1 /*
   2  * Copyright (c) 2003, 2017, 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::G1BarrierSet:
 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::CardTableModRef:
 188       {
 189         if (val == noreg) {
 190           __ store_heap_oop_null(obj);
 191         } else {
 192           __ store_heap_oop(obj, val);
 193           // flatten object address if needed
 194           if (!precise || (obj.index() == noreg && obj.offset() == 0)) {
 195             __ store_check(obj.base());
 196           } else {
 197             __ lea(r3, obj);
 198             __ store_check(r3);
 199           }
 200         }
 201       }
 202       break;
 203     case BarrierSet::ModRef:
 204       if (val == noreg) {
 205         __ store_heap_oop_null(obj);
 206       } else {
 207         __ store_heap_oop(obj, val);
 208       }
 209       break;
 210     default      :
 211       ShouldNotReachHere();
 212 
 213   }
 214 }
 215 
 216 Address TemplateTable::at_bcp(int offset) {
 217   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 218   return Address(rbcp, offset);
 219 }
 220 
 221 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 222                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 223                                    int byte_no)
 224 {
 225   if (!RewriteBytecodes)  return;
 226   Label L_patch_done;
 227 
 228   switch (bc) {
 229   case Bytecodes::_fast_aputfield:
 230   case Bytecodes::_fast_bputfield:
 231   case Bytecodes::_fast_zputfield:
 232   case Bytecodes::_fast_cputfield:
 233   case Bytecodes::_fast_dputfield:
 234   case Bytecodes::_fast_fputfield:
 235   case Bytecodes::_fast_iputfield:
 236   case Bytecodes::_fast_lputfield:
 237   case Bytecodes::_fast_sputfield:
 238     {
 239       // We skip bytecode quickening for putfield instructions when
 240       // the put_code written to the constant pool cache is zero.
 241       // This is required so that every execution of this instruction
 242       // calls out to InterpreterRuntime::resolve_get_put to do
 243       // additional, required work.
 244       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 245       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 246       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 247       __ movw(bc_reg, bc);
 248       __ cbzw(temp_reg, L_patch_done);  // don't patch
 249     }
 250     break;
 251   default:
 252     assert(byte_no == -1, "sanity");
 253     // the pair bytecodes have already done the load.
 254     if (load_bc_into_bc_reg) {
 255       __ movw(bc_reg, bc);
 256     }
 257   }
 258 
 259   if (JvmtiExport::can_post_breakpoint()) {
 260     Label L_fast_patch;
 261     // if a breakpoint is present we can't rewrite the stream directly
 262     __ load_unsigned_byte(temp_reg, at_bcp(0));
 263     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 264     __ br(Assembler::NE, L_fast_patch);
 265     // Let breakpoint table handling rewrite to quicker bytecode
 266     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 267     __ b(L_patch_done);
 268     __ bind(L_fast_patch);
 269   }
 270 
 271 #ifdef ASSERT
 272   Label L_okay;
 273   __ load_unsigned_byte(temp_reg, at_bcp(0));
 274   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 275   __ br(Assembler::EQ, L_okay);
 276   __ cmpw(temp_reg, bc_reg);
 277   __ br(Assembler::EQ, L_okay);
 278   __ stop("patching the wrong bytecode");
 279   __ bind(L_okay);
 280 #endif
 281 
 282   // patch bytecode
 283   __ strb(bc_reg, at_bcp(0));
 284   __ bind(L_patch_done);
 285 }
 286 
 287 
 288 // Individual instructions
 289 
 290 void TemplateTable::nop() {
 291   transition(vtos, vtos);
 292   // nothing to do
 293 }
 294 
 295 void TemplateTable::shouldnotreachhere() {
 296   transition(vtos, vtos);
 297   __ stop("shouldnotreachhere bytecode");
 298 }
 299 
 300 void TemplateTable::aconst_null()
 301 {
 302   transition(vtos, atos);
 303   __ mov(r0, 0);
 304 }
 305 
 306 void TemplateTable::iconst(int value)
 307 {
 308   transition(vtos, itos);
 309   __ mov(r0, value);
 310 }
 311 
 312 void TemplateTable::lconst(int value)
 313 {
 314   __ mov(r0, value);
 315 }
 316 
 317 void TemplateTable::fconst(int value)
 318 {
 319   transition(vtos, ftos);
 320   switch (value) {
 321   case 0:
 322     __ fmovs(v0, zr);
 323     break;
 324   case 1:
 325     __ fmovs(v0, 1.0);
 326     break;
 327   case 2:
 328     __ fmovs(v0, 2.0);
 329     break;
 330   default:
 331     ShouldNotReachHere();
 332     break;
 333   }
 334 }
 335 
 336 void TemplateTable::dconst(int value)
 337 {
 338   transition(vtos, dtos);
 339   switch (value) {
 340   case 0:
 341     __ fmovd(v0, zr);
 342     break;
 343   case 1:
 344     __ fmovd(v0, 1.0);
 345     break;
 346   case 2:
 347     __ fmovd(v0, 2.0);
 348     break;
 349   default:
 350     ShouldNotReachHere();
 351     break;
 352   }
 353 }
 354 
 355 void TemplateTable::bipush()
 356 {
 357   transition(vtos, itos);
 358   __ load_signed_byte32(r0, at_bcp(1));
 359 }
 360 
 361 void TemplateTable::sipush()
 362 {
 363   transition(vtos, itos);
 364   __ load_unsigned_short(r0, at_bcp(1));
 365   __ revw(r0, r0);
 366   __ asrw(r0, r0, 16);
 367 }
 368 
 369 void TemplateTable::ldc(bool wide)
 370 {
 371   transition(vtos, vtos);
 372   Label call_ldc, notFloat, notClass, Done;
 373 
 374   if (wide) {
 375     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 376   } else {
 377     __ load_unsigned_byte(r1, at_bcp(1));
 378   }
 379   __ get_cpool_and_tags(r2, r0);
 380 
 381   const int base_offset = ConstantPool::header_size() * wordSize;
 382   const int tags_offset = Array<u1>::base_offset_in_bytes();
 383 
 384   // get type
 385   __ add(r3, r1, tags_offset);
 386   __ lea(r3, Address(r0, r3));
 387   __ ldarb(r3, r3);
 388 
 389   // unresolved class - get the resolved class
 390   __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
 391   __ br(Assembler::EQ, call_ldc);
 392 
 393   // unresolved class in error state - call into runtime to throw the error
 394   // from the first resolution attempt
 395   __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
 396   __ br(Assembler::EQ, call_ldc);
 397 
 398   // resolved class - need to call vm to get java mirror of the class
 399   __ cmp(r3, JVM_CONSTANT_Class);
 400   __ br(Assembler::NE, notClass);
 401 
 402   __ bind(call_ldc);
 403   __ mov(c_rarg1, wide);
 404   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 405   __ push_ptr(r0);
 406   __ verify_oop(r0);
 407   __ b(Done);
 408 
 409   __ bind(notClass);
 410   __ cmp(r3, JVM_CONSTANT_Float);
 411   __ br(Assembler::NE, notFloat);
 412   // ftos
 413   __ adds(r1, r2, r1, Assembler::LSL, 3);
 414   __ ldrs(v0, Address(r1, base_offset));
 415   __ push_f();
 416   __ b(Done);
 417 
 418   __ bind(notFloat);
 419 #ifdef ASSERT
 420   {
 421     Label L;
 422     __ cmp(r3, JVM_CONSTANT_Integer);
 423     __ br(Assembler::EQ, L);
 424     // String and Object are rewritten to fast_aldc
 425     __ stop("unexpected tag type in ldc");
 426     __ bind(L);
 427   }
 428 #endif
 429   // itos JVM_CONSTANT_Integer only
 430   __ adds(r1, r2, r1, Assembler::LSL, 3);
 431   __ ldrw(r0, Address(r1, base_offset));
 432   __ push_i(r0);
 433   __ bind(Done);
 434 }
 435 
 436 // Fast path for caching oop constants.
 437 void TemplateTable::fast_aldc(bool wide)
 438 {
 439   transition(vtos, atos);
 440 
 441   Register result = r0;
 442   Register tmp = r1;
 443   int index_size = wide ? sizeof(u2) : sizeof(u1);
 444 
 445   Label resolved;
 446 
 447   // We are resolved if the resolved reference cache entry contains a
 448   // non-null object (String, MethodType, etc.)
 449   assert_different_registers(result, tmp);
 450   __ get_cache_index_at_bcp(tmp, 1, index_size);
 451   __ load_resolved_reference_at_index(result, tmp);
 452   __ cbnz(result, resolved);
 453 
 454   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 455 
 456   // first time invocation - must resolve first
 457   __ mov(tmp, (int)bytecode());
 458   __ call_VM(result, entry, tmp);
 459 
 460   __ bind(resolved);
 461 
 462   if (VerifyOops) {
 463     __ verify_oop(result);
 464   }
 465 }
 466 
 467 void TemplateTable::ldc2_w()
 468 {
 469   transition(vtos, vtos);
 470   Label Long, Done;
 471   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 472 
 473   __ get_cpool_and_tags(r1, r2);
 474   const int base_offset = ConstantPool::header_size() * wordSize;
 475   const int tags_offset = Array<u1>::base_offset_in_bytes();
 476 
 477   // get type
 478   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 479   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 480   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 481   __ br(Assembler::NE, Long);
 482   // dtos
 483   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 484   __ ldrd(v0, Address(r2, base_offset));
 485   __ push_d();
 486   __ b(Done);
 487 
 488   __ bind(Long);
 489   // ltos
 490   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 491   __ ldr(r0, Address(r0, base_offset));
 492   __ push_l();
 493 
 494   __ bind(Done);
 495 }
 496 
 497 void TemplateTable::locals_index(Register reg, int offset)
 498 {
 499   __ ldrb(reg, at_bcp(offset));
 500   __ neg(reg, reg);
 501 }
 502 
 503 void TemplateTable::iload() {
 504   iload_internal();
 505 }
 506 
 507 void TemplateTable::nofast_iload() {
 508   iload_internal(may_not_rewrite);
 509 }
 510 
 511 void TemplateTable::iload_internal(RewriteControl rc) {
 512   transition(vtos, itos);
 513   if (RewriteFrequentPairs && rc == may_rewrite) {
 514     Label rewrite, done;
 515     Register bc = r4;
 516 
 517     // get next bytecode
 518     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 519 
 520     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 521     // last two iloads in a pair.  Comparing against fast_iload means that
 522     // the next bytecode is neither an iload or a caload, and therefore
 523     // an iload pair.
