1 /* 2 * Copyright (c) 1998, 2019, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "ci/ciMethodData.hpp" 27 #include "classfile/systemDictionary.hpp" 28 #include "classfile/vmSymbols.hpp" 29 #include "compiler/compileLog.hpp" 30 #include "interpreter/linkResolver.hpp" 31 #include "memory/resourceArea.hpp" 32 #include "memory/universe.hpp" 33 #include "oops/oop.inline.hpp" 34 #include "opto/addnode.hpp" 35 #include "opto/castnode.hpp" 36 #include "opto/convertnode.hpp" 37 #include "opto/divnode.hpp" 38 #include "opto/idealGraphPrinter.hpp" 39 #include "opto/idealKit.hpp" 40 #include "opto/matcher.hpp" 41 #include "opto/memnode.hpp" 42 #include "opto/mulnode.hpp" 43 #include "opto/opaquenode.hpp" 44 #include "opto/parse.hpp" 45 #include "opto/runtime.hpp" 46 #include "opto/valuetypenode.hpp" 47 #include "runtime/deoptimization.hpp" 48 #include "runtime/sharedRuntime.hpp" 49 50 #ifndef PRODUCT 51 extern int explicit_null_checks_inserted, 52 explicit_null_checks_elided; 53 #endif 54 55 //---------------------------------array_load---------------------------------- 56 void Parse::array_load(BasicType bt) { 57 const Type* elemtype = Type::TOP; 58 Node* adr = array_addressing(bt, 0, &elemtype); 59 if (stopped()) return; // guaranteed null or range check 60 61 Node* idx = pop(); 62 Node* ary = pop(); 63 64 // Handle value type arrays 65 const TypeOopPtr* elemptr = elemtype->make_oopptr(); 66 const TypeAryPtr* ary_t = _gvn.type(ary)->is_aryptr(); 67 if (elemtype->isa_valuetype() != NULL) { 68 C->set_flattened_accesses(); 69 // Load from flattened value type array 70 Node* vt = ValueTypeNode::make_from_flattened(this, elemtype->value_klass(), ary, adr); 71 push(vt); 72 return; 73 } else if (elemptr != NULL && elemptr->is_valuetypeptr() && !elemptr->maybe_null()) { 74 // Load from non-flattened but flattenable value type array (elements can never be null) 75 bt = T_VALUETYPE; 76 } else if (!ary_t->is_not_flat()) { 77 // Cannot statically determine if array is flattened, emit runtime check 78 assert(ValueArrayFlatten && elemptr->can_be_value_type() && !ary_t->klass_is_exact() && !ary_t->is_not_null_free() && 79 (!elemptr->is_valuetypeptr() || elemptr->value_klass()->flatten_array()), "array can't be flattened"); 80 Node* ctl = control(); 81 IdealKit ideal(this); 82 IdealVariable res(ideal); 83 ideal.declarations_done(); 84 Node* flattened = gen_flattened_array_test(ary); 85 ideal.if_then(flattened, BoolTest::ne, zerocon(flattened->bottom_type()->basic_type())); { 86 // flattened 87 sync_kit(ideal); 88 if (elemptr->is_valuetypeptr()) { 89 // Element type is known, cast and load from flattened representation 90 ciValueKlass* vk = elemptr->value_klass(); 91 assert(vk->flatten_array() && elemptr->maybe_null(), "must be a flattenable and nullable array"); 92 ciArrayKlass* array_klass = ciArrayKlass::make(vk, /* never_null */ true); 93 const TypeAryPtr* arytype = TypeOopPtr::make_from_klass(array_klass)->isa_aryptr(); 94 Node* cast = _gvn.transform(new CheckCastPPNode(control(), ary, arytype)); 95 Node* casted_adr = array_element_address(cast, idx, T_VALUETYPE, ary_t->size(), control()); 96 Node* vt = ValueTypeNode::make_from_flattened(this, vk, cast, casted_adr)->allocate(this, false, false)->get_oop(); 97 ideal.set(res, vt); 98 ideal.sync_kit(this); 99 } else { 100 Node* kls = load_object_klass(ary); 101 // Element type is unknown, emit runtime call 102 Node* k_adr = basic_plus_adr(kls, in_bytes(ArrayKlass::element_klass_offset())); 103 Node* elem_klass = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), k_adr, TypeInstPtr::KLASS)); 104 Node* obj_size = NULL; 105 kill_dead_locals(); 106 inc_sp(2); 107 Node* alloc_obj = new_instance(elem_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true); 108 dec_sp(2); 109 110 AllocateNode* alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn); 111 assert(alloc->maybe_set_complete(&_gvn), ""); 112 alloc->initialization()->set_complete_with_arraycopy(); 113 114 // This membar keeps this access to an unknown flattened array 115 // correctly ordered with other unknown and known flattened 116 // array accesses. 117 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES)); 118 119 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2(); 120 // Unknown value type might contain reference fields 121 if (!bs->array_copy_requires_gc_barriers(false, T_OBJECT, false, BarrierSetC2::Parsing)) { 122 int base_off = sizeof(instanceOopDesc); 123 Node* dst_base = basic_plus_adr(alloc_obj, base_off); 124 Node* countx = obj_size; 125 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off))); 126 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong))); 127 128 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place"); 129 Node* lhp = basic_plus_adr(kls, in_bytes(Klass::layout_helper_offset())); 130 Node* elem_shift = make_load(NULL, lhp, TypeInt::INT, T_INT, MemNode::unordered); 131 uint header = arrayOopDesc::base_offset_in_bytes(T_VALUETYPE); 132 Node* base = basic_plus_adr(ary, header); 133 idx = Compile::conv_I2X_index(&_gvn, idx, TypeInt::POS, control()); 134 Node* scale = _gvn.transform(new LShiftXNode(idx, elem_shift)); 135 Node* adr = basic_plus_adr(ary, base, scale); 136 137 access_clone(adr, dst_base, countx, false); 138 } else { 139 ideal.sync_kit(this); 140 ideal.make_leaf_call(OptoRuntime::load_unknown_value_Type(), 141 CAST_FROM_FN_PTR(address, OptoRuntime::load_unknown_value), 142 "load_unknown_value", 143 ary, idx, alloc_obj); 144 sync_kit(ideal); 145 } 146 147 // This makes sure no other thread sees a partially initialized buffered value 148 insert_mem_bar_volatile(Op_MemBarStoreStore, Compile::AliasIdxRaw, alloc->proj_out_or_null(AllocateNode::RawAddress)); 149 150 // Same as MemBarCPUOrder above: keep this unknown flattened 151 // array access correctly ordered with other flattened array 152 // access 153 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES)); 154 155 // Prevent any use of the newly allocated value before it is 156 // fully initialized 157 alloc_obj = new CastPPNode(alloc_obj, _gvn.type(alloc_obj), true); 158 alloc_obj->set_req(0, control()); 159 alloc_obj = _gvn.transform(alloc_obj); 160 161 const Type* unknown_value = TypeInstPtr::BOTTOM->cast_to_flat_array(); 162 163 alloc_obj = _gvn.transform(new CheckCastPPNode(control(), alloc_obj, unknown_value)); 164 165 ideal.sync_kit(this); 166 167 ideal.set(res, alloc_obj); 168 } 169 } ideal.else_(); { 170 // non-flattened 171 sync_kit(ideal); 172 const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt); 173 Node* ld = access_load_at(ary, adr, adr_type, elemptr, bt, 174 IN_HEAP | IS_ARRAY | C2_CONTROL_DEPENDENT_LOAD, ctl); 175 ideal.sync_kit(this); 176 ideal.set(res, ld); 177 } ideal.end_if(); 178 sync_kit(ideal); 179 push_node(bt, _gvn.transform(ideal.value(res))); 180 return; 181 } 182 183 if (elemtype == TypeInt::BOOL) { 184 bt = T_BOOLEAN; 185 } else if (bt == T_OBJECT) { 186 elemtype = ary_t->elem()->make_oopptr(); 187 } 188 189 const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt); 190 Node* ld = access_load_at(ary, adr, adr_type, elemtype, bt, 191 IN_HEAP | IS_ARRAY | C2_CONTROL_DEPENDENT_LOAD); 192 if (bt == T_VALUETYPE) { 193 // Loading a non-flattened (but flattenable) value type from an array 194 assert(!gvn().type(ld)->maybe_null(), "value type array elements should never be null"); 195 if (elemptr->value_klass()->is_scalarizable()) { 196 ld = ValueTypeNode::make_from_oop(this, ld, elemptr->value_klass()); 197 } 198 } 199 200 push_node(bt, ld); 201 } 202 203 204 //--------------------------------array_store---------------------------------- 205 void Parse::array_store(BasicType bt) { 206 const Type* elemtype = Type::TOP; 207 Node* adr = array_addressing(bt, type2size[bt], &elemtype); 208 if (stopped()) return; // guaranteed null or range check 209 Node* cast_val = NULL; 210 if (bt == T_OBJECT) { 211 cast_val = array_store_check(); 212 if (stopped()) return; 213 } 214 Node* val = pop_node(bt); // Value to store 215 Node* idx = pop(); // Index in the array 216 Node* ary = pop(); // The array itself 217 218 const TypeAryPtr* ary_t = _gvn.type(ary)->is_aryptr(); 219 const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt); 220 221 if (elemtype == TypeInt::BOOL) { 222 bt = T_BOOLEAN; 223 } else if (bt == T_OBJECT) { 224 elemtype = elemtype->make_oopptr(); 225 const Type* tval = _gvn.type(cast_val); 226 // We may have lost type information for 'val' here due to the casts 227 // emitted by the array_store_check code (see JDK-6312651) 228 // TODO Remove this code once JDK-6312651 is in. 229 const Type* tval_init = _gvn.type(val); 230 bool can_be_value_type = tval->isa_valuetype() || (tval != TypePtr::NULL_PTR && tval_init->is_oopptr()->can_be_value_type() && tval->is_oopptr()->can_be_value_type()); 231 bool not_flattenable = !can_be_value_type || ((tval_init->is_valuetypeptr() || tval_init->isa_valuetype()) && !tval_init->value_klass()->flatten_array()); 232 233 if (!ary_t->is_not_null_free() && !can_be_value_type && (!tval->maybe_null() || !tval_init->maybe_null())) { 234 // Storing a non-inline-type, mark array as not null-free. 235 // This is only legal for non-null stores because the array_store_check passes for null. 236 ary_t = ary_t->cast_to_not_null_free(); 237 Node* cast = _gvn.transform(new CheckCastPPNode(control(), ary, ary_t)); 238 replace_in_map(ary, cast); 239 ary = cast; 240 } else if (!ary_t->is_not_flat() && not_flattenable) { 241 // Storing a non-flattenable value, mark array as not flat. 242 ary_t = ary_t->cast_to_not_flat(); 243 if (tval != TypePtr::NULL_PTR) { 244 // For NULL, this transformation is only valid after the null guard below 245 Node* cast = _gvn.transform(new CheckCastPPNode(control(), ary, ary_t)); 246 replace_in_map(ary, cast); 247 ary = cast; 248 } 249 } 250 251 if (ary_t->elem()->isa_valuetype() != NULL) { 252 // Store to flattened value type array 253 C->set_flattened_accesses(); 254 if (!cast_val->is_ValueType()) { 255 inc_sp(3); 256 cast_val = null_check(cast_val); 257 if (stopped()) return; 258 dec_sp(3); 259 cast_val = ValueTypeNode::make_from_oop(this, cast_val, ary_t->elem()->value_klass()); 260 } 261 cast_val->as_ValueType()->store_flattened(this, ary, adr); 262 return; 263 } else if (elemtype->is_valuetypeptr() && !elemtype->maybe_null()) { 264 // Store to non-flattened but flattenable value type array (elements can never be null) 265 if (!cast_val->is_ValueType() && tval->maybe_null()) { 266 inc_sp(3); 267 cast_val = null_check(cast_val); 268 if (stopped()) return; 269 dec_sp(3); 270 } 271 } else if (!ary_t->is_not_flat()) { 272 // Array might be flattened, emit runtime checks 273 assert(ValueArrayFlatten && !not_flattenable && elemtype->is_oopptr()->can_be_value_type() && 274 !ary_t->klass_is_exact() && !ary_t->is_not_null_free(), "array can't be flattened"); 275 IdealKit ideal(this); 276 Node* flattened = gen_flattened_array_test(ary); 277 ideal.if_then(flattened, BoolTest::ne, zerocon(flattened->bottom_type()->basic_type())); { 278 Node* val = cast_val; 279 // flattened 280 if (!val->is_ValueType() && tval->maybe_null()) { 281 // Add null check 282 sync_kit(ideal); 283 Node* null_ctl = top(); 284 val = null_check_oop(val, &null_ctl); 285 if (null_ctl != top()) { 286 PreserveJVMState pjvms(this); 287 inc_sp(3); 288 set_control(null_ctl); 289 uncommon_trap(Deoptimization::Reason_null_check, Deoptimization::Action_none); 290 dec_sp(3); 291 } 292 ideal.sync_kit(this); 293 } 294 // Try to determine the value klass 295 ciValueKlass* vk = NULL; 296 if (tval->isa_valuetype() || tval->is_valuetypeptr()) { 297 vk = tval->value_klass(); 298 } else if (tval_init->isa_valuetype() || tval_init->is_valuetypeptr()) { 299 vk = tval_init->value_klass(); 300 } else if (elemtype->is_valuetypeptr()) { 301 vk = elemtype->value_klass(); 302 } 303 Node* casted_ary = ary; 304 if (vk != NULL && !stopped()) { 305 // Element type is known, cast and store to flattened representation 306 sync_kit(ideal); 307 assert(vk->flatten_array() && elemtype->maybe_null(), "must be a flattenable and nullable array"); 308 ciArrayKlass* array_klass = ciArrayKlass::make(vk, /* never_null */ true); 309 const TypeAryPtr* arytype = TypeOopPtr::make_from_klass(array_klass)->isa_aryptr(); 310 casted_ary = _gvn.transform(new CheckCastPPNode(control(), casted_ary, arytype)); 311 Node* casted_adr = array_element_address(casted_ary, idx, T_OBJECT, arytype->size(), control()); 312 if (!val->is_ValueType()) { 313 assert(!gvn().type(val)->maybe_null(), "value type array elements should never be null"); 314 val = ValueTypeNode::make_from_oop(this, val, vk); 315 } 316 val->as_ValueType()->store_flattened(this, casted_ary, casted_adr); 317 ideal.sync_kit(this); 318 } else if (!ideal.ctrl()->is_top()) { 319 // Element type is unknown, emit runtime call 320 sync_kit(ideal); 321 322 // This membar keeps this access to an unknown flattened 323 // array correctly ordered with other unknown and known 324 // flattened array accesses. 325 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES)); 326 ideal.sync_kit(this); 327 328 ideal.make_leaf_call(OptoRuntime::store_unknown_value_Type(), 329 CAST_FROM_FN_PTR(address, OptoRuntime::store_unknown_value), 330 "store_unknown_value", 331 val, casted_ary, idx); 332 333 sync_kit(ideal); 334 // Same as MemBarCPUOrder above: keep this unknown 335 // flattened array access correctly ordered with other 336 // flattened array accesses. 337 insert_mem_bar_volatile(Op_MemBarCPUOrder, C->get_alias_index(TypeAryPtr::VALUES)); 338 ideal.sync_kit(this); 339 } 340 } 341 ideal.else_(); 342 { 343 // non-flattened 344 sync_kit(ideal); 345 gen_value_array_null_guard(ary, cast_val, 3); 346 access_store_at(ary, adr, adr_type, cast_val, elemtype, bt, MO_UNORDERED | IN_HEAP | IS_ARRAY, false, false); 347 ideal.sync_kit(this); 348 } 349 ideal.end_if(); 350 sync_kit(ideal); 351 return; 352 } else if (!ary_t->is_not_null_free()) { 353 // Array is not flattened but may be null free 354 assert(elemtype->is_oopptr()->can_be_value_type() && !ary_t->klass_is_exact(), "array can't be null free"); 355 ary = gen_value_array_null_guard(ary, cast_val, 3, true); 356 } 357 } 358 359 access_store_at(ary, adr, adr_type, val, elemtype, bt, MO_UNORDERED | IN_HEAP | IS_ARRAY); 360 } 361 362 363 //------------------------------array_addressing------------------------------- 364 // Pull array and index from the stack. Compute pointer-to-element. 