return true; 4216 } 4217 4218 //------------------------clone_coping----------------------------------- 4219 // Helper function for inline_native_clone. 4220 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) { 4221 assert(obj_size != NULL, ""); 4222 Node* raw_obj = alloc_obj->in(1); 4223 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), ""); 4224 4225 AllocateNode* alloc = NULL; 4226 if (ReduceBulkZeroing) { 4227 // We will be completely responsible for initializing this object - 4228 // mark Initialize node as complete. 4229 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn); 4230 // The object was just allocated - there should be no any stores! 4231 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), ""); 4232 // Mark as complete_with_arraycopy so that on AllocateNode 4233 // expansion, we know this AllocateNode is initialized by an array 4234 // copy and a StoreStore barrier exists after the array copy. 4235 alloc->initialization()->set_complete_with_arraycopy(); 4236 } 4237 4238 Node* size = _gvn.transform(obj_size); 4239 access_clone(obj, alloc_obj, size, is_array); 4240 4241 // Do not let reads from the cloned object float above the arraycopy. 4242 if (alloc != NULL) { 4243 // Do not let stores that initialize this object be reordered with 4244 // a subsequent store that would make this object accessible by 4245 // other threads. 4246 // Record what AllocateNode this StoreStore protects so that 4247 // escape analysis can go from the MemBarStoreStoreNode to the 4248 // AllocateNode and eliminate the MemBarStoreStoreNode if possible 4249 // based on the escape status of the AllocateNode. 4250 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress)); 4251 } else { 4252 insert_mem_bar(Op_MemBarCPUOrder); 4253 } 4254 } 4255 4256 //------------------------inline_native_clone---------------------------- 4257 // protected native Object java.lang.Object.clone(); 4258 // 4259 // Here are the simple edge cases: 4260 // null receiver => normal trap 4261 // virtual and clone was overridden => slow path to out-of-line clone 4262 // not cloneable or finalizer => slow path to out-of-line Object.clone 4263 // 4264 // The general case has two steps, allocation and copying. 4265 // Allocation has two cases, and uses GraphKit::new_instance or new_array. 4266 // 4267 // Copying also has two cases, oop arrays and everything else. 4268 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy). 4269 // Everything else uses the tight inline loop supplied by CopyArrayNode. 4270 // 4271 // These steps fold up nicely if and when the cloned object's klass 4272 // can be sharply typed as an object array, a type array, or an instance. 4273 // 4274 bool LibraryCallKit::inline_native_clone(bool is_virtual) { 4275 PhiNode* result_val; 4276 4277 // Set the reexecute bit for the interpreter to reexecute 4278 // the bytecode that invokes Object.clone if deoptimization happens. 4279 { PreserveReexecuteState preexecs(this); 4280 jvms()->set_should_reexecute(true); 4281 4282 Node* obj = null_check_receiver(); 4283 if (stopped()) return true; 4284 4285 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 4286 4287 // If we are going to clone an instance, we need its exact type to 4288 // know the number and types of fields to convert the clone to 4289 // loads/stores. Maybe a speculative type can help us. 4290 if (!obj_type->klass_is_exact() && 4291 obj_type->speculative_type() != NULL && 4292 obj_type->speculative_type()->is_instance_klass()) { 4293 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass(); 4294 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem && 4295 !spec_ik->has_injected_fields()) { 4296 ciKlass* k = obj_type->klass(); 4297 if (!k->is_instance_klass() || 4298 k->as_instance_klass()->is_interface() || 4299 k->as_instance_klass()->has_subklass()) { 4300 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false); 4301 } 4302 } 4303 } 4304 4305 // Conservatively insert a memory barrier on all memory slices. 4306 // Do not let writes into the original float below the clone. 4307 insert_mem_bar(Op_MemBarCPUOrder); 4308 4309 // paths into result_reg: 4310 enum { 4311 _slow_path = 1, // out-of-line call to clone method (virtual or not) 4312 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy 4313 _array_path, // plain array allocation, plus arrayof_long_arraycopy 4314 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy 4315 PATH_LIMIT 4316 }; 4317 RegionNode* result_reg = new RegionNode(PATH_LIMIT); 4318 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL); 4319 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO); 4320 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM); 4321 record_for_igvn(result_reg); 4322 4323 Node* obj_klass = load_object_klass(obj); 4324 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL); 4325 if (array_ctl != NULL) { 4326 // It's an array. 4327 PreserveJVMState pjvms(this); 4328 set_control(array_ctl); 4329 Node* obj_length = load_array_length(obj); 4330 Node* obj_size = NULL; 4331 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true); 4332 4333 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 4334 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) { 4335 // If it is an oop array, it requires very special treatment, 4336 // because gc barriers are required when accessing the array. 4337 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL); 4338 if (is_obja != NULL) { 4339 PreserveJVMState pjvms2(this); 4340 set_control(is_obja); 4341 // Generate a direct call to the right arraycopy function(s). 4342 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL); 4343 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false); 4344 ac->set_clone_oop_array(); 4345 Node* n = _gvn.transform(ac); 4346 assert(n == ac, "cannot disappear"); 4347 ac->connect_outputs(this, /*deoptimize_on_exception=*/true); 4348 4349 result_reg->init_req(_objArray_path, control()); 4350 result_val->init_req(_objArray_path, alloc_obj); 4351 result_i_o ->set_req(_objArray_path, i_o()); 4352 result_mem ->set_req(_objArray_path, reset_memory()); 4353 } 4354 } 4355 // Otherwise, there are no barriers to worry about. 4356 // (We can dispense with card marks if we know the allocation 4357 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks 4358 // causes the non-eden paths to take compensating steps to 4359 // simulate a fresh allocation, so that no further 4360 // card marks are required in compiled code to initialize 4361 // the object.) 4362 4363 if (!stopped()) { 4364 copy_to_clone(obj, alloc_obj, obj_size, true); 4365 4366 // Present the results of the copy. 4367 result_reg->init_req(_array_path, control()); 4368 result_val->init_req(_array_path, alloc_obj); 4369 result_i_o ->set_req(_array_path, i_o()); 4370 result_mem ->set_req(_array_path, reset_memory()); 4371 } 4372 } 4373 4374 // We only go to the instance fast case code if we pass a number of guards. 4375 // The paths which do not pass are accumulated in the slow_region. 4376 RegionNode* slow_region = new RegionNode(1); 4377 record_for_igvn(slow_region); 4378 if (!stopped()) { 4379 // It's an instance (we did array above). Make the slow-path tests. 4380 // If this is a virtual call, we generate a funny guard. We grab 4381 // the vtable entry corresponding to clone() from the target object. 4382 // If the target method which we are calling happens to be the 4383 // Object clone() method, we pass the guard. We do not need this 4384 // guard for non-virtual calls; the caller is known to be the native 4385 // Object clone(). 4386 if (is_virtual) { 4387 generate_virtual_guard(obj_klass, slow_region); 4388 } 4389 4390 // The object must be easily cloneable and must not have a finalizer. 4391 // Both of these conditions may be checked in a single test. 4392 // We could optimize the test further, but we don't care. 4393 generate_access_flags_guard(obj_klass, 4394 // Test both conditions: 4395 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER, 4396 // Must be cloneable but not finalizer: 4397 JVM_ACC_IS_CLONEABLE_FAST, 4398 slow_region); 4399 } 4400 4401 if (!stopped()) { 4402 // It's an instance, and it passed the slow-path tests. 4403 PreserveJVMState pjvms(this); 4404 Node* obj_size = NULL; 4405 // Need to deoptimize on exception from allocation since Object.clone intrinsic 4406 // is reexecuted if deoptimization occurs and there could be problems when merging 4407 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false). 4408 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true); 4409 4410 copy_to_clone(obj, alloc_obj, obj_size, false); 4411 4412 // Present the results of the slow call. 4413 result_reg->init_req(_instance_path, control()); 4414 result_val->init_req(_instance_path, alloc_obj); 4415 result_i_o ->set_req(_instance_path, i_o()); 4416 result_mem ->set_req(_instance_path, reset_memory()); 4417 } 4418 4419 // Generate code for the slow case. We make a call to clone(). 4420 set_control(_gvn.transform(slow_region)); 4421 if (!stopped()) { 4422 PreserveJVMState pjvms(this); 4423 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual); 4424 // We need to deoptimize on exception (see comment above) 4425 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true); 4426 // this->control() comes from set_results_for_java_call 4427 result_reg->init_req(_slow_path, control()); 4428 result_val->init_req(_slow_path, slow_result); 4429 result_i_o ->set_req(_slow_path, i_o()); 4430 result_mem ->set_req(_slow_path, reset_memory()); 4431 } 4432 4433 // Return the combined state. 4434 set_control( _gvn.transform(result_reg)); 4435 set_i_o( _gvn.transform(result_i_o)); 4436 set_all_memory( _gvn.transform(result_mem)); 4437 } // original reexecute is set back here 4438 4439 set_result(_gvn.transform(result_val)); 4440 return true; 4441 } 4442 4443 // If we have a tightly coupled allocation, the arraycopy may take care 4444 // of the array initialization. If one of the guards we insert between 4445 // the allocation and the arraycopy causes a deoptimization, an 4446 // unitialized array will escape the compiled method. To prevent that 4447 // we set the JVM state for uncommon traps between the allocation and 4448 // the arraycopy to the state before the allocation so, in case of 4449 // deoptimization, we'll reexecute the allocation and the 4450 // initialization. 4451 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) { 4452 if (alloc != NULL) { 4453 ciMethod* trap_method = alloc->jvms()->method(); 4454 int trap_bci = alloc->jvms()->bci(); 4455 4456 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 4457 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) { 4458 // Make sure there's no store between the allocation and the 4459 // arraycopy otherwise visible side effects could be rexecuted 4460 // in case of deoptimization and cause incorrect execution. 4461 bool no_interfering_store = true; 4462 Node* mem = alloc->in(TypeFunc::Memory); 4463 if (mem->is_MergeMem()) { 4464 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) { 4465 Node* n = mms.memory(); 4466 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4467 assert(n->is_Store(), "what else?"); 4468 no_interfering_store = false; 4469 break; 4470 } 4471 } 4472 } else { 4473 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) { 4474 Node* n = mms.memory(); 4475 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) { 4476 assert(n->is_Store(), "what else?"); 4477 no_interfering_store = false; 4478 break; 4479 } 4480 } 4481 } 4482 4483 if (no_interfering_store) { 4484 JVMState* old_jvms = alloc->jvms()->clone_shallow(C); 4485 uint size = alloc->req(); 4486 SafePointNode* sfpt = new SafePointNode(size, old_jvms); 4487 old_jvms->set_map(sfpt); 4488 for (uint i = 0; i < size; i++) { 4489 sfpt->init_req(i, alloc->in(i)); 4490 } 4491 // re-push array length for deoptimization 4492 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength)); 4493 old_jvms->set_sp(old_jvms->sp()+1); 4494 old_jvms->set_monoff(old_jvms->monoff()+1); 4495 old_jvms->set_scloff(old_jvms->scloff()+1); 4496 old_jvms->set_endoff(old_jvms->endoff()+1); 4497 old_jvms->set_should_reexecute(true); 4498 4499 sfpt->set_i_o(map()->i_o()); 4500 sfpt->set_memory(map()->memory()); 4501 sfpt->set_control(map()->control()); 4502 4503 JVMState* saved_jvms = jvms(); 4504 saved_reexecute_sp = _reexecute_sp; 4505 4506 set_jvms(sfpt->jvms()); 4507 _reexecute_sp = jvms()->sp(); 4508 4509 return saved_jvms; 4510 } 4511 } 4512 } 4513 return NULL; 4514 } 4515 4516 // In case of a deoptimization, we restart execution at the 4517 // allocation, allocating a new array. We would leave an uninitialized 4518 // array in the heap that GCs wouldn't expect. Move the allocation 4519 // after the traps so we don't allocate the array if we 4520 // deoptimize. This is possible because tightly_coupled_allocation() 4521 // guarantees there's no observer of the allocated array at this point 4522 // and the control flow is simple enough. 