 524     __ cmpw(r1, Bytecodes::_iload);
 525     __ br(Assembler::EQ, done);
 526 
 527     // if _fast_iload rewrite to _fast_iload2
 528     __ cmpw(r1, Bytecodes::_fast_iload);
 529     __ movw(bc, Bytecodes::_fast_iload2);
 530     __ br(Assembler::EQ, rewrite);
 531 
 532     // if _caload rewrite to _fast_icaload
 533     __ cmpw(r1, Bytecodes::_caload);
 534     __ movw(bc, Bytecodes::_fast_icaload);
 535     __ br(Assembler::EQ, rewrite);
 536 
 537     // else rewrite to _fast_iload
 538     __ movw(bc, Bytecodes::_fast_iload);
 539 
 540     // rewrite
 541     // bc: new bytecode
 542     __ bind(rewrite);
 543     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 544     __ bind(done);
 545 
 546   }
 547 
 548   // do iload, get the local value into tos
 549   locals_index(r1);
 550   __ ldr(r0, iaddress(r1));
 551 
 552 }
 553 
 554 void TemplateTable::fast_iload2()
 555 {
 556   transition(vtos, itos);
 557   locals_index(r1);
 558   __ ldr(r0, iaddress(r1));
 559   __ push(itos);
 560   locals_index(r1, 3);
 561   __ ldr(r0, iaddress(r1));
 562 }
 563 
 564 void TemplateTable::fast_iload()
 565 {
 566   transition(vtos, itos);
 567   locals_index(r1);
 568   __ ldr(r0, iaddress(r1));
 569 }
 570 
 571 void TemplateTable::lload()
 572 {
 573   transition(vtos, ltos);
 574   __ ldrb(r1, at_bcp(1));
 575   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 576   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 577 }
 578 
 579 void TemplateTable::fload()
 580 {
 581   transition(vtos, ftos);
 582   locals_index(r1);
 583   // n.b. we use ldrd here because this is a 64 bit slot
 584   // this is comparable to the iload case
 585   __ ldrd(v0, faddress(r1));
 586 }
 587 
 588 void TemplateTable::dload()
 589 {
 590   transition(vtos, dtos);
 591   __ ldrb(r1, at_bcp(1));
 592   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 593   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 594 }
 595 
 596 void TemplateTable::aload()
 597 {
 598   transition(vtos, atos);
 599   locals_index(r1);
 600   __ ldr(r0, iaddress(r1));
 601 }
 602 
 603 void TemplateTable::locals_index_wide(Register reg) {
 604   __ ldrh(reg, at_bcp(2));
 605   __ rev16w(reg, reg);
 606   __ neg(reg, reg);
 607 }
 608 
 609 void TemplateTable::wide_iload() {
 610   transition(vtos, itos);
 611   locals_index_wide(r1);
 612   __ ldr(r0, iaddress(r1));
 613 }
 614 
 615 void TemplateTable::wide_lload()
 616 {
 617   transition(vtos, ltos);
 618   __ ldrh(r1, at_bcp(2));
 619   __ rev16w(r1, r1);
 620   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 621   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 622 }
 623 
 624 void TemplateTable::wide_fload()
 625 {
 626   transition(vtos, ftos);
 627   locals_index_wide(r1);
 628   // n.b. we use ldrd here because this is a 64 bit slot
 629   // this is comparable to the iload case
 630   __ ldrd(v0, faddress(r1));
 631 }
 632 
 633 void TemplateTable::wide_dload()
 634 {
 635   transition(vtos, dtos);
 636   __ ldrh(r1, at_bcp(2));
 637   __ rev16w(r1, r1);
 638   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 639   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 640 }
 641 
 642 void TemplateTable::wide_aload()
 643 {
 644   transition(vtos, atos);
 645   locals_index_wide(r1);
 646   __ ldr(r0, aaddress(r1));
 647 }
 648 
 649 void TemplateTable::index_check(Register array, Register index)
 650 {
 651   // destroys r1, rscratch1
 652   // check array
 653   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 654   // sign extend index for use by indexed load
 655   // __ movl2ptr(index, index);
 656   // check index
 657   Register length = rscratch1;
 658   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 659   __ cmpw(index, length);
 660   if (index != r1) {
 661     // ??? convention: move aberrant index into r1 for exception message
 662     assert(r1 != array, "different registers");
 663     __ mov(r1, index);
 664   }
 665   Label ok;
 666   __ br(Assembler::LO, ok);
 667   __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 668   __ br(rscratch1);
 669   __ bind(ok);
 670 }
 671 
 672 void TemplateTable::iaload()
 673 {
 674   transition(itos, itos);
 675   __ mov(r1, r0);
 676   __ pop_ptr(r0);
 677   // r0: array
 678   // r1: index
 679   index_check(r0, r1); // leaves index in r1, kills rscratch1
 680   __ lea(r1, Address(r0, r1, Address::uxtw(2)));
 681   __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
 682 }
 683 
 684 void TemplateTable::laload()
 685 {
 686   transition(itos, ltos);
 687   __ mov(r1, r0);
 688   __ pop_ptr(r0);
 689   // r0: array
 690   // r1: index
 691   index_check(r0, r1); // leaves index in r1, kills rscratch1
 692   __ lea(r1, Address(r0, r1, Address::uxtw(3)));
 693   __ ldr(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_LONG)));
 694 }
 695 
 696 void TemplateTable::faload()
 697 {
 698   transition(itos, ftos);
 699   __ mov(r1, r0);
 700   __ pop_ptr(r0);
 701   // r0: array
 702   // r1: index
 703   index_check(r0, r1); // leaves index in r1, kills rscratch1
 704   __ lea(r1,  Address(r0, r1, Address::uxtw(2)));
 705   __ ldrs(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 706 }
 707 
 708 void TemplateTable::daload()
 709 {
 710   transition(itos, dtos);
 711   __ mov(r1, r0);
 712   __ pop_ptr(r0);
 713   // r0: array
 714   // r1: index
 715   index_check(r0, r1); // leaves index in r1, kills rscratch1
 716   __ lea(r1,  Address(r0, r1, Address::uxtw(3)));
 717   __ ldrd(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 718 }
 719 
 720 void TemplateTable::aaload()
 721 {
 722   transition(itos, atos);
 723   __ mov(r1, r0);
 724   __ pop_ptr(r0);
 725   // r0: array
 726   // r1: index
 727   index_check(r0, r1); // leaves index in r1, kills rscratch1
 728   int s = (UseCompressedOops ? 2 : 3);
 729   __ lea(r1, Address(r0, r1, Address::uxtw(s)));
 730   __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 731 }
 732 
 733 void TemplateTable::baload()
 734 {
 735   transition(itos, itos);
 736   __ mov(r1, r0);
 737   __ pop_ptr(r0);
 738   // r0: array
 739   // r1: index
 740   index_check(r0, r1); // leaves index in r1, kills rscratch1
 741   __ lea(r1,  Address(r0, r1, Address::uxtw(0)));
 742   __ load_signed_byte(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 743 }
 744 
 745 void TemplateTable::caload()
 746 {
 747   transition(itos, itos);
 748   __ mov(r1, r0);
 749   __ pop_ptr(r0);
 750   // r0: array
 751   // r1: index
 752   index_check(r0, r1); // leaves index in r1, kills rscratch1
 753   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 754   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 755 }
 756 
 757 // iload followed by caload frequent pair
 758 void TemplateTable::fast_icaload()
 759 {
 760   transition(vtos, itos);
 761   // load index out of locals
 762   locals_index(r2);
 763   __ ldr(r1, iaddress(r2));
 764 
 765   __ pop_ptr(r0);
 766 
 767   // r0: array
 768   // r1: index
 769   index_check(r0, r1); // leaves index in r1, kills rscratch1
 770   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 771   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 772 }
 773 
 774 void TemplateTable::saload()
 775 {
 776   transition(itos, itos);
 777   __ mov(r1, r0);
 778   __ pop_ptr(r0);
 779   // r0: array
 780   // r1: index
 781   index_check(r0, r1); // leaves index in r1, kills rscratch1
 782   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 783   __ load_signed_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 784 }
 785 
 786 void TemplateTable::iload(int n)
 787 {
 788   transition(vtos, itos);
 789   __ ldr(r0, iaddress(n));
 790 }
 791 
 792 void TemplateTable::lload(int n)
 793 {
 794   transition(vtos, ltos);
 795   __ ldr(r0, laddress(n));
 796 }
 797 
 798 void TemplateTable::fload(int n)
 799 {
 800   transition(vtos, ftos);
 801   __ ldrs(v0, faddress(n));
 802 }
 803 
 804 void TemplateTable::dload(int n)
 805 {
 806   transition(vtos, dtos);
 807   __ ldrd(v0, daddress(n));
 808 }
 809 
 810 void TemplateTable::aload(int n)
 811 {
 812   transition(vtos, atos);
 813   __ ldr(r0, iaddress(n));
 814 }
 815 
 816 void TemplateTable::aload_0() {
 817   aload_0_internal();
 818 }
 819 
 820 void TemplateTable::nofast_aload_0() {
 821   aload_0_internal(may_not_rewrite);
 822 }
 823 
 824 void TemplateTable::aload_0_internal(RewriteControl rc) {
 825   // According to bytecode histograms, the pairs:
 826   //
 827   // _aload_0, _fast_igetfield
 828   // _aload_0, _fast_agetfield
 829   // _aload_0, _fast_fgetfield
 830   //
 831   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 832   // _aload_0 bytecode checks if the next bytecode is either
 833   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 834   // rewrites the current bytecode into a pair bytecode; otherwise it
 835   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 836   // the pair check anymore.
 837   //
 838   // Note: If the next bytecode is _getfield, the rewrite must be
 839   //       delayed, otherwise we may miss an opportunity for a pair.
 840   //
 841   // Also rewrite frequent pairs
 842   //   aload_0, aload_1
 843   //   aload_0, iload_1
 844   // These bytecodes with a small amount of code are most profitable
 845   // to rewrite
 846   if (RewriteFrequentPairs && rc == may_rewrite) {
 847     Label rewrite, done;
 848     const Register bc = r4;
 849 
 850     // get next bytecode
 851     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 852 
 853     // if _getfield then wait with rewrite
 854     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 855     __ br(Assembler::EQ, done);
 856 
 857     // if _igetfield then rewrite to _fast_iaccess_0
 858     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 859     __ cmpw(r1, Bytecodes::_fast_igetfield);
 860     __ movw(bc, Bytecodes::_fast_iaccess_0);
 861     __ br(Assembler::EQ, rewrite);
 862 
 863     // if _agetfield then rewrite to _fast_aaccess_0
 864     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 865     __ cmpw(r1, Bytecodes::_fast_agetfield);
 866     __ movw(bc, Bytecodes::_fast_aaccess_0);
 867     __ br(Assembler::EQ, rewrite);
 868 
 869     // if _fgetfield then rewrite to _fast_faccess_0
 870     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 871     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 872     __ movw(bc, Bytecodes::_fast_faccess_0);
 873     __ br(Assembler::EQ, rewrite);
 874 
 875     // else rewrite to _fast_aload0
 876     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 877     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 878 
 879     // rewrite
 880     // bc: new bytecode
 881     __ bind(rewrite);
 882     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 883 
 884     __ bind(done);
 885   }
 886 
 887   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 888   aload(0);
 889 }
 890 
 891 void TemplateTable::istore()
 892 {
 893   transition(itos, vtos);
 894   locals_index(r1);
 895   // FIXME: We're being very pernickerty here storing a jint in a
 896   // local with strw, which costs an extra instruction over what we'd
 897   // be able to do with a simple str.  We should just store the whole
 898   // word.