365 Node* Parse::array_addressing(BasicType type, int vals, const Type* *result2) { 366 Node *idx = peek(0+vals); // Get from stack without popping 367 Node *ary = peek(1+vals); // in case of exception 368 369 // Null check the array base, with correct stack contents 370 ary = null_check(ary, T_ARRAY); 371 // Compile-time detect of null-exception? 372 if (stopped()) return top(); 373 374 const TypeAryPtr* arytype = _gvn.type(ary)->is_aryptr(); 375 const TypeInt* sizetype = arytype->size(); 376 const Type* elemtype = arytype->elem(); 377 378 if (UseUniqueSubclasses && result2 != NULL) { 379 const Type* el = elemtype->make_ptr(); 380 if (el && el->isa_instptr()) { 381 const TypeInstPtr* toop = el->is_instptr(); 382 if (toop->klass()->as_instance_klass()->unique_concrete_subklass()) { 383 // If we load from "AbstractClass[]" we must see "ConcreteSubClass". 384 const Type* subklass = Type::get_const_type(toop->klass()); 385 elemtype = subklass->join_speculative(el); 386 } 387 } 388 } 389 390 // Check for big class initializers with all constant offsets 391 // feeding into a known-size array. 392 const TypeInt* idxtype = _gvn.type(idx)->is_int(); 393 // See if the highest idx value is less than the lowest array bound, 394 // and if the idx value cannot be negative: 395 bool need_range_check = true; 396 if (idxtype->_hi < sizetype->_lo && idxtype->_lo >= 0) { 397 need_range_check = false; 398 if (C->log() != NULL) C->log()->elem("observe that='!need_range_check'"); 399 } 400 401 ciKlass * arytype_klass = arytype->klass(); 402 if ((arytype_klass != NULL) && (!arytype_klass->is_loaded())) { 403 // Only fails for some -Xcomp runs 404 // The class is unloaded. We have to run this bytecode in the interpreter. 405 uncommon_trap(Deoptimization::Reason_unloaded, 406 Deoptimization::Action_reinterpret, 407 arytype->klass(), "!loaded array"); 408 return top(); 409 } 410 411 // Do the range check 412 if (GenerateRangeChecks && need_range_check) { 413 Node* tst; 414 if (sizetype->_hi <= 0) { 415 // The greatest array bound is negative, so we can conclude that we're 416 // compiling unreachable code, but the unsigned compare trick used below 417 // only works with non-negative lengths. Instead, hack "tst" to be zero so 418 // the uncommon_trap path will always be taken. 419 tst = _gvn.intcon(0); 420 } else { 421 // Range is constant in array-oop, so we can use the original state of mem 422 Node* len = load_array_length(ary); 423 424 // Test length vs index (standard trick using unsigned compare) 425 Node* chk = _gvn.transform( new CmpUNode(idx, len) ); 426 BoolTest::mask btest = BoolTest::lt; 427 tst = _gvn.transform( new BoolNode(chk, btest) ); 428 } 429 RangeCheckNode* rc = new RangeCheckNode(control(), tst, PROB_MAX, COUNT_UNKNOWN); 430 _gvn.set_type(rc, rc->Value(&_gvn)); 431 if (!tst->is_Con()) { 432 record_for_igvn(rc); 433 } 434 set_control(_gvn.transform(new IfTrueNode(rc))); 435 // Branch to failure if out of bounds 436 { 437 PreserveJVMState pjvms(this); 438 set_control(_gvn.transform(new IfFalseNode(rc))); 439 if (C->allow_range_check_smearing()) { 440 // Do not use builtin_throw, since range checks are sometimes 441 // made more stringent by an optimistic transformation. 442 // This creates "tentative" range checks at this point, 443 // which are not guaranteed to throw exceptions. 444 // See IfNode::Ideal, is_range_check, adjust_check. 445 uncommon_trap(Deoptimization::Reason_range_check, 446 Deoptimization::Action_make_not_entrant, 447 NULL, "range_check"); 448 } else { 449 // If we have already recompiled with the range-check-widening 450 // heroic optimization turned off, then we must really be throwing 451 // range check exceptions. 452 builtin_throw(Deoptimization::Reason_range_check, idx); 453 } 454 } 455 } 456 // Check for always knowing you are throwing a range-check exception 457 if (stopped()) return top(); 458 459 // Speculate on the array not being null-free 460 if (!arytype->is_not_null_free() && arytype->speculative() != NULL && arytype->speculative()->isa_aryptr() != NULL && 461 arytype->speculative()->is_aryptr()->is_not_null_free() && 462 !too_many_traps_or_recompiles(Deoptimization::Reason_speculate_class_check)) { 463 Node* tst = gen_null_free_array_check(ary); 464 { 465 BuildCutout unless(this, tst, PROB_ALWAYS); 466 uncommon_trap(Deoptimization::Reason_speculate_class_check, 467 Deoptimization::Action_maybe_recompile); 468 } 469 Node* cast = new CheckCastPPNode(control(), ary, arytype->cast_to_not_null_free()); 470 replace_in_map(ary, _gvn.transform(cast)); 471 } 472 473 // Make array address computation control dependent to prevent it 474 // from floating above the range check during loop optimizations. 475 Node* ptr = array_element_address(ary, idx, type, sizetype, control()); 476 477 if (result2 != NULL) *result2 = elemtype; 478 479 assert(ptr != top(), "top should go hand-in-hand with stopped"); 480 481 return ptr; 482 } 483 484 485 // returns IfNode 486 IfNode* Parse::jump_if_fork_int(Node* a, Node* b, BoolTest::mask mask, float prob, float cnt) { 487 Node *cmp = _gvn.transform(new CmpINode(a, b)); // two cases: shiftcount > 32 and shiftcount <= 32 488 Node *tst = _gvn.transform(new BoolNode(cmp, mask)); 489 IfNode *iff = create_and_map_if(control(), tst, prob, cnt); 490 return iff; 491 } 492 493 // return Region node 494 Node* Parse::jump_if_join(Node* iffalse, Node* iftrue) { 495 Node *region = new RegionNode(3); // 2 results 496 record_for_igvn(region); 497 region->init_req(1, iffalse); 498 region->init_req(2, iftrue ); 499 _gvn.set_type(region, Type::CONTROL); 500 region = _gvn.transform(region); 501 set_control (region); 502 return region; 503 } 504 505 // sentinel value for the target bci to mark never taken branches 506 // (according to profiling) 507 static const int never_reached = INT_MAX; 508 509 //------------------------------helper for tableswitch------------------------- 510 void Parse::jump_if_true_fork(IfNode *iff, int dest_bci_if_true, int prof_table_index, bool unc) { 511 // True branch, use existing map info 512 { PreserveJVMState pjvms(this); 513 Node *iftrue = _gvn.transform( new IfTrueNode (iff) ); 514 set_control( iftrue ); 515 if (unc) { 516 repush_if_args(); 517 uncommon_trap(Deoptimization::Reason_unstable_if, 518 Deoptimization::Action_reinterpret, 519 NULL, 520 "taken always"); 521 } else { 522 assert(dest_bci_if_true != never_reached, "inconsistent dest"); 523 profile_switch_case(prof_table_index); 524 merge_new_path(dest_bci_if_true); 525 } 526 } 527 528 // False branch 529 Node *iffalse = _gvn.transform( new IfFalseNode(iff) ); 530 set_control( iffalse ); 531 } 532 533 void Parse::jump_if_false_fork(IfNode *iff, int dest_bci_if_true, int prof_table_index, bool unc) { 534 // True branch, use existing map info 535 { PreserveJVMState pjvms(this); 536 Node *iffalse = _gvn.transform( new IfFalseNode (iff) ); 537 set_control( iffalse ); 538 if (unc) { 539 repush_if_args(); 540 uncommon_trap(Deoptimization::Reason_unstable_if, 541 Deoptimization::Action_reinterpret, 542 NULL, 543 "taken never"); 544 } else { 545 assert(dest_bci_if_true != never_reached, "inconsistent dest"); 546 profile_switch_case(prof_table_index); 547 merge_new_path(dest_bci_if_true); 548 } 549 } 550 551 // False branch 552 Node *iftrue = _gvn.transform( new IfTrueNode(iff) ); 553 set_control( iftrue ); 554 } 555 556 void Parse::jump_if_always_fork(int dest_bci, int prof_table_index, bool unc) { 557 // False branch, use existing map and control() 558 if (unc) { 559 repush_if_args(); 560 uncommon_trap(Deoptimization::Reason_unstable_if, 561 Deoptimization::Action_reinterpret, 562 NULL, 563 "taken never"); 564 } else { 565 assert(dest_bci != never_reached, "inconsistent dest"); 566 profile_switch_case(prof_table_index); 567 merge_new_path(dest_bci); 568 } 569 } 570 571 572 extern "C" { 573 static int jint_cmp(const void *i, const void *j) { 574 int a = *(jint *)i; 575 int b = *(jint *)j; 576 return a > b ? 1 : a < b ? -1 : 0; 577 } 578 } 579 580 581 // Default value for methodData switch indexing. Must be a negative value to avoid 582 // conflict with any legal switch index. 583 #define NullTableIndex -1 584 585 class SwitchRange : public StackObj { 586 // a range of integers coupled with a bci destination 587 jint _lo; // inclusive lower limit 588 jint _hi; // inclusive upper limit 589 int _dest; 590 int _table_index; // index into method data table 591 float _cnt; // how many times this range was hit according to profiling 592 593 public: 594 jint lo() const { return _lo; } 595 jint hi() const { return _hi; } 596 int dest() const { return _dest; } 597 int table_index() const { return _table_index; } 598 bool is_singleton() const { return _lo == _hi; } 599 float cnt() const { return _cnt; } 600 601 void setRange(jint lo, jint hi, int dest, int table_index, float cnt) { 602 assert(lo <= hi, "must be a non-empty range"); 603 _lo = lo, _hi = hi; _dest = dest; _table_index = table_index; _cnt = cnt; 604 assert(_cnt >= 0, ""); 605 } 606 bool adjoinRange(jint lo, jint hi, int dest, int table_index, float cnt, bool trim_ranges) { 607 assert(lo <= hi, "must be a non-empty range"); 608 if (lo == _hi+1 && table_index == _table_index) { 609 // see merge_ranges() comment below 610 if (trim_ranges) { 611 if (cnt == 0) { 612 if (_cnt != 0) { 613 return false; 614 } 615 if (dest != _dest) { 616 _dest = never_reached; 617 } 618 } else { 619 if (_cnt == 0) { 620 return false; 621 } 622 if (dest != _dest) { 623 return false; 624 } 625 } 626 } else { 627 if (dest != _dest) { 628 return false; 629 } 630 } 631 _hi = hi; 632 _cnt += cnt; 633 return true; 634 } 635 return false; 636 } 637 638 void set (jint value, int dest, int table_index, float cnt) { 639 setRange(value, value, dest, table_index, cnt); 640 } 641 bool adjoin(jint value, int dest, int table_index, float cnt, bool trim_ranges) { 642 return adjoinRange(value, value, dest, table_index, cnt, trim_ranges); 643 } 644 bool adjoin(SwitchRange& other) { 645 return adjoinRange(other._lo, other._hi, other._dest, other._table_index, other._cnt, false); 646 } 647 648 void print() { 649 if (is_singleton()) 650 tty->print(" {%d}=>%d (cnt=%f)", lo(), dest(), cnt()); 651 else if (lo() == min_jint) 652 tty->print(" {..%d}=>%d (cnt=%f)", hi(), dest(), cnt()); 653 else if (hi() == max_jint) 654 tty->print(" {%d..}=>%d (cnt=%f)", lo(), dest(), cnt()); 655 else 656 tty->print(" {%d..%d}=>%d (cnt=%f)", lo(), hi(), dest(), cnt()); 657 } 658 }; 659 660 // We try to minimize the number of ranges and the size of the taken 661 // ones using profiling data. When ranges are created, 662 // SwitchRange::adjoinRange() only allows 2 adjoining ranges to merge 663 // if both were never hit or both were hit to build longer unreached 664 // ranges. Here, we now merge adjoining ranges with the same 665 // destination and finally set destination of unreached ranges to the 666 // special value never_reached because it can help minimize the number 667 // of tests that are necessary. 668 // 669 // For instance: 670 // [0, 1] to target1 sometimes taken 671 // [1, 2] to target1 never taken 672 // [2, 3] to target2 never taken 673 // would lead to: 674 // [0, 1] to target1 sometimes taken 675 // [1, 3] never taken 676 // 677 // (first 2 ranges to target1 are not merged) 678 static void merge_ranges(SwitchRange* ranges, int& rp) { 679 if (rp == 0) { 680 return; 681 } 682 int shift = 0; 683 for (int j = 0; j < rp; j++) { 684 SwitchRange& r1 = ranges[j-shift]; 685 SwitchRange& r2 = ranges[j+1]; 686 if (r1.adjoin(r2)) { 687 shift++; 688 } else if (shift > 0) { 689 ranges[j+1-shift] = r2; 690 } 691 } 692 rp -= shift; 693 for (int j = 0; j <= rp; j++) { 694 SwitchRange& r = ranges[j]; 695 if (r.cnt() == 0 && r.dest() != never_reached) { 696 r.setRange(r.lo(), r.hi(), never_reached, r.table_index(), r.cnt()); 697 } 698 } 699 } 700 701 //-------------------------------do_tableswitch-------------------------------- 702 void Parse::do_tableswitch() { 703 Node* lookup = pop(); 704 // Get information about tableswitch 705 int default_dest = iter().get_dest_table(0); 706 int lo_index = iter().get_int_table(1); 707 int hi_index = iter().get_int_table(2); 708 int len = hi_index - lo_index + 1; 709 710 if (len < 1) { 711 // If this is a backward branch, add safepoint 712 maybe_add_safepoint(default_dest); 713 merge(default_dest); 714 return; 715 } 716 717 ciMethodData* methodData = method()->method_data(); 718 ciMultiBranchData* profile = NULL; 719 if (methodData->is_mature() && UseSwitchProfiling) { 720 ciProfileData* data = methodData->bci_to_data(bci()); 721 if (data != NULL && data->is_MultiBranchData()) { 722 profile = (ciMultiBranchData*)data; 723 } 724 } 725 bool trim_ranges = !method_data_update() && !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if); 726 727 // generate decision tree, using trichotomy when possible 728 int rnum = len+2; 729 bool makes_backward_branch = false; 730 SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum); 731 int rp = -1; 732 if (lo_index != min_jint) { 733 uint cnt = 1; 734 if (profile != NULL) { 735 cnt = profile->default_count() / (hi_index != max_jint ? 2 : 1); 736 } 737 ranges[++rp].setRange(min_jint, lo_index-1, default_dest, NullTableIndex, cnt); 738 } 739 for (int j = 0; j < len; j++) { 740 jint match_int = lo_index+j; 741 int dest = iter().get_dest_table(j+3); 742 makes_backward_branch |= (dest <= bci()); 743 int table_index = method_data_update() ? j : NullTableIndex; 744 uint cnt = 1; 745 if (profile != NULL) { 746 cnt = profile->count_at(j); 747 } 748 if (rp < 0 || !ranges[rp].adjoin(match_int, dest, table_index, cnt, trim_ranges)) { 749 ranges[++rp].set(match_int, dest, table_index, cnt); 750 } 751 } 752 jint highest = lo_index+(len-1); 753 assert(ranges[rp].hi() == highest, ""); 754 if (highest != max_jint) { 755 uint cnt = 1; 756 if (profile != NULL) { 757 cnt = profile->default_count() / (lo_index != min_jint ? 