4523 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, 4524 int saved_reexecute_sp, uint new_idx) { 4525 if (saved_jvms != NULL && !stopped()) { 4526 assert(alloc != NULL, "only with a tightly coupled allocation"); 4527 // restore JVM state to the state at the arraycopy 4528 saved_jvms->map()->set_control(map()->control()); 4529 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?"); 4530 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?"); 4531 // If we've improved the types of some nodes (null check) while 4532 // emitting the guards, propagate them to the current state 4533 map()->replaced_nodes().apply(saved_jvms->map(), new_idx); 4534 set_jvms(saved_jvms); 4535 _reexecute_sp = saved_reexecute_sp; 4536 4537 // Remove the allocation from above the guards 4538 CallProjections callprojs; 4539 alloc->extract_projections(&callprojs, true); 4540 InitializeNode* init = alloc->initialization(); 4541 Node* alloc_mem = alloc->in(TypeFunc::Memory); 4542 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O)); 4543 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem); 4544 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0)); 4545 4546 // move the allocation here (after the guards) 4547 _gvn.hash_delete(alloc); 4548 alloc->set_req(TypeFunc::Control, control()); 4549 alloc->set_req(TypeFunc::I_O, i_o()); 4550 Node *mem = reset_memory(); 4551 set_all_memory(mem); 4552 alloc->set_req(TypeFunc::Memory, mem); 4553 set_control(init->proj_out_or_null(TypeFunc::Control)); 4554 set_i_o(callprojs.fallthrough_ioproj); 4555 4556 // Update memory as done in GraphKit::set_output_for_allocation() 4557 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength)); 4558 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type(); 4559 if (ary_type->isa_aryptr() && length_type != NULL) { 4560 ary_type = ary_type->is_aryptr()->cast_to_size(length_type); 4561 } 4562 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot); 4563 int elemidx = C->get_alias_index(telemref); 4564 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw); 4565 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx); 4566 4567 Node* allocx = _gvn.transform(alloc); 4568 assert(allocx == alloc, "where has the allocation gone?"); 4569 assert(dest->is_CheckCastPP(), "not an allocation result?"); 4570 4571 _gvn.hash_delete(dest); 4572 dest->set_req(0, control()); 4573 Node* destx = _gvn.transform(dest); 4574 assert(destx == dest, "where has the allocation result gone?"); 4575 } 4576 } 4577 4578 4579 //------------------------------inline_arraycopy----------------------- 4580 // public static native void java.lang.System.arraycopy(Object src, int srcPos, 4581 // Object dest, int destPos, 4582 // int length); 4583 bool LibraryCallKit::inline_arraycopy() { 4584 // Get the arguments. 4585 Node* src = argument(0); // type: oop 4586 Node* src_offset = argument(1); // type: int 4587 Node* dest = argument(2); // type: oop 4588 Node* dest_offset = argument(3); // type: int 4589 Node* length = argument(4); // type: int 4590 4591 uint new_idx = C->unique(); 4592 4593 // Check for allocation before we add nodes that would confuse 4594 // tightly_coupled_allocation() 4595 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL); 4596 4597 int saved_reexecute_sp = -1; 4598 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp); 4599 // See arraycopy_restore_alloc_state() comment 4600 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards 4601 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation 4602 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards 4603 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL); 4604 4605 // The following tests must be performed 4606 // (1) src and dest are arrays. 4607 // (2) src and dest arrays must have elements of the same BasicType 4608 // (3) src and dest must not be null. 4609 // (4) src_offset must not be negative. 4610 // (5) dest_offset must not be negative. 4611 // (6) length must not be negative. 4612 // (7) src_offset + length must not exceed length of src. 4613 // (8) dest_offset + length must not exceed length of dest. 4614 // (9) each element of an oop array must be assignable 4615 4616 // (3) src and dest must not be null. 4617 // always do this here because we need the JVM state for uncommon traps 4618 Node* null_ctl = top(); 4619 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY); 4620 assert(null_ctl->is_top(), "no null control here"); 4621 dest = null_check(dest, T_ARRAY); 4622 4623 if (!can_emit_guards) { 4624 // if saved_jvms == NULL and alloc != NULL, we don't emit any 4625 // guards but the arraycopy node could still take advantage of a 4626 // tightly allocated allocation. tightly_coupled_allocation() is 4627 // called again to make sure it takes the null check above into 4628 // account: the null check is mandatory and if it caused an 4629 // uncommon trap to be emitted then the allocation can't be 4630 // considered tightly coupled in this context. 4631 alloc = tightly_coupled_allocation(dest, NULL); 4632 } 4633 4634 bool validated = false; 4635 4636 const Type* src_type = _gvn.type(src); 4637 const Type* dest_type = _gvn.type(dest); 4638 const TypeAryPtr* top_src = src_type->isa_aryptr(); 4639 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 4640 4641 // Do we have the type of src? 4642 bool has_src = (top_src != NULL && top_src->klass() != NULL); 4643 // Do we have the type of dest? 4644 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL); 4645 // Is the type for src from speculation? 4646 bool src_spec = false; 4647 // Is the type for dest from speculation? 4648 bool dest_spec = false; 4649 4650 if ((!has_src || !has_dest) && can_emit_guards) { 4651 // We don't have sufficient type information, let's see if 4652 // speculative types can help. We need to have types for both src 4653 // and dest so that it pays off. 4654 4655 // Do we already have or could we have type information for src 4656 bool could_have_src = has_src; 4657 // Do we already have or could we have type information for dest 4658 bool could_have_dest = has_dest; 4659 4660 ciKlass* src_k = NULL; 4661 if (!has_src) { 4662 src_k = src_type->speculative_type_not_null(); 4663 if (src_k != NULL && src_k->is_array_klass()) { 4664 could_have_src = true; 4665 } 4666 } 4667 4668 ciKlass* dest_k = NULL; 4669 if (!has_dest) { 4670 dest_k = dest_type->speculative_type_not_null(); 4671 if (dest_k != NULL && dest_k->is_array_klass()) { 4672 could_have_dest = true; 4673 } 4674 } 4675 4676 if (could_have_src && could_have_dest) { 4677 // This is going to pay off so emit the required guards 4678 if (!has_src) { 4679 src = maybe_cast_profiled_obj(src, src_k, true); 4680 src_type = _gvn.type(src); 4681 top_src = src_type->isa_aryptr(); 4682 has_src = (top_src != NULL && top_src->klass() != NULL); 4683 src_spec = true; 4684 } 4685 if (!has_dest) { 4686 dest = maybe_cast_profiled_obj(dest, dest_k, true); 4687 dest_type = _gvn.type(dest); 4688 top_dest = dest_type->isa_aryptr(); 4689 has_dest = (top_dest != NULL && top_dest->klass() != NULL); 4690 dest_spec = true; 4691 } 4692 } 4693 } 4694 4695 if (has_src && has_dest && can_emit_guards) { 4696 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type(); 4697 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type(); 4698 if (is_reference_type(src_elem)) src_elem = T_OBJECT; 4699 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT; 4700 4701 if (src_elem == dest_elem && src_elem == T_OBJECT) { 4702 // If both arrays are object arrays then having the exact types 4703 // for both will remove the need for a subtype check at runtime 4704 // before the call and may make it possible to pick a faster copy 4705 // routine (without a subtype check on every element) 4706 // Do we have the exact type of src? 4707 bool could_have_src = src_spec; 4708 // Do we have the exact type of dest? 4709 bool could_have_dest = dest_spec; 4710 ciKlass* src_k = top_src->klass(); 4711 ciKlass* dest_k = top_dest->klass(); 4712 if (!src_spec) { 4713 src_k = src_type->speculative_type_not_null(); 4714 if (src_k != NULL && src_k->is_array_klass()) { 4715 could_have_src = true; 4716 } 4717 } 4718 if (!dest_spec) { 4719 dest_k = dest_type->speculative_type_not_null(); 4720 if (dest_k != NULL && dest_k->is_array_klass()) { 4721 could_have_dest = true; 4722 } 4723 } 4724 if (could_have_src && could_have_dest) { 4725 // If we can have both exact types, emit the missing guards 4726 if (could_have_src && !src_spec) { 4727 src = maybe_cast_profiled_obj(src, src_k, true); 4728 } 4729 if (could_have_dest && !dest_spec) { 4730 dest = maybe_cast_profiled_obj(dest, dest_k, true); 4731 } 4732 } 4733 } 4734 } 4735 4736 ciMethod* trap_method = method(); 4737 int trap_bci = bci(); 4738 if (saved_jvms != NULL) { 4739 trap_method = alloc->jvms()->method(); 4740 trap_bci = alloc->jvms()->bci(); 4741 } 4742 4743 bool negative_length_guard_generated = false; 4744 4745 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) && 4746 can_emit_guards && 4747 !src->is_top() && !dest->is_top()) { 4748 // validate arguments: enables transformation the ArrayCopyNode 4749 validated = true; 4750 4751 RegionNode* slow_region = new RegionNode(1); 4752 record_for_igvn(slow_region); 4753 4754 // (1) src and dest are arrays. 4755 generate_non_array_guard(load_object_klass(src), slow_region); 4756 generate_non_array_guard(load_object_klass(dest), slow_region); 4757 4758 // (2) src and dest arrays must have elements of the same BasicType 4759 // done at macro expansion or at Ideal transformation time 4760 4761 // (4) src_offset must not be negative. 4762 generate_negative_guard(src_offset, slow_region); 4763 4764 // (5) dest_offset must not be negative. 4765 generate_negative_guard(dest_offset, slow_region); 4766 4767 // (7) src_offset + length must not exceed length of src. 4768 generate_limit_guard(src_offset, length, 4769 load_array_length(src), 4770 slow_region); 4771 4772 // (8) dest_offset + length must not exceed length of dest. 4773 generate_limit_guard(dest_offset, length, 4774 load_array_length(dest), 4775 slow_region); 4776 4777 // (6) length must not be negative. 4778 // This is also checked in generate_arraycopy() during macro expansion, but 4779 // we also have to check it here for the case where the ArrayCopyNode will 4780 // be eliminated by Escape Analysis. 4781 if (EliminateAllocations) { 4782 generate_negative_guard(length, slow_region); 4783 negative_length_guard_generated = true; 4784 } 4785 4786 // (9) each element of an oop array must be assignable 4787 Node* dest_klass = load_object_klass(dest); 4788 if (src != dest) { 4789 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass); 4790 4791 if (not_subtype_ctrl != top()) { 4792 PreserveJVMState pjvms(this); 4793 set_control(not_subtype_ctrl); 4794 uncommon_trap(Deoptimization::Reason_intrinsic, 4795 Deoptimization::Action_make_not_entrant); 4796 assert(stopped(), "Should be stopped"); 4797 } 4798 } 4799 { 4800 PreserveJVMState pjvms(this); 4801 set_control(_gvn.transform(slow_region)); 4802 uncommon_trap(Deoptimization::Reason_intrinsic, 4803 Deoptimization::Action_make_not_entrant); 4804 assert(stopped(), "Should be stopped"); 4805 } 4806 4807 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr(); 4808 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass()); 4809 src = _gvn.transform(new CheckCastPPNode(control(), src, toop)); 4810 } 4811 4812 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx); 4813 4814 if (stopped()) { 4815 return true; 4816 } 4817 4818 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated, 4819 // Create LoadRange and LoadKlass nodes for use during macro expansion here 4820 // so the compiler has a chance to eliminate them: during macro expansion, 4821 // we have to set their control (CastPP nodes are eliminated). 4822 load_object_klass(src), load_object_klass(dest), 4823 load_array_length(src), load_array_length(dest)); 4824 4825 ac->set_arraycopy(validated); 4826 4827 Node* n = _gvn.transform(ac); 4828 if (n == ac) { 4829 ac->connect_outputs(this); 4830 } else { 4831 assert(validated, "shouldn't transform if all arguments not validated"); 4832 set_all_memory(n); 4833 } 4834 clear_upper_avx(); 4835 4836 4837 return true; 4838 } 4839 4840 4841 // Helper function which determines if an arraycopy immediately follows 4842 // an allocation, with no intervening tests or other escapes for the object. 4843 AllocateArrayNode* 4844 LibraryCallKit::tightly_coupled_allocation(Node* ptr, 4845 RegionNode* slow_region) { 4846 if (stopped()) return NULL; // no fast path 4847 if (C->AliasLevel() == 0) return NULL; // no MergeMems around 4848 4849 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn); 4850 if (alloc == NULL) return NULL; 4851 4852 Node* rawmem = memory(Compile::AliasIdxRaw); 4853 // Is the allocation's memory state untouched? 