1 /* 2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2014, Red Hat Inc. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 #include "precompiled.hpp" 27 #include "asm/macroAssembler.hpp" 28 #include "interpreter/interpreter.hpp" 29 #include "interpreter/interpreterRuntime.hpp" 30 #include "interpreter/interp_masm.hpp" 31 #include "interpreter/templateTable.hpp" 32 #include "memory/universe.inline.hpp" 33 #include "oops/methodData.hpp" 34 #include "oops/method.hpp" 35 #include "oops/objArrayKlass.hpp" 36 #include "oops/oop.inline.hpp" 37 #include "prims/methodHandles.hpp" 38 #include "runtime/sharedRuntime.hpp" 39 #include "runtime/stubRoutines.hpp" 40 #include "runtime/synchronizer.hpp" 41 42 #define __ _masm-> 43 44 // Platform-dependent initialization 45 46 void TemplateTable::pd_initialize() { 47 // No aarch64 specific initialization 48 } 49 50 // Address computation: local variables 51 52 static inline Address iaddress(int n) { 53 return Address(rlocals, Interpreter::local_offset_in_bytes(n)); 54 } 55 56 static inline Address laddress(int n) { 57 return iaddress(n + 1); 58 } 59 60 static inline Address faddress(int n) { 61 return iaddress(n); 62 } 63 64 static inline Address daddress(int n) { 65 return laddress(n); 66 } 67 68 static inline Address aaddress(int n) { 69 return iaddress(n); 70 } 71 72 static inline Address iaddress(Register r) { 73 return Address(rlocals, r, Address::lsl(3)); 74 } 75 76 static inline Address laddress(Register r, Register scratch, 77 InterpreterMacroAssembler* _masm) { 78 __ lea(scratch, Address(rlocals, r, Address::lsl(3))); 79 return Address(scratch, Interpreter::local_offset_in_bytes(1)); 80 } 81 82 static inline Address faddress(Register r) { 83 return iaddress(r); 84 } 85 86 static inline Address daddress(Register r, Register scratch, 87 InterpreterMacroAssembler* _masm) { 88 return laddress(r, scratch, _masm); 89 } 90 91 static inline Address aaddress(Register r) { 92 return iaddress(r); 93 } 94 95 static inline Address at_rsp() { 96 return Address(esp, 0); 97 } 98 99 // At top of Java expression stack which may be different than esp(). It 100 // isn't for category 1 objects. 101 static inline Address at_tos () { 102 return Address(esp, Interpreter::expr_offset_in_bytes(0)); 103 } 104 105 static inline Address at_tos_p1() { 106 return Address(esp, Interpreter::expr_offset_in_bytes(1)); 107 } 108 109 static inline Address at_tos_p2() { 110 return Address(esp, Interpreter::expr_offset_in_bytes(2)); 111 } 112 113 static inline Address at_tos_p3() { 114 return Address(esp, Interpreter::expr_offset_in_bytes(3)); 115 } 116 117 static inline Address at_tos_p4() { 118 return Address(esp, Interpreter::expr_offset_in_bytes(4)); 119 } 120 121 static inline Address at_tos_p5() { 122 return Address(esp, Interpreter::expr_offset_in_bytes(5)); 123 } 124 125 // Condition conversion 126 static Assembler::Condition j_not(TemplateTable::Condition cc) { 127 switch (cc) { 128 case TemplateTable::equal : return Assembler::NE; 129 case TemplateTable::not_equal : return Assembler::EQ; 130 case TemplateTable::less : return Assembler::GE; 131 case TemplateTable::less_equal : return Assembler::GT; 132 case TemplateTable::greater : return Assembler::LE; 133 case TemplateTable::greater_equal: return Assembler::LT; 134 } 135 ShouldNotReachHere(); 136 return Assembler::EQ; 137 } 138 139 140 // Miscelaneous helper routines 141 // Store an oop (or NULL) at the Address described by obj. 142 // If val == noreg this means store a NULL 143 static void do_oop_store(InterpreterMacroAssembler* _masm, 144 Address obj, 145 Register val, 146 BarrierSet::Name barrier, 147 bool precise) { 148 assert(val == noreg || val == r0, "parameter is just for looks"); 149 switch (barrier) { 150 #if INCLUDE_ALL_GCS 151 case BarrierSet::G1SATBCTLogging: 152 { 153 // flatten object address if needed 154 if (obj.index() == noreg && obj.offset() == 0) { 155 if (obj.base() != r3) { 156 __ mov(r3, obj.base()); 157 } 158 } else { 159 __ lea(r3, obj); 160 } 161 __ g1_write_barrier_pre(r3 /* obj */, 162 r1 /* pre_val */, 163 rthread /* thread */, 164 r10 /* tmp */, 165 val != noreg /* tosca_live */, 166 false /* expand_call */); 167 if (val == noreg) { 168 __ store_heap_oop_null(Address(r3, 0)); 169 } else { 170 // G1 barrier needs uncompressed oop for region cross check. 171 Register new_val = val; 172 if (UseCompressedOops) { 173 new_val = rscratch2; 174 __ mov(new_val, val); 175 } 176 __ store_heap_oop(Address(r3, 0), val); 177 __ g1_write_barrier_post(r3 /* store_adr */, 178 new_val /* new_val */, 179 rthread /* thread */, 180 r10 /* tmp */, 181 r1 /* tmp2 */); 182 } 183 184 } 185 break; 186 #endif // INCLUDE_ALL_GCS 187 case BarrierSet::CardTableForRS: 188 case BarrierSet::CardTableExtension: 189 { 190 if (val == noreg) { 191 __ store_heap_oop_null(obj); 192 } else { 193 __ store_heap_oop(obj, val); 194 // flatten object address if needed 195 if (!precise || (obj.index() == noreg && obj.offset() == 0)) { 196 __ store_check(obj.base()); 197 } else { 198 __ lea(r3, obj); 199 __ store_check(r3); 200 } 201 } 202 } 203 break; 204 case BarrierSet::ModRef: 205 if (val == noreg) { 206 __ store_heap_oop_null(obj); 207 } else { 208 __ store_heap_oop(obj, val); 209 } 210 break; 211 default : 212 ShouldNotReachHere(); 213 214 } 215 } 216 217 Address TemplateTable::at_bcp(int offset) { 218 assert(_desc->uses_bcp(), "inconsistent uses_bcp information"); 219 return Address(rbcp, offset); 220 } 221 222 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg, 223 Register temp_reg, bool load_bc_into_bc_reg/*=true*/, 224 int byte_no) 225 { 226 if (!RewriteBytecodes) return; 227 Label L_patch_done; 228 229 switch (bc) { 230 case Bytecodes::_fast_aputfield: 231 case Bytecodes::_fast_bputfield: 232 case Bytecodes::_fast_zputfield: 233 case Bytecodes::_fast_cputfield: 234 case Bytecodes::_fast_dputfield: 235 case Bytecodes::_fast_fputfield: 236 case Bytecodes::_fast_iputfield: 237 case Bytecodes::_fast_lputfield: 238 case Bytecodes::_fast_sputfield: 239 { 240 // We skip bytecode quickening for putfield instructions when 241 // the put_code written to the constant pool cache is zero. 242 // This is required so that every execution of this instruction 243 // calls out to InterpreterRuntime::resolve_get_put to do 244 // additional, required work. 245 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 246 assert(load_bc_into_bc_reg, "we use bc_reg as temp"); 247 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1); 248 __ movw(bc_reg, bc); 249 __ cbzw(temp_reg, L_patch_done); // don't patch 250 } 251 break; 252 default: 253 assert(byte_no == -1, "sanity"); 254 // the pair bytecodes have already done the load. 255 if (load_bc_into_bc_reg) { 256 __ movw(bc_reg, bc); 257 } 258 } 259 260 if (JvmtiExport::can_post_breakpoint()) { 261 Label L_fast_patch; 262 // if a breakpoint is present we can't rewrite the stream directly 263 __ load_unsigned_byte(temp_reg, at_bcp(0)); 264 __ cmpw(temp_reg, Bytecodes::_breakpoint); 265 __ br(Assembler::NE, L_fast_patch); 266 // Let breakpoint table handling rewrite to quicker bytecode 267 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg); 268 __ b(L_patch_done); 269 __ bind(L_fast_patch); 270 } 271 272 #ifdef ASSERT 273 Label L_okay; 274 __ load_unsigned_byte(temp_reg, at_bcp(0)); 275 __ cmpw(temp_reg, (int) Bytecodes::java_code(bc)); 276 __ br(Assembler::EQ, L_okay); 277 __ cmpw(temp_reg, bc_reg); 278 __ br(Assembler::EQ, L_okay); 279 __ stop("patching the wrong bytecode"); 280 __ bind(L_okay); 281 #endif 282 283 // patch bytecode 284 __ strb(bc_reg, at_bcp(0)); 285 __ bind(L_patch_done); 286 } 287 288 289 // Individual instructions 290 291 void TemplateTable::nop() { 292 transition(vtos, vtos); 293 // nothing to do 294 } 295 296 void TemplateTable::shouldnotreachhere() { 297 transition(vtos, vtos); 298 __ stop("shouldnotreachhere bytecode"); 299 } 300 301 void TemplateTable::aconst_null() 302 { 303 transition(vtos, atos); 304 __ mov(r0, 0); 305 } 306 307 void TemplateTable::iconst(int value) 308 { 309 transition(vtos, itos); 310 __ mov(r0, value); 311 } 312 313 void TemplateTable::lconst(int value) 314 { 315 __ mov(r0, value); 316 } 317 318 void TemplateTable::fconst(int value) 319 { 320 transition(vtos, ftos); 321 switch (value) { 322 case 0: 323 __ fmovs(v0, zr); 324 break; 325 case 1: 326 __ fmovs(v0, 1.0); 327 break; 328 case 2: 329 __ fmovs(v0, 2.0); 330 break; 331 default: 332 ShouldNotReachHere(); 333 break; 334 } 335 } 336 337 void TemplateTable::dconst(int value) 338 { 339 transition(vtos, dtos); 340 switch (value) { 341 case 0: 342 __ fmovd(v0, zr); 343 break; 344 case 1: 345 __ fmovd(v0, 1.0); 346 break; 347 case 2: 348 __ fmovd(v0, 2.0); 349 break; 350 default: 351 ShouldNotReachHere(); 352 break; 353 } 354 } 355 356 void TemplateTable::bipush() 357 { 358 transition(vtos, itos); 359 __ load_signed_byte32(r0, at_bcp(1)); 360 } 361 362 void TemplateTable::sipush() 363 { 364 transition(vtos, itos); 365 __ load_unsigned_short(r0, at_bcp(1)); 366 __ revw(r0, r0); 367 __ asrw(r0, r0, 16); 368 } 369 370 void TemplateTable::ldc(bool wide) 371 { 372 transition(vtos, vtos); 373 Label call_ldc, notFloat, notClass, Done; 374 375 if (wide) { 376 __ get_unsigned_2_byte_index_at_bcp(r1, 1); 377 } else { 378 __ load_unsigned_byte(r1, at_bcp(1)); 379 } 380 __ get_cpool_and_tags(r2, r0); 381 382 const int base_offset = ConstantPool::header_size() * wordSize; 383 const int tags_offset = Array<u1>::base_offset_in_bytes(); 384 385 // get type 386 __ add(r3, r1, tags_offset); 387 __ lea(r3, Address(r0, r3)); 388 __ ldarb(r3, r3); 389 390 // unresolved class - get the resolved class 391 __ cmp(r3, JVM_CONSTANT_UnresolvedClass); 392 __ br(Assembler::EQ, call_ldc); 393 394 // unresolved class in error state - call into runtime to throw the error 395 // from the first resolution attempt 396 __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError); 397 __ br(Assembler::EQ, call_ldc); 398 399 // resolved class - need to call vm to get java mirror of the class 400 __ cmp(r3, JVM_CONSTANT_Class); 401 __ br(Assembler::NE, notClass); 402 403 __ bind(call_ldc); 404 __ mov(c_rarg1, wide); 405 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1); 406 __ push_ptr(r0); 407 __ verify_oop(r0); 408 __ b(Done); 409 410 __ bind(notClass); 411 __ cmp(r3, JVM_CONSTANT_Float); 412 __ br(Assembler::NE, notFloat); 413 // ftos 414 __ adds(r1, r2, r1, Assembler::LSL, 3); 415 __ ldrs(v0, Address(r1, base_offset)); 416 __ push_f(); 417 __ b(Done); 418 419 __ bind(notFloat); 420 #ifdef ASSERT 421 { 422 Label L; 423 __ cmp(r3, JVM_CONSTANT_Integer); 424 __ br(Assembler::EQ, L); 425 // String and Object are rewritten to fast_aldc 426 __ stop("unexpected tag type in ldc"); 427 __ bind(L); 428 } 429 #endif 430 // itos JVM_CONSTANT_Integer only 431 __ adds(r1, r2, r1, Assembler::LSL, 3); 432 __ ldrw(r0, Address(r1, base_offset)); 433 __ push_i(r0); 434 __ bind(Done); 435 } 436 437 // Fast path for caching oop constants. 438 void TemplateTable::fast_aldc(bool wide) 439 { 440 transition(vtos, atos); 441 442 Register result = r0; 443 Register tmp = r1; 444 int index_size = wide ? sizeof(u2) : sizeof(u1); 445 446 Label resolved; 447 448 // We are resolved if the resolved reference cache entry contains a 449 // non-null object (String, MethodType, etc.) 450 assert_different_registers(result, tmp); 451 __ get_cache_index_at_bcp(tmp, 1, index_size); 452 __ load_resolved_reference_at_index(result, tmp); 453 __ cbnz(result, resolved); 454 455 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); 456 457 // first time invocation - must resolve first 458 __ mov(tmp, (int)bytecode()); 459 __ call_VM(result, entry, tmp); 460 461 __ bind(resolved); 462 463 if (VerifyOops) { 464 __ verify_oop(result); 465 } 466 } 467 468 void TemplateTable::ldc2_w() 469 { 470 transition(vtos, vtos); 471 Label Long, Done; 472 __ get_unsigned_2_byte_index_at_bcp(r0, 1); 473 474 __ get_cpool_and_tags(r1, r2); 475 const int base_offset = ConstantPool::header_size() * wordSize; 476 const int tags_offset = Array<u1>::base_offset_in_bytes(); 477 478 // get type 479 __ lea(r2, Address(r2, r0, Address::lsl(0))); 480 __ load_unsigned_byte(r2, Address(r2, tags_offset)); 481 __ cmpw(r2, (int)JVM_CONSTANT_Double); 482 __ br(Assembler::NE, Long); 483 // dtos 484 __ lea (r2, Address(r1, r0, Address::lsl(3))); 485 __ ldrd(v0, Address(r2, base_offset)); 486 __ push_d(); 487 __ b(Done); 488 489 __ bind(Long); 490 // ltos 491 __ lea(r0, Address(r1, r0, Address::lsl(3))); 492 __ ldr(r0, Address(r0, base_offset)); 493 __ push_l(); 494 495 __ bind(Done); 496 } 497 498 void TemplateTable::locals_index(Register reg, int offset) 499 { 500 __ ldrb(reg, at_bcp(offset)); 501 __ neg(reg, reg); 502 } 503 504 void TemplateTable::iload() { 505 iload_internal(); 506 } 507 508 void TemplateTable::nofast_iload() { 509 iload_internal(may_not_rewrite); 510 } 511 512 void TemplateTable::iload_internal(RewriteControl rc) { 513 transition(vtos, itos); 514 if (RewriteFrequentPairs && rc == may_rewrite) { 515 Label rewrite, done; 516 Register bc = r4; 517 518 // get next bytecode 519 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload))); 520 521 // if _iload, wait to rewrite to iload2. We only want to rewrite the 522 // last two iloads in a pair. Comparing against fast_iload means that 523 // the next bytecode is neither an iload or a caload, and therefore 524 // an iload pair. 525 __ cmpw(r1, Bytecodes::_iload); 526 __ br(Assembler::EQ, done); 527 528 // if _fast_iload rewrite to _fast_iload2 529 __ cmpw(r1, Bytecodes::_fast_iload); 530 __ movw(bc, Bytecodes::_fast_iload2); 531 __ br(Assembler::EQ, rewrite); 532 533 // if _caload rewrite to _fast_icaload 534 __ cmpw(r1, Bytecodes::_caload); 535 __ movw(bc, Bytecodes::_fast_icaload); 536 __ br(Assembler::EQ, rewrite); 537 538 // else rewrite to _fast_iload 539 __ movw(bc, Bytecodes::_fast_iload); 540 541 // rewrite 542 // bc: new bytecode 543 __ bind(rewrite); 544 patch_bytecode(Bytecodes::_iload, bc, r1, false); 545 __ bind(done); 546 547 } 548 549 // do iload, get the local value into tos 550 locals_index(r1); 551 __ ldr(r0, iaddress(r1)); 552 553 } 554 555 void TemplateTable::fast_iload2() 556 { 557 transition(vtos, itos); 558 locals_index(r1); 559 __ ldr(r0, iaddress(r1)); 560 __ push(itos); 561 locals_index(r1, 3); 562 __ ldr(r0, iaddress(r1)); 563 } 564 565 void TemplateTable::fast_iload() 566 { 567 transition(vtos, itos); 568 locals_index(r1); 569 __ ldr(r0, iaddress(r1)); 570 } 571 572 void TemplateTable::lload() 573 { 574 transition(vtos, ltos); 575 __ ldrb(r1, at_bcp(1)); 576 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord); 577 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1))); 578 } 579 580 