2 : 1); 758 } 759 if (!ranges[rp].adjoinRange(highest+1, max_jint, default_dest, NullTableIndex, cnt, trim_ranges)) { 760 ranges[++rp].setRange(highest+1, max_jint, default_dest, NullTableIndex, cnt); 761 } 762 } 763 assert(rp < len+2, "not too many ranges"); 764 765 if (trim_ranges) { 766 merge_ranges(ranges, rp); 767 } 768 769 // Safepoint in case if backward branch observed 770 if( makes_backward_branch && UseLoopSafepoints ) 771 add_safepoint(); 772 773 jump_switch_ranges(lookup, &ranges[0], &ranges[rp]); 774 } 775 776 777 //------------------------------do_lookupswitch-------------------------------- 778 void Parse::do_lookupswitch() { 779 Node *lookup = pop(); // lookup value 780 // Get information about lookupswitch 781 int default_dest = iter().get_dest_table(0); 782 int len = iter().get_int_table(1); 783 784 if (len < 1) { // If this is a backward branch, add safepoint 785 maybe_add_safepoint(default_dest); 786 merge(default_dest); 787 return; 788 } 789 790 ciMethodData* methodData = method()->method_data(); 791 ciMultiBranchData* profile = NULL; 792 if (methodData->is_mature() && UseSwitchProfiling) { 793 ciProfileData* data = methodData->bci_to_data(bci()); 794 if (data != NULL && data->is_MultiBranchData()) { 795 profile = (ciMultiBranchData*)data; 796 } 797 } 798 bool trim_ranges = !method_data_update() && !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if); 799 800 // generate decision tree, using trichotomy when possible 801 jint* table = NEW_RESOURCE_ARRAY(jint, len*3); 802 { 803 for (int j = 0; j < len; j++) { 804 table[3*j+0] = iter().get_int_table(2+2*j); 805 table[3*j+1] = iter().get_dest_table(2+2*j+1); 806 table[3*j+2] = profile == NULL ? 1 : profile->count_at(j); 807 } 808 qsort(table, len, 3*sizeof(table[0]), jint_cmp); 809 } 810 811 float defaults = 0; 812 jint prev = min_jint; 813 for (int j = 0; j < len; j++) { 814 jint match_int = table[3*j+0]; 815 if (match_int != prev) { 816 defaults += (float)match_int - prev; 817 } 818 prev = match_int+1; 819 } 820 if (prev-1 != max_jint) { 821 defaults += (float)max_jint - prev + 1; 822 } 823 float default_cnt = 1; 824 if (profile != NULL) { 825 default_cnt = profile->default_count()/defaults; 826 } 827 828 int rnum = len*2+1; 829 bool makes_backward_branch = false; 830 SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum); 831 int rp = -1; 832 for (int j = 0; j < len; j++) { 833 jint match_int = table[3*j+0]; 834 int dest = table[3*j+1]; 835 int cnt = table[3*j+2]; 836 int next_lo = rp < 0 ? min_jint : ranges[rp].hi()+1; 837 int table_index = method_data_update() ? j : NullTableIndex; 838 makes_backward_branch |= (dest <= bci()); 839 float c = default_cnt * ((float)match_int - next_lo); 840 if (match_int != next_lo && (rp < 0 || !ranges[rp].adjoinRange(next_lo, match_int-1, default_dest, NullTableIndex, c, trim_ranges))) { 841 assert(default_dest != never_reached, "sentinel value for dead destinations"); 842 ranges[++rp].setRange(next_lo, match_int-1, default_dest, NullTableIndex, c); 843 } 844 if (rp < 0 || !ranges[rp].adjoin(match_int, dest, table_index, cnt, trim_ranges)) { 845 assert(dest != never_reached, "sentinel value for dead destinations"); 846 ranges[++rp].set(match_int, dest, table_index, cnt); 847 } 848 } 849 jint highest = table[3*(len-1)]; 850 assert(ranges[rp].hi() == highest, ""); 851 if (highest != max_jint && 852 !ranges[rp].adjoinRange(highest+1, max_jint, default_dest, NullTableIndex, default_cnt * ((float)max_jint - highest), trim_ranges)) { 853 ranges[++rp].setRange(highest+1, max_jint, default_dest, NullTableIndex, default_cnt * ((float)max_jint - highest)); 854 } 855 assert(rp < rnum, "not too many ranges"); 856 857 if (trim_ranges) { 858 merge_ranges(ranges, rp); 859 } 860 861 // Safepoint in case backward branch observed 862 if (makes_backward_branch && UseLoopSafepoints) 863 add_safepoint(); 864 865 jump_switch_ranges(lookup, &ranges[0], &ranges[rp]); 866 } 867 868 static float if_prob(float taken_cnt, float total_cnt) { 869 assert(taken_cnt <= total_cnt, ""); 870 if (total_cnt == 0) { 871 return PROB_FAIR; 872 } 873 float p = taken_cnt / total_cnt; 874 return MIN2(MAX2(p, PROB_MIN), PROB_MAX); 875 } 876 877 static float if_cnt(float cnt) { 878 if (cnt == 0) { 879 return COUNT_UNKNOWN; 880 } 881 return cnt; 882 } 883 884 static float sum_of_cnts(SwitchRange *lo, SwitchRange *hi) { 885 float total_cnt = 0; 886 for (SwitchRange* sr = lo; sr <= hi; sr++) { 887 total_cnt += sr->cnt(); 888 } 889 return total_cnt; 890 } 891 892 class SwitchRanges : public ResourceObj { 893 public: 894 SwitchRange* _lo; 895 SwitchRange* _hi; 896 SwitchRange* _mid; 897 float _cost; 898 899 enum { 900 Start, 901 LeftDone, 902 RightDone, 903 Done 904 } _state; 905 906 SwitchRanges(SwitchRange *lo, SwitchRange *hi) 907 : _lo(lo), _hi(hi), _mid(NULL), 908 _cost(0), _state(Start) { 909 } 910 911 SwitchRanges() 912 : _lo(NULL), _hi(NULL), _mid(NULL), 913 _cost(0), _state(Start) {} 914 }; 915 916 // Estimate cost of performing a binary search on lo..hi 917 static float compute_tree_cost(SwitchRange *lo, SwitchRange *hi, float total_cnt) { 918 GrowableArray<SwitchRanges> tree; 919 SwitchRanges root(lo, hi); 920 tree.push(root); 921 922 float cost = 0; 923 do { 924 SwitchRanges& r = *tree.adr_at(tree.length()-1); 925 if (r._hi != r._lo) { 926 if (r._mid == NULL) { 927 float r_cnt = sum_of_cnts(r._lo, r._hi); 928 929 if (r_cnt == 0) { 930 tree.pop(); 931 cost = 0; 932 continue; 933 } 934 935 SwitchRange* mid = NULL; 936 mid = r._lo; 937 for (float cnt = 0; ; ) { 938 assert(mid <= r._hi, "out of bounds"); 939 cnt += mid->cnt(); 940 if (cnt > r_cnt / 2) { 941 break; 942 } 943 mid++; 944 } 945 assert(mid <= r._hi, "out of bounds"); 946 r._mid = mid; 947 r._cost = r_cnt / total_cnt; 948 } 949 r._cost += cost; 950 if (r._state < SwitchRanges::LeftDone && r._mid > r._lo) { 951 cost = 0; 952 r._state = SwitchRanges::LeftDone; 953 tree.push(SwitchRanges(r._lo, r._mid-1)); 954 } else if (r._state < SwitchRanges::RightDone) { 955 cost = 0; 956 r._state = SwitchRanges::RightDone; 957 tree.push(SwitchRanges(r._mid == r._lo ? r._mid+1 : r._mid, r._hi)); 958 } else { 959 tree.pop(); 960 cost = r._cost; 961 } 962 } else { 963 tree.pop(); 964 cost = r._cost; 965 } 966 } while (tree.length() > 0); 967 968 969 return cost; 970 } 971 972 // It sometimes pays off to test most common ranges before the binary search 973 void Parse::linear_search_switch_ranges(Node* key_val, SwitchRange*& lo, SwitchRange*& hi) { 974 uint nr = hi - lo + 1; 975 float total_cnt = sum_of_cnts(lo, hi); 976 977 float min = compute_tree_cost(lo, hi, total_cnt); 978 float extra = 1; 979 float sub = 0; 980 981 SwitchRange* array1 = lo; 982 SwitchRange* array2 = NEW_RESOURCE_ARRAY(SwitchRange, nr); 983 984 SwitchRange* ranges = NULL; 985 986 while (nr >= 2) { 987 assert(lo == array1 || lo == array2, "one the 2 already allocated arrays"); 988 ranges = (lo == array1) ? array2 : array1; 989 990 // Find highest frequency range 991 SwitchRange* candidate = lo; 992 for (SwitchRange* sr = lo+1; sr <= hi; sr++) { 993 if (sr->cnt() > candidate->cnt()) { 994 candidate = sr; 995 } 996 } 997 SwitchRange most_freq = *candidate; 998 if (most_freq.cnt() == 0) { 999 break; 1000 } 1001 1002 // Copy remaining ranges into another array 1003 int shift = 0; 1004 for (uint i = 0; i < nr; i++) { 1005 SwitchRange* sr = &lo[i]; 1006 if (sr != candidate) { 1007 ranges[i-shift] = *sr; 1008 } else { 1009 shift++; 1010 if (i > 0 && i < nr-1) { 1011 SwitchRange prev = lo[i-1]; 1012 prev.setRange(prev.lo(), sr->hi(), prev.dest(), prev.table_index(), prev.cnt()); 1013 if (prev.adjoin(lo[i+1])) { 1014 shift++; 1015 i++; 1016 } 1017 ranges[i-shift] = prev; 1018 } 1019 } 1020 } 1021 nr -= shift; 1022 1023 // Evaluate cost of testing the most common range and performing a 1024 // binary search on the other ranges 1025 float cost = extra + compute_tree_cost(&ranges[0], &ranges[nr-1], total_cnt); 1026 if (cost >= min) { 1027 break; 1028 } 1029 // swap arrays 1030 lo = &ranges[0]; 1031 hi = &ranges[nr-1]; 1032 1033 // It pays off: emit the test for the most common range 1034 assert(most_freq.cnt() > 0, "must be taken"); 1035 Node* val = _gvn.transform(new SubINode(key_val, _gvn.intcon(most_freq.lo()))); 1036 Node* cmp = _gvn.transform(new CmpUNode(val, _gvn.intcon(most_freq.hi() - most_freq.lo()))); 1037 Node* tst = _gvn.transform(new BoolNode(cmp, BoolTest::le)); 1038 IfNode* iff = create_and_map_if(control(), tst, if_prob(most_freq.cnt(), total_cnt), if_cnt(most_freq.cnt())); 1039 jump_if_true_fork(iff, most_freq.dest(), most_freq.table_index(), false); 1040 1041 sub += most_freq.cnt() / total_cnt; 1042 extra += 1 - sub; 1043 min = cost; 1044 } 1045 } 1046 1047 //----------------------------create_jump_tables------------------------------- 1048 bool Parse::create_jump_tables(Node* key_val, SwitchRange* lo, SwitchRange* hi) { 1049 // Are jumptables enabled 1050 if (!UseJumpTables) return false; 1051 1052 // Are jumptables supported 1053 if (!Matcher::has_match_rule(Op_Jump)) return false; 1054 1055 // Don't make jump table if profiling 1056 if (method_data_update()) return false; 1057 1058 bool trim_ranges = !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if); 1059 1060 // Decide if a guard is needed to lop off big ranges at either (or 1061 // both) end(s) of the input set. We'll call this the default target 1062 // even though we can't be sure that it is the true "default". 1063 1064 bool needs_guard = false; 1065 int default_dest; 1066 int64_t total_outlier_size = 0; 1067 int64_t hi_size = ((int64_t)hi->hi()) - ((int64_t)hi->lo()) + 1; 1068 int64_t lo_size = ((int64_t)lo->hi()) - ((int64_t)lo->lo()) + 1; 1069 1070 if (lo->dest() == hi->dest()) { 1071 total_outlier_size = hi_size + lo_size; 1072 default_dest = lo->dest(); 1073 } else if (lo_size > hi_size) { 1074 total_outlier_size = lo_size; 1075 default_dest = lo->dest(); 1076 } else { 1077 total_outlier_size = hi_size; 1078 default_dest = hi->dest(); 1079 } 1080 1081 float total = sum_of_cnts(lo, hi); 1082 float cost = compute_tree_cost(lo, hi, total); 1083 1084 // If a guard test will eliminate very sparse end ranges, then 1085 // it is worth the cost of an extra jump. 1086 float trimmed_cnt = 0; 1087 if (total_outlier_size > (MaxJumpTableSparseness * 4)) { 1088 needs_guard = true; 1089 if (default_dest == lo->dest()) { 1090 trimmed_cnt += lo->cnt(); 1091 lo++; 1092 } 1093 if (default_dest == hi->dest()) { 1094 trimmed_cnt += hi->cnt(); 1095 hi--; 1096 } 1097 } 1098 1099 // Find the total number of cases and ranges 1100 int64_t num_cases = ((int64_t)hi->hi()) - ((int64_t)lo->lo()) + 1; 1101 int num_range = hi - lo + 1; 1102 1103 // Don't create table if: too large, too small, or too sparse. 1104 if (num_cases > MaxJumpTableSize) 1105 return false; 1106 if (UseSwitchProfiling) { 1107 // MinJumpTableSize is set so with a well balanced binary tree, 1108 // when the number of ranges is MinJumpTableSize, it's cheaper to 1109 // go through a JumpNode that a tree of IfNodes. Average cost of a 1110 // tree of IfNodes with MinJumpTableSize is 1111 // log2f(MinJumpTableSize) comparisons. So if the cost computed 1112 // from profile data is less than log2f(MinJumpTableSize) then 1113 // going with the binary search is cheaper. 1114 if (cost < log2f(MinJumpTableSize)) { 1115 return false; 1116 } 1117 } else { 1118 if (num_cases < MinJumpTableSize) 1119 return false; 1120 } 1121 if (num_cases > (MaxJumpTableSparseness * num_range)) 1122 return false; 1123 1124 // Normalize table lookups to zero 1125 int lowval = lo->lo(); 1126 key_val = _gvn.transform( new SubINode(key_val, _gvn.intcon(lowval)) ); 1127 1128 // Generate a guard to protect against input keyvals that aren't 1129 // in the switch domain. 1130 if (needs_guard) { 1131 Node* size = _gvn.intcon(num_cases); 1132 Node* cmp = _gvn.transform(new CmpUNode(key_val, size)); 1133 Node* tst = _gvn.transform(new BoolNode(cmp, BoolTest::ge)); 1134 IfNode* iff = create_and_map_if(control(), tst, if_prob(trimmed_cnt, total), if_cnt(trimmed_cnt)); 1135 jump_if_true_fork(iff, default_dest, NullTableIndex, trim_ranges && trimmed_cnt == 0); 1136 1137 total -= trimmed_cnt; 1138 } 1139 1140 // Create an ideal node JumpTable that has projections 1141 // of all possible ranges for a switch statement 1142 // The key_val input must be converted to a pointer offset and scaled. 1143 // Compare Parse::array_addressing above. 1144 1145 // Clean the 32-bit int into a real 64-bit offset. 1146 // Otherwise, the jint value 0 might turn into an offset of 0x0800000000. 1147 const TypeInt* ikeytype = TypeInt::make(0, num_cases, Type::WidenMin); 1148 // Make I2L conversion control dependent to prevent it from 1149 // floating above the range check during loop optimizations. 1150 key_val = C->conv_I2X_index(&_gvn, key_val, ikeytype, control()); 1151 1152 // Shift the value by wordsize so we have an index into the table, rather 1153 // than a switch value 1154 Node *shiftWord = _gvn.MakeConX(wordSize); 1155 key_val = _gvn.transform( new MulXNode( key_val, shiftWord)); 1156 1157 // Create the JumpNode 1158 Arena* arena = C->comp_arena(); 1159 float* probs = (float*)arena->Amalloc(sizeof(float)*num_cases); 1160 int i = 0; 1161 if (total == 0) { 1162 for (SwitchRange* r = lo; r <= hi; r++) { 1163 for (int64_t j = r->lo(); j <= r->hi(); j++, i++) { 1164 probs[i] = 1.0F / num_cases; 1165 } 1166 } 1167 } else { 1168 for (SwitchRange* r = lo; r <= hi; r++) { 1169 float prob = r->cnt()/total; 1170 for (int64_t j = r->lo(); j <= r->hi(); j++, i++) { 1171 probs[i] = prob / (r->hi() - r->lo() + 1); 1172 } 1173 } 1174 } 1175 1176 ciMethodData* methodData = method()->method_data(); 1177 ciMultiBranchData* profile = NULL; 1178 if (methodData->is_mature()) { 1179 ciProfileData* data = methodData->bci_to_data(bci()); 1180 if (data != NULL && data->is_MultiBranchData()) { 1181 profile = (ciMultiBranchData*)data; 1182 } 1183 } 1184 1185 Node* jtn = _gvn.transform(new JumpNode(control(), key_val, num_cases, probs, profile == NULL ? COUNT_UNKNOWN : total)); 1186 1187 // These are the switch destinations hanging off the jumpnode 1188 i = 0; 1189 for (SwitchRange* r = lo; r <= hi; r++) { 1190 for (int64_t j = r->lo(); j <= r->hi(); j++, i++) { 1191 Node* input = _gvn.transform(new JumpProjNode(jtn, i, r->dest(), (int)(j - lowval))); 1192 { 1193 PreserveJVMState pjvms(this); 1194 set_control(input); 1195 jump_if_always_fork(r->dest(), r->table_index(), trim_ranges && r->cnt() == 0); 1196 } 1197 } 1198 } 1199 assert(i == num_cases, "miscount of cases"); 1200 stop_and_kill_map(); // no more uses for this JVMS 1201 return true; 1202 } 1203 1204 //----------------------------jump_switch_ranges------------------------------- 1205 void Parse::jump_switch_ranges(Node* key_val, SwitchRange *lo, SwitchRange *hi, int switch_depth) { 1206 Block* switch_block = block(); 1207 bool trim_ranges = !method_data_update() && !