4854 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) { 4855 // Bail out if there have been raw-memory effects since the allocation. 4856 // (Example: There might have been a call or safepoint.) 4857 return NULL; 4858 } 4859 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw); 4860 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) { 4861 return NULL; 4862 } 4863 4864 // There must be no unexpected observers of this allocation. 4865 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) { 4866 Node* obs = ptr->fast_out(i); 4867 if (obs != this->map()) { 4868 return NULL; 4869 } 4870 } 4871 4872 // This arraycopy must unconditionally follow the allocation of the ptr. 4873 Node* alloc_ctl = ptr->in(0); 4874 Node* ctl = control(); 4875 while (ctl != alloc_ctl) { 4876 // There may be guards which feed into the slow_region. 4877 // Any other control flow means that we might not get a chance 4878 // to finish initializing the allocated object. 4879 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) { 4880 IfNode* iff = ctl->in(0)->as_If(); 4881 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con); 4882 assert(not_ctl != NULL && not_ctl != ctl, "found alternate"); 4883 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) { 4884 ctl = iff->in(0); // This test feeds the known slow_region. 4885 continue; 4886 } 4887 // One more try: Various low-level checks bottom out in 4888 // uncommon traps. If the debug-info of the trap omits 4889 // any reference to the allocation, as we've already 4890 // observed, then there can be no objection to the trap. 4891 bool found_trap = false; 4892 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) { 4893 Node* obs = not_ctl->fast_out(j); 4894 if (obs->in(0) == not_ctl && obs->is_Call() && 4895 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) { 4896 found_trap = true; break; 4897 } 4898 } 4899 if (found_trap) { 4900 ctl = iff->in(0); // This test feeds a harmless uncommon trap. 4901 continue; 4902 } 4903 } 4904 return NULL; 4905 } 4906 4907 // If we get this far, we have an allocation which immediately 4908 // precedes the arraycopy, and we can take over zeroing the new object. 4909 // The arraycopy will finish the initialization, and provide 4910 // a new control state to which we will anchor the destination pointer. 4911 4912 return alloc; 4913 } 4914 4915 //-------------inline_encodeISOArray----------------------------------- 4916 // encode char[] to byte[] in ISO_8859_1 4917 bool LibraryCallKit::inline_encodeISOArray() { 4918 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters"); 4919 // no receiver since it is static method 4920 Node *src = argument(0); 4921 Node *src_offset = argument(1); 4922 Node *dst = argument(2); 4923 Node *dst_offset = argument(3); 4924 Node *length = argument(4); 4925 4926 src = must_be_not_null(src, true); 4927 dst = must_be_not_null(dst, true); 4928 4929 const Type* src_type = src->Value(&_gvn); 4930 const Type* dst_type = dst->Value(&_gvn); 4931 const TypeAryPtr* top_src = src_type->isa_aryptr(); 4932 const TypeAryPtr* top_dest = dst_type->isa_aryptr(); 4933 if (top_src == NULL || top_src->klass() == NULL || 4934 top_dest == NULL || top_dest->klass() == NULL) { 4935 // failed array check 4936 return false; 4937 } 4938 4939 // Figure out the size and type of the elements we will be copying. 4940 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4941 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4942 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) { 4943 return false; 4944 } 4945 4946 Node* src_start = array_element_address(src, src_offset, T_CHAR); 4947 Node* dst_start = array_element_address(dst, dst_offset, dst_elem); 4948 // 'src_start' points to src array + scaled offset 4949 // 'dst_start' points to dst array + scaled offset 4950 4951 const TypeAryPtr* mtype = TypeAryPtr::BYTES; 4952 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length); 4953 enc = _gvn.transform(enc); 4954 Node* res_mem = _gvn.transform(new SCMemProjNode(enc)); 4955 set_memory(res_mem, mtype); 4956 set_result(enc); 4957 clear_upper_avx(); 4958 4959 return true; 4960 } 4961 4962 //-------------inline_multiplyToLen----------------------------------- 4963 bool LibraryCallKit::inline_multiplyToLen() { 4964 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform"); 4965 4966 address stubAddr = StubRoutines::multiplyToLen(); 4967 if (stubAddr == NULL) { 4968 return false; // Intrinsic's stub is not implemented on this platform 4969 } 4970 const char* stubName = "multiplyToLen"; 4971 4972 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters"); 4973 4974 // no receiver because it is a static method 4975 Node* x = argument(0); 4976 Node* xlen = argument(1); 4977 Node* y = argument(2); 4978 Node* ylen = argument(3); 4979 Node* z = argument(4); 4980 4981 x = must_be_not_null(x, true); 4982 y = must_be_not_null(y, true); 4983 4984 const Type* x_type = x->Value(&_gvn); 4985 const Type* y_type = y->Value(&_gvn); 4986 const TypeAryPtr* top_x = x_type->isa_aryptr(); 4987 const TypeAryPtr* top_y = y_type->isa_aryptr(); 4988 if (top_x == NULL || top_x->klass() == NULL || 4989 top_y == NULL || top_y->klass() == NULL) { 4990 // failed array check 4991 return false; 4992 } 4993 4994 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4995 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 4996 if (x_elem != T_INT || y_elem != T_INT) { 4997 return false; 4998 } 4999 5000 // Set the original stack and the reexecute bit for the interpreter to reexecute 5001 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens 5002 // on the return from z array allocation in runtime. 5003 { PreserveReexecuteState preexecs(this); 5004 jvms()->set_should_reexecute(true); 5005 5006 Node* x_start = array_element_address(x, intcon(0), x_elem); 5007 Node* y_start = array_element_address(y, intcon(0), y_elem); 5008 // 'x_start' points to x array + scaled xlen 5009 // 'y_start' points to y array + scaled ylen 5010 5011 // Allocate the result array 5012 Node* zlen = _gvn.transform(new AddINode(xlen, ylen)); 5013 ciKlass* klass = ciTypeArrayKlass::make(T_INT); 5014 Node* klass_node = makecon(TypeKlassPtr::make(klass)); 5015 5016 IdealKit ideal(this); 5017 5018 #define __ ideal. 5019 Node* one = __ ConI(1); 5020 Node* zero = __ ConI(0); 5021 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done(); 5022 __ set(need_alloc, zero); 5023 __ set(z_alloc, z); 5024 __ if_then(z, BoolTest::eq, null()); { 5025 __ increment (need_alloc, one); 5026 } __ else_(); { 5027 // Update graphKit memory and control from IdealKit. 5028 sync_kit(ideal); 5029 Node *cast = new CastPPNode(z, TypePtr::NOTNULL); 5030 cast->init_req(0, control()); 5031 _gvn.set_type(cast, cast->bottom_type()); 5032 C->record_for_igvn(cast); 5033 5034 Node* zlen_arg = load_array_length(cast); 5035 // Update IdealKit memory and control from graphKit. 5036 __ sync_kit(this); 5037 __ if_then(zlen_arg, BoolTest::lt, zlen); { 5038 __ increment (need_alloc, one); 5039 } __ end_if(); 5040 } __ end_if(); 5041 5042 __ if_then(__ value(need_alloc), BoolTest::ne, zero); { 5043 // Update graphKit memory and control from IdealKit. 5044 sync_kit(ideal); 5045 Node * narr = new_array(klass_node, zlen, 1); 5046 // Update IdealKit memory and control from graphKit. 5047 __ sync_kit(this); 5048 __ set(z_alloc, narr); 5049 } __ end_if(); 5050 5051 sync_kit(ideal); 5052 z = __ value(z_alloc); 5053 // Can't use TypeAryPtr::INTS which uses Bottom offset. 5054 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass)); 5055 // Final sync IdealKit and GraphKit. 5056 final_sync(ideal); 5057 #undef __ 5058 5059 Node* z_start = array_element_address(z, intcon(0), T_INT); 5060 5061 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5062 OptoRuntime::multiplyToLen_Type(), 5063 stubAddr, stubName, TypePtr::BOTTOM, 5064 x_start, xlen, y_start, ylen, z_start, zlen); 5065 } // original reexecute is set back here 5066 5067 C->set_has_split_ifs(true); // Has chance for split-if optimization 5068 set_result(z); 5069 return true; 5070 } 5071 5072 //-------------inline_squareToLen------------------------------------ 5073 bool LibraryCallKit::inline_squareToLen() { 5074 assert(UseSquareToLenIntrinsic, "not implemented on this platform"); 5075 5076 address stubAddr = StubRoutines::squareToLen(); 5077 if (stubAddr == NULL) { 5078 return false; // Intrinsic's stub is not implemented on this platform 5079 } 5080 const char* stubName = "squareToLen"; 5081 5082 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters"); 5083 5084 Node* x = argument(0); 5085 Node* len = argument(1); 5086 Node* z = argument(2); 5087 Node* zlen = argument(3); 5088 5089 x = must_be_not_null(x, true); 5090 z = must_be_not_null(z, true); 5091 5092 const Type* x_type = x->Value(&_gvn); 5093 const Type* z_type = z->Value(&_gvn); 5094 const TypeAryPtr* top_x = x_type->isa_aryptr(); 5095 const TypeAryPtr* top_z = z_type->isa_aryptr(); 5096 if (top_x == NULL || top_x->klass() == NULL || 5097 top_z == NULL || top_z->klass() == NULL) { 5098 // failed array check 5099 return false; 5100 } 5101 5102 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5103 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5104 if (x_elem != T_INT || z_elem != T_INT) { 5105 return false; 5106 } 5107 5108 5109 Node* x_start = array_element_address(x, intcon(0), x_elem); 5110 Node* z_start = array_element_address(z, intcon(0), z_elem); 5111 5112 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5113 OptoRuntime::squareToLen_Type(), 5114 stubAddr, stubName, TypePtr::BOTTOM, 5115 x_start, len, z_start, zlen); 5116 5117 set_result(z); 5118 return true; 5119 } 5120 5121 //-------------inline_mulAdd------------------------------------------ 5122 bool LibraryCallKit::inline_mulAdd() { 5123 assert(UseMulAddIntrinsic, "not implemented on this platform"); 5124 5125 address stubAddr = StubRoutines::mulAdd(); 5126 if (stubAddr == NULL) { 5127 return false; // Intrinsic's stub is not implemented on this platform 5128 } 5129 const char* stubName = "mulAdd"; 5130 5131 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters"); 5132 5133 Node* out = argument(0); 5134 Node* in = argument(1); 5135 Node* offset = argument(2); 5136 Node* len = argument(3); 5137 Node* k = argument(4); 5138 5139 out = must_be_not_null(out, true); 5140 5141 const Type* out_type = out->Value(&_gvn); 5142 const Type* in_type = in->Value(&_gvn); 5143 const TypeAryPtr* top_out = out_type->isa_aryptr(); 5144 const TypeAryPtr* top_in = in_type->isa_aryptr(); 5145 if (top_out == NULL || top_out->klass() == NULL || 5146 top_in == NULL || top_in->klass() == NULL) { 5147 // failed array check 5148 return false; 5149 } 5150 5151 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5152 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5153 if (out_elem != T_INT || in_elem != T_INT) { 5154 return false; 5155 } 5156 5157 Node* outlen = load_array_length(out); 5158 Node* new_offset = _gvn.transform(new SubINode(outlen, offset)); 5159 Node* out_start = array_element_address(out, intcon(0), out_elem); 5160 Node* in_start = array_element_address(in, intcon(0), in_elem); 5161 5162 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 5163 OptoRuntime::mulAdd_Type(), 5164 stubAddr, stubName, TypePtr::BOTTOM, 5165 out_start,in_start, new_offset, len, k); 5166 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5167 set_result(result); 5168 return true; 5169 } 5170 5171 //-------------inline_montgomeryMultiply----------------------------------- 5172 bool LibraryCallKit::inline_montgomeryMultiply() { 5173 address stubAddr = StubRoutines::montgomeryMultiply(); 5174 if (stubAddr == NULL) { 5175 return false; // Intrinsic's stub is not implemented on this platform 5176 } 5177 5178 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform"); 5179 const char* stubName = "montgomery_multiply"; 5180 5181 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters"); 5182 5183 Node* a = argument(0); 5184 Node* b = argument(1); 5185 Node* n = argument(2); 5186 Node* len = argument(3); 5187 Node* inv = argument(4); 5188 Node* m = argument(6); 5189 5190 const Type* a_type = a->Value(&_gvn); 5191 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5192 const Type* b_type = b->Value(&_gvn); 5193 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5194 const Type* n_type = a->Value(&_gvn); 5195 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5196 const Type* m_type = a->Value(&_gvn); 5197 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5198 if (top_a == NULL || top_a->klass() == NULL || 5199 top_b == NULL || top_b->klass() == NULL || 5200 top_n == NULL || top_n->klass() == NULL || 5201 top_m == NULL || top_m->klass() == NULL) { 5202 // failed array check 5203 return false; 5204 } 5205 5206 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5207 