 899   __ lea(rscratch1, iaddress(r1));
 900   __ strw(r0, Address(rscratch1));
 901 }
 902 
 903 void TemplateTable::lstore()
 904 {
 905   transition(ltos, vtos);
 906   locals_index(r1);
 907   __ str(r0, laddress(r1, rscratch1, _masm));
 908 }
 909 
 910 void TemplateTable::fstore() {
 911   transition(ftos, vtos);
 912   locals_index(r1);
 913   __ lea(rscratch1, iaddress(r1));
 914   __ strs(v0, Address(rscratch1));
 915 }
 916 
 917 void TemplateTable::dstore() {
 918   transition(dtos, vtos);
 919   locals_index(r1);
 920   __ strd(v0, daddress(r1, rscratch1, _masm));
 921 }
 922 
 923 void TemplateTable::astore()
 924 {
 925   transition(vtos, vtos);
 926   __ pop_ptr(r0);
 927   locals_index(r1);
 928   __ str(r0, aaddress(r1));
 929 }
 930 
 931 void TemplateTable::wide_istore() {
 932   transition(vtos, vtos);
 933   __ pop_i();
 934   locals_index_wide(r1);
 935   __ lea(rscratch1, iaddress(r1));
 936   __ strw(r0, Address(rscratch1));
 937 }
 938 
 939 void TemplateTable::wide_lstore() {
 940   transition(vtos, vtos);
 941   __ pop_l();
 942   locals_index_wide(r1);
 943   __ str(r0, laddress(r1, rscratch1, _masm));
 944 }
 945 
 946 void TemplateTable::wide_fstore() {
 947   transition(vtos, vtos);
 948   __ pop_f();
 949   locals_index_wide(r1);
 950   __ lea(rscratch1, faddress(r1));
 951   __ strs(v0, rscratch1);
 952 }
 953 
 954 void TemplateTable::wide_dstore() {
 955   transition(vtos, vtos);
 956   __ pop_d();
 957   locals_index_wide(r1);
 958   __ strd(v0, daddress(r1, rscratch1, _masm));
 959 }
 960 
 961 void TemplateTable::wide_astore() {
 962   transition(vtos, vtos);
 963   __ pop_ptr(r0);
 964   locals_index_wide(r1);
 965   __ str(r0, aaddress(r1));
 966 }
 967 
 968 void TemplateTable::iastore() {
 969   transition(itos, vtos);
 970   __ pop_i(r1);
 971   __ pop_ptr(r3);
 972   // r0: value
 973   // r1: index
 974   // r3: array
 975   index_check(r3, r1); // prefer index in r1
 976   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
 977   __ strw(r0, Address(rscratch1,
 978                       arrayOopDesc::base_offset_in_bytes(T_INT)));
 979 }
 980 
 981 void TemplateTable::lastore() {
 982   transition(ltos, vtos);
 983   __ pop_i(r1);
 984   __ pop_ptr(r3);
 985   // r0: value
 986   // r1: index
 987   // r3: array
 988   index_check(r3, r1); // prefer index in r1
 989   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
 990   __ str(r0, Address(rscratch1,
 991                       arrayOopDesc::base_offset_in_bytes(T_LONG)));
 992 }
 993 
 994 void TemplateTable::fastore() {
 995   transition(ftos, vtos);
 996   __ pop_i(r1);
 997   __ pop_ptr(r3);
 998   // v0: value
 999   // r1:  index
1000   // r3:  array
1001   index_check(r3, r1); // prefer index in r1
1002   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1003   __ strs(v0, Address(rscratch1,
1004                       arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1005 }
1006 
1007 void TemplateTable::dastore() {
1008   transition(dtos, vtos);
1009   __ pop_i(r1);
1010   __ pop_ptr(r3);
1011   // v0: value
1012   // r1:  index
1013   // r3:  array
1014   index_check(r3, r1); // prefer index in r1
1015   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1016   __ strd(v0, Address(rscratch1,
1017                       arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1018 }
1019 
1020 void TemplateTable::aastore() {
1021   Label is_null, ok_is_subtype, done;
1022   transition(vtos, vtos);
1023   // stack: ..., array, index, value
1024   __ ldr(r0, at_tos());    // value
1025   __ ldr(r2, at_tos_p1()); // index
1026   __ ldr(r3, at_tos_p2()); // array
1027 
1028   Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1029 
1030   index_check(r3, r2);     // kills r1
1031   __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1032 
1033   // do array store check - check for NULL value first
1034   __ cbz(r0, is_null);
1035 
1036   // Move subklass into r1
1037   __ load_klass(r1, r0);
1038   // Move superklass into r0
1039   __ load_klass(r0, r3);
1040   __ ldr(r0, Address(r0,
1041                      ObjArrayKlass::element_klass_offset()));
1042   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1043 
1044   // Generate subtype check.  Blows r2, r5
1045   // Superklass in r0.  Subklass in r1.
1046   __ gen_subtype_check(r1, ok_is_subtype);
1047 
1048   // Come here on failure
1049   // object is at TOS
1050   __ b(Interpreter::_throw_ArrayStoreException_entry);
1051 
1052   // Come here on success
1053   __ bind(ok_is_subtype);
1054 
1055   // Get the value we will store
1056   __ ldr(r0, at_tos());
1057   // Now store using the appropriate barrier
1058   do_oop_store(_masm, element_address, r0, _bs->kind(), true);
1059   __ b(done);
1060 
1061   // Have a NULL in r0, r3=array, r2=index.  Store NULL at ary[idx]
1062   __ bind(is_null);
1063   __ profile_null_seen(r2);
1064 
1065   // Store a NULL
1066   do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1067 
1068   // Pop stack arguments
1069   __ bind(done);
1070   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1071 }
1072 
1073 void TemplateTable::bastore()
1074 {
1075   transition(itos, vtos);
1076   __ pop_i(r1);
1077   __ pop_ptr(r3);
1078   // r0: value
1079   // r1: index
1080   // r3: array
1081   index_check(r3, r1); // prefer index in r1
1082 
1083   // Need to check whether array is boolean or byte
1084   // since both types share the bastore bytecode.
1085   __ load_klass(r2, r3);
1086   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1087   int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1088   Label L_skip;
1089   __ tbz(r2, diffbit_index, L_skip);
1090   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1091   __ bind(L_skip);
1092 
1093   __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1094   __ strb(r0, Address(rscratch1,
1095                       arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1096 }
1097 
1098 void TemplateTable::castore()
1099 {
1100   transition(itos, vtos);
1101   __ pop_i(r1);
1102   __ pop_ptr(r3);
1103   // r0: value
1104   // r1: index
1105   // r3: array
1106   index_check(r3, r1); // prefer index in r1
1107   __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1108   __ strh(r0, Address(rscratch1,
1109                       arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1110 }
1111 
1112 void TemplateTable::sastore()
1113 {
1114   castore();
1115 }
1116 
1117 void TemplateTable::istore(int n)
1118 {
1119   transition(itos, vtos);
1120   __ str(r0, iaddress(n));
1121 }
1122 
1123 void TemplateTable::lstore(int n)
1124 {
1125   transition(ltos, vtos);
1126   __ str(r0, laddress(n));
1127 }
1128 
1129 void TemplateTable::fstore(int n)
1130 {
1131   transition(ftos, vtos);
1132   __ strs(v0, faddress(n));
1133 }
1134 
1135 void TemplateTable::dstore(int n)
1136 {
1137   transition(dtos, vtos);
1138   __ strd(v0, daddress(n));
1139 }
1140 
1141 void TemplateTable::astore(int n)
1142 {
1143   transition(vtos, vtos);
1144   __ pop_ptr(r0);
1145   __ str(r0, iaddress(n));
1146 }
1147 
1148 void TemplateTable::pop()
1149 {
1150   transition(vtos, vtos);
1151   __ add(esp, esp, Interpreter::stackElementSize);
1152 }
1153 
1154 void TemplateTable::pop2()
1155 {
1156   transition(vtos, vtos);
1157   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1158 }
1159 
1160 void TemplateTable::dup()
1161 {
1162   transition(vtos, vtos);
1163   __ ldr(r0, Address(esp, 0));
1164   __ push(r0);
1165   // stack: ..., a, a
1166 }
1167 
1168 void TemplateTable::dup_x1()
1169 {
1170   transition(vtos, vtos);
1171   // stack: ..., a, b
1172   __ ldr(r0, at_tos());  // load b
1173   __ ldr(r2, at_tos_p1());  // load a
1174   __ str(r0, at_tos_p1());  // store b
1175   __ str(r2, at_tos());  // store a
1176   __ push(r0);                  // push b
1177   // stack: ..., b, a, b
1178 }
1179 
1180 void TemplateTable::dup_x2()
1181 {
1182   transition(vtos, vtos);
1183   // stack: ..., a, b, c
1184   __ ldr(r0, at_tos());  // load c
1185   __ ldr(r2, at_tos_p2());  // load a
1186   __ str(r0, at_tos_p2());  // store c in a
1187   __ push(r0);      // push c
1188   // stack: ..., c, b, c, c
1189   __ ldr(r0, at_tos_p2());  // load b
1190   __ str(r2, at_tos_p2());  // store a in b
1191   // stack: ..., c, a, c, c
1192   __ str(r0, at_tos_p1());  // store b in c
1193   // stack: ..., c, a, b, c
1194 }
1195 
1196 void TemplateTable::dup2()
1197 {
1198   transition(vtos, vtos);
1199   // stack: ..., a, b
1200   __ ldr(r0, at_tos_p1());  // load a
1201   __ push(r0);                  // push a
1202   __ ldr(r0, at_tos_p1());  // load b
1203   __ push(r0);                  // push b
1204   // stack: ..., a, b, a, b
1205 }
1206 
1207 void TemplateTable::dup2_x1()
1208 {
1209   transition(vtos, vtos);
1210   // stack: ..., a, b, c
1211   __ ldr(r2, at_tos());  // load c
1212   __ ldr(r0, at_tos_p1());  // load b
1213   __ push(r0);                  // push b
1214   __ push(r2);                  // push c
1215   // stack: ..., a, b, c, b, c
1216   __ str(r2, at_tos_p3());  // store c in b
1217   // stack: ..., a, c, c, b, c
1218   __ ldr(r2, at_tos_p4());  // load a
1219   __ str(r2, at_tos_p2());  // store a in 2nd c
1220   // stack: ..., a, c, a, b, c
1221   __ str(r0, at_tos_p4());  // store b in a
1222   // stack: ..., b, c, a, b, c
1223 }
1224 
1225 void TemplateTable::dup2_x2()
1226 {
1227   transition(vtos, vtos);
1228   // stack: ..., a, b, c, d
1229   __ ldr(r2, at_tos());  // load d
1230   __ ldr(r0, at_tos_p1());  // load c
1231   __ push(r0)            ;      // push c
1232   __ push(r2);                  // push d
1233   // stack: ..., a, b, c, d, c, d
1234   __ ldr(r0, at_tos_p4());  // load b
1235   __ str(r0, at_tos_p2());  // store b in d
1236   __ str(r2, at_tos_p4());  // store d in b
1237   // stack: ..., a, d, c, b, c, d
1238   __ ldr(r2, at_tos_p5());  // load a
1239   __ ldr(r0, at_tos_p3());  // load c
1240   __ str(r2, at_tos_p3());  // store a in c
1241   __ str(r0, at_tos_p5());  // store c in a
1242   // stack: ..., c, d, a, b, c, d
1243 }
1244 
1245 void TemplateTable::swap()
1246 {
1247   transition(vtos, vtos);
1248   // stack: ..., a, b
1249   __ ldr(r2, at_tos_p1());  // load a