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if); 1208 1209 if (switch_depth == 0) { 1210 // Do special processing for the top-level call. 1211 assert(lo->lo() == min_jint, "initial range must exhaust Type::INT"); 1212 assert(hi->hi() == max_jint, "initial range must exhaust Type::INT"); 1213 1214 // Decrement pred-numbers for the unique set of nodes. 1215 #ifdef ASSERT 1216 if (!trim_ranges) { 1217 // Ensure that the block's successors are a (duplicate-free) set. 1218 int successors_counted = 0; // block occurrences in [hi..lo] 1219 int unique_successors = switch_block->num_successors(); 1220 for (int i = 0; i < unique_successors; i++) { 1221 Block* target = switch_block->successor_at(i); 1222 1223 // Check that the set of successors is the same in both places. 1224 int successors_found = 0; 1225 for (SwitchRange* p = lo; p <= hi; p++) { 1226 if (p->dest() == target->start()) successors_found++; 1227 } 1228 assert(successors_found > 0, "successor must be known"); 1229 successors_counted += successors_found; 1230 } 1231 assert(successors_counted == (hi-lo)+1, "no unexpected successors"); 1232 } 1233 #endif 1234 1235 // Maybe prune the inputs, based on the type of key_val. 1236 jint min_val = min_jint; 1237 jint max_val = max_jint; 1238 const TypeInt* ti = key_val->bottom_type()->isa_int(); 1239 if (ti != NULL) { 1240 min_val = ti->_lo; 1241 max_val = ti->_hi; 1242 assert(min_val <= max_val, "invalid int type"); 1243 } 1244 while (lo->hi() < min_val) { 1245 lo++; 1246 } 1247 if (lo->lo() < min_val) { 1248 lo->setRange(min_val, lo->hi(), lo->dest(), lo->table_index(), lo->cnt()); 1249 } 1250 while (hi->lo() > max_val) { 1251 hi--; 1252 } 1253 if (hi->hi() > max_val) { 1254 hi->setRange(hi->lo(), max_val, hi->dest(), hi->table_index(), hi->cnt()); 1255 } 1256 1257 linear_search_switch_ranges(key_val, lo, hi); 1258 } 1259 1260 #ifndef PRODUCT 1261 if (switch_depth == 0) { 1262 _max_switch_depth = 0; 1263 _est_switch_depth = log2_intptr((hi-lo+1)-1)+1; 1264 } 1265 #endif 1266 1267 assert(lo <= hi, "must be a non-empty set of ranges"); 1268 if (lo == hi) { 1269 jump_if_always_fork(lo->dest(), lo->table_index(), trim_ranges && lo->cnt() == 0); 1270 } else { 1271 assert(lo->hi() == (lo+1)->lo()-1, "contiguous ranges"); 1272 assert(hi->lo() == (hi-1)->hi()+1, "contiguous ranges"); 1273 1274 if (create_jump_tables(key_val, lo, hi)) return; 1275 1276 SwitchRange* mid = NULL; 1277 float total_cnt = sum_of_cnts(lo, hi); 1278 1279 int nr = hi - lo + 1; 1280 if (UseSwitchProfiling) { 1281 // Don't keep the binary search tree balanced: pick up mid point 1282 // that split frequencies in half. 1283 float cnt = 0; 1284 for (SwitchRange* sr = lo; sr <= hi; sr++) { 1285 cnt += sr->cnt(); 1286 if (cnt >= total_cnt / 2) { 1287 mid = sr; 1288 break; 1289 } 1290 } 1291 } else { 1292 mid = lo + nr/2; 1293 1294 // if there is an easy choice, pivot at a singleton: 1295 if (nr > 3 && !mid->is_singleton() && (mid-1)->is_singleton()) mid--; 1296 1297 assert(lo < mid && mid <= hi, "good pivot choice"); 1298 assert(nr != 2 || mid == hi, "should pick higher of 2"); 1299 assert(nr != 3 || mid == hi-1, "should pick middle of 3"); 1300 } 1301 1302 1303 Node *test_val = _gvn.intcon(mid == lo ? mid->hi() : mid->lo()); 1304 1305 if (mid->is_singleton()) { 1306 IfNode *iff_ne = jump_if_fork_int(key_val, test_val, BoolTest::ne, 1-if_prob(mid->cnt(), total_cnt), if_cnt(mid->cnt())); 1307 jump_if_false_fork(iff_ne, mid->dest(), mid->table_index(), trim_ranges && mid->cnt() == 0); 1308 1309 // Special Case: If there are exactly three ranges, and the high 1310 // and low range each go to the same place, omit the "gt" test, 1311 // since it will not discriminate anything. 1312 bool eq_test_only = (hi == lo+2 && hi->dest() == lo->dest() && mid == hi-1) || mid == lo; 1313 1314 // if there is a higher range, test for it and process it: 1315 if (mid < hi && !eq_test_only) { 1316 // two comparisons of same values--should enable 1 test for 2 branches 1317 // Use BoolTest::le instead of BoolTest::gt 1318 float cnt = sum_of_cnts(lo, mid-1); 1319 IfNode *iff_le = jump_if_fork_int(key_val, test_val, BoolTest::le, if_prob(cnt, total_cnt), if_cnt(cnt)); 1320 Node *iftrue = _gvn.transform( new IfTrueNode(iff_le) ); 1321 Node *iffalse = _gvn.transform( new IfFalseNode(iff_le) ); 1322 { PreserveJVMState pjvms(this); 1323 set_control(iffalse); 1324 jump_switch_ranges(key_val, mid+1, hi, switch_depth+1); 1325 } 1326 set_control(iftrue); 1327 } 1328 1329 } else { 1330 // mid is a range, not a singleton, so treat mid..hi as a unit 1331 float cnt = sum_of_cnts(mid == lo ? mid+1 : mid, hi); 1332 IfNode *iff_ge = jump_if_fork_int(key_val, test_val, mid == lo ? BoolTest::gt : BoolTest::ge, if_prob(cnt, total_cnt), if_cnt(cnt)); 1333 1334 // if there is a higher range, test for it and process it: 1335 if (mid == hi) { 1336 jump_if_true_fork(iff_ge, mid->dest(), mid->table_index(), trim_ranges && cnt == 0); 1337 } else { 1338 Node *iftrue = _gvn.transform( new IfTrueNode(iff_ge) ); 1339 Node *iffalse = _gvn.transform( new IfFalseNode(iff_ge) ); 1340 { PreserveJVMState pjvms(this); 1341 set_control(iftrue); 1342 jump_switch_ranges(key_val, mid == lo ? mid+1 : mid, hi, switch_depth+1); 1343 } 1344 set_control(iffalse); 1345 } 1346 } 1347 1348 // in any case, process the lower range 1349 if (mid == lo) { 1350 if (mid->is_singleton()) { 1351 jump_switch_ranges(key_val, lo+1, hi, switch_depth+1); 1352 } else { 1353 jump_if_always_fork(lo->dest(), lo->table_index(), trim_ranges && lo->cnt() == 0); 1354 } 1355 } else { 1356 jump_switch_ranges(key_val, lo, mid-1, switch_depth+1); 1357 } 1358 } 1359 1360 // Decrease pred_count for each successor after all is done. 1361 if (switch_depth == 0) { 1362 int unique_successors = switch_block->num_successors(); 1363 for (int i = 0; i < unique_successors; i++) { 1364 Block* target = switch_block->successor_at(i); 1365 // Throw away the pre-allocated path for each unique successor. 1366 target->next_path_num(); 1367 } 1368 } 1369 1370 #ifndef PRODUCT 1371 _max_switch_depth = MAX2(switch_depth, _max_switch_depth); 1372 if (TraceOptoParse && Verbose && WizardMode && switch_depth == 0) { 1373 SwitchRange* r; 1374 int nsing = 0; 1375 for( r = lo; r <= hi; r++ ) { 1376 if( r->is_singleton() ) nsing++; 1377 } 1378 tty->print(">>> "); 1379 _method->print_short_name(); 1380 tty->print_cr(" switch decision tree"); 1381 tty->print_cr(" %d ranges (%d singletons), max_depth=%d, est_depth=%d", 1382 (int) (hi-lo+1), nsing, _max_switch_depth, _est_switch_depth); 1383 if (_max_switch_depth > _est_switch_depth) { 1384 tty->print_cr("******** BAD SWITCH DEPTH ********"); 1385 } 1386 tty->print(" "); 1387 for( r = lo; r <= hi; r++ ) { 1388 r->print(); 1389 } 1390 tty->cr(); 1391 } 1392 #endif 1393 } 1394 1395 void Parse::modf() { 1396 Node *f2 = pop(); 1397 Node *f1 = pop(); 1398 Node* c = make_runtime_call(RC_LEAF, OptoRuntime::modf_Type(), 1399 CAST_FROM_FN_PTR(address, SharedRuntime::frem), 1400 "frem", NULL, //no memory effects 1401 f1, f2); 1402 Node* res = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 0)); 1403 1404 push(res); 1405 } 1406 1407 void Parse::modd() { 1408 Node *d2 = pop_pair(); 1409 Node *d1 = pop_pair(); 1410 Node* c = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), 1411 CAST_FROM_FN_PTR(address, SharedRuntime::drem), 1412 "drem", NULL, //no memory effects 1413 d1, top(), d2, top()); 1414 Node* res_d = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 0)); 1415 1416 #ifdef ASSERT 1417 Node* res_top = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 1)); 1418 assert(res_top == top(), "second value must be top"); 1419 #endif 1420 1421 push_pair(res_d); 1422 } 1423 1424 void Parse::l2f() { 1425 Node* f2 = pop(); 1426 Node* f1 = pop(); 1427 Node* c = make_runtime_call(RC_LEAF, OptoRuntime::l2f_Type(), 1428 CAST_FROM_FN_PTR(address, SharedRuntime::l2f), 1429 "l2f", NULL, //no memory effects 1430 f1, f2); 1431 Node* res = _gvn.transform(new ProjNode(c, TypeFunc::Parms + 0)); 1432 1433 push(res); 1434 } 1435 1436 void Parse::do_irem() { 1437 // Must keep both values on the expression-stack during null-check 1438 zero_check_int(peek()); 1439 // Compile-time detect of null-exception? 1440 if (stopped()) return; 1441 1442 Node* b = pop(); 1443 Node* a = pop(); 1444 1445 const Type *t = _gvn.type(b); 1446 if (t != Type::TOP) { 1447 const TypeInt *ti = t->is_int(); 1448 if (ti->is_con()) { 1449 int divisor = ti->get_con(); 1450 // check for positive power of 2 1451 if (divisor > 0 && 1452 (divisor & ~(divisor-1)) == divisor) { 1453 // yes ! 1454 Node *mask = _gvn.intcon((divisor - 1)); 1455 // Sigh, must handle negative dividends 1456 Node *zero = _gvn.intcon(0); 1457 IfNode *ifff = jump_if_fork_int(a, zero, BoolTest::lt, PROB_FAIR, COUNT_UNKNOWN); 1458 Node *iff = _gvn.transform( new IfFalseNode(ifff) ); 1459 Node *ift = _gvn.transform( new IfTrueNode (ifff) ); 1460 Node *reg = jump_if_join(ift, iff); 1461 Node *phi = PhiNode::make(reg, NULL, TypeInt::INT); 1462 // Negative path; negate/and/negate 1463 Node *neg = _gvn.transform( new SubINode(zero, a) ); 1464 Node *andn= _gvn.transform( new AndINode(neg, mask) ); 1465 Node *negn= _gvn.transform( new SubINode(zero, andn) ); 1466 phi->init_req(1, negn); 1467 // Fast positive case 1468 Node *andx = _gvn.transform( new AndINode(a, mask) ); 1469 phi->init_req(2, andx); 1470 // Push the merge 1471 push( _gvn.transform(phi) ); 1472 return; 1473 } 1474 } 1475 } 1476 // Default case 1477 push( _gvn.transform( new ModINode(control(),a,b) ) ); 1478 } 1479 1480 // Handle jsr and jsr_w bytecode 1481 void Parse::do_jsr() { 1482 assert(bc() == Bytecodes::_jsr || bc() == Bytecodes::_jsr_w, "wrong bytecode"); 1483 1484 // Store information about current state, tagged with new _jsr_bci 1485 int return_bci = iter().next_bci(); 1486 int jsr_bci = (bc() == Bytecodes::_jsr) ? iter().get_dest() : iter().get_far_dest(); 1487 1488 // Update method data 1489 profile_taken_branch(jsr_bci); 1490 1491 // The way we do things now, there is only one successor block 1492 // for the jsr, because the target code is cloned by ciTypeFlow. 1493 Block* target = successor_for_bci(jsr_bci); 1494 1495 // What got pushed? 1496 const Type* ret_addr = target->peek(); 1497 assert(ret_addr->singleton(), "must be a constant (cloned jsr body)"); 1498 1499 // Effect on jsr on stack 1500 push(_gvn.makecon(ret_addr)); 1501 1502 // Flow to the jsr. 1503 merge(jsr_bci); 1504 } 1505 1506 // Handle ret bytecode 1507 void Parse::do_ret() { 1508 // Find to whom we return. 1509 assert(block()->num_successors() == 1, "a ret can only go one place now"); 1510 Block* target = block()->successor_at(0); 1511 assert(!target->is_ready(), "our arrival must be expected"); 1512 profile_ret(target->flow()->start()); 1513 int pnum = target->next_path_num(); 1514 merge_common(target, pnum); 1515 } 1516 1517 static bool has_injected_profile(BoolTest::mask btest, Node* test, int& taken, int& not_taken) { 1518 if (btest != BoolTest::eq && btest != BoolTest::ne) { 1519 // Only ::eq and ::ne are supported for profile injection. 1520 return false; 1521 } 1522 if (test->is_Cmp() && 1523 test->in(1)->Opcode() == Op_ProfileBoolean) { 1524 ProfileBooleanNode* profile = (ProfileBooleanNode*)test->in(1); 1525 int false_cnt = profile->false_count(); 1526 int true_cnt = profile->true_count(); 1527 1528 // Counts matching depends on the actual test operation (::eq or ::ne). 1529 // No need to scale the counts because profile injection was designed 1530 // to feed exact counts into VM. 1531 taken = (btest == BoolTest::eq) ? false_cnt : true_cnt; 1532 not_taken = (btest == BoolTest::eq) ? true_cnt : false_cnt; 1533 1534 profile->consume(); 1535 return true; 1536 } 1537 return false; 1538 } 1539 //--------------------------dynamic_branch_prediction-------------------------- 1540 // Try to gather dynamic branch prediction behavior. Return a probability 1541 // of the branch being taken and set the "cnt" field. Returns a -1.0 1542 // if we need to use static prediction for some reason. 1543 float Parse::dynamic_branch_prediction(float &cnt, BoolTest::mask btest, Node* test) { 1544 ResourceMark rm; 1545 1546 cnt = COUNT_UNKNOWN; 1547 1548 int taken = 0; 1549 int not_taken = 0; 1550 1551 bool use_mdo = !has_injected_profile(btest, test, taken, not_taken); 1552 1553 if (use_mdo) { 1554 // Use MethodData information if it is available 1555 // FIXME: free the ProfileData structure 1556 ciMethodData* methodData = method()->method_data(); 1557 if (!methodData->is_mature()) return PROB_UNKNOWN; 1558 ciProfileData* data = methodData->bci_to_data(bci()); 1559 if (data == NULL) { 1560 return PROB_UNKNOWN; 1561 } 1562 if (!data->is_JumpData()) return PROB_UNKNOWN; 1563 1564 // get taken and not taken values 1565 taken = data->as_JumpData()->taken(); 1566 not_taken = 0; 1567 if (data->is_BranchData()) { 1568 not_taken = data->as_BranchData()->not_taken(); 1569 } 1570 1571 // scale the counts to be commensurate with invocation counts: 1572 taken = method()->scale_count(taken); 1573 not_taken = method()->scale_count(not_taken); 1574 } 1575 1576 // Give up if too few (or too many, in which case the sum will overflow) counts to be meaningful. 1577 // We also check that individual counters are positive first, otherwise the sum can become positive. 1578 if (taken < 0 || not_taken < 0 || taken + not_taken < 40) { 1579 if (C->log() != NULL) { 1580 C->log()->elem("branch target_bci='%d' taken='%d' not_taken='%d'", iter().get_dest(), taken, not_taken); 1581 } 1582 return PROB_UNKNOWN; 1583 } 1584 1585 // Compute frequency that we arrive here 1586 float sum = taken + not_taken; 1587 // Adjust, if this block is a cloned private block but the 1588 // Jump counts are shared. Taken the private counts for 1589 // just this path instead of the shared counts. 