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5208 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5209 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5210 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5211 return false; 5212 } 5213 5214 // Make the call 5215 { 5216 Node* a_start = array_element_address(a, intcon(0), a_elem); 5217 Node* b_start = array_element_address(b, intcon(0), b_elem); 5218 Node* n_start = array_element_address(n, intcon(0), n_elem); 5219 Node* m_start = array_element_address(m, intcon(0), m_elem); 5220 5221 Node* call = make_runtime_call(RC_LEAF, 5222 OptoRuntime::montgomeryMultiply_Type(), 5223 stubAddr, stubName, TypePtr::BOTTOM, 5224 a_start, b_start, n_start, len, inv, top(), 5225 m_start); 5226 set_result(m); 5227 } 5228 5229 return true; 5230 } 5231 5232 bool LibraryCallKit::inline_montgomerySquare() { 5233 address stubAddr = StubRoutines::montgomerySquare(); 5234 if (stubAddr == NULL) { 5235 return false; // Intrinsic's stub is not implemented on this platform 5236 } 5237 5238 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform"); 5239 const char* stubName = "montgomery_square"; 5240 5241 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters"); 5242 5243 Node* a = argument(0); 5244 Node* n = argument(1); 5245 Node* len = argument(2); 5246 Node* inv = argument(3); 5247 Node* m = argument(5); 5248 5249 const Type* a_type = a->Value(&_gvn); 5250 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5251 const Type* n_type = a->Value(&_gvn); 5252 const TypeAryPtr* top_n = n_type->isa_aryptr(); 5253 const Type* m_type = a->Value(&_gvn); 5254 const TypeAryPtr* top_m = m_type->isa_aryptr(); 5255 if (top_a == NULL || top_a->klass() == NULL || 5256 top_n == NULL || top_n->klass() == NULL || 5257 top_m == NULL || top_m->klass() == NULL) { 5258 // failed array check 5259 return false; 5260 } 5261 5262 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5263 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5264 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5265 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) { 5266 return false; 5267 } 5268 5269 // Make the call 5270 { 5271 Node* a_start = array_element_address(a, intcon(0), a_elem); 5272 Node* n_start = array_element_address(n, intcon(0), n_elem); 5273 Node* m_start = array_element_address(m, intcon(0), m_elem); 5274 5275 Node* call = make_runtime_call(RC_LEAF, 5276 OptoRuntime::montgomerySquare_Type(), 5277 stubAddr, stubName, TypePtr::BOTTOM, 5278 a_start, n_start, len, inv, top(), 5279 m_start); 5280 set_result(m); 5281 } 5282 5283 return true; 5284 } 5285 5286 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) { 5287 address stubAddr = NULL; 5288 const char* stubName = NULL; 5289 5290 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift(); 5291 if (stubAddr == NULL) { 5292 return false; // Intrinsic's stub is not implemented on this platform 5293 } 5294 5295 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker"; 5296 5297 assert(callee()->signature()->size() == 5, "expected 5 arguments"); 5298 5299 Node* newArr = argument(0); 5300 Node* oldArr = argument(1); 5301 Node* newIdx = argument(2); 5302 Node* shiftCount = argument(3); 5303 Node* numIter = argument(4); 5304 5305 const Type* newArr_type = newArr->Value(&_gvn); 5306 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr(); 5307 const Type* oldArr_type = oldArr->Value(&_gvn); 5308 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr(); 5309 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL 5310 || top_oldArr->klass() == NULL) { 5311 return false; 5312 } 5313 5314 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5315 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5316 if (newArr_elem != T_INT || oldArr_elem != T_INT) { 5317 return false; 5318 } 5319 5320 // Make the call 5321 { 5322 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem); 5323 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem); 5324 5325 Node* call = make_runtime_call(RC_LEAF, 5326 OptoRuntime::bigIntegerShift_Type(), 5327 stubAddr, 5328 stubName, 5329 TypePtr::BOTTOM, 5330 newArr_start, 5331 oldArr_start, 5332 newIdx, 5333 shiftCount, 5334 numIter); 5335 } 5336 5337 return true; 5338 } 5339 5340 //-------------inline_vectorizedMismatch------------------------------ 5341 bool LibraryCallKit::inline_vectorizedMismatch() { 5342 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform"); 5343 5344 address stubAddr = StubRoutines::vectorizedMismatch(); 5345 if (stubAddr == NULL) { 5346 return false; // Intrinsic's stub is not implemented on this platform 5347 } 5348 const char* stubName = "vectorizedMismatch"; 5349 int size_l = callee()->signature()->size(); 5350 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters"); 5351 5352 Node* obja = argument(0); 5353 Node* aoffset = argument(1); 5354 Node* objb = argument(3); 5355 Node* boffset = argument(4); 5356 Node* length = argument(6); 5357 Node* scale = argument(7); 5358 5359 const Type* a_type = obja->Value(&_gvn); 5360 const Type* b_type = objb->Value(&_gvn); 5361 const TypeAryPtr* top_a = a_type->isa_aryptr(); 5362 const TypeAryPtr* top_b = b_type->isa_aryptr(); 5363 if (top_a == NULL || top_a->klass() == NULL || 5364 top_b == NULL || top_b->klass() == NULL) { 5365 // failed array check 5366 return false; 5367 } 5368 5369 Node* call; 5370 jvms()->set_should_reexecute(true); 5371 5372 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ); 5373 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ); 5374 5375 call = make_runtime_call(RC_LEAF, 5376 OptoRuntime::vectorizedMismatch_Type(), 5377 stubAddr, stubName, TypePtr::BOTTOM, 5378 obja_adr, objb_adr, length, scale); 5379 5380 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5381 set_result(result); 5382 return true; 5383 } 5384 5385 /** 5386 * Calculate CRC32 for byte. 5387 * int java.util.zip.CRC32.update(int crc, int b) 5388 */ 5389 bool LibraryCallKit::inline_updateCRC32() { 5390 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5391 assert(callee()->signature()->size() == 2, "update has 2 parameters"); 5392 // no receiver since it is static method 5393 Node* crc = argument(0); // type: int 5394 Node* b = argument(1); // type: int 5395 5396 /* 5397 * int c = ~ crc; 5398 * b = timesXtoThe32[(b ^ c) & 0xFF]; 5399 * b = b ^ (c >>> 8); 5400 * crc = ~b; 5401 */ 5402 5403 Node* M1 = intcon(-1); 5404 crc = _gvn.transform(new XorINode(crc, M1)); 5405 Node* result = _gvn.transform(new XorINode(crc, b)); 5406 result = _gvn.transform(new AndINode(result, intcon(0xFF))); 5407 5408 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr())); 5409 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2))); 5410 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset)); 5411 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered); 5412 5413 crc = _gvn.transform(new URShiftINode(crc, intcon(8))); 5414 result = _gvn.transform(new XorINode(crc, result)); 5415 result = _gvn.transform(new XorINode(result, M1)); 5416 set_result(result); 5417 return true; 5418 } 5419 5420 /** 5421 * Calculate CRC32 for byte[] array. 5422 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len) 5423 */ 5424 bool LibraryCallKit::inline_updateBytesCRC32() { 5425 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5426 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5427 // no receiver since it is static method 5428 Node* crc = argument(0); // type: int 5429 Node* src = argument(1); // type: oop 5430 Node* offset = argument(2); // type: int 5431 Node* length = argument(3); // type: int 5432 5433 const Type* src_type = src->Value(&_gvn); 5434 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5435 if (top_src == NULL || top_src->klass() == NULL) { 5436 // failed array check 5437 return false; 5438 } 5439 5440 // Figure out the size and type of the elements we will be copying. 5441 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5442 if (src_elem != T_BYTE) { 5443 return false; 5444 } 5445 5446 // 'src_start' points to src array + scaled offset 5447 src = must_be_not_null(src, true); 5448 Node* src_start = array_element_address(src, offset, src_elem); 5449 5450 // We assume that range check is done by caller. 5451 // TODO: generate range check (offset+length < src.length) in debug VM. 5452 5453 // Call the stub. 5454 address stubAddr = StubRoutines::updateBytesCRC32(); 5455 const char *stubName = "updateBytesCRC32"; 5456 5457 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5458 stubAddr, stubName, TypePtr::BOTTOM, 5459 crc, src_start, length); 5460 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5461 set_result(result); 5462 return true; 5463 } 5464 5465 /** 5466 * Calculate CRC32 for ByteBuffer. 5467 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 5468 */ 5469 bool LibraryCallKit::inline_updateByteBufferCRC32() { 5470 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support"); 5471 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5472 // no receiver since it is static method 5473 Node* crc = argument(0); // type: int 5474 Node* src = argument(1); // type: long 5475 Node* offset = argument(3); // type: int 5476 Node* length = argument(4); // type: int 5477 5478 src = ConvL2X(src); // adjust Java long to machine word 5479 Node* base = _gvn.transform(new CastX2PNode(src)); 5480 offset = ConvI2X(offset); 5481 5482 // 'src_start' points to src array + scaled offset 5483 Node* src_start = basic_plus_adr(top(), base, offset); 5484 5485 // Call the stub. 5486 address stubAddr = StubRoutines::updateBytesCRC32(); 5487 const char *stubName = "updateBytesCRC32"; 5488 5489 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(), 5490 stubAddr, stubName, TypePtr::BOTTOM, 5491 crc, src_start, length); 5492 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5493 set_result(result); 5494 return true; 5495 } 5496 5497 //------------------------------get_table_from_crc32c_class----------------------- 5498 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) { 5499 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class); 5500 assert (table != NULL, "wrong version of java.util.zip.CRC32C"); 5501 5502 return table; 5503 } 5504 5505 //------------------------------inline_updateBytesCRC32C----------------------- 5506 // 5507 // Calculate CRC32C for byte[] array. 5508 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end) 5509 // 5510 bool LibraryCallKit::inline_updateBytesCRC32C() { 5511 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5512 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5513 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5514 // no receiver since it is a static method 5515 Node* crc = argument(0); // type: int 5516 Node* src = argument(1); // type: oop 5517 Node* offset = argument(2); // type: int 5518 Node* end = argument(3); // type: int 5519 5520 Node* length = _gvn.transform(new SubINode(end, offset)); 5521 5522 const Type* src_type = src->Value(&_gvn); 5523 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5524 if (top_src == NULL || top_src->klass() == NULL) { 5525 // failed array check 5526 return false; 5527 } 5528 5529 // Figure out the size and type of the elements we will be copying. 5530 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5531 if (src_elem != T_BYTE) { 5532 return false; 5533 } 5534 5535 // 'src_start' points to src array + scaled offset 5536 src = must_be_not_null(src, true); 5537 Node* src_start = array_element_address(src, offset, src_elem); 5538 5539 // static final int[] byteTable in class CRC32C 5540 Node* table = get_table_from_crc32c_class(callee()->holder()); 5541 table = must_be_not_null(table, true); 5542 Node* table_start = array_element_address(table, intcon(0), T_INT); 5543 5544 // We assume that range check is done by caller. 5545 // TODO: generate range check (offset+length < src.length) in debug VM. 5546 5547 // Call the stub. 5548 address stubAddr = StubRoutines::updateBytesCRC32C(); 5549 const char *stubName = "updateBytesCRC32C"; 5550 5551 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5552 stubAddr, stubName, TypePtr::BOTTOM, 5553 crc, src_start, length, table_start); 5554 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5555 set_result(result); 5556 return true; 5557 } 5558 5559 //------------------------------inline_updateDirectByteBufferCRC32C----------------------- 5560 // 5561 // Calculate CRC32C for DirectByteBuffer. 