1250   __ ldr(r0, at_tos());  // load b
1251   __ str(r2, at_tos());  // store a in b
1252   __ str(r0, at_tos_p1());  // store b in a
1253   // stack: ..., b, a
1254 }
1255 
1256 void TemplateTable::iop2(Operation op)
1257 {
1258   transition(itos, itos);
1259   // r0 <== r1 op r0
1260   __ pop_i(r1);
1261   switch (op) {
1262   case add  : __ addw(r0, r1, r0); break;
1263   case sub  : __ subw(r0, r1, r0); break;
1264   case mul  : __ mulw(r0, r1, r0); break;
1265   case _and : __ andw(r0, r1, r0); break;
1266   case _or  : __ orrw(r0, r1, r0); break;
1267   case _xor : __ eorw(r0, r1, r0); break;
1268   case shl  : __ lslvw(r0, r1, r0); break;
1269   case shr  : __ asrvw(r0, r1, r0); break;
1270   case ushr : __ lsrvw(r0, r1, r0);break;
1271   default   : ShouldNotReachHere();
1272   }
1273 }
1274 
1275 void TemplateTable::lop2(Operation op)
1276 {
1277   transition(ltos, ltos);
1278   // r0 <== r1 op r0
1279   __ pop_l(r1);
1280   switch (op) {
1281   case add  : __ add(r0, r1, r0); break;
1282   case sub  : __ sub(r0, r1, r0); break;
1283   case mul  : __ mul(r0, r1, r0); break;
1284   case _and : __ andr(r0, r1, r0); break;
1285   case _or  : __ orr(r0, r1, r0); break;
1286   case _xor : __ eor(r0, r1, r0); break;
1287   default   : ShouldNotReachHere();
1288   }
1289 }
1290 
1291 void TemplateTable::idiv()
1292 {
1293   transition(itos, itos);
1294   // explicitly check for div0
1295   Label no_div0;
1296   __ cbnzw(r0, no_div0);
1297   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1298   __ br(rscratch1);
1299   __ bind(no_div0);
1300   __ pop_i(r1);
1301   // r0 <== r1 idiv r0
1302   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1303 }
1304 
1305 void TemplateTable::irem()
1306 {
1307   transition(itos, itos);
1308   // explicitly check for div0
1309   Label no_div0;
1310   __ cbnzw(r0, no_div0);
1311   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1312   __ br(rscratch1);
1313   __ bind(no_div0);
1314   __ pop_i(r1);
1315   // r0 <== r1 irem r0
1316   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1317 }
1318 
1319 void TemplateTable::lmul()
1320 {
1321   transition(ltos, ltos);
1322   __ pop_l(r1);
1323   __ mul(r0, r0, r1);
1324 }
1325 
1326 void TemplateTable::ldiv()
1327 {
1328   transition(ltos, ltos);
1329   // explicitly check for div0
1330   Label no_div0;
1331   __ cbnz(r0, no_div0);
1332   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1333   __ br(rscratch1);
1334   __ bind(no_div0);
1335   __ pop_l(r1);
1336   // r0 <== r1 ldiv r0
1337   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1338 }
1339 
1340 void TemplateTable::lrem()
1341 {
1342   transition(ltos, ltos);
1343   // explicitly check for div0
1344   Label no_div0;
1345   __ cbnz(r0, no_div0);
1346   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1347   __ br(rscratch1);
1348   __ bind(no_div0);
1349   __ pop_l(r1);
1350   // r0 <== r1 lrem r0
1351   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1352 }
1353 
1354 void TemplateTable::lshl()
1355 {
1356   transition(itos, ltos);
1357   // shift count is in r0
1358   __ pop_l(r1);
1359   __ lslv(r0, r1, r0);
1360 }
1361 
1362 void TemplateTable::lshr()
1363 {
1364   transition(itos, ltos);
1365   // shift count is in r0
1366   __ pop_l(r1);
1367   __ asrv(r0, r1, r0);
1368 }
1369 
1370 void TemplateTable::lushr()
1371 {
1372   transition(itos, ltos);
1373   // shift count is in r0
1374   __ pop_l(r1);
1375   __ lsrv(r0, r1, r0);
1376 }
1377 
1378 void TemplateTable::fop2(Operation op)
1379 {
1380   transition(ftos, ftos);
1381   switch (op) {
1382   case add:
1383     // n.b. use ldrd because this is a 64 bit slot
1384     __ pop_f(v1);
1385     __ fadds(v0, v1, v0);
1386     break;
1387   case sub:
1388     __ pop_f(v1);
1389     __ fsubs(v0, v1, v0);
1390     break;
1391   case mul:
1392     __ pop_f(v1);
1393     __ fmuls(v0, v1, v0);
1394     break;
1395   case div:
1396     __ pop_f(v1);
1397     __ fdivs(v0, v1, v0);
1398     break;
1399   case rem:
1400     __ fmovs(v1, v0);
1401     __ pop_f(v0);
1402     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1403                          0, 2, MacroAssembler::ret_type_float);
1404     break;
1405   default:
1406     ShouldNotReachHere();
1407     break;
1408   }
1409 }
1410 
1411 void TemplateTable::dop2(Operation op)
1412 {
1413   transition(dtos, dtos);
1414   switch (op) {
1415   case add:
1416     // n.b. use ldrd because this is a 64 bit slot
1417     __ pop_d(v1);
1418     __ faddd(v0, v1, v0);
1419     break;
1420   case sub:
1421     __ pop_d(v1);
1422     __ fsubd(v0, v1, v0);
1423     break;
1424   case mul:
1425     __ pop_d(v1);
1426     __ fmuld(v0, v1, v0);
1427     break;
1428   case div:
1429     __ pop_d(v1);
1430     __ fdivd(v0, v1, v0);
1431     break;
1432   case rem:
1433     __ fmovd(v1, v0);
1434     __ pop_d(v0);
1435     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1436                          0, 2, MacroAssembler::ret_type_double);
1437     break;
1438   default:
1439     ShouldNotReachHere();
1440     break;
1441   }
1442 }
1443 
1444 void TemplateTable::ineg()
1445 {
1446   transition(itos, itos);
1447   __ negw(r0, r0);
1448 
1449 }
1450 
1451 void TemplateTable::lneg()
1452 {
1453   transition(ltos, ltos);
1454   __ neg(r0, r0);
1455 }
1456 
1457 void TemplateTable::fneg()
1458 {
1459   transition(ftos, ftos);
1460   __ fnegs(v0, v0);
1461 }
1462 
1463 void TemplateTable::dneg()
1464 {
1465   transition(dtos, dtos);
1466   __ fnegd(v0, v0);
1467 }
1468 
1469 void TemplateTable::iinc()
1470 {
1471   transition(vtos, vtos);
1472   __ load_signed_byte(r1, at_bcp(2)); // get constant
1473   locals_index(r2);
1474   __ ldr(r0, iaddress(r2));
1475   __ addw(r0, r0, r1);
1476   __ str(r0, iaddress(r2));
1477 }
1478 
1479 void TemplateTable::wide_iinc()
1480 {
1481   transition(vtos, vtos);
1482   // __ mov(r1, zr);
1483   __ ldrw(r1, at_bcp(2)); // get constant and index
1484   __ rev16(r1, r1);
1485   __ ubfx(r2, r1, 0, 16);
1486   __ neg(r2, r2);
1487   __ sbfx(r1, r1, 16, 16);
1488   __ ldr(r0, iaddress(r2));
1489   __ addw(r0, r0, r1);
1490   __ str(r0, iaddress(r2));
1491 }
1492 
1493 void TemplateTable::convert()
1494 {
1495   // Checking
1496 #ifdef ASSERT
1497   {
1498     TosState tos_in  = ilgl;
1499     TosState tos_out = ilgl;
1500     switch (bytecode()) {
1501     case Bytecodes::_i2l: // fall through
1502     case Bytecodes::_i2f: // fall through
1503     case Bytecodes::_i2d: // fall through
1504     case Bytecodes::_i2b: // fall through
1505     case Bytecodes::_i2c: // fall through
1506     case Bytecodes::_i2s: tos_in = itos; break;
1507     case Bytecodes::_l2i: // fall through
1508     case Bytecodes::_l2f: // fall through
1509     case Bytecodes::_l2d: tos_in = ltos; break;
1510     case Bytecodes::_f2i: // fall through
1511     case Bytecodes::_f2l: // fall through
1512     case Bytecodes::_f2d: tos_in = ftos; break;
1513     case Bytecodes::_d2i: // fall through
1514     case Bytecodes::_d2l: // fall through
1515     case Bytecodes::_d2f: tos_in = dtos; break;
1516     default             : ShouldNotReachHere();
1517     }
1518     switch (bytecode()) {
1519     case Bytecodes::_l2i: // fall through
1520     case Bytecodes::_f2i: // fall through
1521     case Bytecodes::_d2i: // fall through
1522     case Bytecodes::_i2b: // fall through
1523     case Bytecodes::_i2c: // fall through
1524     case Bytecodes::_i2s: tos_out = itos; break;
1525     case Bytecodes::_i2l: // fall through
1526     case Bytecodes::_f2l: // fall through
1527     case Bytecodes::_d2l: tos_out = ltos; break;
1528     case Bytecodes::_i2f: // fall through
1529     case Bytecodes::_l2f: // fall through
1530     case Bytecodes::_d2f: tos_out = ftos; break;
1531     case Bytecodes::_i2d: // fall through
1532     case Bytecodes::_l2d: // fall through
1533     case Bytecodes::_f2d: tos_out = dtos; break;
1534     default             : ShouldNotReachHere();
1535     }
1536     transition(tos_in, tos_out);
1537   }
1538 #endif // ASSERT
1539   // static const int64_t is_nan = 0x8000000000000000L;
1540 
1541   // Conversion
1542   switch (bytecode()) {
1543   case Bytecodes::_i2l:
1544     __ sxtw(r0, r0);
1545     break;
1546   case Bytecodes::_i2f:
1547     __ scvtfws(v0, r0);
1548     break;
1549   case Bytecodes::_i2d:
1550     __ scvtfwd(v0, r0);
1551     break;
1552   case Bytecodes::_i2b:
1553     __ sxtbw(r0, r0);
1554     break;
1555   case Bytecodes::_i2c:
1556     __ uxthw(r0, r0);
1557     break;
1558   case Bytecodes::_i2s:
1559     __ sxthw(r0, r0);
1560     break;
1561   case Bytecodes::_l2i:
1562     __ uxtw(r0, r0);
1563     break;
1564   case Bytecodes::_l2f:
1565     __ scvtfs(v0, r0);
1566     break;
1567   case Bytecodes::_l2d:
1568     __ scvtfd(v0, r0);
1569     break;
1570   case Bytecodes::_f2i:
1571   {
1572     Label L_Okay;
1573     __ clear_fpsr();
1574     __ fcvtzsw(r0, v0);
1575     __ get_fpsr(r1);
1576     __ cbzw(r1, L_Okay);
1577     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1578                          0, 1, MacroAssembler::ret_type_integral);
1579     __ bind(L_Okay);
1580   }
1581     break;
1582   case Bytecodes::_f2l:
1583   {
1584     Label L_Okay;
1585     __ clear_fpsr();
1586     __ fcvtzs(r0, v0);
1587     __ get_fpsr(r1);
1588     __ cbzw(r1, L_Okay);
1589     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1590                          0, 1, MacroAssembler::ret_type_integral);
1591     __ bind(L_Okay);
1592   }
1593     break;
1594   case Bytecodes::_f2d:
1595     __ fcvts(v0, v0);
1596     break;
1597   case Bytecodes::_d2i:
1598   {
1599     Label L_Okay;
1600     __ clear_fpsr();
1601     __ fcvtzdw(r0, v0);
1602     __ get_fpsr(r1);
1603     __ cbzw(r1, L_Okay);
1604     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1605                          0, 1, MacroAssembler::ret_type_integral);
1606     __ bind(L_Okay);
1607   }
1608     break;
1609   case Bytecodes::_d2l:
1610   {
1611     Label L_Okay;
1612     __ clear_fpsr();
1613     __ fcvtzd(r0, v0);
1614     __ get_fpsr(r1);
1615     __ cbzw(r1, L_Okay);
1616     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1617                          0, 1, MacroAssembler::ret_type_integral);
1618     __ bind(L_Okay);