1590 if( block()->count() > 0 ) 1591 sum = block()->count(); 1592 cnt = sum / FreqCountInvocations; 1593 1594 // Pin probability to sane limits 1595 float prob; 1596 if( !taken ) 1597 prob = (0+PROB_MIN) / 2; 1598 else if( !not_taken ) 1599 prob = (1+PROB_MAX) / 2; 1600 else { // Compute probability of true path 1601 prob = (float)taken / (float)(taken + not_taken); 1602 if (prob > PROB_MAX) prob = PROB_MAX; 1603 if (prob < PROB_MIN) prob = PROB_MIN; 1604 } 1605 1606 assert((cnt > 0.0f) && (prob > 0.0f), 1607 "Bad frequency assignment in if"); 1608 1609 if (C->log() != NULL) { 1610 const char* prob_str = NULL; 1611 if (prob >= PROB_MAX) prob_str = (prob == PROB_MAX) ? "max" : "always"; 1612 if (prob <= PROB_MIN) prob_str = (prob == PROB_MIN) ? "min" : "never"; 1613 char prob_str_buf[30]; 1614 if (prob_str == NULL) { 1615 jio_snprintf(prob_str_buf, sizeof(prob_str_buf), "%20.2f", prob); 1616 prob_str = prob_str_buf; 1617 } 1618 C->log()->elem("branch target_bci='%d' taken='%d' not_taken='%d' cnt='%f' prob='%s'", 1619 iter().get_dest(), taken, not_taken, cnt, prob_str); 1620 } 1621 return prob; 1622 } 1623 1624 //-----------------------------branch_prediction------------------------------- 1625 float Parse::branch_prediction(float& cnt, 1626 BoolTest::mask btest, 1627 int target_bci, 1628 Node* test) { 1629 float prob = dynamic_branch_prediction(cnt, btest, test); 1630 // If prob is unknown, switch to static prediction 1631 if (prob != PROB_UNKNOWN) return prob; 1632 1633 prob = PROB_FAIR; // Set default value 1634 if (btest == BoolTest::eq) // Exactly equal test? 1635 prob = PROB_STATIC_INFREQUENT; // Assume its relatively infrequent 1636 else if (btest == BoolTest::ne) 1637 prob = PROB_STATIC_FREQUENT; // Assume its relatively frequent 1638 1639 // If this is a conditional test guarding a backwards branch, 1640 // assume its a loop-back edge. Make it a likely taken branch. 1641 if (target_bci < bci()) { 1642 if (is_osr_parse()) { // Could be a hot OSR'd loop; force deopt 1643 // Since it's an OSR, we probably have profile data, but since 1644 // branch_prediction returned PROB_UNKNOWN, the counts are too small. 1645 // Let's make a special check here for completely zero counts. 1646 ciMethodData* methodData = method()->method_data(); 1647 if (!methodData->is_empty()) { 1648 ciProfileData* data = methodData->bci_to_data(bci()); 1649 // Only stop for truly zero counts, which mean an unknown part 1650 // of the OSR-ed method, and we want to deopt to gather more stats. 1651 // If you have ANY counts, then this loop is simply 'cold' relative 1652 // to the OSR loop. 1653 if (data == NULL || 1654 (data->as_BranchData()->taken() + data->as_BranchData()->not_taken() == 0)) { 1655 // This is the only way to return PROB_UNKNOWN: 1656 return PROB_UNKNOWN; 1657 } 1658 } 1659 } 1660 prob = PROB_STATIC_FREQUENT; // Likely to take backwards branch 1661 } 1662 1663 assert(prob != PROB_UNKNOWN, "must have some guess at this point"); 1664 return prob; 1665 } 1666 1667 // The magic constants are chosen so as to match the output of 1668 // branch_prediction() when the profile reports a zero taken count. 1669 // It is important to distinguish zero counts unambiguously, because 1670 // some branches (e.g., _213_javac.Assembler.eliminate) validly produce 1671 // very small but nonzero probabilities, which if confused with zero 1672 // counts would keep the program recompiling indefinitely. 1673 bool Parse::seems_never_taken(float prob) const { 1674 return prob < PROB_MIN; 1675 } 1676 1677 // True if the comparison seems to be the kind that will not change its 1678 // statistics from true to false. See comments in adjust_map_after_if. 1679 // This question is only asked along paths which are already 1680 // classifed as untaken (by seems_never_taken), so really, 1681 // if a path is never taken, its controlling comparison is 1682 // already acting in a stable fashion. If the comparison 1683 // seems stable, we will put an expensive uncommon trap 1684 // on the untaken path. 1685 bool Parse::seems_stable_comparison() const { 1686 if (C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if)) { 1687 return false; 1688 } 1689 return true; 1690 } 1691 1692 //-------------------------------repush_if_args-------------------------------- 1693 // Push arguments of an "if" bytecode back onto the stack by adjusting _sp. 1694 inline int Parse::repush_if_args() { 1695 if (PrintOpto && WizardMode) { 1696 tty->print("defending against excessive implicit null exceptions on %s @%d in ", 1697 Bytecodes::name(iter().cur_bc()), iter().cur_bci()); 1698 method()->print_name(); tty->cr(); 1699 } 1700 int bc_depth = - Bytecodes::depth(iter().cur_bc()); 1701 assert(bc_depth == 1 || bc_depth == 2, "only two kinds of branches"); 1702 DEBUG_ONLY(sync_jvms()); // argument(n) requires a synced jvms 1703 assert(argument(0) != NULL, "must exist"); 1704 assert(bc_depth == 1 || argument(1) != NULL, "two must exist"); 1705 inc_sp(bc_depth); 1706 return bc_depth; 1707 } 1708 1709 //----------------------------------do_ifnull---------------------------------- 1710 void Parse::do_ifnull(BoolTest::mask btest, Node *c) { 1711 int target_bci = iter().get_dest(); 1712 1713 Block* branch_block = successor_for_bci(target_bci); 1714 Block* next_block = successor_for_bci(iter().next_bci()); 1715 1716 float cnt; 1717 float prob = branch_prediction(cnt, btest, target_bci, c); 1718 if (prob == PROB_UNKNOWN) { 1719 // (An earlier version of do_ifnull omitted this trap for OSR methods.) 1720 if (PrintOpto && Verbose) { 1721 tty->print_cr("Never-taken edge stops compilation at bci %d", bci()); 1722 } 1723 repush_if_args(); // to gather stats on loop 1724 // We need to mark this branch as taken so that if we recompile we will 1725 // see that it is possible. In the tiered system the interpreter doesn't 1726 // do profiling and by the time we get to the lower tier from the interpreter 1727 // the path may be cold again. Make sure it doesn't look untaken 1728 profile_taken_branch(target_bci, !ProfileInterpreter); 1729 uncommon_trap(Deoptimization::Reason_unreached, 1730 Deoptimization::Action_reinterpret, 1731 NULL, "cold"); 1732 if (C->eliminate_boxing()) { 1733 // Mark the successor blocks as parsed 1734 branch_block->next_path_num(); 1735 next_block->next_path_num(); 1736 } 1737 return; 1738 } 1739 1740 NOT_PRODUCT(explicit_null_checks_inserted++); 1741 1742 // Generate real control flow 1743 Node *tst = _gvn.transform( new BoolNode( c, btest ) ); 1744 1745 // Sanity check the probability value 1746 assert(prob > 0.0f,"Bad probability in Parser"); 1747 // Need xform to put node in hash table 1748 IfNode *iff = create_and_xform_if( control(), tst, prob, cnt ); 1749 assert(iff->_prob > 0.0f,"Optimizer made bad probability in parser"); 1750 // True branch 1751 { PreserveJVMState pjvms(this); 1752 Node* iftrue = _gvn.transform( new IfTrueNode (iff) ); 1753 set_control(iftrue); 1754 1755 if (stopped()) { // Path is dead? 1756 NOT_PRODUCT(explicit_null_checks_elided++); 1757 if (C->eliminate_boxing()) { 1758 // Mark the successor block as parsed 1759 branch_block->next_path_num(); 1760 } 1761 } else { // Path is live. 1762 // Update method data 1763 profile_taken_branch(target_bci); 1764 adjust_map_after_if(btest, c, prob, branch_block); 1765 if (!stopped()) { 1766 merge(target_bci); 1767 } 1768 } 1769 } 1770 1771 // False branch 1772 Node* iffalse = _gvn.transform( new IfFalseNode(iff) ); 1773 set_control(iffalse); 1774 1775 if (stopped()) { // Path is dead? 1776 NOT_PRODUCT(explicit_null_checks_elided++); 1777 if (C->eliminate_boxing()) { 1778 // Mark the successor block as parsed 1779 next_block->next_path_num(); 1780 } 1781 } else { // Path is live. 1782 // Update method data 1783 profile_not_taken_branch(); 1784 adjust_map_after_if(BoolTest(btest).negate(), c, 1.0-prob, next_block); 1785 } 1786 } 1787 1788 //------------------------------------do_if------------------------------------ 1789 void Parse::do_if(BoolTest::mask btest, Node* c, bool new_path, Node** ctrl_taken) { 1790 int target_bci = iter().get_dest(); 1791 1792 Block* branch_block = successor_for_bci(target_bci); 1793 Block* next_block = successor_for_bci(iter().next_bci()); 1794 1795 float cnt; 1796 float prob = branch_prediction(cnt, btest, target_bci, c); 1797 float untaken_prob = 1.0 - prob; 1798 1799 if (prob == PROB_UNKNOWN) { 1800 if (PrintOpto && Verbose) { 1801 tty->print_cr("Never-taken edge stops compilation at bci %d", bci()); 1802 } 1803 repush_if_args(); // to gather stats on loop 1804 // We need to mark this branch as taken so that if we recompile we will 1805 // see that it is possible. In the tiered system the interpreter doesn't 1806 // do profiling and by the time we get to the lower tier from the interpreter 1807 // the path may be cold again. Make sure it doesn't look untaken 1808 profile_taken_branch(target_bci, !ProfileInterpreter); 1809 uncommon_trap(Deoptimization::Reason_unreached, 1810 Deoptimization::Action_reinterpret, 1811 NULL, "cold"); 1812 if (C->eliminate_boxing()) { 1813 // Mark the successor blocks as parsed 1814 branch_block->next_path_num(); 1815 next_block->next_path_num(); 1816 } 1817 return; 1818 } 1819 1820 // Sanity check the probability value 1821 assert(0.0f < prob && prob < 1.0f,"Bad probability in Parser"); 1822 1823 bool taken_if_true = true; 1824 // Convert BoolTest to canonical form: 1825 if (!BoolTest(btest).is_canonical()) { 1826 btest = BoolTest(btest).negate(); 1827 taken_if_true = false; 1828 // prob is NOT updated here; it remains the probability of the taken 1829 // path (as opposed to the prob of the path guarded by an 'IfTrueNode'). 1830 } 1831 assert(btest != BoolTest::eq, "!= is the only canonical exact test"); 1832 1833 Node* tst0 = new BoolNode(c, btest); 1834 Node* tst = _gvn.transform(tst0); 1835 BoolTest::mask taken_btest = BoolTest::illegal; 1836 BoolTest::mask untaken_btest = BoolTest::illegal; 1837 1838 if (tst->is_Bool()) { 1839 // Refresh c from the transformed bool node, since it may be 1840 // simpler than the original c. Also re-canonicalize btest. 1841 // This wins when (Bool ne (Conv2B p) 0) => (Bool ne (CmpP p NULL)). 1842 // That can arise from statements like: if (x instanceof C) ... 1843 if (tst != tst0) { 1844 // Canonicalize one more time since transform can change it. 1845 btest = tst->as_Bool()->_test._test; 1846 if (!BoolTest(btest).is_canonical()) { 1847 // Reverse edges one more time... 1848 tst = _gvn.transform( tst->as_Bool()->negate(&_gvn) ); 1849 btest = tst->as_Bool()->_test._test; 1850 assert(BoolTest(btest).is_canonical(), "sanity"); 1851 taken_if_true = !taken_if_true; 1852 } 1853 c = tst->in(1); 1854 } 1855 BoolTest::mask neg_btest = BoolTest(btest).negate(); 1856 taken_btest = taken_if_true ? btest : neg_btest; 1857 untaken_btest = taken_if_true ? neg_btest : btest; 1858 } 1859 1860 // Generate real control flow 1861 float true_prob = (taken_if_true ? prob : untaken_prob); 1862 IfNode* iff = create_and_map_if(control(), tst, true_prob, cnt); 1863 assert(iff->_prob > 0.0f,"Optimizer made bad probability in parser"); 1864 Node* taken_branch = new IfTrueNode(iff); 1865 Node* untaken_branch = new IfFalseNode(iff); 1866 if (!taken_if_true) { // Finish conversion to canonical form 1867 Node* tmp = taken_branch; 1868 taken_branch = untaken_branch; 1869 untaken_branch = tmp; 1870 } 1871 1872 // Branch is taken: 1873 { PreserveJVMState pjvms(this); 1874 taken_branch = _gvn.transform(taken_branch); 1875 set_control(taken_branch); 1876 1877 if (stopped()) { 1878 if (C->eliminate_boxing() && !new_path) { 1879 // Mark the successor block as parsed (if we haven't created a new path) 1880 branch_block->next_path_num(); 1881 } 1882 } else { 1883 // Update method data 1884 profile_taken_branch(target_bci); 1885 adjust_map_after_if(taken_btest, c, prob, branch_block); 1886 if (!stopped()) { 1887 if (new_path) { 1888 // Merge by using a new path 1889 merge_new_path(target_bci); 1890 } else if (ctrl_taken != NULL) { 1891 // Don't merge but save taken branch to be wired by caller 1892 *ctrl_taken = control(); 1893 } else { 1894 merge(target_bci); 1895 } 1896 } 1897 } 1898 } 1899 1900 untaken_branch = _gvn.transform(untaken_branch); 1901 set_control(untaken_branch); 1902 1903 // Branch not taken. 