5562 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 5563 // 5564 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() { 5565 assert(UseCRC32CIntrinsics, "need CRC32C instruction support"); 5566 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long"); 5567 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded"); 5568 // no receiver since it is a static method 5569 Node* crc = argument(0); // type: int 5570 Node* src = argument(1); // type: long 5571 Node* offset = argument(3); // type: int 5572 Node* end = argument(4); // type: int 5573 5574 Node* length = _gvn.transform(new SubINode(end, offset)); 5575 5576 src = ConvL2X(src); // adjust Java long to machine word 5577 Node* base = _gvn.transform(new CastX2PNode(src)); 5578 offset = ConvI2X(offset); 5579 5580 // 'src_start' points to src array + scaled offset 5581 Node* src_start = basic_plus_adr(top(), base, offset); 5582 5583 // static final int[] byteTable in class CRC32C 5584 Node* table = get_table_from_crc32c_class(callee()->holder()); 5585 table = must_be_not_null(table, true); 5586 Node* table_start = array_element_address(table, intcon(0), T_INT); 5587 5588 // Call the stub. 5589 address stubAddr = StubRoutines::updateBytesCRC32C(); 5590 const char *stubName = "updateBytesCRC32C"; 5591 5592 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(), 5593 stubAddr, stubName, TypePtr::BOTTOM, 5594 crc, src_start, length, table_start); 5595 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5596 set_result(result); 5597 return true; 5598 } 5599 5600 //------------------------------inline_updateBytesAdler32---------------------- 5601 // 5602 // Calculate Adler32 checksum for byte[] array. 5603 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len) 5604 // 5605 bool LibraryCallKit::inline_updateBytesAdler32() { 5606 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 5607 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters"); 5608 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 5609 // no receiver since it is static method 5610 Node* crc = argument(0); // type: int 5611 Node* src = argument(1); // type: oop 5612 Node* offset = argument(2); // type: int 5613 Node* length = argument(3); // type: int 5614 5615 const Type* src_type = src->Value(&_gvn); 5616 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5617 if (top_src == NULL || top_src->klass() == NULL) { 5618 // failed array check 5619 return false; 5620 } 5621 5622 // Figure out the size and type of the elements we will be copying. 5623 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 5624 if (src_elem != T_BYTE) { 5625 return false; 5626 } 5627 5628 // 'src_start' points to src array + scaled offset 5629 Node* src_start = array_element_address(src, offset, src_elem); 5630 5631 // We assume that range check is done by caller. 5632 // TODO: generate range check (offset+length < src.length) in debug VM. 5633 5634 // Call the stub. 5635 address stubAddr = StubRoutines::updateBytesAdler32(); 5636 const char *stubName = "updateBytesAdler32"; 5637 5638 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 5639 stubAddr, stubName, TypePtr::BOTTOM, 5640 crc, src_start, length); 5641 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5642 set_result(result); 5643 return true; 5644 } 5645 5646 //------------------------------inline_updateByteBufferAdler32--------------- 5647 // 5648 // Calculate Adler32 checksum for DirectByteBuffer. 5649 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len) 5650 // 5651 bool LibraryCallKit::inline_updateByteBufferAdler32() { 5652 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one 5653 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long"); 5654 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded"); 5655 // no receiver since it is static method 5656 Node* crc = argument(0); // type: int 5657 Node* src = argument(1); // type: long 5658 Node* offset = argument(3); // type: int 5659 Node* length = argument(4); // type: int 5660 5661 src = ConvL2X(src); // adjust Java long to machine word 5662 Node* base = _gvn.transform(new CastX2PNode(src)); 5663 offset = ConvI2X(offset); 5664 5665 // 'src_start' points to src array + scaled offset 5666 Node* src_start = basic_plus_adr(top(), base, offset); 5667 5668 // Call the stub. 5669 address stubAddr = StubRoutines::updateBytesAdler32(); 5670 const char *stubName = "updateBytesAdler32"; 5671 5672 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(), 5673 stubAddr, stubName, TypePtr::BOTTOM, 5674 crc, src_start, length); 5675 5676 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 5677 set_result(result); 5678 return true; 5679 } 5680 5681 //----------------------------inline_reference_get---------------------------- 5682 // public T java.lang.ref.Reference.get(); 5683 bool LibraryCallKit::inline_reference_get() { 5684 const int referent_offset = java_lang_ref_Reference::referent_offset(); 5685 5686 // Get the argument: 5687 Node* reference_obj = null_check_receiver(); 5688 if (stopped()) return true; 5689 5690 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr(); 5691 assert(tinst != NULL, "obj is null"); 5692 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 5693 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass(); 5694 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"), 5695 ciSymbol::make("Ljava/lang/Object;"), 5696 false); 5697 assert (field != NULL, "undefined field"); 5698 5699 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset); 5700 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 5701 5702 ciInstanceKlass* klass = env()->Object_klass(); 5703 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass); 5704 5705 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF; 5706 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators); 5707 // Add memory barrier to prevent commoning reads from this field 5708 // across safepoint since GC can change its value. 5709 insert_mem_bar(Op_MemBarCPUOrder); 5710 5711 set_result(result); 5712 return true; 5713 } 5714 5715 5716 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 5717 bool is_exact=true, bool is_static=false, 5718 ciInstanceKlass * fromKls=NULL) { 5719 if (fromKls == NULL) { 5720 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 5721 assert(tinst != NULL, "obj is null"); 5722 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 5723 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 5724 fromKls = tinst->klass()->as_instance_klass(); 5725 } else { 5726 assert(is_static, "only for static field access"); 5727 } 5728 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 5729 ciSymbol::make(fieldTypeString), 5730 is_static); 5731 5732 assert (field != NULL, "undefined field"); 5733 if (field == NULL) return (Node *) NULL; 5734 5735 if (is_static) { 5736 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 5737 fromObj = makecon(tip); 5738 } 5739 5740 // Next code copied from Parse::do_get_xxx(): 5741 5742 // Compute address and memory type. 5743 int offset = field->offset_in_bytes(); 5744 bool is_vol = field->is_volatile(); 5745 ciType* field_klass = field->type(); 5746 assert(field_klass->is_loaded(), "should be loaded"); 5747 const TypePtr* adr_type = C->alias_type(field)->adr_type(); 5748 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 5749 BasicType bt = field->layout_type(); 5750 5751 // Build the resultant type of the load 5752 const Type *type; 5753 if (bt == T_OBJECT) { 5754 type = TypeOopPtr::make_from_klass(field_klass->as_klass()); 5755 } else { 5756 type = Type::get_const_basic_type(bt); 5757 } 5758 5759 DecoratorSet decorators = IN_HEAP; 5760 5761 if (is_vol) { 5762 decorators |= MO_SEQ_CST; 5763 } 5764 5765 return access_load_at(fromObj, adr, adr_type, type, bt, decorators); 5766 } 5767 5768 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, 5769 bool is_exact = true, bool is_static = false, 5770 ciInstanceKlass * fromKls = NULL) { 5771 if (fromKls == NULL) { 5772 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr(); 5773 assert(tinst != NULL, "obj is null"); 5774 assert(tinst->klass()->is_loaded(), "obj is not loaded"); 5775 assert(!is_exact || tinst->klass_is_exact(), "klass not exact"); 5776 fromKls = tinst->klass()->as_instance_klass(); 5777 } 5778 else { 5779 assert(is_static, "only for static field access"); 5780 } 5781 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName), 5782 ciSymbol::make(fieldTypeString), 5783 is_static); 5784 5785 assert(field != NULL, "undefined field"); 5786 assert(!field->is_volatile(), "not defined for volatile fields"); 5787 5788 if (is_static) { 5789 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror()); 5790 fromObj = makecon(tip); 5791 } 5792 5793 // Next code copied from Parse::do_get_xxx(): 5794 5795 // Compute address and memory type. 5796 int offset = field->offset_in_bytes(); 5797 Node *adr = basic_plus_adr(fromObj, fromObj, offset); 5798 5799 return adr; 5800 } 5801 5802 //------------------------------inline_aescrypt_Block----------------------- 5803 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { 5804 address stubAddr = NULL; 5805 const char *stubName; 5806 assert(UseAES, "need AES instruction support"); 5807 5808 switch(id) { 5809 case vmIntrinsics::_aescrypt_encryptBlock: 5810 stubAddr = StubRoutines::aescrypt_encryptBlock(); 5811 stubName = "aescrypt_encryptBlock"; 5812 break; 5813 case vmIntrinsics::_aescrypt_decryptBlock: 5814 stubAddr = StubRoutines::aescrypt_decryptBlock(); 5815 stubName = "aescrypt_decryptBlock"; 5816 break; 5817 default: 5818 break; 5819 } 5820 if (stubAddr == NULL) return false; 5821 5822 Node* aescrypt_object = argument(0); 5823 Node* src = argument(1); 5824 Node* src_offset = argument(2); 5825 Node* dest = argument(3); 5826 Node* dest_offset = argument(4); 5827 5828 src = must_be_not_null(src, true); 5829 dest = must_be_not_null(dest, true); 5830 5831 // (1) src and dest are arrays. 5832 const Type* src_type = src->Value(&_gvn); 5833 const Type* dest_type = dest->Value(&_gvn); 5834 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5835 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5836 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 5837 5838 // for the quick and dirty code we will skip all the checks. 5839 // we are just trying to get the call to be generated. 5840 Node* src_start = src; 5841 Node* dest_start = dest; 5842 if (src_offset != NULL || dest_offset != NULL) { 5843 assert(src_offset != NULL && dest_offset != NULL, ""); 5844 src_start = array_element_address(src, src_offset, T_BYTE); 5845 dest_start = array_element_address(dest, dest_offset, T_BYTE); 5846 } 5847 5848 // now need to get the start of its expanded key array 5849 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 5850 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 5851 if (k_start == NULL) return false; 5852 5853 if (Matcher::pass_original_key_for_aes()) { 5854 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 5855 // compatibility issues between Java key expansion and SPARC crypto instructions 5856 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 5857 if (original_k_start == NULL) return false; 5858 5859 // Call the stub. 5860 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 5861 stubAddr, stubName, TypePtr::BOTTOM, 5862 src_start, dest_start, k_start, original_k_start); 5863 } else { 5864 // Call the stub. 5865 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), 5866 stubAddr, stubName, TypePtr::BOTTOM, 5867 src_start, dest_start, k_start); 5868 } 5869 5870 return true; 5871 } 5872 5873 //------------------------------inline_cipherBlockChaining_AESCrypt----------------------- 5874 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { 5875 address stubAddr = NULL; 5876 const char *stubName = NULL; 5877 5878 assert(UseAES, "need AES instruction support"); 5879 5880 switch(id) { 5881 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt: 5882 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt(); 5883 stubName = "cipherBlockChaining_encryptAESCrypt"; 5884 break; 5885 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt: 5886 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt(); 5887 stubName = "cipherBlockChaining_decryptAESCrypt"; 5888 break; 5889 default: 5890 break; 5891 } 5892 if (stubAddr == NULL) return false; 5893 5894 Node* cipherBlockChaining_object = argument(0); 5895 Node* src = argument(1); 5896 Node* src_offset = argument(2); 5897 Node* len = argument(3); 5898 Node* dest = argument(4); 5899 Node* dest_offset = argument(5); 5900 5901 src = must_be_not_null(src, false); 5902 dest = must_be_not_null(dest, false); 5903 5904 // (1) src and dest are arrays. 5905 const Type* src_type = src->Value(&_gvn); 5906 const Type* dest_type = dest->Value(&_gvn); 5907 const TypeAryPtr* top_src = src_type->isa_aryptr(); 5908 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 5909 assert (top_src != NULL && top_src->klass() != NULL 5910 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 5911 5912 // checks are the responsibility of the caller 5913 Node* src_start = src; 5914 Node* dest_start = dest; 5915 if (src_offset != NULL || dest_offset != NULL) { 5916 assert(src_offset != NULL && dest_offset != NULL, ""); 5917 src_start = array_element_address(src, src_offset, T_BYTE); 5918 dest_start = array_element_address(dest, dest_offset, T_BYTE); 5919 } 5920 5921 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 5922 // (because of the predicated logic executed earlier). 