1619   }
1620     break;
1621   case Bytecodes::_d2f:
1622     __ fcvtd(v0, v0);
1623     break;
1624   default:
1625     ShouldNotReachHere();
1626   }
1627 }
1628 
1629 void TemplateTable::lcmp()
1630 {
1631   transition(ltos, itos);
1632   Label done;
1633   __ pop_l(r1);
1634   __ cmp(r1, r0);
1635   __ mov(r0, (u_int64_t)-1L);
1636   __ br(Assembler::LT, done);
1637   // __ mov(r0, 1UL);
1638   // __ csel(r0, r0, zr, Assembler::NE);
1639   // and here is a faster way
1640   __ csinc(r0, zr, zr, Assembler::EQ);
1641   __ bind(done);
1642 }
1643 
1644 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1645 {
1646   Label done;
1647   if (is_float) {
1648     // XXX get rid of pop here, use ... reg, mem32
1649     __ pop_f(v1);
1650     __ fcmps(v1, v0);
1651   } else {
1652     // XXX get rid of pop here, use ... reg, mem64
1653     __ pop_d(v1);
1654     __ fcmpd(v1, v0);
1655   }
1656   if (unordered_result < 0) {
1657     // we want -1 for unordered or less than, 0 for equal and 1 for
1658     // greater than.
1659     __ mov(r0, (u_int64_t)-1L);
1660     // for FP LT tests less than or unordered
1661     __ br(Assembler::LT, done);
1662     // install 0 for EQ otherwise 1
1663     __ csinc(r0, zr, zr, Assembler::EQ);
1664   } else {
1665     // we want -1 for less than, 0 for equal and 1 for unordered or
1666     // greater than.
1667     __ mov(r0, 1L);
1668     // for FP HI tests greater than or unordered
1669     __ br(Assembler::HI, done);
1670     // install 0 for EQ otherwise ~0
1671     __ csinv(r0, zr, zr, Assembler::EQ);
1672 
1673   }
1674   __ bind(done);
1675 }
1676 
1677 void TemplateTable::branch(bool is_jsr, bool is_wide)
1678 {
1679   // We might be moving to a safepoint.  The thread which calls
1680   // Interpreter::notice_safepoints() will effectively flush its cache
1681   // when it makes a system call, but we need to do something to
1682   // ensure that we see the changed dispatch table.
1683   __ membar(MacroAssembler::LoadLoad);
1684 
1685   __ profile_taken_branch(r0, r1);
1686   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1687                              InvocationCounter::counter_offset();
1688   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1689                               InvocationCounter::counter_offset();
1690 
1691   // load branch displacement
1692   if (!is_wide) {
1693     __ ldrh(r2, at_bcp(1));
1694     __ rev16(r2, r2);
1695     // sign extend the 16 bit value in r2
1696     __ sbfm(r2, r2, 0, 15);
1697   } else {
1698     __ ldrw(r2, at_bcp(1));
1699     __ revw(r2, r2);
1700     // sign extend the 32 bit value in r2
1701     __ sbfm(r2, r2, 0, 31);
1702   }
1703 
1704   // Handle all the JSR stuff here, then exit.
1705   // It's much shorter and cleaner than intermingling with the non-JSR
1706   // normal-branch stuff occurring below.
1707 
1708   if (is_jsr) {
1709     // Pre-load the next target bytecode into rscratch1
1710     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1711     // compute return address as bci
1712     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1713     __ add(rscratch2, rscratch2,
1714            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1715     __ sub(r1, rbcp, rscratch2);
1716     __ push_i(r1);
1717     // Adjust the bcp by the 16-bit displacement in r2
1718     __ add(rbcp, rbcp, r2);
1719     __ dispatch_only(vtos);
1720     return;
1721   }
1722 
1723   // Normal (non-jsr) branch handling
1724 
1725   // Adjust the bcp by the displacement in r2
1726   __ add(rbcp, rbcp, r2);
1727 
1728   assert(UseLoopCounter || !UseOnStackReplacement,
1729          "on-stack-replacement requires loop counters");
1730   Label backedge_counter_overflow;
1731   Label profile_method;
1732   Label dispatch;
1733   if (UseLoopCounter) {
1734     // increment backedge counter for backward branches
1735     // r0: MDO
1736     // w1: MDO bumped taken-count
1737     // r2: target offset
1738     __ cmp(r2, zr);
1739     __ br(Assembler::GT, dispatch); // count only if backward branch
1740 
1741     // ECN: FIXME: This code smells
1742     // check if MethodCounters exists
1743     Label has_counters;
1744     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1745     __ cbnz(rscratch1, has_counters);
1746     __ push(r0);
1747     __ push(r1);
1748     __ push(r2);
1749     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1750             InterpreterRuntime::build_method_counters), rmethod);
1751     __ pop(r2);
1752     __ pop(r1);
1753     __ pop(r0);
1754     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1755     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1756     __ bind(has_counters);
1757 
1758     if (TieredCompilation) {
1759       Label no_mdo;
1760       int increment = InvocationCounter::count_increment;
1761       if (ProfileInterpreter) {
1762         // Are we profiling?
1763         __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1764         __ cbz(r1, no_mdo);
1765         // Increment the MDO backedge counter
1766         const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1767                                            in_bytes(InvocationCounter::counter_offset()));
1768         const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1769         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1770                                    r0, rscratch1, false, Assembler::EQ, &backedge_counter_overflow);
1771         __ b(dispatch);
1772       }
1773       __ bind(no_mdo);
1774       // Increment backedge counter in MethodCounters*
1775       __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1776       const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1777       __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1778                                  r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1779     } else { // not TieredCompilation
1780       // increment counter
1781       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1782       __ ldrw(r0, Address(rscratch2, be_offset));        // load backedge counter
1783       __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1784       __ strw(rscratch1, Address(rscratch2, be_offset));        // store counter
1785 
1786       __ ldrw(r0, Address(rscratch2, inv_offset));    // load invocation counter
1787       __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1788       __ addw(r0, r0, rscratch1);        // add both counters
1789 
1790       if (ProfileInterpreter) {
1791         // Test to see if we should create a method data oop
1792         __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1793         __ cmpw(r0, rscratch1);
1794         __ br(Assembler::LT, dispatch);
1795 
1796         // if no method data exists, go to profile method
1797         __ test_method_data_pointer(r0, profile_method);
1798 
1799         if (UseOnStackReplacement) {
1800           // check for overflow against w1 which is the MDO taken count
1801           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1802           __ cmpw(r1, rscratch1);
1803           __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1804 
1805           // When ProfileInterpreter is on, the backedge_count comes
1806           // from the MethodData*, which value does not get reset on
1807           // the call to frequency_counter_overflow().  To avoid
1808           // excessive calls to the overflow routine while the method is
1809           // being compiled, add a second test to make sure the overflow
1810           // function is called only once every overflow_frequency.
1811           const int overflow_frequency = 1024;
1812           __ andsw(r1, r1, overflow_frequency - 1);
1813           __ br(Assembler::EQ, backedge_counter_overflow);
1814 
1815         }
1816       } else {
1817         if (UseOnStackReplacement) {
1818           // check for overflow against w0, which is the sum of the
1819           // counters
1820           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1821           __ cmpw(r0, rscratch1);
1822           __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1823         }
1824       }
1825     }
1826   }
1827   __ bind(dispatch);
1828 
1829   // Pre-load the next target bytecode into rscratch1
1830   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1831 
1832   // continue with the bytecode @ target
1833   // rscratch1: target bytecode
1834   // rbcp: target bcp
1835   __ dispatch_only(vtos);
1836 
1837   if (UseLoopCounter) {
1838     if (ProfileInterpreter) {
1839       // Out-of-line code to allocate method data oop.
1840       __ bind(profile_method);
1841       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1842       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1843       __ set_method_data_pointer_for_bcp();
1844       __ b(dispatch);
1845     }
1846 
1847     if (TieredCompilation || UseOnStackReplacement) {
1848       // invocation counter overflow
1849       __ bind(backedge_counter_overflow);
1850       __ neg(r2, r2);
1851       __ add(r2, r2, rbcp);     // branch bcp
1852       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1853       __ call_VM(noreg,
1854                  CAST_FROM_FN_PTR(address,
1855                                   InterpreterRuntime::frequency_counter_overflow),
1856                  r2);
1857       if (!UseOnStackReplacement)
1858         __ b(dispatch);
1859     }
1860 
1861     if (UseOnStackReplacement) {
1862       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1863 
1864       // r0: osr nmethod (osr ok) or NULL (osr not possible)
1865       // w1: target bytecode
1866       // r2: scratch
1867       __ cbz(r0, dispatch);     // test result -- no osr if null
1868       // nmethod may have been invalidated (VM may block upon call_VM return)
1869       __ ldrb(r2, Address(r0, nmethod::state_offset()));
1870       if (nmethod::in_use != 0)
1871         __ sub(r2, r2, nmethod::in_use);
1872       __ cbnz(r2, dispatch);
1873 
1874       // We have the address of an on stack replacement routine in r0
1875       // We need to prepare to execute the OSR method. First we must
1876       // migrate the locals and monitors off of the stack.