1904 if (stopped() && ctrl_taken == NULL) { 1905 if (C->eliminate_boxing()) { 1906 // Mark the successor block as parsed (if caller does not re-wire control flow) 1907 next_block->next_path_num(); 1908 } 1909 } else { 1910 // Update method data 1911 profile_not_taken_branch(); 1912 adjust_map_after_if(untaken_btest, c, untaken_prob, next_block); 1913 } 1914 } 1915 1916 void Parse::do_acmp(BoolTest::mask btest, Node* a, Node* b) { 1917 ciMethod* subst_method = ciEnv::current()->ValueBootstrapMethods_klass()->find_method(ciSymbol::isSubstitutable_name(), ciSymbol::object_object_boolean_signature()); 1918 // If current method is ValueBootstrapMethods::isSubstitutable(), 1919 // compile the acmp as a regular pointer comparison otherwise we 1920 // could call ValueBootstrapMethods::isSubstitutable() back 1921 if (!EnableValhalla || (method() == subst_method)) { 1922 Node* cmp = CmpP(a, b); 1923 cmp = optimize_cmp_with_klass(cmp); 1924 do_if(btest, cmp); 1925 return; 1926 } 1927 1928 // Substitutability test 1929 if (a->is_ValueType()) { 1930 inc_sp(2); 1931 a = a->as_ValueType()->allocate(this, true)->get_oop(); 1932 dec_sp(2); 1933 } 1934 if (b->is_ValueType()) { 1935 inc_sp(2); 1936 b = b->as_ValueType()->allocate(this, true)->get_oop(); 1937 dec_sp(2); 1938 } 1939 1940 const TypeOopPtr* ta = _gvn.type(a)->isa_oopptr(); 1941 const TypeOopPtr* tb = _gvn.type(b)->isa_oopptr(); 1942 1943 if (ta == NULL || !ta->can_be_value_type_raw() || 1944 tb == NULL || !tb->can_be_value_type_raw()) { 1945 Node* cmp = CmpP(a, b); 1946 cmp = optimize_cmp_with_klass(cmp); 1947 do_if(btest, cmp); 1948 return; 1949 } 1950 1951 Node* cmp = CmpP(a, b); 1952 cmp = optimize_cmp_with_klass(cmp); 1953 Node* eq_region = NULL; 1954 if (btest == BoolTest::eq) { 1955 do_if(btest, cmp, true); 1956 if (stopped()) { 1957 return; 1958 } 1959 } else { 1960 assert(btest == BoolTest::ne, "only eq or ne"); 1961 Node* is_not_equal = NULL; 1962 eq_region = new RegionNode(3); 1963 { 1964 PreserveJVMState pjvms(this); 1965 do_if(btest, cmp, false, &is_not_equal); 1966 if (!stopped()) { 1967 eq_region->init_req(1, control()); 1968 } 1969 } 1970 if (is_not_equal == NULL || is_not_equal->is_top()) { 1971 record_for_igvn(eq_region); 1972 set_control(_gvn.transform(eq_region)); 1973 return; 1974 } 1975 set_control(is_not_equal); 1976 } 1977 // Pointers not equal, check for values 1978 Node* ne_region = new RegionNode(6); 1979 inc_sp(2); 1980 Node* null_ctl = top(); 1981 Node* not_null_a = null_check_oop(a, &null_ctl, !too_many_traps(Deoptimization::Reason_null_check), false, false); 1982 dec_sp(2); 1983 ne_region->init_req(1, null_ctl); 1984 if (stopped()) { 1985 record_for_igvn(ne_region); 1986 set_control(_gvn.transform(ne_region)); 1987 if (btest == BoolTest::ne) { 1988 { 1989 PreserveJVMState pjvms(this); 1990 int target_bci = iter().get_dest(); 1991 merge(target_bci); 1992 } 1993 record_for_igvn(eq_region); 1994 set_control(_gvn.transform(eq_region)); 1995 } 1996 return; 1997 } 1998 1999 Node* is_value = is_always_locked(not_null_a); 2000 Node* value_mask = _gvn.MakeConX(markWord::always_locked_pattern); 2001 Node* is_value_cmp = _gvn.transform(new CmpXNode(is_value, value_mask)); 2002 Node* is_value_bol = _gvn.transform(new BoolNode(is_value_cmp, BoolTest::ne)); 2003 IfNode* is_value_iff = create_and_map_if(control(), is_value_bol, PROB_FAIR, COUNT_UNKNOWN); 2004 Node* not_value = _gvn.transform(new IfTrueNode(is_value_iff)); 2005 set_control(_gvn.transform(new IfFalseNode(is_value_iff))); 2006 ne_region->init_req(2, not_value); 2007 2008 // One of the 2 pointers refers to a value, check if both are of 2009 // the same class 2010 inc_sp(2); 2011 null_ctl = top(); 2012 Node* not_null_b = null_check_oop(b, &null_ctl, !too_many_traps(Deoptimization::Reason_null_check), false, false); 2013 dec_sp(2); 2014 ne_region->init_req(3, null_ctl); 2015 if (stopped()) { 2016 record_for_igvn(ne_region); 2017 set_control(_gvn.transform(ne_region)); 2018 if (btest == BoolTest::ne) { 2019 { 2020 PreserveJVMState pjvms(this); 2021 int target_bci = iter().get_dest(); 2022 merge(target_bci); 2023 } 2024 record_for_igvn(eq_region); 2025 set_control(_gvn.transform(eq_region)); 2026 } 2027 return; 2028 } 2029 Node* kls_a = load_object_klass(not_null_a); 2030 Node* kls_b = load_object_klass(not_null_b); 2031 Node* kls_cmp = CmpP(kls_a, kls_b); 2032 Node* kls_bol = _gvn.transform(new BoolNode(kls_cmp, BoolTest::ne)); 2033 IfNode* kls_iff = create_and_map_if(control(), kls_bol, PROB_FAIR, COUNT_UNKNOWN); 2034 Node* kls_ne = _gvn.transform(new IfTrueNode(kls_iff)); 2035 set_control(_gvn.transform(new IfFalseNode(kls_iff))); 2036 ne_region->init_req(4, kls_ne); 2037 2038 if (stopped()) { 2039 record_for_igvn(ne_region); 2040 set_control(_gvn.transform(ne_region)); 2041 if (btest == BoolTest::ne) { 2042 { 2043 PreserveJVMState pjvms(this); 2044 int target_bci = iter().get_dest(); 2045 merge(target_bci); 2046 } 2047 record_for_igvn(eq_region); 2048 set_control(_gvn.transform(eq_region)); 2049 } 2050 return; 2051 } 2052 // Both are values of the same class, we need to perform a 2053 // substitutability test. Delegate to 2054 // ValueBootstrapMethods::isSubstitutable(). 2055 2056 Node* ne_io_phi = PhiNode::make(ne_region, i_o()); 2057 Node* mem = reset_memory(); 2058 Node* ne_mem_phi = PhiNode::make(ne_region, mem); 2059 2060 Node* eq_io_phi = NULL; 2061 Node* eq_mem_phi = NULL; 2062 if (eq_region != NULL) { 2063 eq_io_phi = PhiNode::make(eq_region, i_o()); 2064 eq_mem_phi = PhiNode::make(eq_region, mem); 2065 } 2066 2067 set_all_memory(mem); 2068 2069 kill_dead_locals(); 2070 CallStaticJavaNode *call = new CallStaticJavaNode(C, TypeFunc::make(subst_method), SharedRuntime::get_resolve_static_call_stub(), subst_method, bci()); 2071 call->set_override_symbolic_info(true); 2072 call->init_req(TypeFunc::Parms, not_null_a); 2073 call->init_req(TypeFunc::Parms+1, not_null_b); 2074 inc_sp(2); 2075 set_edges_for_java_call(call, false, false); 2076 Node* ret = set_results_for_java_call(call, false, true); 2077 dec_sp(2); 2078 2079 // Test the return value of ValueBootstrapMethods::isSubstitutable() 2080 Node* subst_cmp = _gvn.transform(new CmpINode(ret, intcon(1))); 2081 Node* ctl = C->top(); 2082 if (btest == BoolTest::eq) { 2083 PreserveJVMState pjvms(this); 2084 do_if(btest, subst_cmp); 2085 if (!stopped()) { 2086 ctl = control(); 2087 } 2088 } else { 2089 assert(btest == BoolTest::ne, "only eq or ne"); 2090 PreserveJVMState pjvms(this); 2091 do_if(btest, subst_cmp, false, &ctl); 2092 if (!stopped()) { 2093 eq_region->init_req(2, control()); 2094 eq_io_phi->init_req(2, i_o()); 2095 eq_mem_phi->init_req(2, reset_memory()); 2096 } 2097 } 2098 ne_region->init_req(5, ctl); 2099 ne_io_phi->init_req(5, i_o()); 2100 ne_mem_phi->init_req(5, reset_memory()); 2101 2102 record_for_igvn(ne_region); 2103 set_control(_gvn.transform(ne_region)); 2104 set_i_o(_gvn.transform(ne_io_phi)); 2105 set_all_memory(_gvn.transform(ne_mem_phi)); 2106 2107 if (btest == BoolTest::ne) { 2108 { 2109 PreserveJVMState pjvms(this); 2110 int target_bci = iter().get_dest(); 2111 merge(target_bci); 2112 } 2113 2114 record_for_igvn(eq_region); 2115 set_control(_gvn.transform(eq_region)); 2116 set_i_o(_gvn.transform(eq_io_phi)); 2117 set_all_memory(_gvn.transform(eq_mem_phi)); 2118 } 2119 } 2120 2121 bool Parse::path_is_suitable_for_uncommon_trap(float prob) const { 2122 // Don't want to speculate on uncommon traps when running with -Xcomp 2123 if (!UseInterpreter) { 2124 return false; 2125 } 2126 return (seems_never_taken(prob) && seems_stable_comparison()); 2127 } 2128 2129 void Parse::maybe_add_predicate_after_if(Block* path) { 2130 if (path->is_SEL_head() && path->preds_parsed() == 0) { 2131 // Add predicates at bci of if dominating the loop so traps can be 2132 // recorded on the if's profile data 2133 int bc_depth = repush_if_args(); 2134 add_predicate(); 2135 dec_sp(bc_depth); 2136 path->set_has_predicates(); 2137 } 2138 } 2139 2140 2141 //----------------------------adjust_map_after_if------------------------------ 2142 // Adjust the JVM state to reflect the result of taking this path. 2143 // Basically, it means inspecting the CmpNode controlling this 2144 // branch, seeing how it constrains a tested value, and then 2145 // deciding if it's worth our while to encode this constraint 2146 // as graph nodes in the current abstract interpretation map. 2147 void Parse::adjust_map_after_if(BoolTest::mask btest, Node* c, float prob, Block* path) { 2148 if (!c->is_Cmp()) { 2149 maybe_add_predicate_after_if(path); 2150 return; 2151 } 2152 2153 if (stopped() || btest == BoolTest::illegal) { 2154 return; // nothing to do 2155 } 2156 2157 bool is_fallthrough = (path == successor_for_bci(iter().next_bci())); 2158 2159 if (path_is_suitable_for_uncommon_trap(prob)) { 2160 repush_if_args(); 2161 uncommon_trap(Deoptimization::Reason_unstable_if, 2162 Deoptimization::Action_reinterpret, 2163 NULL, 2164 (is_fallthrough ? "taken always" : "taken never")); 2165 return; 2166 } 2167 2168 Node* val = c->in(1); 2169 Node* con = c->in(2); 2170 const Type* tcon = _gvn.type(con); 2171 const Type* tval = _gvn.type(val); 2172 bool have_con = tcon->singleton(); 2173 if (tval->singleton()) { 2174 if (!have_con) { 2175 // Swap, so constant is in con. 2176 con = val; 2177 tcon = tval; 2178 val = c->in(2); 2179 tval = _gvn.type(val); 2180 btest = BoolTest(btest).commute(); 2181 have_con = true; 2182 } else { 2183 // Do we have two constants? Then leave well enough alone. 2184 have_con = false; 2185 } 2186 } 2187 if (!have_con) { // remaining adjustments need a con 2188 maybe_add_predicate_after_if(path); 2189 return; 2190 } 2191 2192 sharpen_type_after_if(btest, con, tcon, val, tval); 2193 maybe_add_predicate_after_if(path); 2194 } 2195 2196 2197 static Node* extract_obj_from_klass_load(PhaseGVN* gvn, Node* n) { 2198 Node* ldk; 2199 if (n->is_DecodeNKlass()) { 2200 if (n->in(1)->Opcode() != Op_LoadNKlass) { 2201 return NULL; 2202 } else { 2203 ldk = n->in(1); 2204 } 2205 } else if (n->Opcode() != Op_LoadKlass) { 2206 return NULL; 2207 } else { 2208 ldk = n; 2209 } 2210 assert(ldk != NULL && ldk->is_Load(), "should have found a LoadKlass or LoadNKlass node"); 2211 2212 Node* adr = ldk->in(MemNode::Address); 2213 intptr_t off = 0; 2214 Node* obj = AddPNode::Ideal_base_and_offset(adr, gvn, off); 2215 if (obj == NULL || off != oopDesc::klass_offset_in_bytes()) // loading oopDesc::_klass? 2216 return NULL; 2217 const TypePtr* tp = gvn->type(obj)->is_ptr(); 2218 if (tp == NULL || !(tp->isa_instptr() || tp->isa_aryptr())) // is obj a Java object ptr? 2219 return NULL; 2220 2221 return obj; 2222 } 2223 2224 void Parse::sharpen_type_after_if(BoolTest::mask btest, 2225 Node* con, const Type* tcon, 2226 Node* val, const Type* tval) { 2227 // Look for opportunities to sharpen the type of a node 2228 // whose klass is compared with a constant klass. 2229 if (btest == BoolTest::eq && tcon->isa_klassptr()) { 2230 Node* obj = extract_obj_from_klass_load(&_gvn, val); 2231 const TypeOopPtr* con_type = tcon->isa_klassptr()->as_instance_type(); 2232 if (obj != NULL && (con_type->isa_instptr() || con_type->isa_aryptr())) { 2233 // Found: 2234 // Bool(CmpP(LoadKlass(obj._klass), ConP(Foo.klass)), [eq]) 2235 // or the narrowOop equivalent. 2236 const Type* obj_type = _gvn.type(obj); 2237 const TypeOopPtr* tboth = obj_type->join_speculative(con_type)->isa_oopptr(); 2238 if (tboth != NULL && tboth->klass_is_exact() && tboth != obj_type && 2239 tboth->higher_equal(obj_type)) { 2240 // obj has to be of the exact type Foo if the CmpP succeeds. 2241 int obj_in_map = map()->find_edge(obj); 2242 JVMState* jvms = this->jvms(); 2243 if (obj_in_map >= 0 && 2244 (jvms->is_loc(obj_in_map) || jvms->is_stk(obj_in_map))) { 2245 TypeNode* ccast = new CheckCastPPNode(control(), obj, tboth); 2246 const Type* tcc = ccast->as_Type()->type(); 2247 assert(tcc != obj_type && tcc->higher_equal(obj_type), "must improve"); 2248 // Delay transform() call to allow recovery of pre-cast value 2249 // at the control merge. 2250 _gvn.set_type_bottom(ccast); 2251 record_for_igvn(ccast); 2252 // Here's the payoff. 2253 replace_in_map(obj, ccast); 2254 } 2255 } 2256 } 2257 } 2258 2259 int val_in_map = map()->find_edge(val); 2260 if (val_in_map < 0) return; // replace_in_map would be useless 2261 { 2262 JVMState* jvms = this->jvms(); 2263 if (!(jvms->is_loc(val_in_map) || 2264 jvms->is_stk(val_in_map))) 2265 return; // again, it would be useless 2266 } 2267 2268 // Check for a comparison to a constant, and "know" that the compared 2269 // value is constrained on this path. 2270 assert(tcon->singleton(), ""); 2271 ConstraintCastNode* ccast = NULL; 2272 Node* cast = NULL; 2273 2274 switch (btest) { 2275 case BoolTest::eq: // Constant test? 2276 { 2277 const Type* tboth = tcon->join_speculative(tval); 2278 if (tboth == tval) break; // Nothing to gain. 2279 if (tcon->isa_int()) { 2280 ccast = new CastIINode(val, tboth); 2281 } else if (tcon == TypePtr::NULL_PTR) { 2282 // Cast to null, but keep the pointer identity temporarily live. 2283 ccast = new CastPPNode(val, tboth); 2284 } else { 2285 const TypeF* tf = tcon->isa_float_constant(); 2286 const TypeD* td = tcon->isa_double_constant(); 2287 // Exclude tests vs float/double 0 as these could be 2288 // either +0 or -0. Just because you are equal to +0 2289 // doesn't mean you ARE +0! 2290 // Note, following code also replaces Long and Oop values. 2291 if ((!tf || tf->_f != 0.0) && 2292 (!td || td->_d != 0.0)) 2293 cast = con; // Replace non-constant val by con. 2294 } 2295 } 2296 break; 2297 2298 case BoolTest::ne: 2299 if (tcon == TypePtr::NULL_PTR) { 2300 cast = cast_not_null(val, false); 2301 } 2302 break; 2303 2304 default: 2305 // (At this point we could record int range types with CastII.) 2306 break; 2307 } 2308 2309 if (ccast != NULL) { 2310 const Type* tcc = ccast->as_Type()->type(); 2311 assert(tcc != tval && tcc->higher_equal(tval), "must improve"); 2312 // Delay transform() call to allow recovery of pre-cast value 2313 // at the control merge. 2314 ccast->set_req(0, control()); 2315 _gvn.set_type_bottom(ccast); 2316 record_for_igvn(ccast); 2317 cast = ccast; 2318 } 2319 2320 if (cast != NULL) { // Here's the payoff. 