5923 // so we cast it here safely. 5924 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 5925 5926 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 5927 if (embeddedCipherObj == NULL) return false; 5928 5929 // cast it to what we know it will be at runtime 5930 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr(); 5931 assert(tinst != NULL, "CBC obj is null"); 5932 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded"); 5933 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 5934 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 5935 5936 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 5937 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 5938 const TypeOopPtr* xtype = aklass->as_instance_type(); 5939 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 5940 aescrypt_object = _gvn.transform(aescrypt_object); 5941 5942 // we need to get the start of the aescrypt_object's expanded key array 5943 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 5944 if (k_start == NULL) return false; 5945 5946 // similarly, get the start address of the r vector 5947 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false); 5948 if (objRvec == NULL) return false; 5949 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); 5950 5951 Node* cbcCrypt; 5952 if (Matcher::pass_original_key_for_aes()) { 5953 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to 5954 // compatibility issues between Java key expansion and SPARC crypto instructions 5955 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); 5956 if (original_k_start == NULL) return false; 5957 5958 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start 5959 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 5960 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 5961 stubAddr, stubName, TypePtr::BOTTOM, 5962 src_start, dest_start, k_start, r_start, len, original_k_start); 5963 } else { 5964 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 5965 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 5966 OptoRuntime::cipherBlockChaining_aescrypt_Type(), 5967 stubAddr, stubName, TypePtr::BOTTOM, 5968 src_start, dest_start, k_start, r_start, len); 5969 } 5970 5971 // return cipher length (int) 5972 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms)); 5973 set_result(retvalue); 5974 return true; 5975 } 5976 5977 //------------------------------inline_electronicCodeBook_AESCrypt----------------------- 5978 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) { 5979 address stubAddr = NULL; 5980 const char *stubName = NULL; 5981 5982 assert(UseAES, "need AES instruction support"); 5983 5984 switch (id) { 5985 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt: 5986 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt(); 5987 stubName = "electronicCodeBook_encryptAESCrypt"; 5988 break; 5989 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt: 5990 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt(); 5991 stubName = "electronicCodeBook_decryptAESCrypt"; 5992 break; 5993 default: 5994 break; 5995 } 5996 5997 if (stubAddr == NULL) return false; 5998 5999 Node* electronicCodeBook_object = argument(0); 6000 Node* src = argument(1); 6001 Node* src_offset = argument(2); 6002 Node* len = argument(3); 6003 Node* dest = argument(4); 6004 Node* dest_offset = argument(5); 6005 6006 // (1) src and dest are arrays. 6007 const Type* src_type = src->Value(&_gvn); 6008 const Type* dest_type = dest->Value(&_gvn); 6009 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6010 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6011 assert(top_src != NULL && top_src->klass() != NULL 6012 && top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6013 6014 // checks are the responsibility of the caller 6015 Node* src_start = src; 6016 Node* dest_start = dest; 6017 if (src_offset != NULL || dest_offset != NULL) { 6018 assert(src_offset != NULL && dest_offset != NULL, ""); 6019 src_start = array_element_address(src, src_offset, T_BYTE); 6020 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6021 } 6022 6023 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6024 // (because of the predicated logic executed earlier). 6025 // so we cast it here safely. 6026 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6027 6028 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6029 if (embeddedCipherObj == NULL) return false; 6030 6031 // cast it to what we know it will be at runtime 6032 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr(); 6033 assert(tinst != NULL, "ECB obj is null"); 6034 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded"); 6035 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6036 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6037 6038 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6039 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6040 const TypeOopPtr* xtype = aklass->as_instance_type(); 6041 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6042 aescrypt_object = _gvn.transform(aescrypt_object); 6043 6044 // we need to get the start of the aescrypt_object's expanded key array 6045 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6046 if (k_start == NULL) return false; 6047 6048 Node* ecbCrypt; 6049 if (Matcher::pass_original_key_for_aes()) { 6050 // no SPARC version for AES/ECB intrinsics now. 6051 return false; 6052 } 6053 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6054 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP, 6055 OptoRuntime::electronicCodeBook_aescrypt_Type(), 6056 stubAddr, stubName, TypePtr::BOTTOM, 6057 src_start, dest_start, k_start, len); 6058 6059 // return cipher length (int) 6060 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms)); 6061 set_result(retvalue); 6062 return true; 6063 } 6064 6065 //------------------------------inline_counterMode_AESCrypt----------------------- 6066 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) { 6067 assert(UseAES, "need AES instruction support"); 6068 if (!UseAESCTRIntrinsics) return false; 6069 6070 address stubAddr = NULL; 6071 const char *stubName = NULL; 6072 if (id == vmIntrinsics::_counterMode_AESCrypt) { 6073 stubAddr = StubRoutines::counterMode_AESCrypt(); 6074 stubName = "counterMode_AESCrypt"; 6075 } 6076 if (stubAddr == NULL) return false; 6077 6078 Node* counterMode_object = argument(0); 6079 Node* src = argument(1); 6080 Node* src_offset = argument(2); 6081 Node* len = argument(3); 6082 Node* dest = argument(4); 6083 Node* dest_offset = argument(5); 6084 6085 // (1) src and dest are arrays. 6086 const Type* src_type = src->Value(&_gvn); 6087 const Type* dest_type = dest->Value(&_gvn); 6088 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6089 const TypeAryPtr* top_dest = dest_type->isa_aryptr(); 6090 assert(top_src != NULL && top_src->klass() != NULL && 6091 top_dest != NULL && top_dest->klass() != NULL, "args are strange"); 6092 6093 // checks are the responsibility of the caller 6094 Node* src_start = src; 6095 Node* dest_start = dest; 6096 if (src_offset != NULL || dest_offset != NULL) { 6097 assert(src_offset != NULL && dest_offset != NULL, ""); 6098 src_start = array_element_address(src, src_offset, T_BYTE); 6099 dest_start = array_element_address(dest, dest_offset, T_BYTE); 6100 } 6101 6102 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object 6103 // (because of the predicated logic executed earlier). 6104 // so we cast it here safely. 6105 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java 6106 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6107 if (embeddedCipherObj == NULL) return false; 6108 // cast it to what we know it will be at runtime 6109 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr(); 6110 assert(tinst != NULL, "CTR obj is null"); 6111 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded"); 6112 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6113 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded"); 6114 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6115 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt); 6116 const TypeOopPtr* xtype = aklass->as_instance_type(); 6117 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype); 6118 aescrypt_object = _gvn.transform(aescrypt_object); 6119 // we need to get the start of the aescrypt_object's expanded key array 6120 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); 6121 if (k_start == NULL) return false; 6122 // similarly, get the start address of the r vector 6123 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false); 6124 if (obj_counter == NULL) return false; 6125 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE); 6126 6127 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false); 6128 if (saved_encCounter == NULL) return false; 6129 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE); 6130 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false); 6131 6132 Node* ctrCrypt; 6133 if (Matcher::pass_original_key_for_aes()) { 6134 // no SPARC version for AES/CTR intrinsics now. 6135 return false; 6136 } 6137 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len 6138 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, 6139 OptoRuntime::counterMode_aescrypt_Type(), 6140 stubAddr, stubName, TypePtr::BOTTOM, 6141 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used); 6142 6143 // return cipher length (int) 6144 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms)); 6145 set_result(retvalue); 6146 return true; 6147 } 6148 6149 //------------------------------get_key_start_from_aescrypt_object----------------------- 6150 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) { 6151 #if defined(PPC64) || defined(S390) 6152 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys. 6153 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns. 6154 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption. 6155 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]). 6156 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false); 6157 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6158 if (objSessionK == NULL) { 6159 return (Node *) NULL; 6160 } 6161 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS); 6162 #else 6163 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false); 6164 #endif // PPC64 6165 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6166 if (objAESCryptKey == NULL) return (Node *) NULL; 6167 6168 // now have the array, need to get the start address of the K array 6169 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT); 6170 return k_start; 6171 } 6172 6173 //------------------------------get_original_key_start_from_aescrypt_object----------------------- 6174 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) { 6175 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false); 6176 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); 6177 if (objAESCryptKey == NULL) return (Node *) NULL; 6178 6179 // now have the array, need to get the start address of the lastKey array 6180 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE); 6181 return original_k_start; 6182 } 6183 6184 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- 6185 // Return node representing slow path of predicate check. 6186 // the pseudo code we want to emulate with this predicate is: 6187 // for encryption: 6188 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6189 // for decryption: 6190 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6191 // note cipher==plain is more conservative than the original java code but that's OK 6192 // 6193 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) { 6194 // The receiver was checked for NULL already. 6195 Node* objCBC = argument(0); 6196 6197 Node* src = argument(1); 6198 Node* dest = argument(4); 6199 6200 // Load embeddedCipher field of CipherBlockChaining object. 6201 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6202 6203 // get AESCrypt klass for instanceOf check 6204 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6205 // will have same classloader as CipherBlockChaining object 6206 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr(); 6207 assert(tinst != NULL, "CBCobj is null"); 6208 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded"); 6209 6210 // we want to do an instanceof comparison against the AESCrypt class 6211 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6212 if (!klass_AESCrypt->is_loaded()) { 6213 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6214 Node* ctrl = control(); 6215 set_control(top()); // no regular fast path 6216 return ctrl; 6217 } 6218 6219 src = must_be_not_null(src, true); 6220 dest = must_be_not_null(dest, true); 6221 6222 // Resolve oops to stable for CmpP below. 