1877 
1878       __ mov(r19, r0);                             // save the nmethod
1879 
1880       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1881 
1882       // r0 is OSR buffer, move it to expected parameter location
1883       __ mov(j_rarg0, r0);
1884 
1885       // remove activation
1886       // get sender esp
1887       __ ldr(esp,
1888           Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1889       // remove frame anchor
1890       __ leave();
1891       // Ensure compiled code always sees stack at proper alignment
1892       __ andr(sp, esp, -16);
1893 
1894       // and begin the OSR nmethod
1895       __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1896       __ br(rscratch1);
1897     }
1898   }
1899 }
1900 
1901 
1902 void TemplateTable::if_0cmp(Condition cc)
1903 {
1904   transition(itos, vtos);
1905   // assume branch is more often taken than not (loops use backward branches)
1906   Label not_taken;
1907   if (cc == equal)
1908     __ cbnzw(r0, not_taken);
1909   else if (cc == not_equal)
1910     __ cbzw(r0, not_taken);
1911   else {
1912     __ andsw(zr, r0, r0);
1913     __ br(j_not(cc), not_taken);
1914   }
1915 
1916   branch(false, false);
1917   __ bind(not_taken);
1918   __ profile_not_taken_branch(r0);
1919 }
1920 
1921 void TemplateTable::if_icmp(Condition cc)
1922 {
1923   transition(itos, vtos);
1924   // assume branch is more often taken than not (loops use backward branches)
1925   Label not_taken;
1926   __ pop_i(r1);
1927   __ cmpw(r1, r0, Assembler::LSL);
1928   __ br(j_not(cc), not_taken);
1929   branch(false, false);
1930   __ bind(not_taken);
1931   __ profile_not_taken_branch(r0);
1932 }
1933 
1934 void TemplateTable::if_nullcmp(Condition cc)
1935 {
1936   transition(atos, vtos);
1937   // assume branch is more often taken than not (loops use backward branches)
1938   Label not_taken;
1939   if (cc == equal)
1940     __ cbnz(r0, not_taken);
1941   else
1942     __ cbz(r0, not_taken);
1943   branch(false, false);
1944   __ bind(not_taken);
1945   __ profile_not_taken_branch(r0);
1946 }
1947 
1948 void TemplateTable::if_acmp(Condition cc)
1949 {
1950   transition(atos, vtos);
1951   // assume branch is more often taken than not (loops use backward branches)
1952   Label not_taken;
1953   __ pop_ptr(r1);
1954   __ cmp(r1, r0);
1955   __ br(j_not(cc), not_taken);
1956   branch(false, false);
1957   __ bind(not_taken);
1958   __ profile_not_taken_branch(r0);
1959 }
1960 
1961 void TemplateTable::ret() {
1962   transition(vtos, vtos);
1963   // We might be moving to a safepoint.  The thread which calls
1964   // Interpreter::notice_safepoints() will effectively flush its cache
1965   // when it makes a system call, but we need to do something to
1966   // ensure that we see the changed dispatch table.
1967   __ membar(MacroAssembler::LoadLoad);
1968 
1969   locals_index(r1);
1970   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1971   __ profile_ret(r1, r2);
1972   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1973   __ lea(rbcp, Address(rbcp, r1));
1974   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1975   __ dispatch_next(vtos);
1976 }
1977 
1978 void TemplateTable::wide_ret() {
1979   transition(vtos, vtos);
1980   locals_index_wide(r1);
1981   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1982   __ profile_ret(r1, r2);
1983   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1984   __ lea(rbcp, Address(rbcp, r1));
1985   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1986   __ dispatch_next(vtos);
1987 }
1988 
1989 
1990 void TemplateTable::tableswitch() {
1991   Label default_case, continue_execution;
1992   transition(itos, vtos);
1993   // align rbcp
1994   __ lea(r1, at_bcp(BytesPerInt));
1995   __ andr(r1, r1, -BytesPerInt);
1996   // load lo & hi
1997   __ ldrw(r2, Address(r1, BytesPerInt));
1998   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
1999   __ rev32(r2, r2);
2000   __ rev32(r3, r3);
2001   // check against lo & hi
2002   __ cmpw(r0, r2);
2003   __ br(Assembler::LT, default_case);
2004   __ cmpw(r0, r3);
2005   __ br(Assembler::GT, default_case);
2006   // lookup dispatch offset
2007   __ subw(r0, r0, r2);
2008   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2009   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2010   __ profile_switch_case(r0, r1, r2);
2011   // continue execution
2012   __ bind(continue_execution);
2013   __ rev32(r3, r3);
2014   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2015   __ add(rbcp, rbcp, r3, ext::sxtw);
2016   __ dispatch_only(vtos);
2017   // handle default
2018   __ bind(default_case);
2019   __ profile_switch_default(r0);
2020   __ ldrw(r3, Address(r1, 0));
2021   __ b(continue_execution);
2022 }
2023 
2024 void TemplateTable::lookupswitch() {
2025   transition(itos, itos);
2026   __ stop("lookupswitch bytecode should have been rewritten");
2027 }
2028 
2029 void TemplateTable::fast_linearswitch() {
2030   transition(itos, vtos);
2031   Label loop_entry, loop, found, continue_execution;
2032   // bswap r0 so we can avoid bswapping the table entries
2033   __ rev32(r0, r0);
2034   // align rbcp
2035   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2036                                     // this instruction (change offsets
2037                                     // below)
2038   __ andr(r19, r19, -BytesPerInt);
2039   // set counter
2040   __ ldrw(r1, Address(r19, BytesPerInt));
2041   __ rev32(r1, r1);
2042   __ b(loop_entry);
2043   // table search
2044   __ bind(loop);
2045   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2046   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2047   __ cmpw(r0, rscratch1);
2048   __ br(Assembler::EQ, found);
2049   __ bind(loop_entry);
2050   __ subs(r1, r1, 1);
2051   __ br(Assembler::PL, loop);
2052   // default case
2053   __ profile_switch_default(r0);
2054   __ ldrw(r3, Address(r19, 0));
2055   __ b(continue_execution);
2056   // entry found -> get offset
2057   __ bind(found);
2058   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2059   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2060   __ profile_switch_case(r1, r0, r19);
2061   // continue execution
2062   __ bind(continue_execution);
2063   __ rev32(r3, r3);
2064   __ add(rbcp, rbcp, r3, ext::sxtw);
2065   __ ldrb(rscratch1, Address(rbcp, 0));
2066   __ dispatch_only(vtos);
2067 }
2068 
2069 void TemplateTable::fast_binaryswitch() {
2070   transition(itos, vtos);
2071   // Implementation using the following core algorithm:
2072   //
2073   // int binary_search(int key, LookupswitchPair* array, int n) {
2074   //   // Binary search according to "Methodik des Programmierens" by
2075   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2076   //   int i = 0;
2077   //   int j = n;
2078   //   while (i+1 < j) {
2079   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2080   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2081   //     // where a stands for the array and assuming that the (inexisting)
2082   //     // element a[n] is infinitely big.
2083   //     int h = (i + j) >> 1;
2084   //     // i < h < j
2085   //     if (key < array[h].fast_match()) {
2086   //       j = h;
2087   //     } else {
2088   //       i = h;
2089   //     }
2090   //   }
2091   //   // R: a[i] <= key < a[i+1] or Q
2092   //   // (i.e., if key is within array, i is the correct index)
2093   //   return i;
2094   // }
2095 
2096   // Register allocation
2097   const Register key   = r0; // already set (tosca)
2098   const Register array = r1;
2099   const Register i     = r2;
2100   const Register j     = r3;
2101   const Register h     = rscratch1;
2102   const Register temp  = rscratch2;
2103 
2104   // Find array start
2105   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2106                                           // get rid of this
2107                                           // instruction (change
2108                                           // offsets below)
2109   __ andr(array, array, -BytesPerInt);
2110 
2111   // Initialize i & j
2112   __ mov(i, 0);                            // i = 0;
2113   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2114 
2115   // Convert j into native byteordering
2116   __ rev32(j, j);
2117 
2118   // And start
2119   Label entry;
2120   __ b(entry);
2121 
2122   // binary search loop
2123   {
2124     Label loop;
2125     __ bind(loop);
2126     // int h = (i + j) >> 1;
2127     __ addw(h, i, j);                           // h = i + j;
2128     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2129     // if (key < array[h].fast_match()) {
2130     //   j = h;
2131     // } else {
2132     //   i = h;
2133     // }
2134     // Convert array[h].match to native byte-ordering before compare
2135     __ ldr(temp, Address(array, h, Address::lsl(3)));
2136     __ rev32(temp, temp);
2137     __ cmpw(key, temp);
2138     // j = h if (key <  array[h].fast_match())
2139     __ csel(j, h, j, Assembler::LT);
2140     // i = h if (key >= array[h].fast_match())
2141     __ csel(i, h, i, Assembler::GE);
2142     // while (i+1 < j)
2143     __ bind(entry);
2144     __ addw(h, i, 1);          // i+1
2145     __ cmpw(h, j);             // i+1 < j
2146     __ br(Assembler::LT, loop);
2147   }
2148 
2149   // end of binary search, result index is i (must check again!)
2150   Label default_case;
2151   // Convert array[i].match to native byte-ordering before compare
2152   __ ldr(temp, Address(array, i, Address::lsl(3)));
2153   __ rev32(temp, temp);
2154   __ cmpw(key, temp);
2155   __ br(Assembler::NE, default_case);
2156 
2157   // entry found -> j = offset
2158   __ add(j, array, i, ext::uxtx, 3);
2159   __ ldrw(j, Address(j, BytesPerInt));
2160   __ profile_switch_case(i, key, array);
2161   __ rev32(j, j);
2162   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2163   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2164   __ dispatch_only(vtos);
2165 
2166   // default case -> j = default offset
2167   __ bind(default_case);
2168   __ profile_switch_default(i);
2169   __ ldrw(j, Address(array, -2 * BytesPerInt));
2170   __ rev32(j, j);
2171   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2172   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2173   __ dispatch_only(vtos);
2174 }
2175 
2176 
2177 void TemplateTable::_return(TosState state)
2178 {
2179   transition(state, state);
2180   assert(_desc->calls_vm(),
2181          "inconsistent calls_vm information"); // call in remove_activation
2182 
2183   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2184     assert(state == vtos, "only valid state");
2185 
2186     __ ldr(c_rarg1, aaddress(0));
2187     __ load_klass(r3, c_rarg1);
2188     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2189     Label skip_register_finalizer;
2190     __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2191 
2192     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2193 
2194     __ bind(skip_register_finalizer);
2195   }
2196 
2197   // Issue a StoreStore barrier after all stores but before return
2198   // from any constructor for any class with a final field.  We don't
2199   // know if this is a finalizer, so we always do so.
2200   if (_desc->bytecode() == Bytecodes::_return)
2201     __ membar(MacroAssembler::StoreStore);
2202 
2203   // Narrow result if state is itos but result type is smaller.