void TemplateTable::fload() 581 { 582 transition(vtos, ftos); 583 locals_index(r1); 584 // n.b. we use ldrd here because this is a 64 bit slot 585 // this is comparable to the iload case 586 __ ldrd(v0, faddress(r1)); 587 } 588 589 void TemplateTable::dload() 590 { 591 transition(vtos, dtos); 592 __ ldrb(r1, at_bcp(1)); 593 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord); 594 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1))); 595 } 596 597 void TemplateTable::aload() 598 { 599 transition(vtos, atos); 600 locals_index(r1); 601 __ ldr(r0, iaddress(r1)); 602 } 603 604 void TemplateTable::locals_index_wide(Register reg) { 605 __ ldrh(reg, at_bcp(2)); 606 __ rev16w(reg, reg); 607 __ neg(reg, reg); 608 } 609 610 void TemplateTable::wide_iload() { 611 transition(vtos, itos); 612 locals_index_wide(r1); 613 __ ldr(r0, iaddress(r1)); 614 } 615 616 void TemplateTable::wide_lload() 617 { 618 transition(vtos, ltos); 619 __ ldrh(r1, at_bcp(2)); 620 __ rev16w(r1, r1); 621 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord); 622 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1))); 623 } 624 625 void TemplateTable::wide_fload() 626 { 627 transition(vtos, ftos); 628 locals_index_wide(r1); 629 // n.b. we use ldrd here because this is a 64 bit slot 630 // this is comparable to the iload case 631 __ ldrd(v0, faddress(r1)); 632 } 633 634 void TemplateTable::wide_dload() 635 { 636 transition(vtos, dtos); 637 __ ldrh(r1, at_bcp(2)); 638 __ rev16w(r1, r1); 639 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord); 640 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1))); 641 } 642 643 void TemplateTable::wide_aload() 644 { 645 transition(vtos, atos); 646 locals_index_wide(r1); 647 __ ldr(r0, aaddress(r1)); 648 } 649 650 void TemplateTable::index_check(Register array, Register index) 651 { 652 // destroys r1, rscratch1 653 // check array 654 __ null_check(array, arrayOopDesc::length_offset_in_bytes()); 655 // sign extend index for use by indexed load 656 // __ movl2ptr(index, index); 657 // check index 658 Register length = rscratch1; 659 __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes())); 660 __ cmpw(index, length); 661 if (index != r1) { 662 // ??? convention: move aberrant index into r1 for exception message 663 assert(r1 != array, "different registers"); 664 __ mov(r1, index); 665 } 666 Label ok; 667 __ br(Assembler::LO, ok); 668 __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 669 __ br(rscratch1); 670 __ bind(ok); 671 } 672 673 void TemplateTable::iaload() 674 { 675 transition(itos, itos); 676 __ mov(r1, r0); 677 __ pop_ptr(r0); 678 // r0: array 679 // r1: index 680 index_check(r0, r1); // leaves index in r1, kills rscratch1 681 __ lea(r1, Address(r0, r1, Address::uxtw(2))); 682 __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT))); 683 } 684 685 void TemplateTable::laload() 686 { 687 transition(itos, ltos); 688 __ mov(r1, r0); 689 __ pop_ptr(r0); 690 // r0: array 691 // r1: index 692 index_check(r0, r1); // leaves index in r1, kills rscratch1 693 __ lea(r1, Address(r0, r1, Address::uxtw(3))); 694 __ ldr(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_LONG))); 695 } 696 697 void TemplateTable::faload() 698 { 699 transition(itos, ftos); 700 __ mov(r1, r0); 701 __ pop_ptr(r0); 702 // r0: array 703 // r1: index 704 index_check(r0, r1); // leaves index in r1, kills rscratch1 705 __ lea(r1, Address(r0, r1, Address::uxtw(2))); 706 __ ldrs(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 707 } 708 709 void TemplateTable::daload() 710 { 711 transition(itos, dtos); 712 __ mov(r1, r0); 713 __ pop_ptr(r0); 714 // r0: array 715 // r1: index 716 index_check(r0, r1); // leaves index in r1, kills rscratch1 717 __ lea(r1, Address(r0, r1, Address::uxtw(3))); 718 __ ldrd(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 719 } 720 721 void TemplateTable::aaload() 722 { 723 transition(itos, atos); 724 __ mov(r1, r0); 725 __ pop_ptr(r0); 726 // r0: array 727 // r1: index 728 index_check(r0, r1); // leaves index in r1, kills rscratch1 729 int s = (UseCompressedOops ? 2 : 3); 730 __ lea(r1, Address(r0, r1, Address::uxtw(s))); 731 __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 732 } 733 734 void TemplateTable::baload() 735 { 736 transition(itos, itos); 737 __ mov(r1, r0); 738 __ pop_ptr(r0); 739 // r0: array 740 // r1: index 741 index_check(r0, r1); // leaves index in r1, kills rscratch1 742 __ lea(r1, Address(r0, r1, Address::uxtw(0))); 743 __ load_signed_byte(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_BYTE))); 744 } 745 746 void TemplateTable::caload() 747 { 748 transition(itos, itos); 749 __ mov(r1, r0); 750 __ pop_ptr(r0); 751 // r0: array 752 // r1: index 753 index_check(r0, r1); // leaves index in r1, kills rscratch1 754 __ lea(r1, Address(r0, r1, Address::uxtw(1))); 755 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 756 } 757 758 // iload followed by caload frequent pair 759 void TemplateTable::fast_icaload() 760 { 761 transition(vtos, itos); 762 // load index out of locals 763 locals_index(r2); 764 __ ldr(r1, iaddress(r2)); 765 766 __ pop_ptr(r0); 767 768 // r0: array 769 // r1: index 770 index_check(r0, r1); // leaves index in r1, kills rscratch1 771 __ lea(r1, Address(r0, r1, Address::uxtw(1))); 772 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 773 } 774 775 void TemplateTable::saload() 776 { 777 transition(itos, itos); 778 __ mov(r1, r0); 779 __ pop_ptr(r0); 780 // r0: array 781 // r1: index 782 index_check(r0, r1); // leaves index in r1, kills rscratch1 783 __ lea(r1, Address(r0, r1, Address::uxtw(1))); 784 __ load_signed_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_SHORT))); 785 } 786 787 void TemplateTable::iload(int n) 788 { 789 transition(vtos, itos); 790 __ ldr(r0, iaddress(n)); 791 } 792 793 void TemplateTable::lload(int n) 794 { 795 transition(vtos, ltos); 796 __ ldr(r0, laddress(n)); 797 } 798 799 void TemplateTable::fload(int n) 800 { 801 transition(vtos, ftos); 802 __ ldrs(v0, faddress(n)); 803 } 804 805 void TemplateTable::dload(int n) 806 { 807 transition(vtos, dtos); 808 __ ldrd(v0, daddress(n)); 809 } 810 811 void TemplateTable::aload(int n) 812 { 813 transition(vtos, atos); 814 __ ldr(r0, iaddress(n)); 815 } 816 817 void TemplateTable::aload_0() { 818 aload_0_internal(); 819 } 820 821 void TemplateTable::nofast_aload_0() { 822 aload_0_internal(may_not_rewrite); 823 } 824 825 void TemplateTable::aload_0_internal(RewriteControl rc) { 826 // According to bytecode histograms, the pairs: 827 // 828 // _aload_0, _fast_igetfield 829 // _aload_0, _fast_agetfield 830 // _aload_0, _fast_fgetfield 831 // 832 // occur frequently. If RewriteFrequentPairs is set, the (slow) 833 // _aload_0 bytecode checks if the next bytecode is either 834 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then 835 // rewrites the current bytecode into a pair bytecode; otherwise it 836 // rewrites the current bytecode into _fast_aload_0 that doesn't do 837 // the pair check anymore. 838 // 839 // Note: If the next bytecode is _getfield, the rewrite must be 840 // delayed, otherwise we may miss an opportunity for a pair. 841 // 842 // Also rewrite frequent pairs 843 // aload_0, aload_1 844 // aload_0, iload_1 845 // These bytecodes with a small amount of code are most profitable 846 // to rewrite 847 if (RewriteFrequentPairs && rc == may_rewrite) { 848 Label rewrite, done; 849 const Register bc = r4; 850 851 // get next bytecode 852 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0))); 853 854 // if _getfield then wait with rewrite 855 __ cmpw(r1, Bytecodes::Bytecodes::_getfield); 856 __ br(Assembler::EQ, done); 857 858 // if _igetfield then rewrite to _fast_iaccess_0 859 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 860 __ cmpw(r1, Bytecodes::_fast_igetfield); 861 __ movw(bc, Bytecodes::_fast_iaccess_0); 862 __ br(Assembler::EQ, rewrite); 863 864 // if _agetfield then rewrite to _fast_aaccess_0 865 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 866 __ cmpw(r1, Bytecodes::_fast_agetfield); 867 __ movw(bc, Bytecodes::_fast_aaccess_0); 868 __ br(Assembler::EQ, rewrite); 869 870 // if _fgetfield then rewrite to _fast_faccess_0 871 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 872 __ cmpw(r1, Bytecodes::_fast_fgetfield); 873 __ movw(bc, Bytecodes::_fast_faccess_0); 874 __ br(Assembler::EQ, rewrite); 875 876 // else rewrite to _fast_aload0 877 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition"); 878 __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0); 879 880 // rewrite 881 // bc: new bytecode 882 __ bind(rewrite); 883 patch_bytecode(Bytecodes::_aload_0, bc, r1, false); 884 885 __ bind(done); 886 } 887 888 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop). 889 aload(0); 890 } 891 892 void TemplateTable::istore() 893 { 894 transition(itos, vtos); 895 locals_index(r1); 896 // FIXME: We're being very pernickerty here storing a jint in a 897 // local with strw, which costs an extra instruction over what we'd 898 // be able to do with a simple str. We should just store the whole 899 // word. 900 __ lea(rscratch1, iaddress(r1)); 901 __ strw(r0, Address(rscratch1)); 902 } 903 904 void TemplateTable::lstore() 905 { 906 transition(ltos, vtos); 907 locals_index(r1); 908 __ str(r0, laddress(r1, rscratch1, _masm)); 909 } 910 911 void TemplateTable::fstore() { 912 transition(ftos, vtos); 913 locals_index(r1); 914 __ lea(rscratch1, iaddress(r1)); 915 __ strs(v0, Address(rscratch1)); 916 } 917 918 void TemplateTable::dstore() { 919 transition(dtos, vtos); 920 locals_index(r1); 921 __ strd(v0, daddress(r1, rscratch1, _masm)); 922 } 923 924 void TemplateTable::astore() 925 { 926 transition(vtos, vtos); 927 __ pop_ptr(r0); 928 locals_index(r1); 929 __ str(r0, aaddress(r1)); 930 } 931 932 void TemplateTable::wide_istore() { 933 transition(vtos, vtos); 934 __ pop_i(); 935 locals_index_wide(r1); 936 __ lea(rscratch1, iaddress(r1)); 937 __ strw(r0, Address(rscratch1)); 938 } 939 940 void TemplateTable::wide_lstore() { 941 transition(vtos, vtos); 942 __ pop_l(); 943 locals_index_wide(r1); 944 __ str(r0, laddress(r1, rscratch1, _masm)); 945 } 946 947 void TemplateTable::wide_fstore() { 948 transition(vtos, vtos); 949 __ pop_f(); 950 locals_index_wide(r1); 951 __ lea(rscratch1, faddress(r1)); 952 __ strs(v0, rscratch1); 953 } 954 955 void TemplateTable::wide_dstore() { 956 transition(vtos, vtos); 957 __ pop_d(); 958 locals_index_wide(r1); 959 __ strd(v0, daddress(r1, rscratch1, _masm)); 960 } 961 962 void TemplateTable::wide_astore() { 963 transition(vtos, vtos); 964 __ pop_ptr(r0); 965 locals_index_wide(r1); 966 __ str(r0, aaddress(r1)); 967 } 968 969 void TemplateTable::iastore() { 970 transition(itos, vtos); 971 __ pop_i(r1); 972 __ pop_ptr(r3); 973 // r0: value 974 // r1: index 975 // r3: array 976 index_check(r3, r1); // prefer index in r1 977 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2))); 978 __ strw(r0, Address(rscratch1, 979 arrayOopDesc::base_offset_in_bytes(T_INT))); 980 } 981 982 void TemplateTable::lastore() { 983 transition(ltos, vtos); 984 __ pop_i(r1); 985 __ pop_ptr(r3); 986 // r0: value 987 // r1: index 988 // r3: array 989 index_check(r3, r1); // prefer index in r1 990 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3))); 991 __ str(r0, Address(rscratch1, 992 arrayOopDesc::base_offset_in_bytes(T_LONG))); 993 } 994 995 void TemplateTable::fastore() { 996 transition(ftos, vtos); 997 __ pop_i(r1); 998 __ pop_ptr(r3); 999 // v0: value 1000 // r1: index 1001 // r3: array 1002 index_check(r3, r1); // prefer index in r1 1003 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2))); 1004 __ strs(v0, Address(rscratch1, 1005 arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 1006 } 1007 1008 void TemplateTable::dastore() { 1009 transition(dtos, vtos); 1010 __ pop_i(r1); 1011 __ pop_ptr(r3); 1012 // v0: value 1013 // r1: index 1014 // r3: array 1015 index_check(r3, r1); // prefer index in r1 1016 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3))); 1017 __ strd(v0, Address(rscratch1, 1018 arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 1019 } 1020 1021 void TemplateTable::aastore() { 1022 Label is_null, ok_is_subtype, done; 1023 transition(vtos, vtos); 1024 // stack: ..., array, index, value 1025 __ ldr(r0, at_tos()); // value 1026 __ ldr(r2, at_tos_p1()); // index 1027 __ ldr(r3, at_tos_p2()); // array 1028 1029 Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 1030 1031 index_check(r3, r2); // kills r1 1032 __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3))); 1033 1034 // do array store check - check for NULL value first 1035 __ cbz(r0, is_null); 1036 1037 // Move subklass into r1 1038 __ load_klass(r1, r0); 1039 // Move superklass into r0 1040 __ load_klass(r0, r3); 1041 __ ldr(r0, Address(r0, 1042 ObjArrayKlass::element_klass_offset())); 1043 // Compress array + index*oopSize + 12 into a single register. Frees r2. 1044 1045 // Generate subtype check. Blows r2, r5 1046 // Superklass in r0. Subklass in r1. 1047 __ gen_subtype_check(r1, ok_is_subtype); 1048 1049 // Come here on failure 1050 // object is at TOS 1051 __ b(Interpreter::_throw_ArrayStoreException_entry); 1052 1053 // Come here on success 1054 __ bind(ok_is_subtype); 1055 1056 // Get the value we will store 1057 __ ldr(r0, at_tos()); 1058 // Now store using the appropriate barrier 1059 do_oop_store(_masm, element_address, r0, _bs->kind(), true); 1060 __ b(done); 1061 1062 // Have a NULL in r0, r3=array, r2=index. Store NULL at ary[idx] 1063 __ bind(is_null); 1064 __ profile_null_seen(r2); 1065 1066 // Store a NULL 1067 do_oop_store(_masm, element_address, noreg, _bs->kind(), true); 1068 1069 // Pop stack arguments 1070 __ bind(done); 1071 __ add(esp, esp, 3 * Interpreter::stackElementSize); 1072 } 1073 1074 void TemplateTable::bastore() 1075 { 1076 transition(itos, vtos); 1077 __ pop_i(r1); 1078 __ pop_ptr(r3); 1079 // r0: value 1080 // r1: index 1081 // r3: array 1082 index_check(r3, r1); // prefer index in r1 1083 1084 // Need to check whether array is boolean or byte 1085 // since both types share the bastore bytecode. 