6223 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6224 6225 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6226 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6227 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6228 6229 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6230 6231 // for encryption, we are done 6232 if (!decrypting) 6233 return instof_false; // even if it is NULL 6234 6235 // for decryption, we need to add a further check to avoid 6236 // taking the intrinsic path when cipher and plain are the same 6237 // see the original java code for why. 6238 RegionNode* region = new RegionNode(3); 6239 region->init_req(1, instof_false); 6240 6241 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6242 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6243 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6244 region->init_req(2, src_dest_conjoint); 6245 6246 record_for_igvn(region); 6247 return _gvn.transform(region); 6248 } 6249 6250 //----------------------------inline_electronicCodeBook_AESCrypt_predicate---------------------------- 6251 // Return node representing slow path of predicate check. 6252 // the pseudo code we want to emulate with this predicate is: 6253 // for encryption: 6254 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6255 // for decryption: 6256 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6257 // note cipher==plain is more conservative than the original java code but that's OK 6258 // 6259 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) { 6260 // The receiver was checked for NULL already. 6261 Node* objECB = argument(0); 6262 6263 // Load embeddedCipher field of ElectronicCodeBook object. 6264 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6265 6266 // get AESCrypt klass for instanceOf check 6267 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6268 // will have same classloader as ElectronicCodeBook object 6269 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr(); 6270 assert(tinst != NULL, "ECBobj is null"); 6271 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded"); 6272 6273 // we want to do an instanceof comparison against the AESCrypt class 6274 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6275 if (!klass_AESCrypt->is_loaded()) { 6276 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6277 Node* ctrl = control(); 6278 set_control(top()); // no regular fast path 6279 return ctrl; 6280 } 6281 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6282 6283 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6284 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6285 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6286 6287 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6288 6289 // for encryption, we are done 6290 if (!decrypting) 6291 return instof_false; // even if it is NULL 6292 6293 // for decryption, we need to add a further check to avoid 6294 // taking the intrinsic path when cipher and plain are the same 6295 // see the original java code for why. 6296 RegionNode* region = new RegionNode(3); 6297 region->init_req(1, instof_false); 6298 Node* src = argument(1); 6299 Node* dest = argument(4); 6300 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest)); 6301 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq)); 6302 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN); 6303 region->init_req(2, src_dest_conjoint); 6304 6305 record_for_igvn(region); 6306 return _gvn.transform(region); 6307 } 6308 6309 //----------------------------inline_counterMode_AESCrypt_predicate---------------------------- 6310 // Return node representing slow path of predicate check. 6311 // the pseudo code we want to emulate with this predicate is: 6312 // for encryption: 6313 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath 6314 // for decryption: 6315 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath 6316 // note cipher==plain is more conservative than the original java code but that's OK 6317 // 6318 6319 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() { 6320 // The receiver was checked for NULL already. 6321 Node* objCTR = argument(0); 6322 6323 // Load embeddedCipher field of CipherBlockChaining object. 6324 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false); 6325 6326 // get AESCrypt klass for instanceOf check 6327 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point 6328 // will have same classloader as CipherBlockChaining object 6329 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr(); 6330 assert(tinst != NULL, "CTRobj is null"); 6331 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded"); 6332 6333 // we want to do an instanceof comparison against the AESCrypt class 6334 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt")); 6335 if (!klass_AESCrypt->is_loaded()) { 6336 // if AESCrypt is not even loaded, we never take the intrinsic fast path 6337 Node* ctrl = control(); 6338 set_control(top()); // no regular fast path 6339 return ctrl; 6340 } 6341 6342 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass(); 6343 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt))); 6344 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6345 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6346 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6347 6348 return instof_false; // even if it is NULL 6349 } 6350 6351 //------------------------------inline_ghash_processBlocks 6352 bool LibraryCallKit::inline_ghash_processBlocks() { 6353 address stubAddr; 6354 const char *stubName; 6355 assert(UseGHASHIntrinsics, "need GHASH intrinsics support"); 6356 6357 stubAddr = StubRoutines::ghash_processBlocks(); 6358 stubName = "ghash_processBlocks"; 6359 6360 Node* data = argument(0); 6361 Node* offset = argument(1); 6362 Node* len = argument(2); 6363 Node* state = argument(3); 6364 Node* subkeyH = argument(4); 6365 6366 state = must_be_not_null(state, true); 6367 subkeyH = must_be_not_null(subkeyH, true); 6368 data = must_be_not_null(data, true); 6369 6370 Node* state_start = array_element_address(state, intcon(0), T_LONG); 6371 assert(state_start, "state is NULL"); 6372 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG); 6373 assert(subkeyH_start, "subkeyH is NULL"); 6374 Node* data_start = array_element_address(data, offset, T_BYTE); 6375 assert(data_start, "data is NULL"); 6376 6377 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP, 6378 OptoRuntime::ghash_processBlocks_Type(), 6379 stubAddr, stubName, TypePtr::BOTTOM, 6380 state_start, subkeyH_start, data_start, len); 6381 return true; 6382 } 6383 6384 bool LibraryCallKit::inline_base64_encodeBlock() { 6385 address stubAddr; 6386 const char *stubName; 6387 assert(UseBASE64Intrinsics, "need Base64 intrinsics support"); 6388 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters"); 6389 stubAddr = StubRoutines::base64_encodeBlock(); 6390 stubName = "encodeBlock"; 6391 6392 if (!stubAddr) return false; 6393 Node* base64obj = argument(0); 6394 Node* src = argument(1); 6395 Node* offset = argument(2); 6396 Node* len = argument(3); 6397 Node* dest = argument(4); 6398 Node* dp = argument(5); 6399 Node* isURL = argument(6); 6400 6401 src = must_be_not_null(src, true); 6402 dest = must_be_not_null(dest, true); 6403 6404 Node* src_start = array_element_address(src, intcon(0), T_BYTE); 6405 assert(src_start, "source array is NULL"); 6406 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE); 6407 assert(dest_start, "destination array is NULL"); 6408 6409 Node* base64 = make_runtime_call(RC_LEAF, 6410 OptoRuntime::base64_encodeBlock_Type(), 6411 stubAddr, stubName, TypePtr::BOTTOM, 6412 src_start, offset, len, dest_start, dp, isURL); 6413 return true; 6414 } 6415 6416 //------------------------------inline_digestBase_implCompress----------------------- 6417 // 6418 // Calculate MD5 for single-block byte[] array. 6419 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs) 6420 // 6421 // Calculate SHA (i.e., SHA-1) for single-block byte[] array. 6422 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs) 6423 // 6424 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array. 6425 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs) 6426 // 6427 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array. 6428 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs) 6429 // 6430 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) { 6431 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters"); 6432 6433 Node* digestBase_obj = argument(0); 6434 Node* src = argument(1); // type oop 6435 Node* ofs = argument(2); // type int 6436 6437 const Type* src_type = src->Value(&_gvn); 6438 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6439 if (top_src == NULL || top_src->klass() == NULL) { 6440 // failed array check 6441 return false; 6442 } 6443 // Figure out the size and type of the elements we will be copying. 6444 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6445 if (src_elem != T_BYTE) { 6446 return false; 6447 } 6448 // 'src_start' points to src array + offset 6449 src = must_be_not_null(src, true); 6450 Node* src_start = array_element_address(src, ofs, src_elem); 6451 Node* state = NULL; 6452 address stubAddr; 6453 const char *stubName; 6454 6455 switch(id) { 6456 case vmIntrinsics::_md5_implCompress: 6457 assert(UseMD5Intrinsics, "need MD5 instruction support"); 6458 state = get_state_from_digest_object(digestBase_obj); 6459 stubAddr = StubRoutines::md5_implCompress(); 6460 stubName = "md5_implCompress"; 6461 break; 6462 case vmIntrinsics::_sha_implCompress: 6463 assert(UseSHA1Intrinsics, "need SHA1 instruction support"); 6464 state = get_state_from_digest_object(digestBase_obj); 6465 stubAddr = StubRoutines::sha1_implCompress(); 6466 stubName = "sha1_implCompress"; 6467 break; 6468 case vmIntrinsics::_sha2_implCompress: 6469 assert(UseSHA256Intrinsics, "need SHA256 instruction support"); 6470 state = get_state_from_digest_object(digestBase_obj); 6471 stubAddr = StubRoutines::sha256_implCompress(); 6472 stubName = "sha256_implCompress"; 6473 break; 6474 case vmIntrinsics::_sha5_implCompress: 6475 assert(UseSHA512Intrinsics, "need SHA512 instruction support"); 6476 state = get_long_state_from_digest_object(digestBase_obj); 6477 stubAddr = StubRoutines::sha512_implCompress(); 6478 stubName = "sha512_implCompress"; 6479 break; 6480 default: 6481 fatal_unexpected_iid(id); 6482 return false; 6483 } 6484 if (state == NULL) return false; 6485 6486 assert(stubAddr != NULL, "Stub is generated"); 6487 if (stubAddr == NULL) return false; 6488 6489 // Call the stub. 6490 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(), 6491 stubAddr, stubName, TypePtr::BOTTOM, 6492 src_start, state); 6493 6494 return true; 6495 } 6496 6497 //------------------------------inline_digestBase_implCompressMB----------------------- 6498 // 6499 // Calculate MD5/SHA/SHA2/SHA5 for multi-block byte[] array. 6500 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit) 6501 // 6502 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) { 6503 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 6504 "need MD5/SHA1/SHA256/SHA512 instruction support"); 6505 assert((uint)predicate < 4, "sanity"); 6506 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters"); 6507 6508 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already. 6509 Node* src = argument(1); // byte[] array 6510 Node* ofs = argument(2); // type int 6511 Node* limit = argument(3); // type int 6512 6513 const Type* src_type = src->Value(&_gvn); 6514 const TypeAryPtr* top_src = src_type->isa_aryptr(); 6515 if (top_src == NULL || top_src->klass() == NULL) { 6516 // failed array check 6517 return false; 6518 } 6519 // Figure out the size and type of the elements we will be copying. 