2321 replace_in_map(val, cast); 2322 } 2323 } 2324 2325 /** 2326 * Use speculative type to optimize CmpP node: if comparison is 2327 * against the low level class, cast the object to the speculative 2328 * type if any. CmpP should then go away. 2329 * 2330 * @param c expected CmpP node 2331 * @return result of CmpP on object casted to speculative type 2332 * 2333 */ 2334 Node* Parse::optimize_cmp_with_klass(Node* c) { 2335 // If this is transformed by the _gvn to a comparison with the low 2336 // level klass then we may be able to use speculation 2337 if (c->Opcode() == Op_CmpP && 2338 (c->in(1)->Opcode() == Op_LoadKlass || c->in(1)->Opcode() == Op_DecodeNKlass) && 2339 c->in(2)->is_Con()) { 2340 Node* load_klass = NULL; 2341 Node* decode = NULL; 2342 if (c->in(1)->Opcode() == Op_DecodeNKlass) { 2343 decode = c->in(1); 2344 load_klass = c->in(1)->in(1); 2345 } else { 2346 load_klass = c->in(1); 2347 } 2348 if (load_klass->in(2)->is_AddP()) { 2349 Node* addp = load_klass->in(2); 2350 Node* obj = addp->in(AddPNode::Address); 2351 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr(); 2352 if (obj_type->speculative_type_not_null() != NULL) { 2353 ciKlass* k = obj_type->speculative_type(); 2354 inc_sp(2); 2355 obj = maybe_cast_profiled_obj(obj, k); 2356 dec_sp(2); 2357 if (obj->is_ValueType()) { 2358 assert(obj->as_ValueType()->is_allocated(&_gvn), "must be allocated"); 2359 obj = obj->as_ValueType()->get_oop(); 2360 } 2361 // Make the CmpP use the casted obj 2362 addp = basic_plus_adr(obj, addp->in(AddPNode::Offset)); 2363 load_klass = load_klass->clone(); 2364 load_klass->set_req(2, addp); 2365 load_klass = _gvn.transform(load_klass); 2366 if (decode != NULL) { 2367 decode = decode->clone(); 2368 decode->set_req(1, load_klass); 2369 load_klass = _gvn.transform(decode); 2370 } 2371 c = c->clone(); 2372 c->set_req(1, load_klass); 2373 c = _gvn.transform(c); 2374 } 2375 } 2376 } 2377 return c; 2378 } 2379 2380 //------------------------------do_one_bytecode-------------------------------- 2381 // Parse this bytecode, and alter the Parsers JVM->Node mapping 2382 void Parse::do_one_bytecode() { 2383 Node *a, *b, *c, *d; // Handy temps 2384 BoolTest::mask btest; 2385 int i; 2386 2387 assert(!has_exceptions(), "bytecode entry state must be clear of throws"); 2388 2389 if (C->check_node_count(NodeLimitFudgeFactor * 5, 2390 "out of nodes parsing method")) { 2391 return; 2392 } 2393 2394 #ifdef ASSERT 2395 // for setting breakpoints 2396 if (TraceOptoParse) { 2397 tty->print(" @"); 2398 dump_bci(bci()); 2399 tty->cr(); 2400 } 2401 #endif 2402 2403 switch (bc()) { 2404 case Bytecodes::_nop: 2405 // do nothing 2406 break; 2407 case Bytecodes::_lconst_0: 2408 push_pair(longcon(0)); 2409 break; 2410 2411 case Bytecodes::_lconst_1: 2412 push_pair(longcon(1)); 2413 break; 2414 2415 case Bytecodes::_fconst_0: 2416 push(zerocon(T_FLOAT)); 2417 break; 2418 2419 case Bytecodes::_fconst_1: 2420 push(makecon(TypeF::ONE)); 2421 break; 2422 2423 case Bytecodes::_fconst_2: 2424 push(makecon(TypeF::make(2.0f))); 2425 break; 2426 2427 case Bytecodes::_dconst_0: 2428 push_pair(zerocon(T_DOUBLE)); 2429 break; 2430 2431 case Bytecodes::_dconst_1: 2432 push_pair(makecon(TypeD::ONE)); 2433 break; 2434 2435 case Bytecodes::_iconst_m1:push(intcon(-1)); break; 2436 case Bytecodes::_iconst_0: push(intcon( 0)); break; 2437 case Bytecodes::_iconst_1: push(intcon( 1)); break; 2438 case Bytecodes::_iconst_2: push(intcon( 2)); break; 2439 case Bytecodes::_iconst_3: push(intcon( 3)); break; 2440 case Bytecodes::_iconst_4: push(intcon( 4)); break; 2441 case Bytecodes::_iconst_5: push(intcon( 5)); break; 2442 case Bytecodes::_bipush: push(intcon(iter().get_constant_u1())); break; 2443 case Bytecodes::_sipush: push(intcon(iter().get_constant_u2())); break; 2444 case Bytecodes::_aconst_null: push(null()); break; 2445 case Bytecodes::_ldc: 2446 case Bytecodes::_ldc_w: 2447 case Bytecodes::_ldc2_w: 2448 // If the constant is unresolved, run this BC once in the interpreter. 2449 { 2450 ciConstant constant = iter().get_constant(); 2451 if (!constant.is_valid() || 2452 (constant.basic_type() == T_OBJECT && 2453 !constant.as_object()->is_loaded())) { 2454 int index = iter().get_constant_pool_index(); 2455 constantTag tag = iter().get_constant_pool_tag(index); 2456 uncommon_trap(Deoptimization::make_trap_request 2457 (Deoptimization::Reason_unloaded, 2458 Deoptimization::Action_reinterpret, 2459 index), 2460 NULL, tag.internal_name()); 2461 break; 2462 } 2463 assert(constant.basic_type() != T_OBJECT || constant.as_object()->is_instance(), 2464 "must be java_mirror of klass"); 2465 const Type* con_type = Type::make_from_constant(constant); 2466 if (con_type != NULL) { 2467 push_node(con_type->basic_type(), makecon(con_type)); 2468 } 2469 } 2470 2471 break; 2472 2473 case Bytecodes::_aload_0: 2474 push( local(0) ); 2475 break; 2476 case Bytecodes::_aload_1: 2477 push( local(1) ); 2478 break; 2479 case Bytecodes::_aload_2: 2480 push( local(2) ); 2481 break; 2482 case Bytecodes::_aload_3: 2483 push( local(3) ); 2484 break; 2485 case Bytecodes::_aload: 2486 push( local(iter().get_index()) ); 2487 break; 2488 2489 case Bytecodes::_fload_0: 2490 case Bytecodes::_iload_0: 2491 push( local(0) ); 2492 break; 2493 case Bytecodes::_fload_1: 2494 case Bytecodes::_iload_1: 2495 push( local(1) ); 2496 break; 2497 case Bytecodes::_fload_2: 2498 case Bytecodes::_iload_2: 2499 push( local(2) ); 2500 break; 2501 case Bytecodes::_fload_3: 2502 case Bytecodes::_iload_3: 2503 push( local(3) ); 2504 break; 2505 case Bytecodes::_fload: 2506 case Bytecodes::_iload: 2507 push( local(iter().get_index()) ); 2508 break; 2509 case Bytecodes::_lload_0: 2510 push_pair_local( 0 ); 2511 break; 2512 case Bytecodes::_lload_1: 2513 push_pair_local( 1 ); 2514 break; 2515 case Bytecodes::_lload_2: 2516 push_pair_local( 2 ); 2517 break; 2518 case Bytecodes::_lload_3: 2519 push_pair_local( 3 ); 2520 break; 2521 case Bytecodes::_lload: 2522 push_pair_local( iter().get_index() ); 2523 break; 2524 2525 case Bytecodes::_dload_0: 2526 push_pair_local(0); 2527 break; 2528 case Bytecodes::_dload_1: 2529 push_pair_local(1); 2530 break; 2531 case Bytecodes::_dload_2: 2532 push_pair_local(2); 2533 break; 2534 case Bytecodes::_dload_3: 2535 push_pair_local(3); 2536 break; 2537 case Bytecodes::_dload: 2538 push_pair_local(iter().get_index()); 2539 break; 2540 case Bytecodes::_fstore_0: 2541 case Bytecodes::_istore_0: 2542 case Bytecodes::_astore_0: 2543 set_local( 0, pop() ); 2544 break; 2545 case Bytecodes::_fstore_1: 2546 case Bytecodes::_istore_1: 2547 case Bytecodes::_astore_1: 2548 set_local( 1, pop() ); 2549 break; 2550 case Bytecodes::_fstore_2: 2551 case Bytecodes::_istore_2: 2552 case Bytecodes::_astore_2: 2553 set_local( 2, pop() ); 2554 break; 2555 case Bytecodes::_fstore_3: 2556 case Bytecodes::_istore_3: 2557 case Bytecodes::_astore_3: 2558 set_local( 3, pop() ); 2559 break; 2560 case Bytecodes::_fstore: 2561 case Bytecodes::_istore: 2562 case Bytecodes::_astore: 2563 set_local( iter().get_index(), pop() ); 2564 break; 2565 // long stores 2566 case Bytecodes::_lstore_0: 2567 set_pair_local( 0, pop_pair() ); 2568 break; 2569 case Bytecodes::_lstore_1: 2570 set_pair_local( 1, pop_pair() ); 2571 break; 2572 case Bytecodes::_lstore_2: 2573 set_pair_local( 2, pop_pair() ); 2574 break; 2575 case Bytecodes::_lstore_3: 2576 set_pair_local( 3, pop_pair() ); 2577 break; 2578 case Bytecodes::_lstore: 2579 set_pair_local( iter().get_index(), pop_pair() ); 2580 break; 2581 2582 // double stores 2583 case Bytecodes::_dstore_0: 2584 set_pair_local( 0, dstore_rounding(pop_pair()) ); 2585 break; 2586 case Bytecodes::_dstore_1: 2587 set_pair_local( 1, dstore_rounding(pop_pair()) ); 2588 break; 2589 case Bytecodes::_dstore_2: 2590 set_pair_local( 2, dstore_rounding(pop_pair()) ); 2591 break; 2592 case Bytecodes::_dstore_3: 2593 set_pair_local( 3, dstore_rounding(pop_pair()) ); 2594 break; 2595 case Bytecodes::_dstore: 2596 set_pair_local( iter().get_index(), dstore_rounding(pop_pair()) ); 2597 break; 2598 2599 case Bytecodes::_pop: dec_sp(1); break; 2600 case Bytecodes::_pop2: dec_sp(2); break; 2601 case Bytecodes::_swap: 2602 a = pop(); 2603 b = pop(); 2604 push(a); 2605 push(b); 2606 break; 2607 case Bytecodes::_dup: 2608 a = pop(); 2609 push(a); 2610 push(a); 2611 break; 2612 case Bytecodes::_dup_x1: 2613 a = pop(); 2614 b = pop(); 2615 push( a ); 2616 push( b ); 2617 push( a ); 2618 break; 2619 case Bytecodes::_dup_x2: 2620 a = pop(); 2621 b = pop(); 2622 c = pop(); 2623 push( a ); 2624 push( c ); 2625 push( b ); 2626 push( a ); 2627 break; 2628 case Bytecodes::_dup2: 2629 a = pop(); 2630 b = pop(); 2631 push( b ); 2632 push( a ); 2633 push( b ); 2634 push( a ); 2635 break; 2636 2637 case Bytecodes::_dup2_x1: 2638 // before: .. c, b, a 2639 // after: .. b, a, c, b, a 2640 // not tested 2641 a = pop(); 2642 b = pop(); 2643 c = pop(); 2644 push( b ); 2645 push( a ); 2646 push( c ); 2647 push( b ); 2648 push( a ); 2649 break; 2650 case Bytecodes::_dup2_x2: 2651 // before: .. d, c, b, a 2652 // after: .. b, a, d, c, b, a 2653 // not tested 2654 a = pop(); 2655 b = pop(); 2656 c = pop(); 2657 d = pop(); 2658 push( b ); 2659 push( a ); 2660 push( d ); 2661 push( c ); 2662 push( b ); 2663 push( a ); 2664 break; 2665 2666 case Bytecodes::_arraylength: { 2667 // Must do null-check with value on expression stack 2668 Node *ary = null_check(peek(), T_ARRAY); 2669 // Compile-time detect of null-exception? 2670 if (stopped()) return; 2671 a = pop(); 2672 push(load_array_length(a)); 2673 break; 2674 } 2675 2676 case Bytecodes::_baload: array_load(T_BYTE); break; 2677 case Bytecodes::_caload: array_load(T_CHAR); break; 2678 case Bytecodes::_iaload: array_load(T_INT); break; 2679 case Bytecodes::_saload: array_load(T_SHORT); break; 2680 case Bytecodes::_faload: array_load(T_FLOAT); break; 2681 case Bytecodes::_aaload: array_load(T_OBJECT); break; 2682 case Bytecodes::_laload: array_load(T_LONG); break; 2683 case Bytecodes::_daload: array_load(T_DOUBLE); break; 2684 case Bytecodes::_bastore: array_store(T_BYTE); break; 2685 case Bytecodes::_castore: array_store(T_CHAR); break; 2686 case Bytecodes::_iastore: array_store(T_INT); break; 2687 case Bytecodes::_sastore: array_store(T_SHORT); break; 2688 case Bytecodes::_fastore: array_store(T_FLOAT); break; 2689 case Bytecodes::_aastore: array_store(T_OBJECT); break; 2690 case Bytecodes::_lastore: array_store(T_LONG); break; 2691 case Bytecodes::_dastore: array_store(T_DOUBLE); break; 2692 2693 case Bytecodes::_getfield: 2694 do_getfield(); 2695 break; 2696 2697 case Bytecodes::_getstatic: 2698 do_getstatic(); 2699 break; 2700 2701 case Bytecodes::_putfield: 2702 do_putfield(); 2703 break; 2704 2705 case Bytecodes::_putstatic: 2706 do_putstatic(); 2707 break; 2708 2709 case Bytecodes::_irem: 2710 do_irem(); 2711 break; 2712 case Bytecodes::_idiv: 2713 // Must keep both values on the expression-stack during null-check 2714 zero_check_int(peek()); 2715 // Compile-time detect of null-exception? 2716 if (stopped()) return; 2717 b = pop(); 2718 a = pop(); 2719 push( _gvn.transform( new DivINode(control(),a,b) ) ); 2720 break; 2721 case Bytecodes::_imul: 2722 b = pop(); a = pop(); 2723 push( _gvn.transform( new MulINode(a,b) ) ); 2724 break; 2725 case Bytecodes::_iadd: 2726 b = pop(); a = pop(); 2727 push( _gvn.transform( new AddINode(a,b) ) ); 2728 break; 2729 case Bytecodes::_ineg: 2730 a = pop(); 2731 push( _gvn.transform( new SubINode(_gvn.intcon(0),a)) ); 2732 break; 2733 case Bytecodes::_isub: 2734 b = pop(); a = pop(); 2735 push( _gvn.transform( new SubINode(a,b) ) ); 2736 break; 2737 case Bytecodes::_iand: 2738 b = pop(); a = pop(); 2739 push( _gvn.transform( new AndINode(a,b) ) ); 2740 break; 2741 case Bytecodes::_ior: 2742 b = pop(); a = pop(); 2743 push( _gvn.transform( new OrINode(a,b) ) ); 2744 break; 2745 case Bytecodes::_ixor: 2746 b = pop(); a = pop(); 2747 push( _gvn.transform( new XorINode(a,b) ) ); 2748 break; 2749 case Bytecodes::_ishl: 2750 b = pop(); a = pop(); 2751 push( _gvn.transform( new LShiftINode(a,b) ) ); 2752 break; 2753 case Bytecodes::_ishr: 2754 b = pop(); a = pop(); 2755 push( _gvn.transform( new RShiftINode(a,b) ) ); 2756 break; 2757 case Bytecodes::_iushr: 2758 b = pop(); a = pop(); 2759 push( _gvn.transform( new URShiftINode(a,b) ) ); 2760 break; 2761 2762 case Bytecodes::_fneg: 2763 a = pop(); 2764 b = _gvn.transform(new NegFNode (a)); 2765 push(b); 2766 break; 2767 2768 case Bytecodes::_fsub: 2769 b = pop(); 2770 a = pop(); 2771 c = _gvn.transform( new SubFNode(a,b) ); 2772 d = precision_rounding(c); 2773 push( d ); 2774 break; 2775 2776 case Bytecodes::_fadd: 2777 b = pop(); 2778 a = pop(); 2779 c = _gvn.transform( new AddFNode(a,b) ); 2780 d = precision_rounding(c); 2781 push( d ); 2782 break; 2783 2784 case Bytecodes::_fmul: 2785 b = pop(); 2786 a = pop(); 2787 c = _gvn.transform( new MulFNode(a,b) ); 2788 d = precision_rounding(c); 2789 push( d ); 2790 break; 2791 2792 case Bytecodes::_fdiv: 2793 b = pop(); 2794 a = pop(); 2795 c = _gvn.transform( new DivFNode(0,a,b) ); 2796 d = precision_rounding(c); 2797 push( d ); 2798 break; 2799 2800 case Bytecodes::_frem: 2801 if (Matcher::has_match_rule(Op_ModF)) { 2802 // Generate a ModF node. 2803 b = pop(); 2804 a = pop(); 2805 c = _gvn.transform( new ModFNode(0,a,b) ); 2806 d = precision_rounding(c); 2807 push( d ); 2808 } 2809 else { 2810 // Generate a call. 2811 modf(); 2812 } 2813 break; 2814 2815 case Bytecodes::_fcmpl: 2816 b = pop(); 2817 a = pop(); 2818 c = _gvn.transform( new CmpF3Node( a, b)); 2819 push(c); 2820 break; 2821 case Bytecodes::_fcmpg: 2822 b = pop(); 2823 a = pop(); 2824 2825 // Same as fcmpl but need to flip the unordered case. Swap the inputs, 2826 // which negates the result sign except for unordered. Flip the unordered 2827 // as well by using CmpF3 which implements unordered-lesser instead of 2828 // unordered-greater semantics. Finally, commute the result bits. Result 2829 // is same as using a CmpF3Greater except we did it with CmpF3 alone. 2830 c = _gvn.transform( new CmpF3Node( b, a)); 2831 c = _gvn.transform( new SubINode(_gvn.intcon(0),c) ); 2832 push(c); 2833 break; 2834 2835 case Bytecodes::_f2i: 2836 a = pop(); 2837 push(_gvn.transform(new ConvF2INode(a))); 2838 break; 2839 2840 case Bytecodes::_d2i: 2841 a = pop_pair(); 2842 b = _gvn.transform(new ConvD2INode(a)); 2843 push( b ); 2844 break; 2845 2846 case Bytecodes::_f2d: 2847 a = pop(); 2848 b = _gvn.transform( new ConvF2DNode(a)); 2849 push_pair( b ); 2850 break; 2851 2852 case Bytecodes::_d2f: 2853 a = pop_pair(); 2854 b = _gvn.transform( new ConvD2FNode(a)); 2855 // This breaks _227_mtrt (speed & correctness) and _222_mpegaudio (speed) 2856 //b = _gvn.transform(new RoundFloatNode(0, b) ); 2857 push( b ); 2858 break; 2859 2860 case Bytecodes::_l2f: 2861 if (Matcher::convL2FSupported()) { 2862 a = pop_pair(); 2863 b = _gvn.transform( new ConvL2FNode(a)); 2864 // For i486.ad, FILD doesn't restrict precision to 24 or 53 bits. 2865 // Rather than storing the result into an FP register then pushing 2866 // out to memory to round, the machine instruction that implements 2867 // ConvL2D is responsible for rounding. 2868 // c = precision_rounding(b); 2869 c = _gvn.transform(b); 2870 push(c); 2871 } else { 2872 l2f(); 2873 } 2874 break; 2875 2876 case Bytecodes::_l2d: 2877 a = pop_pair(); 2878 b = _gvn.transform( new ConvL2DNode(a)); 2879 // For i486.ad, rounding is always necessary (see _l2f above). 