1086 __ load_klass(r2, r3); 1087 __ ldrw(r2, Address(r2, Klass::layout_helper_offset())); 1088 int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit()); 1089 Label L_skip; 1090 __ tbz(r2, diffbit_index, L_skip); 1091 __ andw(r0, r0, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1 1092 __ bind(L_skip); 1093 1094 __ lea(rscratch1, Address(r3, r1, Address::uxtw(0))); 1095 __ strb(r0, Address(rscratch1, 1096 arrayOopDesc::base_offset_in_bytes(T_BYTE))); 1097 } 1098 1099 void TemplateTable::castore() 1100 { 1101 transition(itos, vtos); 1102 __ pop_i(r1); 1103 __ pop_ptr(r3); 1104 // r0: value 1105 // r1: index 1106 // r3: array 1107 index_check(r3, r1); // prefer index in r1 1108 __ lea(rscratch1, Address(r3, r1, Address::uxtw(1))); 1109 __ strh(r0, Address(rscratch1, 1110 arrayOopDesc::base_offset_in_bytes(T_CHAR))); 1111 } 1112 1113 void TemplateTable::sastore() 1114 { 1115 castore(); 1116 } 1117 1118 void TemplateTable::istore(int n) 1119 { 1120 transition(itos, vtos); 1121 __ str(r0, iaddress(n)); 1122 } 1123 1124 void TemplateTable::lstore(int n) 1125 { 1126 transition(ltos, vtos); 1127 __ str(r0, laddress(n)); 1128 } 1129 1130 void TemplateTable::fstore(int n) 1131 { 1132 transition(ftos, vtos); 1133 __ strs(v0, faddress(n)); 1134 } 1135 1136 void TemplateTable::dstore(int n) 1137 { 1138 transition(dtos, vtos); 1139 __ strd(v0, daddress(n)); 1140 } 1141 1142 void TemplateTable::astore(int n) 1143 { 1144 transition(vtos, vtos); 1145 __ pop_ptr(r0); 1146 __ str(r0, iaddress(n)); 1147 } 1148 1149 void TemplateTable::pop() 1150 { 1151 transition(vtos, vtos); 1152 __ add(esp, esp, Interpreter::stackElementSize); 1153 } 1154 1155 void TemplateTable::pop2() 1156 { 1157 transition(vtos, vtos); 1158 __ add(esp, esp, 2 * Interpreter::stackElementSize); 1159 } 1160 1161 void TemplateTable::dup() 1162 { 1163 transition(vtos, vtos); 1164 __ ldr(r0, Address(esp, 0)); 1165 __ push(r0); 1166 // stack: ..., a, a 1167 } 1168 1169 void TemplateTable::dup_x1() 1170 { 1171 transition(vtos, vtos); 1172 // stack: ..., a, b 1173 __ ldr(r0, at_tos()); // load b 1174 __ ldr(r2, at_tos_p1()); // load a 1175 __ str(r0, at_tos_p1()); // store b 1176 __ str(r2, at_tos()); // store a 1177 __ push(r0); // push b 1178 // stack: ..., b, a, b 1179 } 1180 1181 void TemplateTable::dup_x2() 1182 { 1183 transition(vtos, vtos); 1184 // stack: ..., a, b, c 1185 __ ldr(r0, at_tos()); // load c 1186 __ ldr(r2, at_tos_p2()); // load a 1187 __ str(r0, at_tos_p2()); // store c in a 1188 __ push(r0); // push c 1189 // stack: ..., c, b, c, c 1190 __ ldr(r0, at_tos_p2()); // load b 1191 __ str(r2, at_tos_p2()); // store a in b 1192 // stack: ..., c, a, c, c 1193 __ str(r0, at_tos_p1()); // store b in c 1194 // stack: ..., c, a, b, c 1195 } 1196 1197 void TemplateTable::dup2() 1198 { 1199 transition(vtos, vtos); 1200 // stack: ..., a, b 1201 __ ldr(r0, at_tos_p1()); // load a 1202 __ push(r0); // push a 1203 __ ldr(r0, at_tos_p1()); // load b 1204 __ push(r0); // push b 1205 // stack: ..., a, b, a, b 1206 } 1207 1208 void TemplateTable::dup2_x1() 1209 { 1210 transition(vtos, vtos); 1211 // stack: ..., a, b, c 1212 __ ldr(r2, at_tos()); // load c 1213 __ ldr(r0, at_tos_p1()); // load b 1214 __ push(r0); // push b 1215 __ push(r2); // push c 1216 // stack: ..., a, b, c, b, c 1217 __ str(r2, at_tos_p3()); // store c in b 1218 // stack: ..., a, c, c, b, c 1219 __ ldr(r2, at_tos_p4()); // load a 1220 __ str(r2, at_tos_p2()); // store a in 2nd c 1221 // stack: ..., a, c, a, b, c 1222 __ str(r0, at_tos_p4()); // store b in a 1223 // stack: ..., b, c, a, b, c 1224 } 1225 1226 void TemplateTable::dup2_x2() 1227 { 1228 transition(vtos, vtos); 1229 // stack: ..., a, b, c, d 1230 __ ldr(r2, at_tos()); // load d 1231 __ ldr(r0, at_tos_p1()); // load c 1232 __ push(r0) ; // push c 1233 __ push(r2); // push d 1234 // stack: ..., a, b, c, d, c, d 1235 __ ldr(r0, at_tos_p4()); // load b 1236 __ str(r0, at_tos_p2()); // store b in d 1237 __ str(r2, at_tos_p4()); // store d in b 1238 // stack: ..., a, d, c, b, c, d 1239 __ ldr(r2, at_tos_p5()); // load a 1240 __ ldr(r0, at_tos_p3()); // load c 1241 __ str(r2, at_tos_p3()); // store a in c 1242 __ str(r0, at_tos_p5()); // store c in a 1243 // stack: ..., c, d, a, b, c, d 1244 } 1245 1246 void TemplateTable::swap() 1247 { 1248 transition(vtos, vtos); 1249 // stack: ..., a, b 1250 __ ldr(r2, at_tos_p1()); // load a 1251 __ ldr(r0, at_tos()); // load b 1252 __ str(r2, at_tos()); // store a in b 1253 __ str(r0, at_tos_p1()); // store b in a 1254 // stack: ..., b, a 1255 } 1256 1257 void TemplateTable::iop2(Operation op) 1258 { 1259 transition(itos, itos); 1260 // r0 <== r1 op r0 1261 __ pop_i(r1); 1262 switch (op) { 1263 case add : __ addw(r0, r1, r0); break; 1264 case sub : __ subw(r0, r1, r0); break; 1265 case mul : __ mulw(r0, r1, r0); break; 1266 case _and : __ andw(r0, r1, r0); break; 1267 case _or : __ orrw(r0, r1, r0); break; 1268 case _xor : __ eorw(r0, r1, r0); break; 1269 case shl : __ lslvw(r0, r1, r0); break; 1270 case shr : __ asrvw(r0, r1, r0); break; 1271 case ushr : __ lsrvw(r0, r1, r0);break; 1272 default : ShouldNotReachHere(); 1273 } 1274 } 1275 1276 void TemplateTable::lop2(Operation op) 1277 { 1278 transition(ltos, ltos); 1279 // r0 <== r1 op r0 1280 __ pop_l(r1); 1281 switch (op) { 1282 case add : __ add(r0, r1, r0); break; 1283 case sub : __ sub(r0, r1, r0); break; 1284 case mul : __ mul(r0, r1, r0); break; 1285 case _and : __ andr(r0, r1, r0); break; 1286 case _or : __ orr(r0, r1, r0); break; 1287 case _xor : __ eor(r0, r1, r0); break; 1288 default : ShouldNotReachHere(); 1289 } 1290 } 1291 1292 void TemplateTable::idiv() 1293 { 1294 transition(itos, itos); 1295 // explicitly check for div0 1296 Label no_div0; 1297 __ cbnzw(r0, no_div0); 1298 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry); 1299 __ br(rscratch1); 1300 __ bind(no_div0); 1301 __ pop_i(r1); 1302 // r0 <== r1 idiv r0 1303 __ corrected_idivl(r0, r1, r0, /* want_remainder */ false); 1304 } 1305 1306 void TemplateTable::irem() 1307 { 1308 transition(itos, itos); 1309 // explicitly check for div0 1310 Label no_div0; 1311 __ cbnzw(r0, no_div0); 1312 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry); 1313 __ br(rscratch1); 1314 __ bind(no_div0); 1315 __ pop_i(r1); 1316 // r0 <== r1 irem r0 1317 __ corrected_idivl(r0, r1, r0, /* want_remainder */ true); 1318 } 1319 1320 void TemplateTable::lmul() 1321 { 1322 transition(ltos, ltos); 1323 __ pop_l(r1); 1324 __ mul(r0, r0, r1); 1325 } 1326 1327 void TemplateTable::ldiv() 1328 { 1329 transition(ltos, ltos); 1330 // explicitly check for div0 1331 Label no_div0; 1332 __ cbnz(r0, no_div0); 1333 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry); 1334 __ br(rscratch1); 1335 __ bind(no_div0); 1336 __ pop_l(r1); 1337 // r0 <== r1 ldiv r0 1338 __ corrected_idivq(r0, r1, r0, /* want_remainder */ false); 1339 } 1340 1341 void TemplateTable::lrem() 1342 { 1343 transition(ltos, ltos); 1344 // explicitly check for div0 1345 Label no_div0; 1346 __ cbnz(r0, no_div0); 1347 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry); 1348 __ br(rscratch1); 1349 __ bind(no_div0); 1350 __ pop_l(r1); 1351 // r0 <== r1 lrem r0 1352 __ corrected_idivq(r0, r1, r0, /* want_remainder */ true); 1353 } 1354 1355 void TemplateTable::lshl() 1356 { 1357 transition(itos, ltos); 1358 // shift count is in r0 1359 __ pop_l(r1); 1360 __ lslv(r0, r1, r0); 1361 } 1362 1363 void TemplateTable::lshr() 1364 { 1365 transition(itos, ltos); 1366 // shift count is in r0 1367 __ pop_l(r1); 1368 __ asrv(r0, r1, r0); 1369 } 1370 1371 void TemplateTable::lushr() 1372 { 1373 transition(itos, ltos); 1374 // shift count is in r0 1375 __ pop_l(r1); 1376 __ lsrv(r0, r1, r0); 1377 } 1378 1379 void TemplateTable::fop2(Operation op) 1380 { 1381 transition(ftos, ftos); 1382 switch (op) { 1383 case add: 1384 // n.b. use ldrd because this is a 64 bit slot 1385 __ pop_f(v1); 1386 __ fadds(v0, v1, v0); 1387 break; 1388 case sub: 1389 __ pop_f(v1); 1390 __ fsubs(v0, v1, v0); 1391 break; 1392 case mul: 1393 __ pop_f(v1); 1394 __ fmuls(v0, v1, v0); 1395 break; 1396 case div: 1397 __ pop_f(v1); 1398 __ fdivs(v0, v1, v0); 1399 break; 1400 case rem: 1401 __ fmovs(v1, v0); 1402 __ pop_f(v0); 1403 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 1404 0, 2, MacroAssembler::ret_type_float); 1405 break; 1406 default: 1407 ShouldNotReachHere(); 1408 break; 1409 } 1410 } 1411 1412 void TemplateTable::dop2(Operation op) 1413 { 1414 transition(dtos, dtos); 1415 switch (op) { 1416 case add: 1417 // n.b. use ldrd because this is a 64 bit slot 1418 __ pop_d(v1); 1419 __ faddd(v0, v1, v0); 1420 break; 1421 case sub: 1422 __ pop_d(v1); 1423 __ fsubd(v0, v1, v0); 1424 break; 1425 case mul: 1426 __ pop_d(v1); 1427 __ fmuld(v0, v1, v0); 1428 break; 1429 case div: 1430 __ pop_d(v1); 1431 __ fdivd(v0, v1, v0); 1432 break; 1433 case rem: 1434 __ fmovd(v1, v0); 1435 __ pop_d(v0); 1436 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 1437 0, 2, MacroAssembler::ret_type_double); 1438 break; 1439 default: 1440 ShouldNotReachHere(); 1441 break; 1442 } 1443 } 1444 1445 void TemplateTable::ineg() 1446 { 1447 transition(itos, itos); 1448 __ negw(r0, r0); 1449 1450 } 1451 1452 void TemplateTable::lneg() 1453 { 1454 transition(ltos, ltos); 1455 __ neg(r0, r0); 1456 } 1457 1458 void TemplateTable::fneg() 1459 { 1460 transition(ftos, ftos); 1461 __ fnegs(v0, v0); 1462 } 1463 1464 void TemplateTable::dneg() 1465 { 1466 transition(dtos, dtos); 1467 __ fnegd(v0, v0); 1468 } 1469 1470 void TemplateTable::iinc() 1471 { 1472 transition(vtos, vtos); 1473 __ load_signed_byte(r1, at_bcp(2)); // get constant 1474 locals_index(r2); 1475 __ ldr(r0, iaddress(r2)); 1476 __ addw(r0, r0, r1); 1477 __ str(r0, iaddress(r2)); 1478 } 1479 1480 void TemplateTable::wide_iinc() 1481 { 1482 transition(vtos, vtos); 1483 // __ mov(r1, zr); 1484 __ ldrw(r1, at_bcp(2)); // get constant and index 1485 __ rev16(r1, r1); 1486 __ ubfx(r2, r1, 0, 16); 1487 __ neg(r2, r2); 1488 __ sbfx(r1, r1, 16, 16); 1489 __ ldr(r0, iaddress(r2)); 1490 __ addw(r0, r0, r1); 1491 __ str(r0, iaddress(r2)); 1492 } 1493 1494 void TemplateTable::convert() 1495 { 1496 // Checking 1497 #ifdef ASSERT 1498 { 1499 TosState tos_in = ilgl; 1500 TosState tos_out = ilgl; 1501 switch (bytecode()) { 1502 case Bytecodes::_i2l: // fall through 1503 case Bytecodes::_i2f: // fall through 1504 case Bytecodes::_i2d: // fall through 1505 case Bytecodes::_i2b: // fall through 1506 case Bytecodes::_i2c: // fall through 1507 case Bytecodes::_i2s: tos_in = itos; break; 1508 case Bytecodes::_l2i: // fall through 1509 case Bytecodes::_l2f: // fall through 1510 case Bytecodes::_l2d: tos_in = ltos; break; 1511 case Bytecodes::_f2i: // fall through 1512 case Bytecodes::_f2l: // fall through 1513 case Bytecodes::_f2d: tos_in = ftos; break; 1514 case Bytecodes::_d2i: // fall through 1515 case Bytecodes::_d2l: // fall through 1516 case Bytecodes::_d2f: tos_in = dtos; break; 1517 default : ShouldNotReachHere(); 1518 } 1519 switch (bytecode()) { 1520 case Bytecodes::_l2i: // fall through 1521 case Bytecodes::_f2i: // fall through 1522 case Bytecodes::_d2i: // fall through 1523 case Bytecodes::_i2b: // fall through 1524 case Bytecodes::_i2c: // fall through 1525 case Bytecodes::_i2s: tos_out = itos; break; 1526 case Bytecodes::_i2l: // fall through 1527 case Bytecodes::_f2l: // fall through 1528 case Bytecodes::_d2l: tos_out = ltos; break; 1529 case Bytecodes::_i2f: // fall through 1530 case Bytecodes::_l2f: // fall through 1531 case Bytecodes::_d2f: tos_out = ftos; break; 1532 case Bytecodes::_i2d: // fall through 1533 case Bytecodes::_l2d: // fall through 1534 case Bytecodes::_f2d: tos_out = dtos; break; 1535 default : ShouldNotReachHere(); 1536 } 1537 transition(tos_in, tos_out); 1538 } 1539 #endif // ASSERT 1540 // static const int64_t is_nan = 0x8000000000000000L; 1541 1542 // Conversion 1543 switch (bytecode()) { 1544 case Bytecodes::_i2l: 1545 __ sxtw(r0, r0); 1546 break; 1547 case Bytecodes::_i2f: 1548 __ scvtfws(v0, r0); 1549 break; 1550 case Bytecodes::_i2d: 1551 __ scvtfwd(v0, r0); 1552 break; 1553 case Bytecodes::_i2b: 1554 __ sxtbw(r0, r0); 1555 break; 1556 case Bytecodes::_i2c: 1557 __ uxthw(r0, r0); 1558 break; 1559 case Bytecodes::_i2s: 1560 __ sxthw(r0, r0); 1561 break; 1562 case Bytecodes::_l2i: 1563 __ uxtw(r0, r0); 1564 break; 1565 case Bytecodes::_l2f: 1566 __ scvtfs(v0, r0); 1567 break; 1568 case Bytecodes::_l2d: 1569 __ scvtfd(v0, r0); 1570 break; 1571 case Bytecodes::_f2i: 1572 { 1573 Label L_Okay; 1574 __ clear_fpsr(); 1575 __ fcvtzsw(r0, v0); 1576 __ get_fpsr(r1); 1577 __ cbzw(r1, L_Okay); 1578 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1579 0, 1, MacroAssembler::ret_type_integral); 1580 __ bind(L_Okay); 1581 } 1582 break; 1583 case Bytecodes::_f2l: 1584 { 1585 Label L_Okay; 1586 __ clear_fpsr(); 1587 __ fcvtzs(r0, v0); 1588 __ get_fpsr(r1); 1589 __ cbzw(r1, L_Okay); 1590 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1591 0, 1, MacroAssembler::ret_type_integral); 1592 __ bind(L_Okay); 1593 } 1594 break; 1595 case Bytecodes::_f2d: 1596 __ fcvts(v0, v0); 1597 break; 1598 case Bytecodes::_d2i: 1599 { 1600 Label L_Okay; 1601 __ clear_fpsr(); 1602 __ fcvtzdw(r0, v0); 1603 __ get_fpsr(r1); 1604 __ cbzw(r1, L_Okay); 1605 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1606 0, 1, MacroAssembler::ret_type_integral); 1607 __ bind(L_Okay); 1608 } 1609 break; 1610 case Bytecodes::_d2l: 1611 { 1612 Label L_Okay; 1613 __ clear_fpsr(); 1614 __ fcvtzd(r0, v0); 1615 __ get_fpsr(r1); 1616 __ cbzw(r1, L_Okay); 1617 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1618 0, 1, MacroAssembler::ret_type_integral); 1619 __ bind(L_Okay); 1620 } 1621 break; 1622 case Bytecodes::_d2f: 1623 __ fcvtd(v0, v0); 1624 break; 1625 default: 1626 ShouldNotReachHere(); 1627 } 1628 } 1629 1630 void TemplateTable::lcmp() 1631 { 1632 transition(ltos, itos); 1633 Label done; 1634 __ pop_l(r1); 1635 __ cmp(r1, r0); 1636 __ mov(r0, (u_int64_t)-1L); 1637 __ br(Assembler::LT, done); 1638 // __ mov(r0, 1UL); 1639 // __ csel(r0, r0, zr, Assembler::NE); 1640 // and here is a faster way 1641 __ csinc(r0, zr, zr, Assembler::EQ); 1642 __ bind(done); 1643 } 1644 1645 void TemplateTable::float_cmp(bool is_float, int unordered_result) 1646 { 1647 Label done; 1648 if (is_float) { 1649 // XXX get rid of pop here, use ... reg, mem32 1650 __ pop_f(v1); 1651 __ fcmps(v1, v0); 1652 } else { 1653 // XXX get rid of pop here, use ... reg, mem64 1654 __ pop_d(v1); 1655 __ fcmpd(v1, v0); 1656 } 1657 if (unordered_result < 0) { 1658 // we want -1 for unordered or less than, 0 for equal and 1 for 1659 // greater than. 1660 __ mov(r0, (u_int64_t)-1L); 1661 // for FP LT tests less than or unordered 1662 __ br(Assembler::LT, done); 1663 // install 0 for EQ otherwise 1 1664 __ csinc(r0, zr, zr, Assembler::EQ); 1665 } else { 1666 // we want -1 for less than, 0 for equal and 1 for unordered or 1667 // greater than. 1668 __ mov(r0, 1L); 1669 // for FP HI tests greater than or unordered 1670 __ br(Assembler::HI, done); 1671 // install 0 for EQ otherwise ~0 1672 __ csinv(r0, zr, zr, Assembler::EQ); 1673 1674 } 1675 __ bind(done); 1676 } 1677 1678 void TemplateTable::branch(bool is_jsr, bool is_wide) 1679 { 1680 // We might be moving to a safepoint. The thread which calls 1681 // Interpreter::notice_safepoints() will effectively flush its cache 1682 // when it makes a system call, but we need to do something to 1683 // ensure that we see the changed dispatch table. 