6520 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type(); 6521 if (src_elem != T_BYTE) { 6522 return false; 6523 } 6524 // 'src_start' points to src array + offset 6525 src = must_be_not_null(src, false); 6526 Node* src_start = array_element_address(src, ofs, src_elem); 6527 6528 const char* klass_digestBase_name = NULL; 6529 const char* stub_name = NULL; 6530 address stub_addr = NULL; 6531 bool long_state = false; 6532 6533 switch (predicate) { 6534 case 0: 6535 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) { 6536 klass_digestBase_name = "sun/security/provider/MD5"; 6537 stub_name = "md5_implCompressMB"; 6538 stub_addr = StubRoutines::md5_implCompressMB(); 6539 } 6540 break; 6541 case 1: 6542 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) { 6543 klass_digestBase_name = "sun/security/provider/SHA"; 6544 stub_name = "sha1_implCompressMB"; 6545 stub_addr = StubRoutines::sha1_implCompressMB(); 6546 } 6547 break; 6548 case 2: 6549 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) { 6550 klass_digestBase_name = "sun/security/provider/SHA2"; 6551 stub_name = "sha256_implCompressMB"; 6552 stub_addr = StubRoutines::sha256_implCompressMB(); 6553 } 6554 break; 6555 case 3: 6556 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) { 6557 klass_digestBase_name = "sun/security/provider/SHA5"; 6558 stub_name = "sha512_implCompressMB"; 6559 stub_addr = StubRoutines::sha512_implCompressMB(); 6560 long_state = true; 6561 } 6562 break; 6563 default: 6564 fatal("unknown DigestBase intrinsic predicate: %d", predicate); 6565 } 6566 if (klass_digestBase_name != NULL) { 6567 assert(stub_addr != NULL, "Stub is generated"); 6568 if (stub_addr == NULL) return false; 6569 6570 // get DigestBase klass to lookup for SHA klass 6571 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr(); 6572 assert(tinst != NULL, "digestBase_obj is not instance???"); 6573 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 6574 6575 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name)); 6576 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded"); 6577 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass(); 6578 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, long_state, stub_addr, stub_name, src_start, ofs, limit); 6579 } 6580 return false; 6581 } 6582 6583 //------------------------------inline_digestBase_implCompressMB----------------------- 6584 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase, 6585 bool long_state, address stubAddr, const char *stubName, 6586 Node* src_start, Node* ofs, Node* limit) { 6587 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase); 6588 const TypeOopPtr* xtype = aklass->as_instance_type(); 6589 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype); 6590 digest_obj = _gvn.transform(digest_obj); 6591 6592 Node* state; 6593 if (long_state) { 6594 state = get_long_state_from_digest_object(digest_obj); 6595 } else { 6596 state = get_state_from_digest_object(digest_obj); 6597 } 6598 if (state == NULL) return false; 6599 6600 // Call the stub. 6601 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, 6602 OptoRuntime::digestBase_implCompressMB_Type(), 6603 stubAddr, stubName, TypePtr::BOTTOM, 6604 src_start, state, ofs, limit); 6605 // return ofs (int) 6606 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms)); 6607 set_result(result); 6608 6609 return true; 6610 } 6611 6612 //------------------------------get_state_from_digest_object----------------------- 6613 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object) { 6614 Node* digest_state = load_field_from_object(digest_object, "state", "[I", /*is_exact*/ false); 6615 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2"); 6616 if (digest_state == NULL) return (Node *) NULL; 6617 6618 // now have the array, need to get the start address of the state array 6619 Node* state = array_element_address(digest_state, intcon(0), T_INT); 6620 return state; 6621 } 6622 6623 //------------------------------get_long_state_from_digest_object----------------------- 6624 Node * LibraryCallKit::get_long_state_from_digest_object(Node *digest_object) { 6625 Node* digest_state = load_field_from_object(digest_object, "state", "[J", /*is_exact*/ false); 6626 assert (digest_state != NULL, "wrong version of sun.security.provider.SHA5"); 6627 if (digest_state == NULL) return (Node *) NULL; 6628 6629 // now have the array, need to get the start address of the state array 6630 Node* state = array_element_address(digest_state, intcon(0), T_LONG); 6631 return state; 6632 } 6633 6634 //----------------------------inline_digestBase_implCompressMB_predicate---------------------------- 6635 // Return node representing slow path of predicate check. 6636 // the pseudo code we want to emulate with this predicate is: 6637 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5) do_intrinsic, else do_javapath 6638 // 6639 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) { 6640 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics, 6641 "need MD5/SHA1/SHA256/SHA512 instruction support"); 6642 assert((uint)predicate < 4, "sanity"); 6643 6644 // The receiver was checked for NULL already. 6645 Node* digestBaseObj = argument(0); 6646 6647 // get DigestBase klass for instanceOf check 6648 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr(); 6649 assert(tinst != NULL, "digestBaseObj is null"); 6650 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded"); 6651 6652 const char* klass_name = NULL; 6653 switch (predicate) { 6654 case 0: 6655 if (UseMD5Intrinsics) { 6656 // we want to do an instanceof comparison against the MD5 class 6657 klass_name = "sun/security/provider/MD5"; 6658 } 6659 break; 6660 case 1: 6661 if (UseSHA1Intrinsics) { 6662 // we want to do an instanceof comparison against the SHA class 6663 klass_name = "sun/security/provider/SHA"; 6664 } 6665 break; 6666 case 2: 6667 if (UseSHA256Intrinsics) { 6668 // we want to do an instanceof comparison against the SHA2 class 6669 klass_name = "sun/security/provider/SHA2"; 6670 } 6671 break; 6672 case 3: 6673 if (UseSHA512Intrinsics) { 6674 // we want to do an instanceof comparison against the SHA5 class 6675 klass_name = "sun/security/provider/SHA5"; 6676 } 6677 break; 6678 default: 6679 fatal("unknown SHA intrinsic predicate: %d", predicate); 6680 } 6681 6682 ciKlass* klass = NULL; 6683 if (klass_name != NULL) { 6684 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name)); 6685 } 6686 if ((klass == NULL) || !klass->is_loaded()) { 6687 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path 6688 Node* ctrl = control(); 6689 set_control(top()); // no intrinsic path 6690 return ctrl; 6691 } 6692 ciInstanceKlass* instklass = klass->as_instance_klass(); 6693 6694 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass))); 6695 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1))); 6696 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne)); 6697 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN); 6698 6699 return instof_false; // even if it is NULL 6700 } 6701 6702 //-------------inline_fma----------------------------------- 6703 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) { 6704 Node *a = NULL; 6705 Node *b = NULL; 6706 Node *c = NULL; 6707 Node* result = NULL; 6708 switch (id) { 6709 case vmIntrinsics::_fmaD: 6710 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each."); 6711 // no receiver since it is static method 6712 a = round_double_node(argument(0)); 6713 b = round_double_node(argument(2)); 6714 c = round_double_node(argument(4)); 6715 result = _gvn.transform(new FmaDNode(control(), a, b, c)); 6716 break; 6717 case vmIntrinsics::_fmaF: 6718 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each."); 6719 a = argument(0); 6720 b = argument(1); 6721 c = argument(2); 6722 result = _gvn.transform(new FmaFNode(control(), a, b, c)); 6723 break; 6724 default: 6725 fatal_unexpected_iid(id); break; 6726 } 6727 set_result(result); 6728 return true; 6729 } 6730 6731 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) { 6732 // argument(0) is receiver 6733 Node* codePoint = argument(1); 6734 Node* n = NULL; 6735 6736 switch (id) { 6737 case vmIntrinsics::_isDigit : 6738 n = new DigitNode(control(), codePoint); 6739 break; 6740 case vmIntrinsics::_isLowerCase : 6741 n = new LowerCaseNode(control(), codePoint); 6742 break; 6743 case vmIntrinsics::_isUpperCase : 6744 n = new UpperCaseNode(control(), codePoint); 6745 break; 6746 case vmIntrinsics::_isWhitespace : 6747 n = new WhitespaceNode(control(), codePoint); 6748 break; 6749 default: 6750 fatal_unexpected_iid(id); 6751 } 6752 6753 set_result(_gvn.transform(n)); 6754 return true; 6755 } 6756 6757 //------------------------------inline_fp_min_max------------------------------ 6758 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) { 6759 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416. 6760 6761 // The intrinsic should be used only when the API branches aren't predictable, 6762 // the last one performing the most important comparison. The following heuristic 6763 // uses the branch statistics to eventually bail out if necessary. 6764 6765 ciMethodData *md = callee()->method_data(); 6766 6767 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) { 6768 ciCallProfile cp = caller()->call_profile_at_bci(bci()); 6769 6770 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) { 6771 // Bail out if the call-site didn't contribute enough to the statistics. 6772 return false; 6773 } 6774 6775 uint taken = 0, not_taken = 0; 6776 6777 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) { 6778 if (p->is_BranchData()) { 6779 taken = ((ciBranchData*)p)->taken(); 6780 not_taken = ((ciBranchData*)p)->not_taken(); 6781 } 6782 } 6783 6784 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count()); 6785 balance = balance < 0 ? -balance : balance; 6786 if ( balance > 0.2 ) { 6787 // Bail out if the most important branch is predictable enough. 6788 return false; 6789 } 6790 } 6791 */ 6792 6793 Node *a = NULL; 6794 Node *b = NULL; 6795 Node *n = NULL; 6796 switch (id) { 6797 case vmIntrinsics::_maxF: 6798 case vmIntrinsics::_minF: 6799 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each."); 6800 a = argument(0); 6801 b = argument(1); 6802 break; 6803 case vmIntrinsics::_maxD: 6804 case vmIntrinsics::_minD: 6805 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each."); 6806 a = round_double_node(argument(0)); 6807 b = round_double_node(argument(2)); 6808 break; 6809 default: 6810 fatal_unexpected_iid(id); 6811 break; 6812 } 6813 switch (id) { 6814 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break; 6815 case vmIntrinsics::_minF: n = new MinFNode(a, b); break; 6816 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break; 6817 case vmIntrinsics::_minD: n = new MinDNode(a, b); break; 6818 default: fatal_unexpected_iid(id); break; 6819 } 6820 set_result(_gvn.transform(n)); 6821 return true; 6822 } 6823 6824 bool LibraryCallKit::inline_profileBoolean() { 6825 Node* counts = argument(1); 6826 const TypeAryPtr* ary = NULL; 6827 ciArray* aobj = NULL; 6828 if (counts->is_Con() 6829 && (ary = counts->bottom_type()->isa_aryptr()) != NULL 6830 && (aobj = ary->const_oop()->as_array()) != NULL 6831 && (aobj->length() == 2)) { 6832 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively. 6833 jint false_cnt = aobj->element_value(0).as_int(); 6834 jint true_cnt = aobj->element_value(1).as_int(); 6835 6836 if (C->log() != NULL) { 6837 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'", 6838 false_cnt, true_cnt); 6839 } 6840 6841 if (false_cnt + true_cnt == 0) { 6842 // According to profile, never executed. 6843 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 6844 Deoptimization::Action_reinterpret); 6845 return true; 6846 } 6847 6848 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt) 6849 // is a number of each value occurrences. 6850 Node* result = argument(0); 6851 if (false_cnt == 0 || true_cnt == 0) { 6852 // According to profile, one value has been never seen. 6853 int expected_val = (false_cnt == 0) ? 1 : 0; 6854 6855 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val))); 6856 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq)); 6857 6858 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN); 6859 Node* fast_path = _gvn.transform(new IfTrueNode(check)); 6860 Node* slow_path = _gvn.transform(new IfFalseNode(check)); 6861 6862 { // Slow path: uncommon trap for never seen value and then reexecute 6863 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows 6864 // the value has been seen at least once. 6865 PreserveJVMState pjvms(this); 6866 PreserveReexecuteState preexecs(this); 6867 jvms()->set_should_reexecute(true); 6868 6869 set_control(slow_path); 6870 set_i_o(i_o()); 6871 6872 uncommon_trap_exact(Deoptimization::Reason_intrinsic, 6873 Deoptimization::Action_reinterpret); 6874 } 6875 // The guard for never seen value enables sharpening of the result and 6876 // returning a constant. It allows to eliminate branches on the same value 6877 // later on. 6878 set_control(fast_path); 6879 result = intcon(expected_val); 6880 } 6881 // Stop profiling. 6882 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode. 6883 // By replacing method body with profile data (represented as ProfileBooleanNode 6884 // on IR level) we effectively disable profiling. 6885 // It enables full speed execution once optimized code is generated. 6886 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt)); 6887 C->record_for_igvn(profile); 6888 set_result(profile); 6889 return true; 6890 } else { 6891 // Continue profiling. 6892 // Profile data isn't available at the moment. So, execute method's bytecode version. 6893 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod 6894 // is compiled and counters aren't available since corresponding MethodHandle 6895 // isn't a compile-time constant. 6896 return false; 6897 } 6898 } 6899 6900 bool LibraryCallKit::inline_isCompileConstant() { 6901 Node* n = argument(0); 6902 set_result(n->is_Con() ? intcon(1) : intcon(0)); 6903 return true; 6904 }