2204   // Need to narrow in the return bytecode rather than in generate_return_entry
2205   // since compiled code callers expect the result to already be narrowed.
2206   if (state == itos) {
2207     __ narrow(r0);
2208   }
2209 
2210   __ remove_activation(state);
2211   __ ret(lr);
2212 }
2213 
2214 // ----------------------------------------------------------------------------
2215 // Volatile variables demand their effects be made known to all CPU's
2216 // in order.  Store buffers on most chips allow reads & writes to
2217 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2218 // without some kind of memory barrier (i.e., it's not sufficient that
2219 // the interpreter does not reorder volatile references, the hardware
2220 // also must not reorder them).
2221 //
2222 // According to the new Java Memory Model (JMM):
2223 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2224 //     writes act as aquire & release, so:
2225 // (2) A read cannot let unrelated NON-volatile memory refs that
2226 //     happen after the read float up to before the read.  It's OK for
2227 //     non-volatile memory refs that happen before the volatile read to
2228 //     float down below it.
2229 // (3) Similar a volatile write cannot let unrelated NON-volatile
2230 //     memory refs that happen BEFORE the write float down to after the
2231 //     write.  It's OK for non-volatile memory refs that happen after the
2232 //     volatile write to float up before it.
2233 //
2234 // We only put in barriers around volatile refs (they are expensive),
2235 // not _between_ memory refs (that would require us to track the
2236 // flavor of the previous memory refs).  Requirements (2) and (3)
2237 // require some barriers before volatile stores and after volatile
2238 // loads.  These nearly cover requirement (1) but miss the
2239 // volatile-store-volatile-load case.  This final case is placed after
2240 // volatile-stores although it could just as well go before
2241 // volatile-loads.
2242 
2243 void TemplateTable::resolve_cache_and_index(int byte_no,
2244                                             Register Rcache,
2245                                             Register index,
2246                                             size_t index_size) {
2247   const Register temp = r19;
2248   assert_different_registers(Rcache, index, temp);
2249 
2250   Label resolved;
2251 
2252   Bytecodes::Code code = bytecode();
2253   switch (code) {
2254   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2255   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2256   }
2257 
2258   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2259   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2260   __ cmp(temp, (int) code);  // have we resolved this bytecode?
2261   __ br(Assembler::EQ, resolved);
2262 
2263   // resolve first time through
2264   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2265   __ mov(temp, (int) code);
2266   __ call_VM(noreg, entry, temp);
2267 
2268   // Update registers with resolved info
2269   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2270   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2271   // so all clients ofthis method must be modified accordingly
2272   __ bind(resolved);
2273 }
2274 
2275 // The Rcache and index registers must be set before call
2276 // n.b unlike x86 cache already includes the index offset
2277 void TemplateTable::load_field_cp_cache_entry(Register obj,
2278                                               Register cache,
2279                                               Register index,
2280                                               Register off,
2281                                               Register flags,
2282                                               bool is_static = false) {
2283   assert_different_registers(cache, index, flags, off);
2284 
2285   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2286   // Field offset
2287   __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2288                                           ConstantPoolCacheEntry::f2_offset())));
2289   // Flags
2290   __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2291                                            ConstantPoolCacheEntry::flags_offset())));
2292 
2293   // klass overwrite register
2294   if (is_static) {
2295     __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2296                                         ConstantPoolCacheEntry::f1_offset())));
2297     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2298     __ ldr(obj, Address(obj, mirror_offset));
2299   }
2300 }
2301 
2302 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2303                                                Register method,
2304                                                Register itable_index,
2305                                                Register flags,
2306                                                bool is_invokevirtual,
2307                                                bool is_invokevfinal, /*unused*/
2308                                                bool is_invokedynamic) {
2309   // setup registers
2310   const Register cache = rscratch2;
2311   const Register index = r4;
2312   assert_different_registers(method, flags);
2313   assert_different_registers(method, cache, index);
2314   assert_different_registers(itable_index, flags);
2315   assert_different_registers(itable_index, cache, index);
2316   // determine constant pool cache field offsets
2317   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2318   const int method_offset = in_bytes(
2319     ConstantPoolCache::base_offset() +
2320       (is_invokevirtual
2321        ? ConstantPoolCacheEntry::f2_offset()
2322        : ConstantPoolCacheEntry::f1_offset()));
2323   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2324                                     ConstantPoolCacheEntry::flags_offset());
2325   // access constant pool cache fields
2326   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2327                                     ConstantPoolCacheEntry::f2_offset());
2328 
2329   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2330   resolve_cache_and_index(byte_no, cache, index, index_size);
2331   __ ldr(method, Address(cache, method_offset));
2332 
2333   if (itable_index != noreg) {
2334     __ ldr(itable_index, Address(cache, index_offset));
2335   }
2336   __ ldrw(flags, Address(cache, flags_offset));
2337 }
2338 
2339 
2340 // The registers cache and index expected to be set before call.
2341 // Correct values of the cache and index registers are preserved.
2342 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2343                                             bool is_static, bool has_tos) {
2344   // do the JVMTI work here to avoid disturbing the register state below
2345   // We use c_rarg registers here because we want to use the register used in
2346   // the call to the VM
2347   if (JvmtiExport::can_post_field_access()) {
2348     // Check to see if a field access watch has been set before we
2349     // take the time to call into the VM.
2350     Label L1;
2351     assert_different_registers(cache, index, r0);
2352     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2353     __ ldrw(r0, Address(rscratch1));
2354     __ cbzw(r0, L1);
2355 
2356     __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2357     __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2358 
2359     if (is_static) {
2360       __ mov(c_rarg1, zr); // NULL object reference
2361     } else {
2362       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2363       __ verify_oop(c_rarg1);
2364     }
2365     // c_rarg1: object pointer or NULL
2366     // c_rarg2: cache entry pointer
2367     // c_rarg3: jvalue object on the stack
2368     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2369                                        InterpreterRuntime::post_field_access),
2370                c_rarg1, c_rarg2, c_rarg3);
2371     __ get_cache_and_index_at_bcp(cache, index, 1);
2372     __ bind(L1);
2373   }
2374 }
2375 
2376 void TemplateTable::pop_and_check_object(Register r)
2377 {
2378   __ pop_ptr(r);
2379   __ null_check(r);  // for field access must check obj.
2380   __ verify_oop(r);
2381 }
2382 
2383 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2384 {
2385   const Register cache = r2;
2386   const Register index = r3;
2387   const Register obj   = r4;
2388   const Register off   = r19;
2389   const Register flags = r0;
2390   const Register raw_flags = r6;
2391   const Register bc    = r4; // uses same reg as obj, so don't mix them
2392 
2393   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2394   jvmti_post_field_access(cache, index, is_static, false);
2395   load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2396 
2397   if (!is_static) {
2398     // obj is on the stack
2399     pop_and_check_object(obj);
2400   }
2401 
2402   // 8179954: We need to make sure that the code generated for
2403   // volatile accesses forms a sequentially-consistent set of
2404   // operations when combined with STLR and LDAR.  Without a leading
2405   // membar it's possible for a simple Dekker test to fail if loads
2406   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2407   // the stores in one method and we interpret the loads in another.
2408   if (! UseBarriersForVolatile) {
2409     Label notVolatile;
2410     __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2411     __ membar(MacroAssembler::AnyAny);
2412     __ bind(notVolatile);
2413   }
2414 
2415   const Address field(obj, off);
2416 
2417   Label Done, notByte, notBool, notInt, notShort, notChar,
2418               notLong, notFloat, notObj, notDouble;
2419 
2420   // x86 uses a shift and mask or wings it with a shift plus assert
2421   // the mask is not needed. aarch64 just uses bitfield extract
2422   __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2423            ConstantPoolCacheEntry::tos_state_bits);
2424 
2425   assert(btos == 0, "change code, btos != 0");
2426   __ cbnz(flags, notByte);
2427 
2428   // Don't rewrite getstatic, only getfield
2429   if (is_static) rc = may_not_rewrite;
2430 
2431   // btos
2432   __ load_signed_byte(r0, field);
2433   __ push(btos);
2434   // Rewrite bytecode to be faster
2435   if (rc == may_rewrite) {
2436     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2437   }
2438   __ b(Done);
2439 
2440   __ bind(notByte);
2441   __ cmp(flags, ztos);
2442   __ br(Assembler::NE, notBool);
2443 
2444   // ztos (same code as btos)
2445   __ ldrsb(r0, field);
2446   __ push(ztos);
2447   // Rewrite bytecode to be faster
2448   if (rc == may_rewrite) {
2449     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2450     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2451   }
2452   __ b(Done);
2453 
2454   __ bind(notBool);
2455   __ cmp(flags, atos);
2456   __ br(Assembler::NE, notObj);
2457   // atos
2458   __ load_heap_oop(r0, field);
2459   __ push(atos);
2460   if (rc == may_rewrite) {
2461     patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2462   }
2463   __ b(Done);
2464 
2465   __ bind(notObj);
2466   __ cmp(flags, itos);
2467   __ br(Assembler::NE, notInt);
2468   // itos
2469   __ ldrw(r0, field);
2470   __ push(itos);
2471   // Rewrite bytecode to be faster
2472   if (rc == may_rewrite) {
2473     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2474   }
2475   __ b(Done);
2476 
2477   __ bind(notInt);
2478   __ cmp(flags, ctos);
2479   __ br(Assembler::NE, notChar);
2480   // ctos
2481   __ load_unsigned_short(r0, field);
2482   __ push(ctos);
2483   // Rewrite bytecode to be faster
2484   if (rc == may_rewrite) {
2485     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2486   }
2487   __ b(Done);
2488 
2489   __ bind(notChar);
2490   __ cmp(flags, stos);
2491   __ br(Assembler::NE, notShort);
2492   // stos
2493   __ load_signed_short(r0, field);
2494   __ push(stos);
2495   // Rewrite bytecode to be faster
2496   if (rc == may_rewrite) {
2497     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2498   }
2499   __ b(Done);
2500 
2501   __ bind(notShort);
2502   __ cmp(flags, ltos);
2503   __ br(Assembler::NE, notLong);
2504   // ltos
2505   __ ldr(r0, field);
2506   __ push(ltos);
2507   // Rewrite bytecode to be faster
2508   if (rc == may_rewrite) {
2509     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2510   }
2511   __ b(Done);
2512 
2513   __ bind(notLong);
2514   __ cmp(flags, ftos);
2515   __ br(Assembler::NE, notFloat);
2516   // ftos
2517   __ ldrs(v0, field);
2518   __ push(ftos);
2519   // Rewrite bytecode to be faster
2520   if (rc == may_rewrite) {
2521     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2522   }
2523   __ b(Done);
2524 
2525   __ bind(notFloat);
2526 #ifdef ASSERT
2527   __ cmp(flags, dtos);
2528   __ br(Assembler::NE, notDouble);
2529 #endif
2530   // dtos
2531   __ ldrd(v0, field);
2532   __ push(dtos);
2533   // Rewrite bytecode to be faster
2534   if (rc == may_rewrite) {
2535     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2536   }
2537 #ifdef ASSERT
2538   __ b(Done);
2539 
2540   __ bind(notDouble);
2541   __ stop("Bad state");
2542 #endif
2543 
2544   __ bind(Done);
2545 
2546   Label notVolatile;
2547   __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2548   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2549   __ bind(notVolatile);
2550 }
2551 
2552 
2553 void TemplateTable::getfield(int byte_no)
2554 {
2555   getfield_or_static(byte_no, false);
2556 }
2557 
2558 void TemplateTable::nofast_getfield(int byte_no) {
2559   getfield_or_static(byte_no, false, may_not_rewrite);
2560 }
2561 
2562 void TemplateTable::getstatic(int byte_no)
2563 {
2564   getfield_or_static(byte_no, true);
2565 }
2566 
2567 // The registers cache and index expected to be set before call.