1684 __ membar(MacroAssembler::LoadLoad); 1685 1686 __ profile_taken_branch(r0, r1); 1687 const ByteSize be_offset = MethodCounters::backedge_counter_offset() + 1688 InvocationCounter::counter_offset(); 1689 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() + 1690 InvocationCounter::counter_offset(); 1691 1692 // load branch displacement 1693 if (!is_wide) { 1694 __ ldrh(r2, at_bcp(1)); 1695 __ rev16(r2, r2); 1696 // sign extend the 16 bit value in r2 1697 __ sbfm(r2, r2, 0, 15); 1698 } else { 1699 __ ldrw(r2, at_bcp(1)); 1700 __ revw(r2, r2); 1701 // sign extend the 32 bit value in r2 1702 __ sbfm(r2, r2, 0, 31); 1703 } 1704 1705 // Handle all the JSR stuff here, then exit. 1706 // It's much shorter and cleaner than intermingling with the non-JSR 1707 // normal-branch stuff occurring below. 1708 1709 if (is_jsr) { 1710 // Pre-load the next target bytecode into rscratch1 1711 __ load_unsigned_byte(rscratch1, Address(rbcp, r2)); 1712 // compute return address as bci 1713 __ ldr(rscratch2, Address(rmethod, Method::const_offset())); 1714 __ add(rscratch2, rscratch2, 1715 in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3)); 1716 __ sub(r1, rbcp, rscratch2); 1717 __ push_i(r1); 1718 // Adjust the bcp by the 16-bit displacement in r2 1719 __ add(rbcp, rbcp, r2); 1720 __ dispatch_only(vtos, /*generate_poll*/true); 1721 return; 1722 } 1723 1724 // Normal (non-jsr) branch handling 1725 1726 // Adjust the bcp by the displacement in r2 1727 __ add(rbcp, rbcp, r2); 1728 1729 assert(UseLoopCounter || !UseOnStackReplacement, 1730 "on-stack-replacement requires loop counters"); 1731 Label backedge_counter_overflow; 1732 Label profile_method; 1733 Label dispatch; 1734 if (UseLoopCounter) { 1735 // increment backedge counter for backward branches 1736 // r0: MDO 1737 // w1: MDO bumped taken-count 1738 // r2: target offset 1739 __ cmp(r2, zr); 1740 __ br(Assembler::GT, dispatch); // count only if backward branch 1741 1742 // ECN: FIXME: This code smells 1743 // check if MethodCounters exists 1744 Label has_counters; 1745 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset())); 1746 __ cbnz(rscratch1, has_counters); 1747 __ push(r0); 1748 __ push(r1); 1749 __ push(r2); 1750 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 1751 InterpreterRuntime::build_method_counters), rmethod); 1752 __ pop(r2); 1753 __ pop(r1); 1754 __ pop(r0); 1755 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset())); 1756 __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory 1757 __ bind(has_counters); 1758 1759 if (TieredCompilation) { 1760 Label no_mdo; 1761 int increment = InvocationCounter::count_increment; 1762 if (ProfileInterpreter) { 1763 // Are we profiling? 1764 __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset()))); 1765 __ cbz(r1, no_mdo); 1766 // Increment the MDO backedge counter 1767 const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) + 1768 in_bytes(InvocationCounter::counter_offset())); 1769 const Address mask(r1, in_bytes(MethodData::backedge_mask_offset())); 1770 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, 1771 r0, rscratch1, false, Assembler::EQ, &backedge_counter_overflow); 1772 __ b(dispatch); 1773 } 1774 __ bind(no_mdo); 1775 // Increment backedge counter in MethodCounters* 1776 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset())); 1777 const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset())); 1778 __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask, 1779 r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow); 1780 } else { // not TieredCompilation 1781 // increment counter 1782 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset())); 1783 __ ldrw(r0, Address(rscratch2, be_offset)); // load backedge counter 1784 __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter 1785 __ strw(rscratch1, Address(rscratch2, be_offset)); // store counter 1786 1787 __ ldrw(r0, Address(rscratch2, inv_offset)); // load invocation counter 1788 __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits 1789 __ addw(r0, r0, rscratch1); // add both counters 1790 1791 if (ProfileInterpreter) { 1792 // Test to see if we should create a method data oop 1793 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset()))); 1794 __ cmpw(r0, rscratch1); 1795 __ br(Assembler::LT, dispatch); 1796 1797 // if no method data exists, go to profile method 1798 __ test_method_data_pointer(r0, profile_method); 1799 1800 if (UseOnStackReplacement) { 1801 // check for overflow against w1 which is the MDO taken count 1802 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()))); 1803 __ cmpw(r1, rscratch1); 1804 __ br(Assembler::LO, dispatch); // Intel == Assembler::below 1805 1806 // When ProfileInterpreter is on, the backedge_count comes 1807 // from the MethodData*, which value does not get reset on 1808 // the call to frequency_counter_overflow(). To avoid 1809 // excessive calls to the overflow routine while the method is 1810 // being compiled, add a second test to make sure the overflow 1811 // function is called only once every overflow_frequency. 1812 const int overflow_frequency = 1024; 1813 __ andsw(r1, r1, overflow_frequency - 1); 1814 __ br(Assembler::EQ, backedge_counter_overflow); 1815 1816 } 1817 } else { 1818 if (UseOnStackReplacement) { 1819 // check for overflow against w0, which is the sum of the 1820 // counters 1821 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()))); 1822 __ cmpw(r0, rscratch1); 1823 __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual 1824 } 1825 } 1826 } 1827 } 1828 __ bind(dispatch); 1829 1830 // Pre-load the next target bytecode into rscratch1 1831 __ load_unsigned_byte(rscratch1, Address(rbcp, 0)); 1832 1833 // continue with the bytecode @ target 1834 // rscratch1: target bytecode 1835 // rbcp: target bcp 1836 __ dispatch_only(vtos, /*generate_poll*/true); 1837 1838 if (UseLoopCounter) { 1839 if (ProfileInterpreter) { 1840 // Out-of-line code to allocate method data oop. 1841 __ bind(profile_method); 1842 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1843 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode 1844 __ set_method_data_pointer_for_bcp(); 1845 __ b(dispatch); 1846 } 1847 1848 if (TieredCompilation || UseOnStackReplacement) { 1849 // invocation counter overflow 1850 __ bind(backedge_counter_overflow); 1851 __ neg(r2, r2); 1852 __ add(r2, r2, rbcp); // branch bcp 1853 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp) 1854 __ call_VM(noreg, 1855 CAST_FROM_FN_PTR(address, 1856 InterpreterRuntime::frequency_counter_overflow), 1857 r2); 1858 if (!UseOnStackReplacement) 1859 __ b(dispatch); 1860 } 1861 1862 if (UseOnStackReplacement) { 1863 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode 1864 1865 // r0: osr nmethod (osr ok) or NULL (osr not possible) 1866 // w1: target bytecode 1867 // r2: scratch 1868 __ cbz(r0, dispatch); // test result -- no osr if null 1869 // nmethod may have been invalidated (VM may block upon call_VM return) 1870 __ ldrb(r2, Address(r0, nmethod::state_offset())); 1871 if (nmethod::in_use != 0) 1872 __ sub(r2, r2, nmethod::in_use); 1873 __ cbnz(r2, dispatch); 1874 1875 // We have the address of an on stack replacement routine in r0 1876 // We need to prepare to execute the OSR method. First we must 1877 // migrate the locals and monitors off of the stack. 1878 1879 __ mov(r19, r0); // save the nmethod 1880 1881 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin)); 1882 1883 // r0 is OSR buffer, move it to expected parameter location 1884 __ mov(j_rarg0, r0); 1885 1886 // remove activation 1887 // get sender esp 1888 __ ldr(esp, 1889 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1890 // remove frame anchor 1891 __ leave(); 1892 // Ensure compiled code always sees stack at proper alignment 1893 __ andr(sp, esp, -16); 1894 1895 // and begin the OSR nmethod 1896 __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset())); 1897 __ br(rscratch1); 1898 } 1899 } 1900 } 1901 1902 1903 void TemplateTable::if_0cmp(Condition cc) 1904 { 1905 transition(itos, vtos); 1906 // assume branch is more often taken than not (loops use backward branches) 1907 Label not_taken; 1908 if (cc == equal) 1909 __ cbnzw(r0, not_taken); 1910 else if (cc == not_equal) 1911 __ cbzw(r0, not_taken); 1912 else { 1913 __ andsw(zr, r0, r0); 1914 __ br(j_not(cc), not_taken); 1915 } 1916 1917 branch(false, false); 1918 __ bind(not_taken); 1919 __ profile_not_taken_branch(r0); 1920 } 1921 1922 void TemplateTable::if_icmp(Condition cc) 1923 { 1924 transition(itos, vtos); 1925 // assume branch is more often taken than not (loops use backward branches) 1926 Label not_taken; 1927 __ pop_i(r1); 1928 __ cmpw(r1, r0, Assembler::LSL); 1929 __ br(j_not(cc), not_taken); 1930 branch(false, false); 1931 __ bind(not_taken); 1932 __ profile_not_taken_branch(r0); 1933 } 1934 1935 void TemplateTable::if_nullcmp(Condition cc) 1936 { 1937 transition(atos, vtos); 1938 // assume branch is more often taken than not (loops use backward branches) 1939 Label not_taken; 1940 if (cc == equal) 1941 __ cbnz(r0, not_taken); 1942 else 1943 __ cbz(r0, not_taken); 1944 branch(false, false); 1945 __ bind(not_taken); 1946 __ profile_not_taken_branch(r0); 1947 } 1948 1949 void TemplateTable::if_acmp(Condition cc) 1950 { 1951 transition(atos, vtos); 1952 // assume branch is more often taken than not (loops use backward branches) 1953 Label not_taken; 1954 __ pop_ptr(r1); 1955 __ cmp(r1, r0); 1956 __ br(j_not(cc), not_taken); 1957 branch(false, false); 1958 __ bind(not_taken); 1959 __ profile_not_taken_branch(r0); 1960 } 1961 1962 void TemplateTable::ret() { 1963 transition(vtos, vtos); 1964 // We might be moving to a safepoint. The thread which calls 1965 // Interpreter::notice_safepoints() will effectively flush its cache 1966 // when it makes a system call, but we need to do something to 1967 // ensure that we see the changed dispatch table. 1968 __ membar(MacroAssembler::LoadLoad); 1969 1970 locals_index(r1); 1971 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp 1972 __ profile_ret(r1, r2); 1973 __ ldr(rbcp, Address(rmethod, Method::const_offset())); 1974 __ lea(rbcp, Address(rbcp, r1)); 1975 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); 1976 __ dispatch_next(vtos, 0, /*generate_poll*/true); 1977 } 1978 1979 void TemplateTable::wide_ret() { 1980 transition(vtos, vtos); 1981 locals_index_wide(r1); 1982 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp 1983 __ profile_ret(r1, r2); 1984 __ ldr(rbcp, Address(rmethod, Method::const_offset())); 1985 __ lea(rbcp, Address(rbcp, r1)); 1986 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); 1987 __ dispatch_next(vtos, 0, /*generate_poll*/true); 1988 } 1989 1990 1991 void TemplateTable::tableswitch() { 1992 Label default_case, continue_execution; 1993 transition(itos, vtos); 1994 // align rbcp 1995 __ lea(r1, at_bcp(BytesPerInt)); 1996 __ andr(r1, r1, -BytesPerInt); 1997 // load lo & hi 1998 __ ldrw(r2, Address(r1, BytesPerInt)); 1999 __ ldrw(r3, Address(r1, 2 * BytesPerInt)); 2000 __ rev32(r2, r2); 2001 __ rev32(r3, r3); 2002 // check against lo & hi 2003 __ cmpw(r0, r2); 2004 __ br(Assembler::LT, default_case); 2005 __ cmpw(r0, r3); 2006 __ br(Assembler::GT, default_case); 2007 // lookup dispatch offset 2008 __ subw(r0, r0, r2); 2009 __ lea(r3, Address(r1, r0, Address::uxtw(2))); 2010 __ ldrw(r3, Address(r3, 3 * BytesPerInt)); 2011 __ profile_switch_case(r0, r1, r2); 2012 // continue execution 2013 __ bind(continue_execution); 2014 __ rev32(r3, r3); 2015 __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0))); 2016 __ add(rbcp, rbcp, r3, ext::sxtw); 2017 __ dispatch_only(vtos, /*generate_poll*/true); 2018 // handle default 2019 __ bind(default_case); 2020 __ profile_switch_default(r0); 2021 __ ldrw(r3, Address(r1, 0)); 2022 __ b(continue_execution); 2023 } 2024 2025 void TemplateTable::lookupswitch() { 2026 transition(itos, itos); 2027 __ stop("lookupswitch bytecode should have been rewritten"); 2028 } 2029 2030 void TemplateTable::fast_linearswitch() { 2031 transition(itos, vtos); 2032 Label loop_entry, loop, found, continue_execution; 2033 // bswap r0 so we can avoid bswapping the table entries 2034 __ rev32(r0, r0); 2035 // align rbcp 2036 __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of 2037 // this instruction (change offsets 2038 // below) 2039 __ andr(r19, r19, -BytesPerInt); 2040 // set counter 2041 __ ldrw(r1, Address(r19, BytesPerInt)); 2042 __ rev32(r1, r1); 2043 __ b(loop_entry); 2044 // table search 2045 __ bind(loop); 2046 __ lea(rscratch1, Address(r19, r1, Address::lsl(3))); 2047 __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt)); 2048 __ cmpw(r0, rscratch1); 2049 __ br(Assembler::EQ, found); 2050 __ bind(loop_entry); 2051 __ subs(r1, r1, 1); 2052 __ br(Assembler::PL, loop); 2053 // default case 2054 __ profile_switch_default(r0); 2055 __ ldrw(r3, Address(r19, 0)); 2056 __ b(continue_execution); 2057 // entry found -> get offset 2058 __ bind(found); 2059 __ lea(rscratch1, Address(r19, r1, Address::lsl(3))); 2060 __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt)); 2061 __ profile_switch_case(r1, r0, r19); 2062 // continue execution 2063 __ bind(continue_execution); 2064 __ rev32(r3, r3); 2065 __ add(rbcp, rbcp, r3, ext::sxtw); 2066 __ ldrb(rscratch1, Address(rbcp, 0)); 2067 __ dispatch_only(vtos, /*generate_poll*/true); 2068 } 2069 2070 void TemplateTable::fast_binaryswitch() { 2071 transition(itos, vtos); 2072 // Implementation using the following core algorithm: 2073 // 2074 // int binary_search(int key, LookupswitchPair* array, int n) { 2075 // // Binary search according to "Methodik des Programmierens" by 2076 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985. 