2880 // c = dprecision_rounding(b); 2881 c = _gvn.transform(b); 2882 push_pair(c); 2883 break; 2884 2885 case Bytecodes::_f2l: 2886 a = pop(); 2887 b = _gvn.transform( new ConvF2LNode(a)); 2888 push_pair(b); 2889 break; 2890 2891 case Bytecodes::_d2l: 2892 a = pop_pair(); 2893 b = _gvn.transform( new ConvD2LNode(a)); 2894 push_pair(b); 2895 break; 2896 2897 case Bytecodes::_dsub: 2898 b = pop_pair(); 2899 a = pop_pair(); 2900 c = _gvn.transform( new SubDNode(a,b) ); 2901 d = dprecision_rounding(c); 2902 push_pair( d ); 2903 break; 2904 2905 case Bytecodes::_dadd: 2906 b = pop_pair(); 2907 a = pop_pair(); 2908 c = _gvn.transform( new AddDNode(a,b) ); 2909 d = dprecision_rounding(c); 2910 push_pair( d ); 2911 break; 2912 2913 case Bytecodes::_dmul: 2914 b = pop_pair(); 2915 a = pop_pair(); 2916 c = _gvn.transform( new MulDNode(a,b) ); 2917 d = dprecision_rounding(c); 2918 push_pair( d ); 2919 break; 2920 2921 case Bytecodes::_ddiv: 2922 b = pop_pair(); 2923 a = pop_pair(); 2924 c = _gvn.transform( new DivDNode(0,a,b) ); 2925 d = dprecision_rounding(c); 2926 push_pair( d ); 2927 break; 2928 2929 case Bytecodes::_dneg: 2930 a = pop_pair(); 2931 b = _gvn.transform(new NegDNode (a)); 2932 push_pair(b); 2933 break; 2934 2935 case Bytecodes::_drem: 2936 if (Matcher::has_match_rule(Op_ModD)) { 2937 // Generate a ModD node. 2938 b = pop_pair(); 2939 a = pop_pair(); 2940 // a % b 2941 2942 c = _gvn.transform( new ModDNode(0,a,b) ); 2943 d = dprecision_rounding(c); 2944 push_pair( d ); 2945 } 2946 else { 2947 // Generate a call. 2948 modd(); 2949 } 2950 break; 2951 2952 case Bytecodes::_dcmpl: 2953 b = pop_pair(); 2954 a = pop_pair(); 2955 c = _gvn.transform( new CmpD3Node( a, b)); 2956 push(c); 2957 break; 2958 2959 case Bytecodes::_dcmpg: 2960 b = pop_pair(); 2961 a = pop_pair(); 2962 // Same as dcmpl but need to flip the unordered case. 2963 // Commute the inputs, which negates the result sign except for unordered. 2964 // Flip the unordered as well by using CmpD3 which implements 2965 // unordered-lesser instead of unordered-greater semantics. 2966 // Finally, negate the result bits. Result is same as using a 2967 // CmpD3Greater except we did it with CmpD3 alone. 2968 c = _gvn.transform( new CmpD3Node( b, a)); 2969 c = _gvn.transform( new SubINode(_gvn.intcon(0),c) ); 2970 push(c); 2971 break; 2972 2973 2974 // Note for longs -> lo word is on TOS, hi word is on TOS - 1 2975 case Bytecodes::_land: 2976 b = pop_pair(); 2977 a = pop_pair(); 2978 c = _gvn.transform( new AndLNode(a,b) ); 2979 push_pair(c); 2980 break; 2981 case Bytecodes::_lor: 2982 b = pop_pair(); 2983 a = pop_pair(); 2984 c = _gvn.transform( new OrLNode(a,b) ); 2985 push_pair(c); 2986 break; 2987 case Bytecodes::_lxor: 2988 b = pop_pair(); 2989 a = pop_pair(); 2990 c = _gvn.transform( new XorLNode(a,b) ); 2991 push_pair(c); 2992 break; 2993 2994 case Bytecodes::_lshl: 2995 b = pop(); // the shift count 2996 a = pop_pair(); // value to be shifted 2997 c = _gvn.transform( new LShiftLNode(a,b) ); 2998 push_pair(c); 2999 break; 3000 case Bytecodes::_lshr: 3001 b = pop(); // the shift count 3002 a = pop_pair(); // value to be shifted 3003 c = _gvn.transform( new RShiftLNode(a,b) ); 3004 push_pair(c); 3005 break; 3006 case Bytecodes::_lushr: 3007 b = pop(); // the shift count 3008 a = pop_pair(); // value to be shifted 3009 c = _gvn.transform( new URShiftLNode(a,b) ); 3010 push_pair(c); 3011 break; 3012 case Bytecodes::_lmul: 3013 b = pop_pair(); 3014 a = pop_pair(); 3015 c = _gvn.transform( new MulLNode(a,b) ); 3016 push_pair(c); 3017 break; 3018 3019 case Bytecodes::_lrem: 3020 // Must keep both values on the expression-stack during null-check 3021 assert(peek(0) == top(), "long word order"); 3022 zero_check_long(peek(1)); 3023 // Compile-time detect of null-exception? 3024 if (stopped()) return; 3025 b = pop_pair(); 3026 a = pop_pair(); 3027 c = _gvn.transform( new ModLNode(control(),a,b) ); 3028 push_pair(c); 3029 break; 3030 3031 case Bytecodes::_ldiv: 3032 // Must keep both values on the expression-stack during null-check 3033 assert(peek(0) == top(), "long word order"); 3034 zero_check_long(peek(1)); 3035 // Compile-time detect of null-exception? 3036 if (stopped()) return; 3037 b = pop_pair(); 3038 a = pop_pair(); 3039 c = _gvn.transform( new DivLNode(control(),a,b) ); 3040 push_pair(c); 3041 break; 3042 3043 case Bytecodes::_ladd: 3044 b = pop_pair(); 3045 a = pop_pair(); 3046 c = _gvn.transform( new AddLNode(a,b) ); 3047 push_pair(c); 3048 break; 3049 case Bytecodes::_lsub: 3050 b = pop_pair(); 3051 a = pop_pair(); 3052 c = _gvn.transform( new SubLNode(a,b) ); 3053 push_pair(c); 3054 break; 3055 case Bytecodes::_lcmp: 3056 // Safepoints are now inserted _before_ branches. The long-compare 3057 // bytecode painfully produces a 3-way value (-1,0,+1) which requires a 3058 // slew of control flow. These are usually followed by a CmpI vs zero and 3059 // a branch; this pattern then optimizes to the obvious long-compare and 3060 // branch. However, if the branch is backwards there's a Safepoint 3061 // inserted. The inserted Safepoint captures the JVM state at the 3062 // pre-branch point, i.e. it captures the 3-way value. Thus if a 3063 // long-compare is used to control a loop the debug info will force 3064 // computation of the 3-way value, even though the generated code uses a 3065 // long-compare and branch. We try to rectify the situation by inserting 3066 // a SafePoint here and have it dominate and kill the safepoint added at a 3067 // following backwards branch. At this point the JVM state merely holds 2 3068 // longs but not the 3-way value. 3069 if( UseLoopSafepoints ) { 3070 switch( iter().next_bc() ) { 3071 case Bytecodes::_ifgt: 3072 case Bytecodes::_iflt: 3073 case Bytecodes::_ifge: 3074 case Bytecodes::_ifle: 3075 case Bytecodes::_ifne: 3076 case Bytecodes::_ifeq: 3077 // If this is a backwards branch in the bytecodes, add Safepoint 3078 maybe_add_safepoint(iter().next_get_dest()); 3079 default: 3080 break; 3081 } 3082 } 3083 b = pop_pair(); 3084 a = pop_pair(); 3085 c = _gvn.transform( new CmpL3Node( a, b )); 3086 push(c); 3087 break; 3088 3089 case Bytecodes::_lneg: 3090 a = pop_pair(); 3091 b = _gvn.transform( new SubLNode(longcon(0),a)); 3092 push_pair(b); 3093 break; 3094 case Bytecodes::_l2i: 3095 a = pop_pair(); 3096 push( _gvn.transform( new ConvL2INode(a))); 3097 break; 3098 case Bytecodes::_i2l: 3099 a = pop(); 3100 b = _gvn.transform( new ConvI2LNode(a)); 3101 push_pair(b); 3102 break; 3103 case Bytecodes::_i2b: 3104 // Sign extend 3105 a = pop(); 3106 a = _gvn.transform( new LShiftINode(a,_gvn.intcon(24)) ); 3107 a = _gvn.transform( new RShiftINode(a,_gvn.intcon(24)) ); 3108 push( a ); 3109 break; 3110 case Bytecodes::_i2s: 3111 a = pop(); 3112 a = _gvn.transform( new LShiftINode(a,_gvn.intcon(16)) ); 3113 a = _gvn.transform( new RShiftINode(a,_gvn.intcon(16)) ); 3114 push( a ); 3115 break; 3116 case Bytecodes::_i2c: 3117 a = pop(); 3118 push( _gvn.transform( new AndINode(a,_gvn.intcon(0xFFFF)) ) ); 3119 break; 3120 3121 case Bytecodes::_i2f: 3122 a = pop(); 3123 b = _gvn.transform( new ConvI2FNode(a) ) ; 3124 c = precision_rounding(b); 3125 push (b); 3126 break; 3127 3128 case Bytecodes::_i2d: 3129 a = pop(); 3130 b = _gvn.transform( new ConvI2DNode(a)); 3131 push_pair(b); 3132 break; 3133 3134 case Bytecodes::_iinc: // Increment local 3135 i = iter().get_index(); // Get local index 3136 set_local( i, _gvn.transform( new AddINode( _gvn.intcon(iter().get_iinc_con()), local(i) ) ) ); 3137 break; 3138 3139 // Exit points of synchronized methods must have an unlock node 3140 case Bytecodes::_return: 3141 return_current(NULL); 3142 break; 3143 3144 case Bytecodes::_ireturn: 3145 case Bytecodes::_areturn: 3146 case Bytecodes::_freturn: 3147 return_current(pop()); 3148 break; 3149 case Bytecodes::_lreturn: 3150 return_current(pop_pair()); 3151 break; 3152 case Bytecodes::_dreturn: 3153 return_current(pop_pair()); 3154 break; 3155 3156 case Bytecodes::_athrow: 3157 // null exception oop throws NULL pointer exception 3158 null_check(peek()); 3159 if (stopped()) return; 3160 // Hook the thrown exception directly to subsequent handlers. 3161 if (BailoutToInterpreterForThrows) { 3162 // Keep method interpreted from now on. 3163 uncommon_trap(Deoptimization::Reason_unhandled, 3164 Deoptimization::Action_make_not_compilable); 3165 return; 3166 } 3167 if (env()->jvmti_can_post_on_exceptions()) { 3168 // check if we must post exception events, take uncommon trap if so (with must_throw = false) 3169 uncommon_trap_if_should_post_on_exceptions(Deoptimization::Reason_unhandled, false); 3170 } 3171 // Here if either can_post_on_exceptions or should_post_on_exceptions is false 3172 add_exception_state(make_exception_state(peek())); 3173 break; 3174 3175 case Bytecodes::_goto: // fall through 3176 case Bytecodes::_goto_w: { 3177 int target_bci = (bc() == Bytecodes::_goto) ? iter().get_dest() : iter().get_far_dest(); 3178 3179 // If this is a backwards branch in the bytecodes, add Safepoint 3180 maybe_add_safepoint(target_bci); 3181 3182 // Update method data 3183 profile_taken_branch(target_bci); 3184 3185 // Merge the current control into the target basic block 3186 merge(target_bci); 3187 3188 // See if we can get some profile data and hand it off to the next block 3189 Block *target_block = block()->successor_for_bci(target_bci); 3190 if (target_block->pred_count() != 1) break; 3191 ciMethodData* methodData = method()->method_data(); 3192 if (!methodData->is_mature()) break; 3193 ciProfileData* data = methodData->bci_to_data(bci()); 3194 assert(data != NULL && data->is_JumpData(), "need JumpData for taken branch"); 3195 int taken = ((ciJumpData*)data)->taken(); 3196 taken = method()->scale_count(taken); 3197 target_block->set_count(taken); 3198 break; 3199 } 3200 3201 case Bytecodes::_ifnull: btest = BoolTest::eq; goto handle_if_null; 3202 case Bytecodes::_ifnonnull: btest = BoolTest::ne; goto handle_if_null; 3203 handle_if_null: 3204 // If this is a backwards branch in the bytecodes, add Safepoint 3205 maybe_add_safepoint(iter().get_dest()); 3206 a = null(); 3207 b = pop(); 3208 if (b->is_ValueType()) { 3209 // Return constant false because 'b' is always non-null 3210 c = _gvn.makecon(TypeInt::CC_GT); 3211 } else { 3212 if (!_gvn.type(b)->speculative_maybe_null() && 3213 !too_many_traps(Deoptimization::Reason_speculate_null_check)) { 3214 inc_sp(1); 3215 Node* null_ctl = top(); 3216 b = null_check_oop(b, &null_ctl, true, true, true); 3217 assert(null_ctl->is_top(), "no null control here"); 3218 dec_sp(1); 3219 } else if (_gvn.type(b)->speculative_always_null() && 3220 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) { 3221 inc_sp(1); 3222 b = null_assert(b); 3223 dec_sp(1); 3224 } 3225 c = _gvn.transform( new CmpPNode(b, a) ); 3226 } 3227 do_ifnull(btest, c); 3228 break; 3229 3230 case Bytecodes::_if_acmpeq: btest = BoolTest::eq; goto handle_if_acmp; 3231 case Bytecodes::_if_acmpne: btest = BoolTest::ne; goto handle_if_acmp; 3232 handle_if_acmp: 3233 // If this is a backwards branch in the bytecodes, add Safepoint 3234 maybe_add_safepoint(iter().get_dest()); 3235 a = access_resolve(pop(), 0); 3236 b = access_resolve(pop(), 0); 3237 do_acmp(btest, a, b); 3238 break; 3239 3240 case Bytecodes::_ifeq: btest = BoolTest::eq; goto handle_ifxx; 3241 case Bytecodes::_ifne: btest = BoolTest::ne; goto handle_ifxx; 3242 case Bytecodes::_iflt: btest = BoolTest::lt; goto handle_ifxx; 3243 case Bytecodes::_ifle: btest = BoolTest::le; goto handle_ifxx; 3244 case Bytecodes::_ifgt: btest = BoolTest::gt; goto handle_ifxx; 3245 case Bytecodes::_ifge: btest = BoolTest::ge; goto handle_ifxx; 3246 handle_ifxx: 3247 // If this is a backwards branch in the bytecodes, add Safepoint 3248 maybe_add_safepoint(iter().get_dest()); 3249 a = _gvn.intcon(0); 3250 b = pop(); 3251 c = _gvn.transform( new CmpINode(b, a) ); 3252 do_if(btest, c); 3253 break; 3254 3255 case Bytecodes::_if_icmpeq: btest = BoolTest::eq; goto handle_if_icmp; 3256 case Bytecodes::_if_icmpne: btest = BoolTest::ne; goto handle_if_icmp; 3257 case Bytecodes::_if_icmplt: btest = BoolTest::lt; goto handle_if_icmp; 3258 case Bytecodes::_if_icmple: btest = BoolTest::le; goto handle_if_icmp; 3259 case Bytecodes::_if_icmpgt: btest = BoolTest::gt; goto handle_if_icmp; 3260 case Bytecodes::_if_icmpge: btest = BoolTest::ge; goto handle_if_icmp; 3261 handle_if_icmp: 3262 // If this is a backwards branch in the bytecodes, add Safepoint 3263 maybe_add_safepoint(iter().get_dest()); 3264 a = pop(); 3265 b = pop(); 3266 c = _gvn.transform( new CmpINode( b, a ) ); 3267 do_if(btest, c); 3268 break; 3269 3270 case Bytecodes::_tableswitch: 3271 do_tableswitch(); 3272 break; 3273 3274 case Bytecodes::_lookupswitch: 3275 do_lookupswitch(); 3276 break; 3277 3278 case Bytecodes::_invokestatic: 3279 case Bytecodes::_invokedynamic: 3280 case Bytecodes::_invokespecial: 3281 case Bytecodes::_invokevirtual: 3282 case Bytecodes::_invokeinterface: 3283 do_call(); 3284 break; 3285 case Bytecodes::_checkcast: 3286 do_checkcast(); 3287 break; 3288 case Bytecodes::_instanceof: 3289 do_instanceof(); 3290 break; 3291 case Bytecodes::_anewarray: 3292 do_newarray(); 3293 break; 3294 case Bytecodes::_newarray: 3295 do_newarray((BasicType)iter().get_index()); 3296 break; 3297 case Bytecodes::_multianewarray: 3298 do_multianewarray(); 3299 break; 3300 case Bytecodes::_new: 3301 do_new(); 3302 break; 3303 case Bytecodes::_defaultvalue: 3304 do_defaultvalue(); 3305 break; 3306 case Bytecodes::_withfield: 3307 do_withfield(); 3308 break; 3309 3310 case Bytecodes::_jsr: 3311 case Bytecodes::_jsr_w: 3312 do_jsr(); 3313 break; 3314 3315 case Bytecodes::_ret: 3316 do_ret(); 3317 break; 3318 3319 3320 case Bytecodes::_monitorenter: 3321 do_monitor_enter(); 3322 break; 3323 3324 case Bytecodes::_monitorexit: 3325 do_monitor_exit(); 3326 break; 3327 3328 case Bytecodes::_breakpoint: 3329 // Breakpoint set concurrently to compile 3330 // %%% use an uncommon trap? 3331 C->record_failure("breakpoint in method"); 3332 return; 3333 3334 default: 3335 #ifndef PRODUCT 3336 map()->dump(99); 3337 #endif 3338 tty->print("\nUnhandled bytecode %s\n", Bytecodes::name(bc()) ); 3339 ShouldNotReachHere(); 3340 } 3341 3342 #ifndef PRODUCT 3343 IdealGraphPrinter *printer = C->printer(); 3344 if (printer && printer->should_print(1)) { 3345 char buffer[256]; 3346 jio_snprintf(buffer, sizeof(buffer), "Bytecode %d: %s", bci(), Bytecodes::name(bc())); 3347 bool old = printer->traverse_outs(); 3348 printer->set_traverse_outs(true); 3349 printer->print_method(buffer, 4); 3350 printer->set_traverse_outs(old); 3351 } 3352 #endif 3353 }