2077 // int i = 0; 2078 // int j = n; 2079 // while (i+1 < j) { 2080 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q) 2081 // // with Q: for all i: 0 <= i < n: key < a[i] 2082 // // where a stands for the array and assuming that the (inexisting) 2083 // // element a[n] is infinitely big. 2084 // int h = (i + j) >> 1; 2085 // // i < h < j 2086 // if (key < array[h].fast_match()) { 2087 // j = h; 2088 // } else { 2089 // i = h; 2090 // } 2091 // } 2092 // // R: a[i] <= key < a[i+1] or Q 2093 // // (i.e., if key is within array, i is the correct index) 2094 // return i; 2095 // } 2096 2097 // Register allocation 2098 const Register key = r0; // already set (tosca) 2099 const Register array = r1; 2100 const Register i = r2; 2101 const Register j = r3; 2102 const Register h = rscratch1; 2103 const Register temp = rscratch2; 2104 2105 // Find array start 2106 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to 2107 // get rid of this 2108 // instruction (change 2109 // offsets below) 2110 __ andr(array, array, -BytesPerInt); 2111 2112 // Initialize i & j 2113 __ mov(i, 0); // i = 0; 2114 __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array); 2115 2116 // Convert j into native byteordering 2117 __ rev32(j, j); 2118 2119 // And start 2120 Label entry; 2121 __ b(entry); 2122 2123 // binary search loop 2124 { 2125 Label loop; 2126 __ bind(loop); 2127 // int h = (i + j) >> 1; 2128 __ addw(h, i, j); // h = i + j; 2129 __ lsrw(h, h, 1); // h = (i + j) >> 1; 2130 // if (key < array[h].fast_match()) { 2131 // j = h; 2132 // } else { 2133 // i = h; 2134 // } 2135 // Convert array[h].match to native byte-ordering before compare 2136 __ ldr(temp, Address(array, h, Address::lsl(3))); 2137 __ rev32(temp, temp); 2138 __ cmpw(key, temp); 2139 // j = h if (key < array[h].fast_match()) 2140 __ csel(j, h, j, Assembler::LT); 2141 // i = h if (key >= array[h].fast_match()) 2142 __ csel(i, h, i, Assembler::GE); 2143 // while (i+1 < j) 2144 __ bind(entry); 2145 __ addw(h, i, 1); // i+1 2146 __ cmpw(h, j); // i+1 < j 2147 __ br(Assembler::LT, loop); 2148 } 2149 2150 // end of binary search, result index is i (must check again!) 2151 Label default_case; 2152 // Convert array[i].match to native byte-ordering before compare 2153 __ ldr(temp, Address(array, i, Address::lsl(3))); 2154 __ rev32(temp, temp); 2155 __ cmpw(key, temp); 2156 __ br(Assembler::NE, default_case); 2157 2158 // entry found -> j = offset 2159 __ add(j, array, i, ext::uxtx, 3); 2160 __ ldrw(j, Address(j, BytesPerInt)); 2161 __ profile_switch_case(i, key, array); 2162 __ rev32(j, j); 2163 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0))); 2164 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0))); 2165 __ dispatch_only(vtos, /*generate_poll*/true); 2166 2167 // default case -> j = default offset 2168 __ bind(default_case); 2169 __ profile_switch_default(i); 2170 __ ldrw(j, Address(array, -2 * BytesPerInt)); 2171 __ rev32(j, j); 2172 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0))); 2173 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0))); 2174 __ dispatch_only(vtos, /*generate_poll*/true); 2175 } 2176 2177 2178 void TemplateTable::_return(TosState state) 2179 { 2180 transition(state, state); 2181 assert(_desc->calls_vm(), 2182 "inconsistent calls_vm information"); // call in remove_activation 2183 2184 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) { 2185 assert(state == vtos, "only valid state"); 2186 2187 __ ldr(c_rarg1, aaddress(0)); 2188 __ load_klass(r3, c_rarg1); 2189 __ ldrw(r3, Address(r3, Klass::access_flags_offset())); 2190 Label skip_register_finalizer; 2191 __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer); 2192 2193 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1); 2194 2195 __ bind(skip_register_finalizer); 2196 } 2197 2198 // Issue a StoreStore barrier after all stores but before return 2199 // from any constructor for any class with a final field. We don't 2200 // know if this is a finalizer, so we always do so. 2201 if (_desc->bytecode() == Bytecodes::_return) 2202 __ membar(MacroAssembler::StoreStore); 2203 2204 // Narrow result if state is itos but result type is smaller. 2205 // Need to narrow in the return bytecode rather than in generate_return_entry 2206 // since compiled code callers expect the result to already be narrowed. 2207 if (state == itos) { 2208 __ narrow(r0); 2209 } 2210 2211 __ remove_activation(state); 2212 __ ret(lr); 2213 } 2214 2215 // ---------------------------------------------------------------------------- 2216 // Volatile variables demand their effects be made known to all CPU's 2217 // in order. Store buffers on most chips allow reads & writes to 2218 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode 2219 // without some kind of memory barrier (i.e., it's not sufficient that 2220 // the interpreter does not reorder volatile references, the hardware 2221 // also must not reorder them). 2222 // 2223 // According to the new Java Memory Model (JMM): 2224 // (1) All volatiles are serialized wrt to each other. ALSO reads & 2225 // writes act as aquire & release, so: 2226 // (2) A read cannot let unrelated NON-volatile memory refs that 2227 // happen after the read float up to before the read. It's OK for 2228 // non-volatile memory refs that happen before the volatile read to 2229 // float down below it. 2230 // (3) Similar a volatile write cannot let unrelated NON-volatile 2231 // memory refs that happen BEFORE the write float down to after the 2232 // write. It's OK for non-volatile memory refs that happen after the 2233 // volatile write to float up before it. 2234 // 2235 // We only put in barriers around volatile refs (they are expensive), 2236 // not _between_ memory refs (that would require us to track the 2237 // flavor of the previous memory refs). Requirements (2) and (3) 2238 // require some barriers before volatile stores and after volatile 2239 // loads. These nearly cover requirement (1) but miss the 2240 // volatile-store-volatile-load case. This final case is placed after 2241 // volatile-stores although it could just as well go before 2242 // volatile-loads. 2243 2244 void TemplateTable::resolve_cache_and_index(int byte_no, 2245 Register Rcache, 2246 Register index, 2247 size_t index_size) { 2248 const Register temp = r19; 2249 assert_different_registers(Rcache, index, temp); 2250 2251 Label resolved; 2252 2253 Bytecodes::Code code = bytecode(); 2254 switch (code) { 2255 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break; 2256 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break; 2257 } 2258 2259 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 2260 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size); 2261 __ cmp(temp, (int) code); // have we resolved this bytecode? 2262 __ br(Assembler::EQ, resolved); 2263 2264 // resolve first time through 2265 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache); 2266 __ mov(temp, (int) code); 2267 __ call_VM(noreg, entry, temp); 2268 2269 // Update registers with resolved info 2270 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size); 2271 // n.b. unlike x86 Rcache is now rcpool plus the indexed offset 2272 // so all clients ofthis method must be modified accordingly 2273 __ bind(resolved); 2274 } 2275 2276 // The Rcache and index registers must be set before call 2277 // n.b unlike x86 cache already includes the index offset 2278 void TemplateTable::load_field_cp_cache_entry(Register obj, 2279 Register cache, 2280 Register index, 2281 Register off, 2282 Register flags, 2283 bool is_static = false) { 2284 assert_different_registers(cache, index, flags, off); 2285 2286 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2287 // Field offset 2288 __ ldr(off, Address(cache, in_bytes(cp_base_offset + 2289 ConstantPoolCacheEntry::f2_offset()))); 2290 // Flags 2291 __ ldrw(flags, Address(cache, in_bytes(cp_base_offset + 2292 ConstantPoolCacheEntry::flags_offset()))); 2293 2294 // klass overwrite register 2295 if (is_static) { 2296 __ ldr(obj, Address(cache, in_bytes(cp_base_offset + 2297 ConstantPoolCacheEntry::f1_offset()))); 2298 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 2299 __ ldr(obj, Address(obj, mirror_offset)); 2300 __ resolve_oop_handle(obj); 2301 } 2302 } 2303 2304 void TemplateTable::load_invoke_cp_cache_entry(int byte_no, 2305 Register method, 2306 Register itable_index, 2307 Register flags, 2308 bool is_invokevirtual, 2309 bool is_invokevfinal, /*unused*/ 2310 bool is_invokedynamic) { 2311 // setup registers 2312 const Register cache = rscratch2; 2313 const Register index = r4; 2314 assert_different_registers(method, flags); 2315 assert_different_registers(method, cache, index); 2316 assert_different_registers(itable_index, flags); 2317 assert_different_registers(itable_index, cache, index); 2318 // determine constant pool cache field offsets 2319 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant"); 2320 const int method_offset = in_bytes( 2321 ConstantPoolCache::base_offset() + 2322 (is_invokevirtual 2323 ? ConstantPoolCacheEntry::f2_offset() 2324 : ConstantPoolCacheEntry::f1_offset())); 2325 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() + 2326 ConstantPoolCacheEntry::flags_offset()); 2327 // access constant pool cache fields 2328 const int index_offset = in_bytes(ConstantPoolCache::base_offset() + 2329 ConstantPoolCacheEntry::f2_offset()); 2330 2331 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2)); 2332 resolve_cache_and_index(byte_no, cache, index, index_size); 2333 __ ldr(method, Address(cache, method_offset)); 2334 2335 if (itable_index != noreg) { 2336 __ ldr(itable_index, Address(cache, index_offset)); 2337 } 2338 __ ldrw(flags, Address(cache, flags_offset)); 2339 } 2340 2341 2342 // The registers cache and index expected to be set before call. 2343 // Correct values of the cache and index registers are preserved. 2344 void TemplateTable::jvmti_post_field_access(Register cache, Register index, 2345 bool is_static, bool has_tos) { 2346 // do the JVMTI work here to avoid disturbing the register state below 2347 // We use c_rarg registers here because we want to use the register used in 2348 // the call to the VM 2349 if (JvmtiExport::can_post_field_access()) { 2350 // Check to see if a field access watch has been set before we 2351 // take the time to call into the VM. 2352 Label L1; 2353 assert_different_registers(cache, index, r0); 2354 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2355 __ ldrw(r0, Address(rscratch1)); 2356 __ cbzw(r0, L1); 2357 2358 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1); 2359 __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset()))); 2360 2361 if (is_static) { 2362 __ mov(c_rarg1, zr); // NULL object reference 2363 } else { 2364 __ ldr(c_rarg1, at_tos()); // get object pointer without popping it 2365 __ verify_oop(c_rarg1); 2366 } 2367 // c_rarg1: object pointer or NULL 2368 // c_rarg2: cache entry pointer 2369 // c_rarg3: jvalue object on the stack 2370 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 2371 InterpreterRuntime::post_field_access), 2372 c_rarg1, c_rarg2, c_rarg3); 2373 __ get_cache_and_index_at_bcp(cache, index, 1); 2374 __ bind(L1); 2375 } 2376 } 2377 2378 void TemplateTable::pop_and_check_object(Register r) 2379 { 2380 __ pop_ptr(r); 2381 __ null_check(r); // for field access must check obj. 2382 __ verify_oop(r); 2383 } 2384 2385 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) 2386 { 2387 const Register cache = r2; 2388 const Register index = r3; 2389 const Register obj = r4; 2390 const Register off = r19; 2391 const Register flags = r0; 2392 const Register raw_flags = r6; 2393 const Register bc = r4; // uses same reg as obj, so don't mix them 2394 2395 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2396 jvmti_post_field_access(cache, index, is_static, false); 2397 load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static); 2398 2399 if (!is_static) { 2400 // obj is on the stack 2401 pop_and_check_object(obj); 2402 } 2403 2404 // 8179954: We need to make sure that the code generated for 2405 // volatile accesses forms a sequentially-consistent set of 2406 // operations when combined with STLR and LDAR. Without a leading 2407 // membar it's possible for a simple Dekker test to fail if loads 2408 // use LDR;DMB but stores use STLR. This can happen if C2 compiles 2409 // the stores in one method and we interpret the loads in another. 2410 if (! UseBarriersForVolatile) { 2411 Label notVolatile; 2412 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 2413 __ membar(MacroAssembler::AnyAny); 2414 __ bind(notVolatile); 2415 } 2416 2417 const Address field(obj, off); 2418 2419 Label Done, notByte, notBool, notInt, notShort, notChar, 2420 notLong, notFloat, notObj, notDouble; 2421 2422 // x86 uses a shift and mask or wings it with a shift plus assert 2423 // the mask is not needed. aarch64 just uses bitfield extract 2424 __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift, 2425 ConstantPoolCacheEntry::tos_state_bits); 2426 2427 assert(btos == 0, "change code, btos != 0"); 2428 __ cbnz(flags, notByte); 2429 2430 // Don't rewrite getstatic, only getfield 2431 if (is_static) rc = may_not_rewrite; 2432 2433 // btos 2434 __ load_signed_byte(r0, field); 2435 __ push(btos); 2436 // Rewrite bytecode to be faster 2437 if (rc == may_rewrite) { 2438 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1); 2439 } 2440 __ b(Done); 2441 2442 __ bind(notByte); 2443 __ cmp(flags, ztos); 2444 __ br(Assembler::NE, notBool); 2445 2446 // ztos (same code as btos) 2447 __ ldrsb(r0, field); 2448 __ push(ztos); 2449 // Rewrite bytecode to be faster 2450 if (rc == may_rewrite) { 2451 // use btos rewriting, no truncating to t/f bit is needed for getfield. 2452 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1); 2453 } 2454 __ b(Done); 2455 2456 __ bind(notBool); 2457 __ cmp(flags, atos); 2458 __ br(Assembler::NE, notObj); 2459 // atos 2460 __ load_heap_oop(r0, field); 2461 __ push(atos); 2462 if (rc == may_rewrite) { 2463 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1); 2464 } 2465 __ b(Done); 2466 2467 __ bind(notObj); 2468 __ cmp(flags, itos); 2469 __ br(Assembler::NE, notInt); 2470 // itos 2471 __ ldrw(r0, field); 2472 __ push(itos); 2473 // Rewrite bytecode to be faster 2474 if (rc == may_rewrite) { 2475 patch_bytecode(Bytecodes::_fast_igetfield, bc, r1); 2476 } 2477 __ b(Done); 2478 2479 __ bind(notInt); 2480 __ cmp(flags, ctos); 2481 __ br(Assembler::NE, notChar); 2482 // ctos 2483 __ load_unsigned_short(r0, field); 2484 __ push(ctos); 2485 // Rewrite bytecode to be faster 2486 if (rc == may_rewrite) { 2487 patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1); 2488 } 2489 __ b(Done); 2490 2491 __ bind(notChar); 2492 __ cmp(flags, stos); 2493 __ br(Assembler::NE, notShort); 2494 // stos 2495 __ load_signed_short(r0, field); 2496 __ push(stos); 2497 // Rewrite bytecode to be faster 2498 if (rc == may_rewrite) { 2499 patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1); 2500 } 2501 __ b(Done); 2502 2503 __ bind(notShort); 2504 __ cmp(flags, ltos); 2505 __ br(Assembler::NE, notLong); 2506 // ltos 2507 __ ldr(r0, field); 2508 __ push(ltos); 2509 // Rewrite bytecode to be faster 2510 if (rc == may_rewrite) { 2511 patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1); 2512 } 2513 __ b(Done); 2514 2515 __ bind(notLong); 2516 __ cmp(flags, ftos); 2517 __ br(Assembler::NE, notFloat); 2518 // ftos 2519 __ ldrs(v0, field); 2520 __ push(ftos); 2521 // Rewrite bytecode to be faster 2522 if (rc == may_rewrite) { 2523 patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1); 2524 } 2525 __ b(Done); 2526 2527 __ bind(notFloat); 2528 #ifdef ASSERT 2529 __ cmp(flags, dtos); 2530 __ br(Assembler::NE, notDouble); 2531 #endif 2532 // dtos 2533 __ ldrd(v0, field); 2534 __ push(dtos); 2535 // Rewrite bytecode to be faster 2536 if (rc == may_rewrite) { 2537 patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1); 2538 } 2539 #ifdef ASSERT 2540 __ b(Done); 2541 2542 __ bind(notDouble); 2543 __ stop("Bad state"); 2544 #endif 2545 2546 __ bind(Done); 2547 2548 Label notVolatile; 2549 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 2550 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore); 2551 __ bind(notVolatile); 2552 } 2553 2554 2555 void TemplateTable::getfield(int byte_no) 2556 { 2557 getfield_or_static(byte_no, false); 2558 } 2559 2560 void TemplateTable::nofast_getfield(int byte_no) { 2561 getfield_or_static(byte_no, false, may_not_rewrite); 2562 } 2563 2564 void TemplateTable::getstatic(int byte_no) 2565 { 2566 getfield_or_static(byte_no, true); 2567 } 2568 2569 // The registers cache and index expected to be set before call. 2570 // The function may destroy various registers, just not the cache and index registers. 2571 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) { 2572 transition(vtos, vtos); 2573 2574 ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 2575 2576 if (JvmtiExport::can_post_field_modification()) { 2577 // Check to see if a field modification watch has been set before 2578 // we take the time to call into the VM. 2579 Label L1; 2580 assert_different_registers(cache, index, r0); 2581 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2582 __ ldrw(r0, Address(rscratch1)); 2583 __ cbz(r0, L1); 2584 2585 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1); 2586 2587 if (is_static) { 2588 // Life is simple. Null out the object pointer. 2589 __ mov(c_rarg1, zr); 2590 } else { 2591 // Life is harder. The stack holds the value on top, followed by 2592 // the object. We don't know the size of the value, though; it 2593 // could be one or two words depending on its type. As a result, 2594 // we must find the type to determine where the object is. 2595 __ ldrw(c_rarg3, Address(c_rarg2, 2596 in_bytes(cp_base_offset + 2597 ConstantPoolCacheEntry::flags_offset()))); 2598 __ lsr(c_rarg3, c_rarg3, 2599 ConstantPoolCacheEntry::tos_state_shift); 2600 ConstantPoolCacheEntry::verify_tos_state_shift(); 2601 Label nope2, done, ok; 2602 __ ldr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue 2603 __ cmpw(c_rarg3, ltos); 2604 __ br(Assembler::EQ, ok); 2605 __ cmpw(c_rarg3, dtos); 2606 __ br(Assembler::NE, nope2); 2607 __ bind(ok); 2608 __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue) 2609 __ bind(nope2); 2610 } 2611 // cache entry pointer 2612 __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset)); 2613 // object (tos) 2614 __ mov(c_rarg3, esp); 2615 // c_rarg1: object pointer set up above (NULL if static) 2616 // c_rarg2: cache entry pointer 2617 // c_rarg3: jvalue object on the stack 2618 __ call_VM(noreg, 2619 CAST_FROM_FN_PTR(address, 2620 InterpreterRuntime::post_field_modification), 2621 c_rarg1, c_rarg2, c_rarg3); 2622 __ get_cache_and_index_at_bcp(cache, index, 1); 2623 __ bind(L1); 2624 } 2625 } 2626 2627 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) { 2628 transition(vtos, vtos); 2629 2630 const Register cache = r2; 2631 const Register index = r3; 2632 const Register obj = r2; 2633 const Register off = r19; 2634 const Register flags = r0; 2635 const Register bc = r4; 2636 2637 resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 2638 jvmti_post_field_mod(cache, index, is_static); 2639 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 2640 2641 Label Done; 2642 __ mov(r5, flags); 2643 2644 { 2645 Label notVolatile; 2646 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 2647 __ membar(MacroAssembler::StoreStore); 2648 __ bind(notVolatile); 2649 } 2650 2651 // field address 2652 const Address field(obj, off); 2653 2654 Label notByte, notBool, notInt, notShort, notChar, 2655 notLong, notFloat, notObj, notDouble; 2656 2657 // x86 uses a shift and mask or wings it with a shift plus assert 2658 // the mask is not needed. aarch64 just uses bitfield extract 2659 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits); 2660 2661 assert(btos == 0, "change code, btos != 0"); 2662 __ cbnz(flags, notByte); 2663 2664 // Don't rewrite putstatic, only putfield 2665 if (is_static) rc = may_not_rewrite; 2666 2667 // btos 2668 { 2669 __ pop(btos); 2670 if (!is_static) pop_and_check_object(obj); 2671 __ strb(r0, field); 2672 if (rc == may_rewrite) { 2673 patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no); 2674 } 2675 __ b(Done); 2676 } 2677 2678 __ bind(notByte); 2679 __ cmp(flags, ztos); 2680 __ br(Assembler::NE, notBool); 2681 2682 // ztos 2683 { 2684 __ pop(ztos); 2685 if (!is_static) pop_and_check_object(obj); 2686 __ andw(r0, r0, 0x1); 2687 __ strb(r0, field); 2688 if (rc == may_rewrite) { 2689 patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no); 2690 } 2691 __ b(Done); 2692 } 2693 2694 __ bind(notBool); 2695 __ cmp(flags, atos); 2696 __ br(Assembler::NE, notObj); 2697 2698 // atos 2699 { 2700 __ pop(atos); 2701 if (!is_static) pop_and_check_object(obj); 2702 // Store into the field 2703 do_oop_store(_masm, field, r0, _bs->kind(), false); 2704 if (rc == may_rewrite) { 2705 patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no); 2706 } 2707 __ b(Done); 2708 } 2709 2710 __ bind(notObj); 2711 __ cmp(flags, itos); 2712 __ br(Assembler::NE, notInt); 2713 2714 // itos 2715 { 2716 __ pop(itos); 2717 if (!is_static) pop_and_check_object(obj); 2718 __ strw(r0, field); 2719 if (rc == may_rewrite) { 2720 patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no); 2721 } 2722 __ b(Done); 2723 } 2724 2725 __ bind(notInt); 2726 __ cmp(flags, ctos); 2727 __ br(Assembler::NE, notChar); 2728 2729 // ctos 2730 { 2731 __ pop(ctos); 2732 if (!is_static) pop_and_check_object(obj); 2733 __ strh(r0, field); 2734 if (rc == may_rewrite) { 2735 patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no); 2736 } 2737 __ b(Done); 2738 } 2739 2740 __ bind(notChar); 2741 __ cmp(flags, stos); 2742 __ br(Assembler::NE, notShort); 2743 2744 // stos 2745 { 2746 __ pop(stos); 2747 if (!is_static) pop_and_check_object(obj); 2748 __ strh(r0, field); 2749 if (rc == may_rewrite) { 2750 patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no); 2751 } 2752 __ b(Done); 2753 } 2754 2755 __ bind(notShort); 2756 __ cmp(flags, ltos); 2757 __ br(Assembler::NE, notLong); 2758 2759 // ltos 2760 { 2761 __ pop(ltos); 2762 if (!is_static) pop_and_check_object(obj); 2763 __ str(r0, field); 2764 if (rc == may_rewrite) { 2765 patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no); 2766 } 2767 __ b(Done); 2768 } 2769 2770 __ bind(notLong); 2771 __ cmp(flags, ftos); 2772 __ br(Assembler::NE, notFloat); 2773 2774 // ftos 2775 { 2776 __ pop(ftos); 2777 if (!is_static) pop_and_check_object(obj); 2778 __ strs(v0, field); 2779 if (rc == may_rewrite) { 2780 patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no); 2781 } 2782 __ b(Done); 2783 } 2784 2785 __ bind(notFloat); 2786 #ifdef ASSERT 2787 __ cmp(flags, dtos); 2788 __ br(Assembler::NE, notDouble); 2789 #endif 2790 2791 // dtos 2792 { 2793 __ pop(dtos); 2794 if (!is_static) pop_and_check_object(obj); 2795 __ strd(v0, field); 2796 if (rc == may_rewrite) { 2797 patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no); 2798 } 2799 } 2800 2801 #ifdef ASSERT 2802 __ b(Done); 2803 2804 __ bind(notDouble); 2805 __ stop("Bad state"); 2806 #endif 2807 2808 __ bind(Done); 2809 2810 { 2811 Label notVolatile; 2812 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 2813 __ membar(MacroAssembler::StoreLoad); 2814 __ bind(notVolatile); 2815 } 2816 } 2817 2818 void TemplateTable::putfield(int byte_no) 2819 { 2820 putfield_or_static(byte_no, false); 2821 } 2822 2823 void TemplateTable::nofast_putfield(int byte_no) { 2824 putfield_or_static(byte_no, false, may_not_rewrite); 2825 } 2826 2827 void TemplateTable::putstatic(int byte_no) { 2828 putfield_or_static(byte_no, true); 2829 } 2830 2831 void TemplateTable::jvmti_post_fast_field_mod() 2832 { 2833 if (JvmtiExport::can_post_field_modification()) { 2834 // Check to see if a field modification watch has been set before 2835 // we take the time to call into the VM. 2836 Label L2; 2837 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 2838 __ ldrw(c_rarg3, Address(rscratch1)); 2839 __ cbzw(c_rarg3, L2); 2840 __ pop_ptr(r19); // copy the object pointer from tos 2841 __ verify_oop(r19); 2842 __ push_ptr(r19); // put the object pointer back on tos 2843 // Save tos values before call_VM() clobbers them. Since we have 2844 // to do it for every data type, we use the saved values as the 2845 // jvalue object. 2846 switch (bytecode()) { // load values into the jvalue object 2847 case Bytecodes::_fast_aputfield: __ push_ptr(r0); break; 2848 case Bytecodes::_fast_bputfield: // fall through 2849 case Bytecodes::_fast_zputfield: // fall through 2850 case Bytecodes::_fast_sputfield: // fall through 2851 case Bytecodes::_fast_cputfield: // fall through 2852 case Bytecodes::_fast_iputfield: __ push_i(r0); break; 2853 case Bytecodes::_fast_dputfield: __ push_d(); break; 2854 case Bytecodes::_fast_fputfield: __ push_f(); break; 2855 case Bytecodes::_fast_lputfield: __ push_l(r0); break; 2856 2857 default: 2858 ShouldNotReachHere(); 2859 } 2860 __ mov(c_rarg3, esp); // points to jvalue on the stack 2861 // access constant pool cache entry 2862 __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1); 2863 __ verify_oop(r19); 2864 // r19: object pointer copied above 2865 // c_rarg2: cache entry pointer 2866 // c_rarg3: jvalue object on the stack 2867 __ call_VM(noreg, 2868 CAST_FROM_FN_PTR(address, 2869 InterpreterRuntime::post_field_modification), 2870 r19, c_rarg2, c_rarg3); 2871 2872 switch (bytecode()) { // restore tos values 2873 case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break; 2874 case Bytecodes::_fast_bputfield: // fall through 2875 case Bytecodes::_fast_zputfield: // fall through 2876 case Bytecodes::_fast_sputfield: // fall through 2877 case Bytecodes::_fast_cputfield: // fall through 2878 case Bytecodes::_fast_iputfield: __ pop_i(r0); break; 2879 case Bytecodes::_fast_dputfield: __ pop_d(); break; 2880 case Bytecodes::_fast_fputfield: __ pop_f(); break; 2881 case Bytecodes::_fast_lputfield: __ pop_l(r0); break; 2882 } 2883 __ bind(L2); 2884 } 2885 } 2886 2887 void TemplateTable::fast_storefield(TosState state) 2888 { 2889 transition(state, vtos); 2890 2891 ByteSize base = ConstantPoolCache::base_offset(); 2892 2893 jvmti_post_fast_field_mod(); 2894 2895 // access constant pool cache 2896 __ get_cache_and_index_at_bcp(r2, r1, 1); 2897 2898 // test for volatile with r3 2899 __ ldrw(r3, Address(r2, in_bytes(base + 2900 ConstantPoolCacheEntry::flags_offset()))); 2901 2902 // replace index with field offset from cache entry 2903 __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset()))); 2904 2905 { 2906 Label notVolatile; 2907 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 2908 __ membar(MacroAssembler::StoreStore); 2909 __ bind(notVolatile); 2910 } 2911 2912 Label notVolatile; 2913 2914 // Get object from stack 2915 pop_and_check_object(r2); 2916 2917 // field address 2918 const Address field(r2, r1); 2919 2920 // access field 2921 switch (bytecode()) { 2922 case Bytecodes::_fast_aputfield: 2923 do_oop_store(_masm, field, r0, _bs->kind(), false); 2924 break; 2925 case Bytecodes::_fast_lputfield: 2926 __ str(r0, field); 2927 break; 2928 case Bytecodes::_fast_iputfield: 2929 __ strw(r0, field); 2930 break; 2931 case Bytecodes::_fast_zputfield: 2932 __ andw(r0, r0, 0x1); // boolean is true if LSB is 1 2933 // fall through to bputfield 2934 case Bytecodes::_fast_bputfield: 2935 __ strb(r0, field); 2936 break; 2937 case Bytecodes::_fast_sputfield: 2938 // fall through 2939 case Bytecodes::_fast_cputfield: 2940 __ strh(r0, field); 2941 break; 2942 case Bytecodes::_fast_fputfield: 2943 __ strs(v0, field); 2944 break; 2945 case Bytecodes::_fast_dputfield: 2946 __ strd(v0, field); 2947 break; 2948 default: 2949 ShouldNotReachHere(); 2950 } 2951 2952 { 2953 Label notVolatile; 2954 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 2955 __ membar(MacroAssembler::StoreLoad); 2956 __ bind(notVolatile); 2957 } 2958 } 2959 2960 2961 void TemplateTable::fast_accessfield(TosState state) 2962 { 2963 transition(atos, state); 2964 // Do the JVMTI work here to avoid disturbing the register state below 2965 if (JvmtiExport::can_post_field_access()) { 2966 // Check to see if a field access watch has been set before we 2967 // take the time to call into the VM. 2968 Label L1; 2969 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 2970 __ ldrw(r2, Address(rscratch1)); 2971 __ cbzw(r2, L1); 2972 // access constant pool cache entry 2973 __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1); 2974 __ verify_oop(r0); 2975 __ push_ptr(r0); // save object pointer before call_VM() clobbers it 2976 __ mov(c_rarg1, r0); 2977 // c_rarg1: object pointer copied above 2978 // c_rarg2: cache entry pointer 2979 __ call_VM(noreg, 2980 CAST_FROM_FN_PTR(address, 2981 InterpreterRuntime::post_field_access), 2982 c_rarg1, c_rarg2); 2983 __ pop_ptr(r0); // restore object pointer 2984 __ bind(L1); 2985 } 2986 2987 // access constant pool cache 2988 __ get_cache_and_index_at_bcp(r2, r1, 1); 2989 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() + 2990 ConstantPoolCacheEntry::f2_offset()))); 2991 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() + 2992 ConstantPoolCacheEntry::flags_offset()))); 2993 2994 // r0: object 2995 __ verify_oop(r0); 2996 __ null_check(r0); 2997 const Address field(r0, r1); 2998 2999 // 8179954: We need to make sure that the code generated for 3000 // volatile accesses forms a sequentially-consistent set of 3001 // operations when combined with STLR and LDAR. Without a leading 3002 // membar it's possible for a simple Dekker test to fail if loads 3003 // use LDR;DMB but stores use STLR. This can happen if C2 compiles 3004 // the stores in one method and we interpret the loads in another. 3005 if (! UseBarriersForVolatile) { 3006 Label notVolatile; 3007 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 3008 __ membar(MacroAssembler::AnyAny); 3009 __ bind(notVolatile); 3010 } 3011 3012 // access field 3013 switch (bytecode()) { 3014 case Bytecodes::_fast_agetfield: 3015 __ load_heap_oop(r0, field); 3016 __ verify_oop(r0); 3017 break; 3018 case Bytecodes::_fast_lgetfield: 3019 __ ldr(r0, field); 3020 break; 3021 case Bytecodes::_fast_igetfield: 3022 __ ldrw(r0, field); 3023 break; 3024 case Bytecodes::_fast_bgetfield: 3025 __ load_signed_byte(r0, field); 3026 break; 3027 case Bytecodes::_fast_sgetfield: 3028 __ load_signed_short(r0, field); 3029 break; 3030 case Bytecodes::_fast_cgetfield: 3031 __ load_unsigned_short(r0, field); 3032 break; 3033 case Bytecodes::_fast_fgetfield: 3034 __ ldrs(v0, field); 3035 break; 3036 case Bytecodes::_fast_dgetfield: 3037 __ ldrd(v0, field); 3038 break; 3039 default: 3040 ShouldNotReachHere(); 3041 } 3042 { 3043 Label notVolatile; 3044 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 3045 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore); 3046 __ bind(notVolatile); 3047 } 3048 } 3049 3050 void TemplateTable::fast_xaccess(TosState state) 3051 { 3052 transition(vtos, state); 3053 3054 // get receiver 3055 __ ldr(r0, aaddress(0)); 3056 // access constant pool cache 3057 __ get_cache_and_index_at_bcp(r2, r3, 2); 3058 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() + 3059 ConstantPoolCacheEntry::f2_offset()))); 3060 3061 // 8179954: We need to make sure that the code generated for 3062 // volatile accesses forms a sequentially-consistent set of 3063 // operations when combined with STLR and LDAR. Without a leading 3064 // membar it's possible for a simple Dekker test to fail if loads 3065 // use LDR;DMB but stores use STLR. This can happen if C2 compiles 3066 // the stores in one method and we interpret the loads in another. 3067 if (! UseBarriersForVolatile) { 3068 Label notVolatile; 3069 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() + 3070 ConstantPoolCacheEntry::flags_offset()))); 3071 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 3072 __ membar(MacroAssembler::AnyAny); 3073 __ bind(notVolatile); 3074 } 3075 3076 // make sure exception is reported in correct bcp range (getfield is 3077 // next instruction) 3078 __ increment(rbcp); 3079 __ null_check(r0); 3080 switch (state) { 3081 case itos: 3082 __ ldrw(r0, Address(r0, r1, Address::lsl(0))); 3083 break; 3084 case atos: 3085 __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0))); 3086 __ verify_oop(r0); 3087 break; 3088 case ftos: 3089 __ ldrs(v0, Address(r0, r1, Address::lsl(0))); 3090 break; 3091 default: 3092 ShouldNotReachHere(); 3093 } 3094 3095 { 3096 Label notVolatile; 3097 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() + 3098 ConstantPoolCacheEntry::flags_offset()))); 3099 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile); 3100 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore); 3101 __ bind(notVolatile); 3102 } 3103 3104 __ decrement(rbcp); 3105 } 3106 3107 3108 3109 //----------------------------------------------------------------------------- 3110 // Calls 3111 3112 void TemplateTable::count_calls(Register method, Register temp) 3113 { 3114 __ call_Unimplemented(); 3115 } 3116 3117 void TemplateTable::prepare_invoke(int byte_no, 3118 Register method, // linked method (or i-klass) 3119 Register index, // itable index, MethodType, etc. 3120 Register recv, // if caller wants to see it 3121 Register flags // if caller wants to test it 3122 ) { 3123 // determine flags 3124 Bytecodes::Code code = bytecode(); 3125 const bool is_invokeinterface = code == Bytecodes::_invokeinterface; 3126 const bool is_invokedynamic = code == Bytecodes::_invokedynamic; 3127 const bool is_invokehandle = code == Bytecodes::_invokehandle; 3128 const bool is_invokevirtual = code == Bytecodes::_invokevirtual; 3129 const bool is_invokespecial = code == Bytecodes::_invokespecial; 3130 const bool load_receiver = (recv != noreg); 3131 const bool save_flags = (flags != noreg); 3132 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), ""); 3133 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal"); 3134 assert(flags == noreg || flags == r3, ""); 3135 assert(recv == noreg || recv == r2, ""); 3136 3137 // setup registers & access constant pool cache 3138 if (recv == noreg) recv = r2; 3139 if (flags == noreg) flags = r3; 3140 assert_different_registers(method, index, recv, flags); 3141 3142 // save 'interpreter return address' 3143 __ save_bcp(); 3144 3145 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic); 3146 3147 // maybe push appendix to arguments (just before return address) 3148 if (is_invokedynamic || is_invokehandle) { 3149 Label L_no_push; 3150 __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push); 3151 // Push the appendix as a trailing parameter. 3152 // This must be done before we get the receiver, 3153 // since the parameter_size includes it. 3154 __ push(r19); 3155 __ mov(r19, index); 3156 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0"); 3157 __ load_resolved_reference_at_index(index, r19); 3158 __ pop(r19); 3159 __ push(index); // push appendix (MethodType, CallSite, etc.) 3160 __ bind(L_no_push); 3161 } 3162 3163 // load receiver if needed (note: no return address pushed yet) 3164 if (load_receiver) { 3165 __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask); 3166 // FIXME -- is this actually correct? looks like it should be 2 3167 // const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address 3168 // const int receiver_is_at_end = -1; // back off one slot to get receiver 3169 // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end); 3170 // __ movptr(recv, recv_addr); 3171 __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here? 3172 __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1))); 3173 __ verify_oop(recv); 3174 } 3175 3176 // compute return type 3177 // x86 uses a shift and mask or wings it with a shift plus assert 3178 // the mask is not needed. aarch64 just uses bitfield extract 3179 __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits); 3180 // load return address 3181 { 3182 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code); 3183 __ mov(rscratch1, table_addr); 3184 __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3))); 3185 } 3186 } 3187 3188 3189 void TemplateTable::invokevirtual_helper(Register index, 3190 Register recv, 3191 Register flags) 3192 { 3193 // Uses temporary registers r0, r3 3194 assert_different_registers(index, recv, r0, r3); 3195 // Test for an invoke of a final method 3196 Label notFinal; 3197 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal); 3198 3199 const Register method = index; // method must be rmethod 3200 assert(method == rmethod, 3201 "methodOop must be rmethod for interpreter calling convention"); 3202 3203 // do the call - the index is actually the method to call 3204 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method* 3205 3206 // It's final, need a null check here! 3207 __ null_check(recv); 3208 3209 // profile this call 3210 __ profile_final_call(r0); 3211 __ profile_arguments_type(r0, method, r4, true); 3212 3213 __ jump_from_interpreted(method, r0); 3214 3215 __ bind(notFinal); 3216 3217 // get receiver klass 3218 __ null_check(recv, oopDesc::klass_offset_in_bytes()); 3219 __ load_klass(r0, recv); 3220 3221 // profile this call 3222 __ profile_virtual_call(r0, rlocals, r3); 3223 3224 // get target methodOop & entry point 3225 __ lookup_virtual_method(r0, index, method); 3226 __ profile_arguments_type(r3, method, r4, true); 3227 // FIXME -- this looks completely redundant. is it? 3228 // __ ldr(r3, Address(method, Method::interpreter_entry_offset())); 3229 __ jump_from_interpreted(method, r3); 3230 } 3231 3232 void TemplateTable::invokevirtual(int byte_no) 3233 { 3234 transition(vtos, vtos); 3235 assert(byte_no == f2_byte, "use this argument"); 3236 3237 prepare_invoke(byte_no, rmethod, noreg, r2, r3); 3238 3239 // rmethod: index (actually a Method*) 3240 // r2: receiver 3241 // r3: flags 3242 3243 invokevirtual_helper(rmethod, r2, r3); 3244 } 3245 3246 void TemplateTable::invokespecial(int byte_no) 3247 { 3248 transition(vtos, vtos); 3249 assert(byte_no == f1_byte, "use this argument"); 3250 3251 prepare_invoke(byte_no, rmethod, noreg, // get f1 Method* 3252 r2); // get receiver also for null check 3253 __ verify_oop(r2); 3254 __ null_check(r2); 3255 // do the call 3256 __ profile_call(r0); 3257 __ profile_arguments_type(r0, rmethod, rbcp, false); 3258 __ jump_from_interpreted(rmethod, r0); 3259 } 3260 3261 void TemplateTable::invokestatic(int byte_no) 3262 { 3263 transition(vtos, vtos); 3264 assert(byte_no == f1_byte, "use this argument"); 3265 3266 prepare_invoke(byte_no, rmethod); // get f1 Method* 3267 // do the call 3268 __ profile_call(r0); 3269 __ profile_arguments_type(r0, rmethod, r4, false); 3270 __ jump_from_interpreted(rmethod, r0); 3271 } 3272 3273 void TemplateTable::fast_invokevfinal(int byte_no) 3274 { 3275 __ call_Unimplemented(); 3276 } 3277 3278 void TemplateTable::invokeinterface(int byte_no) { 3279 transition(vtos, vtos); 3280 assert(byte_no == f1_byte, "use this argument"); 3281 3282 prepare_invoke(byte_no, r0, rmethod, // get f1 Klass*, f2 itable index 3283 r2, r3); // recv, flags 3284 3285 // r0: interface klass (from f1) 3286 // rmethod: itable index (from f2) or Method* 3287 // r2: receiver 3288 // r3: flags 3289 3290 // Check for private method invocation - indicated by vfinal 3291 Label notVFinal; 3292 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notVFinal); 3293 3294 // do the call - the index is actually the method to call 3295 3296 __ null_check(r2); 3297 3298 __ profile_final_call(r0); 3299 __ profile_arguments_type(r0, rmethod, r4, true); 3300 __ jump_from_interpreted(rmethod, r0); 3301 3302 __ bind(notVFinal); 3303 3304 // Special case of invokeinterface called for virtual method of 3305 // java.lang.Object. See cpCacheOop.cpp for details. 3306 // This code isn't produced by javac, but could be produced by 3307 // another compliant java compiler. 3308 Label notMethod; 3309 __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod); 3310 3311 invokevirtual_helper(rmethod, r2, r3); 3312 __ bind(notMethod); 3313 3314 // Get receiver klass into r3 - also a null check 3315 __ restore_locals(); 3316 __ null_check(r2, oopDesc::klass_offset_in_bytes()); 3317 __ load_klass(r3, r2); 3318 3319 // profile this call 3320 __ profile_virtual_call(r3, r13, r19); 3321 3322 Label no_such_interface, no_such_method; 3323 3324 __ lookup_interface_method(// inputs: rec. class, interface, itable index 3325 r3, r0, rmethod, 3326 // outputs: method, scan temp. reg 3327 rmethod, r13, 3328 no_such_interface); 3329 3330 // rmethod,: methodOop to call 3331 // r2: receiver 3332 // Check for abstract method error 3333 // Note: This should be done more efficiently via a throw_abstract_method_error 3334 // interpreter entry point and a conditional jump to it in case of a null 3335 // method. 3336 __ cbz(rmethod, no_such_method); 3337 3338 __ profile_arguments_type(r3, rmethod, r13, true); 3339 3340 // do the call 3341 // r2: receiver 3342 // rmethod,: methodOop 3343 __ jump_from_interpreted(rmethod, r3); 3344 __ should_not_reach_here(); 3345 3346 // exception handling code follows... 3347 // note: must restore interpreter registers to canonical 3348 // state for exception handling to work correctly! 3349 3350 __ bind(no_such_method); 3351 // throw exception 3352 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed) 3353 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3354 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 3355 // the call_VM checks for exception, so we should never return here. 3356 __ should_not_reach_here(); 3357 3358 __ bind(no_such_interface); 3359 // throw exception 3360 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed) 3361 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 3362 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 3363 InterpreterRuntime::throw_IncompatibleClassChangeError)); 3364 // the call_VM checks for exception, so we should never return here. 3365 __ should_not_reach_here(); 3366 return; 3367 } 3368 3369 void TemplateTable::invokehandle(int byte_no) { 3370 transition(vtos, vtos); 3371 assert(byte_no == f1_byte, "use this argument"); 3372 3373 prepare_invoke(byte_no, rmethod, r0, r2); 3374 __ verify_method_ptr(r2); 3375 __ verify_oop(r2); 3376 __ null_check(r2); 3377 3378 // FIXME: profile the LambdaForm also 3379 3380 // r13 is safe to use here as a scratch reg because it is about to 3381 // be clobbered by jump_from_interpreted(). 3382 __ profile_final_call(r13); 3383 __ profile_arguments_type(r13, rmethod, r4, true); 3384 3385 __ jump_from_interpreted(rmethod, r0); 3386 } 3387 3388 void TemplateTable::invokedynamic(int byte_no) { 3389 transition(vtos, vtos); 3390 assert(byte_no == f1_byte, "use this argument"); 3391 3392 prepare_invoke(byte_no, rmethod, r0); 3393 3394 // r0: CallSite object (from cpool->resolved_references[]) 3395 // rmethod: MH.linkToCallSite method (from f2) 3396 3397 // Note: r0_callsite is already pushed by prepare_invoke 3398 3399 // %%% should make a type profile for any invokedynamic that takes a ref argument 3400 // profile this call 3401 __ profile_call(rbcp); 3402 __ profile_arguments_type(r3, rmethod, r13, false); 3403 3404 __ verify_oop(r0); 3405 3406 __ jump_from_interpreted(rmethod, r0); 3407 } 3408 3409 3410 //----------------------------------------------------------------------------- 3411 // Allocation 3412 3413 void TemplateTable::_new() { 3414 transition(vtos, atos); 3415 3416 __ get_unsigned_2_byte_index_at_bcp(r3, 1); 3417 Label slow_case; 3418 Label done; 3419 Label initialize_header; 3420 Label initialize_object; // including clearing the fields 3421 Label allocate_shared; 3422 3423 __ get_cpool_and_tags(r4, r0); 3424 // Make sure the class we're about to instantiate has been resolved. 3425 // This is done before loading InstanceKlass to be consistent with the order 3426 // how Constant Pool is updated (see ConstantPool::klass_at_put) 3427 const int tags_offset = Array<u1>::base_offset_in_bytes(); 3428 __ lea(rscratch1, Address(r0, r3, Address::lsl(0))); 3429 __ lea(rscratch1, Address(rscratch1, tags_offset)); 3430 __ ldarb(rscratch1, rscratch1); 3431 __ cmp(rscratch1, JVM_CONSTANT_Class); 3432 __ br(Assembler::NE, slow_case); 3433 3434 // get InstanceKlass 3435 __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1); 3436 3437 // make sure klass is initialized & doesn't have finalizer 3438 // make sure klass is fully initialized 3439 __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset())); 3440 __ cmp(rscratch1, InstanceKlass::fully_initialized); 3441 __ br(Assembler::NE, slow_case); 3442 3443 // get instance_size in InstanceKlass (scaled to a count of bytes) 3444 __ ldrw(r3, 3445 Address(r4, 3446 Klass::layout_helper_offset())); 3447 // test to see if it has a finalizer or is malformed in some way 3448 __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case); 3449 3450 // Allocate the instance 3451 // 1) Try to allocate in the TLAB 3452 // 2) if fail and the object is large allocate in the shared Eden 3453 // 3) if the above fails (or is not applicable), go to a slow case 3454 // (creates a new TLAB, etc.) 3455 3456 const bool allow_shared_alloc = 3457 Universe::heap()->supports_inline_contig_alloc(); 3458 3459 if (UseTLAB) { 3460 __ tlab_allocate(r0, r3, 0, noreg, r1, 3461 allow_shared_alloc ? allocate_shared : slow_case); 3462 3463 if (ZeroTLAB) { 3464 // the fields have been already cleared 3465 __ b(initialize_header); 3466 } else { 3467 // initialize both the header and fields 3468 __ b(initialize_object); 3469 } 3470 } 3471 3472 // Allocation in the shared Eden, if allowed. 3473 // 3474 // r3: instance size in bytes 3475 if (allow_shared_alloc) { 3476 __ bind(allocate_shared); 3477 3478 __ eden_allocate(r0, r3, 0, r10, slow_case); 3479 __ incr_allocated_bytes(rthread, r3, 0, rscratch1); 3480 } 3481 3482 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) { 3483 // The object is initialized before the header. If the object size is 3484 // zero, go directly to the header initialization. 3485 __ bind(initialize_object); 3486 __ sub(r3, r3, sizeof(oopDesc)); 3487 __ cbz(r3, initialize_header); 3488 3489 // Initialize object fields 3490 { 3491 __ add(r2, r0, sizeof(oopDesc)); 3492 Label loop; 3493 __ bind(loop); 3494 __ str(zr, Address(__ post(r2, BytesPerLong))); 3495 __ sub(r3, r3, BytesPerLong); 3496 __ cbnz(r3, loop); 3497 } 3498 3499 // initialize object header only. 3500 __ bind(initialize_header); 3501 if (UseBiasedLocking) { 3502 __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset())); 3503 } else { 3504 __ mov(rscratch1, (intptr_t)markOopDesc::prototype()); 3505 } 3506 __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes())); 3507 __ store_klass_gap(r0, zr); // zero klass gap for compressed oops 3508 __ store_klass(r0, r4); // store klass last 3509 3510 { 3511 SkipIfEqual skip(_masm, &DTraceAllocProbes, false); 3512 // Trigger dtrace event for fastpath 3513 __ push(atos); // save the return value 3514 __ call_VM_leaf( 3515 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0); 3516 __ pop(atos); // restore the return value 3517 3518 } 3519 __ b(done); 3520 } 3521 3522 // slow case 3523 __ bind(slow_case); 3524 __ get_constant_pool(c_rarg1); 3525 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1); 3526 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2); 3527 __ verify_oop(r0); 3528 3529 // continue 3530 __ bind(done); 3531 // Must prevent reordering of stores for object initialization with stores that publish the new object. 3532 __ membar(Assembler::StoreStore); 3533 } 3534 3535 void TemplateTable::newarray() { 3536 transition(itos, atos); 3537 __ load_unsigned_byte(c_rarg1, at_bcp(1)); 3538 __ mov(c_rarg2, r0); 3539 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), 3540 c_rarg1, c_rarg2); 3541 // Must prevent reordering of stores for object initialization with stores that publish the new object. 3542 __ membar(Assembler::StoreStore); 3543 } 3544 3545 void TemplateTable::anewarray() { 3546 transition(itos, atos); 3547 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1); 3548 __ get_constant_pool(c_rarg1); 3549 __ mov(c_rarg3, r0); 3550 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), 3551 c_rarg1, c_rarg2, c_rarg3); 3552 // Must prevent reordering of stores for object initialization with stores that publish the new object. 3553 __ membar(Assembler::StoreStore); 3554 } 3555 3556 void TemplateTable::arraylength() { 3557 transition(atos, itos); 3558 __ null_check(r0, arrayOopDesc::length_offset_in_bytes()); 3559 __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes())); 3560 } 3561 3562 void TemplateTable::checkcast() 3563 { 3564 transition(atos, atos); 3565 Label done, is_null, ok_is_subtype, quicked, resolved; 3566 __ cbz(r0, is_null); 3567 3568 // Get cpool & tags index 3569 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array 3570 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index 3571 // See if bytecode has already been quicked 3572 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes()); 3573 __ lea(r1, Address(rscratch1, r19)); 3574 __ ldarb(r1, r1); 3575 __ cmp(r1, JVM_CONSTANT_Class); 3576 __ br(Assembler::EQ, quicked); 3577 3578 __ push(atos); // save receiver for result, and for GC 3579 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc)); 3580 // vm_result_2 has metadata result 3581 __ get_vm_result_2(r0, rthread); 3582 __ pop(r3); // restore receiver 3583 __ b(resolved); 3584 3585 // Get superklass in r0 and subklass in r3 3586 __ bind(quicked); 3587 __ mov(r3, r0); // Save object in r3; r0 needed for subtype check 3588 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass 3589 3590 __ bind(resolved); 3591 __ load_klass(r19, r3); 3592 3593 // Generate subtype check. Blows r2, r5. Object in r3. 3594 // Superklass in r0. Subklass in r19. 3595 __ gen_subtype_check(r19, ok_is_subtype); 3596 3597 // Come here on failure 3598 __ push(r3); 3599 // object is at TOS 3600 __ b(Interpreter::_throw_ClassCastException_entry); 3601 3602 // Come here on success 3603 __ bind(ok_is_subtype); 3604 __ mov(r0, r3); // Restore object in r3 3605 3606 // Collect counts on whether this test sees NULLs a lot or not. 3607 if (ProfileInterpreter) { 3608 __ b(done); 3609 __ bind(is_null); 3610 __ profile_null_seen(r2); 3611 } else { 3612 __ bind(is_null); // same as 'done' 3613 } 3614 __ bind(done); 3615 } 3616 3617 void TemplateTable::instanceof() { 3618 transition(atos, itos); 3619 Label done, is_null, ok_is_subtype, quicked, resolved; 3620 __ cbz(r0, is_null); 3621 3622 // Get cpool & tags index 3623 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array 3624 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index 3625 // See if bytecode has already been quicked 3626 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes()); 3627 __ lea(r1, Address(rscratch1, r19)); 3628 __ ldarb(r1, r1); 3629 __ cmp(r1, JVM_CONSTANT_Class); 3630 __ br(Assembler::EQ, quicked); 3631 3632 __ push(atos); // save receiver for result, and for GC 3633 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc)); 3634 // vm_result_2 has metadata result 3635 __ get_vm_result_2(r0, rthread); 3636 __ pop(r3); // restore receiver 3637 __ verify_oop(r3); 3638 __ load_klass(r3, r3); 3639 __ b(resolved); 3640 3641 // Get superklass in r0 and subklass in r3 3642 __ bind(quicked); 3643 __ load_klass(r3, r0); 3644 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); 3645 3646 __ bind(resolved); 3647 3648 // Generate subtype check. Blows r2, r5 3649 // Superklass in r0. Subklass in r3. 3650 __ gen_subtype_check(r3, ok_is_subtype); 3651 3652 // Come here on failure 3653 __ mov(r0, 0); 3654 __ b(done); 3655 // Come here on success 3656 __ bind(ok_is_subtype); 3657 __ mov(r0, 1); 3658 3659 // Collect counts on whether this test sees NULLs a lot or not. 3660 if (ProfileInterpreter) { 3661 __ b(done); 3662 __ bind(is_null); 3663 __ profile_null_seen(r2); 3664 } else { 3665 __ bind(is_null); // same as 'done' 3666 } 3667 __ bind(done); 3668 // r0 = 0: obj == NULL or obj is not an instanceof the specified klass 3669 // r0 = 1: obj != NULL and obj is an instanceof the specified klass 3670 } 3671 3672 //----------------------------------------------------------------------------- 3673 // Breakpoints 3674 void TemplateTable::_breakpoint() { 3675 // Note: We get here even if we are single stepping.. 3676 // jbug inists on setting breakpoints at every bytecode 3677 // even if we are in single step mode. 3678 3679 transition(vtos, vtos); 3680 3681 // get the unpatched byte code 3682 __ get_method(c_rarg1); 3683 __ call_VM(noreg, 3684 CAST_FROM_FN_PTR(address, 3685 InterpreterRuntime::get_original_bytecode_at), 3686 c_rarg1, rbcp); 3687 __ mov(r19, r0); 3688 3689 // post the breakpoint event 3690 __ call_VM(noreg, 3691 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), 3692 rmethod, rbcp); 3693 3694 // complete the execution of original bytecode 3695 __ mov(rscratch1, r19); 3696 __ dispatch_only_normal(vtos); 3697 } 3698 3699 //----------------------------------------------------------------------------- 3700 // Exceptions 3701 3702 void TemplateTable::athrow() { 3703 transition(atos, vtos); 3704 __ null_check(r0); 3705 __ b(Interpreter::throw_exception_entry()); 3706 } 3707 3708 //----------------------------------------------------------------------------- 3709 // Synchronization 3710 // 3711 // Note: monitorenter & exit are symmetric routines; which is reflected 3712 // in the assembly code structure as well 3713 // 3714 // Stack layout: 3715 // 3716 // [expressions ] <--- esp = expression stack top 3717 // .. 3718 // [expressions ] 3719 // [monitor entry] <--- monitor block top = expression stack bot 3720 // .. 3721 // [monitor entry] 3722 // [frame data ] <--- monitor block bot 3723 // ... 3724 // [saved rbp ] <--- rbp 3725 void TemplateTable::monitorenter() 3726 { 3727 transition(atos, vtos); 3728 3729 // check for NULL object 3730 __ null_check(r0); 3731 3732 const Address monitor_block_top( 3733 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 3734 const Address monitor_block_bot( 3735 rfp, frame::interpreter_frame_initial_sp_offset * wordSize); 3736 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 3737 3738 Label allocated; 3739 3740 // initialize entry pointer 3741 __ mov(c_rarg1, zr); // points to free slot or NULL 3742 3743 // find a free slot in the monitor block (result in c_rarg1) 3744 { 3745 Label entry, loop, exit; 3746 __ ldr(c_rarg3, monitor_block_top); // points to current entry, 3747 // starting with top-most entry 3748 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom 3749 3750 __ b(entry); 3751 3752 __ bind(loop); 3753 // check if current entry is used 3754 // if not used then remember entry in c_rarg1 3755 __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes())); 3756 __ cmp(zr, rscratch1); 3757 __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ); 3758 // check if current entry is for same object 3759 __ cmp(r0, rscratch1); 3760 // if same object then stop searching 3761 __ br(Assembler::EQ, exit); 3762 // otherwise advance to next entry 3763 __ add(c_rarg3, c_rarg3, entry_size); 3764 __ bind(entry); 3765 // check if bottom reached 3766 __ cmp(c_rarg3, c_rarg2); 3767 // if not at bottom then check this entry 3768 __ br(Assembler::NE, loop); 3769 __ bind(exit); 3770 } 3771 3772 __ cbnz(c_rarg1, allocated); // check if a slot has been found and 3773 // if found, continue with that on 3774 3775 // allocate one if there's no free slot 3776 { 3777 Label entry, loop; 3778 // 1. compute new pointers // rsp: old expression stack top 3779 __ ldr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom 3780 __ sub(esp, esp, entry_size); // move expression stack top 3781 __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom 3782 __ mov(c_rarg3, esp); // set start value for copy loop 3783 __ str(c_rarg1, monitor_block_bot); // set new monitor block bottom 3784 3785 __ sub(sp, sp, entry_size); // make room for the monitor 3786 3787 __ b(entry); 3788 // 2. move expression stack contents 3789 __ bind(loop); 3790 __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack 3791 // word from old location 3792 __ str(c_rarg2, Address(c_rarg3, 0)); // and store it at new location 3793 __ add(c_rarg3, c_rarg3, wordSize); // advance to next word 3794 __ bind(entry); 3795 __ cmp(c_rarg3, c_rarg1); // check if bottom reached 3796 __ br(Assembler::NE, loop); // if not at bottom then 3797 // copy next word 3798 } 3799 3800 // call run-time routine 3801 // c_rarg1: points to monitor entry 3802 __ bind(allocated); 3803 3804 // Increment bcp to point to the next bytecode, so exception 3805 // handling for async. exceptions work correctly. 3806 // The object has already been poped from the stack, so the 3807 // expression stack looks correct. 3808 __ increment(rbcp); 3809 3810 // store object 3811 __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 3812 __ lock_object(c_rarg1); 3813 3814 // check to make sure this monitor doesn't cause stack overflow after locking 3815 __ save_bcp(); // in case of exception 3816 __ generate_stack_overflow_check(0); 3817 3818 // The bcp has already been incremented. Just need to dispatch to 3819 // next instruction. 3820 __ dispatch_next(vtos); 3821 } 3822 3823 3824 void TemplateTable::monitorexit() 3825 { 3826 transition(atos, vtos); 3827 3828 // check for NULL object 3829 __ null_check(r0); 3830 3831 const Address monitor_block_top( 3832 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 3833 const Address monitor_block_bot( 3834 rfp, frame::interpreter_frame_initial_sp_offset * wordSize); 3835 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 3836 3837 Label found; 3838 3839 // find matching slot 3840 { 3841 Label entry, loop; 3842 __ ldr(c_rarg1, monitor_block_top); // points to current entry, 3843 // starting with top-most entry 3844 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom 3845 // of monitor block 3846 __ b(entry); 3847 3848 __ bind(loop); 3849 // check if current entry is for same object 3850 __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 3851 __ cmp(r0, rscratch1); 3852 // if same object then stop searching 3853 __ br(Assembler::EQ, found); 3854 // otherwise advance to next entry 3855 __ add(c_rarg1, c_rarg1, entry_size); 3856 __ bind(entry); 3857 // check if bottom reached 3858 __ cmp(c_rarg1, c_rarg2); 3859 // if not at bottom then check this entry 3860 __ br(Assembler::NE, loop); 3861 } 3862 3863 // error handling. Unlocking was not block-structured 3864 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 3865 InterpreterRuntime::throw_illegal_monitor_state_exception)); 3866 __ should_not_reach_here(); 3867 3868 // call run-time routine 3869 __ bind(found); 3870 __ push_ptr(r0); // make sure object is on stack (contract with oopMaps) 3871 __ unlock_object(c_rarg1); 3872 __ pop_ptr(r0); // discard object 3873 } 3874 3875 3876 // Wide instructions 3877 void TemplateTable::wide() 3878 { 3879 __ load_unsigned_byte(r19, at_bcp(1)); 3880 __ mov(rscratch1, (address)Interpreter::_wentry_point); 3881 __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3))); 3882 __ br(rscratch1); 3883 } 3884 3885 3886 // Multi arrays 3887 void TemplateTable::multianewarray() { 3888 transition(vtos, atos); 3889 __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions 3890 // last dim is on top of stack; we want address of first one: 3891 // first_addr = last_addr + (ndims - 1) * wordSize 3892 __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3))); 3893 __ sub(c_rarg1, c_rarg1, wordSize); 3894 call_VM(r0, 3895 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), 3896 c_rarg1); 3897 __ load_unsigned_byte(r1, at_bcp(3)); 3898 __ lea(esp, Address(esp, r1, Address::uxtw(3))); 3899 }