2568 // The function may destroy various registers, just not the cache and index registers.
2569 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2570   transition(vtos, vtos);
2571 
2572   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2573 
2574   if (JvmtiExport::can_post_field_modification()) {
2575     // Check to see if a field modification watch has been set before
2576     // we take the time to call into the VM.
2577     Label L1;
2578     assert_different_registers(cache, index, r0);
2579     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2580     __ ldrw(r0, Address(rscratch1));
2581     __ cbz(r0, L1);
2582 
2583     __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2584 
2585     if (is_static) {
2586       // Life is simple.  Null out the object pointer.
2587       __ mov(c_rarg1, zr);
2588     } else {
2589       // Life is harder. The stack holds the value on top, followed by
2590       // the object.  We don't know the size of the value, though; it
2591       // could be one or two words depending on its type. As a result,
2592       // we must find the type to determine where the object is.
2593       __ ldrw(c_rarg3, Address(c_rarg2,
2594                                in_bytes(cp_base_offset +
2595                                         ConstantPoolCacheEntry::flags_offset())));
2596       __ lsr(c_rarg3, c_rarg3,
2597              ConstantPoolCacheEntry::tos_state_shift);
2598       ConstantPoolCacheEntry::verify_tos_state_shift();
2599       Label nope2, done, ok;
2600       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2601       __ cmpw(c_rarg3, ltos);
2602       __ br(Assembler::EQ, ok);
2603       __ cmpw(c_rarg3, dtos);
2604       __ br(Assembler::NE, nope2);
2605       __ bind(ok);
2606       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2607       __ bind(nope2);
2608     }
2609     // cache entry pointer
2610     __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2611     // object (tos)
2612     __ mov(c_rarg3, esp);
2613     // c_rarg1: object pointer set up above (NULL if static)
2614     // c_rarg2: cache entry pointer
2615     // c_rarg3: jvalue object on the stack
2616     __ call_VM(noreg,
2617                CAST_FROM_FN_PTR(address,
2618                                 InterpreterRuntime::post_field_modification),
2619                c_rarg1, c_rarg2, c_rarg3);
2620     __ get_cache_and_index_at_bcp(cache, index, 1);
2621     __ bind(L1);
2622   }
2623 }
2624 
2625 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2626   transition(vtos, vtos);
2627 
2628   const Register cache = r2;
2629   const Register index = r3;
2630   const Register obj   = r2;
2631   const Register off   = r19;
2632   const Register flags = r0;
2633   const Register bc    = r4;
2634 
2635   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2636   jvmti_post_field_mod(cache, index, is_static);
2637   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2638 
2639   Label Done;
2640   __ mov(r5, flags);
2641 
2642   {
2643     Label notVolatile;
2644     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2645     __ membar(MacroAssembler::StoreStore);
2646     __ bind(notVolatile);
2647   }
2648 
2649   // field address
2650   const Address field(obj, off);
2651 
2652   Label notByte, notBool, notInt, notShort, notChar,
2653         notLong, notFloat, notObj, notDouble;
2654 
2655   // x86 uses a shift and mask or wings it with a shift plus assert
2656   // the mask is not needed. aarch64 just uses bitfield extract
2657   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2658 
2659   assert(btos == 0, "change code, btos != 0");
2660   __ cbnz(flags, notByte);
2661 
2662   // Don't rewrite putstatic, only putfield
2663   if (is_static) rc = may_not_rewrite;
2664 
2665   // btos
2666   {
2667     __ pop(btos);
2668     if (!is_static) pop_and_check_object(obj);
2669     __ strb(r0, field);
2670     if (rc == may_rewrite) {
2671       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2672     }
2673     __ b(Done);
2674   }
2675 
2676   __ bind(notByte);
2677   __ cmp(flags, ztos);
2678   __ br(Assembler::NE, notBool);
2679 
2680   // ztos
2681   {
2682     __ pop(ztos);
2683     if (!is_static) pop_and_check_object(obj);
2684     __ andw(r0, r0, 0x1);
2685     __ strb(r0, field);
2686     if (rc == may_rewrite) {
2687       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2688     }
2689     __ b(Done);
2690   }
2691 
2692   __ bind(notBool);
2693   __ cmp(flags, atos);
2694   __ br(Assembler::NE, notObj);
2695 
2696   // atos
2697   {
2698     __ pop(atos);
2699     if (!is_static) pop_and_check_object(obj);
2700     // Store into the field
2701     do_oop_store(_masm, field, r0, _bs->kind(), false);
2702     if (rc == may_rewrite) {
2703       patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2704     }
2705     __ b(Done);
2706   }
2707 
2708   __ bind(notObj);
2709   __ cmp(flags, itos);
2710   __ br(Assembler::NE, notInt);
2711 
2712   // itos
2713   {
2714     __ pop(itos);
2715     if (!is_static) pop_and_check_object(obj);
2716     __ strw(r0, field);
2717     if (rc == may_rewrite) {
2718       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2719     }
2720     __ b(Done);
2721   }
2722 
2723   __ bind(notInt);
2724   __ cmp(flags, ctos);
2725   __ br(Assembler::NE, notChar);
2726 
2727   // ctos
2728   {
2729     __ pop(ctos);
2730     if (!is_static) pop_and_check_object(obj);
2731     __ strh(r0, field);
2732     if (rc == may_rewrite) {
2733       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2734     }
2735     __ b(Done);
2736   }
2737 
2738   __ bind(notChar);
2739   __ cmp(flags, stos);
2740   __ br(Assembler::NE, notShort);
2741 
2742   // stos
2743   {
2744     __ pop(stos);
2745     if (!is_static) pop_and_check_object(obj);
2746     __ strh(r0, field);
2747     if (rc == may_rewrite) {
2748       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2749     }
2750     __ b(Done);
2751   }
2752 
2753   __ bind(notShort);
2754   __ cmp(flags, ltos);
2755   __ br(Assembler::NE, notLong);
2756 
2757   // ltos
2758   {
2759     __ pop(ltos);
2760     if (!is_static) pop_and_check_object(obj);
2761     __ str(r0, field);
2762     if (rc == may_rewrite) {
2763       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2764     }
2765     __ b(Done);
2766   }
2767 
2768   __ bind(notLong);
2769   __ cmp(flags, ftos);
2770   __ br(Assembler::NE, notFloat);
2771 
2772   // ftos
2773   {
2774     __ pop(ftos);
2775     if (!is_static) pop_and_check_object(obj);
2776     __ strs(v0, field);
2777     if (rc == may_rewrite) {
2778       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2779     }
2780     __ b(Done);
2781   }
2782 
2783   __ bind(notFloat);
2784 #ifdef ASSERT
2785   __ cmp(flags, dtos);
2786   __ br(Assembler::NE, notDouble);
2787 #endif
2788 
2789   // dtos
2790   {
2791     __ pop(dtos);
2792     if (!is_static) pop_and_check_object(obj);
2793     __ strd(v0, field);
2794     if (rc == may_rewrite) {
2795       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2796     }
2797   }
2798 
2799 #ifdef ASSERT
2800   __ b(Done);
2801 
2802   __ bind(notDouble);
2803   __ stop("Bad state");
2804 #endif
2805 
2806   __ bind(Done);
2807 
2808   {
2809     Label notVolatile;
2810     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2811     __ membar(MacroAssembler::StoreLoad);
2812     __ bind(notVolatile);
2813   }
2814 }
2815 
2816 void TemplateTable::putfield(int byte_no)
2817 {
2818   putfield_or_static(byte_no, false);
2819 }
2820 
2821 void TemplateTable::nofast_putfield(int byte_no) {
2822   putfield_or_static(byte_no, false, may_not_rewrite);
2823 }
2824 
2825 void TemplateTable::putstatic(int byte_no) {
2826   putfield_or_static(byte_no, true);
2827 }
2828 
2829 void TemplateTable::jvmti_post_fast_field_mod()
2830 {
2831   if (JvmtiExport::can_post_field_modification()) {
2832     // Check to see if a field modification watch has been set before
2833     // we take the time to call into the VM.
2834     Label L2;
2835     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2836     __ ldrw(c_rarg3, Address(rscratch1));
2837     __ cbzw(c_rarg3, L2);
2838     __ pop_ptr(r19);                  // copy the object pointer from tos
2839     __ verify_oop(r19);
2840     __ push_ptr(r19);                 // put the object pointer back on tos
2841     // Save tos values before call_VM() clobbers them. Since we have
2842     // to do it for every data type, we use the saved values as the
2843     // jvalue object.
2844     switch (bytecode()) {          // load values into the jvalue object
2845     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2846     case Bytecodes::_fast_bputfield: // fall through
2847     case Bytecodes::_fast_zputfield: // fall through
2848     case Bytecodes::_fast_sputfield: // fall through
2849     case Bytecodes::_fast_cputfield: // fall through
2850     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2851     case Bytecodes::_fast_dputfield: __ push_d(); break;
2852     case Bytecodes::_fast_fputfield: __ push_f(); break;
2853     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2854 
2855     default:
2856       ShouldNotReachHere();
2857     }
2858     __ mov(c_rarg3, esp);             // points to jvalue on the stack
2859     // access constant pool cache entry
2860     __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2861     __ verify_oop(r19);
2862     // r19: object pointer copied above
2863     // c_rarg2: cache entry pointer
2864     // c_rarg3: jvalue object on the stack
2865     __ call_VM(noreg,
2866                CAST_FROM_FN_PTR(address,
2867                                 InterpreterRuntime::post_field_modification),
2868                r19, c_rarg2, c_rarg3);
2869 
2870     switch (bytecode()) {             // restore tos values
2871     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2872     case Bytecodes::_fast_bputfield: // fall through
2873     case Bytecodes::_fast_zputfield: // fall through
2874     case Bytecodes::_fast_sputfield: // fall through
2875     case Bytecodes::_fast_cputfield: // fall through
2876     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2877     case Bytecodes::_fast_dputfield: __ pop_d(); break;
2878     case Bytecodes::_fast_fputfield: __ pop_f(); break;
2879     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2880     }
2881     __ bind(L2);
2882   }
2883 }
2884 
2885 void TemplateTable::fast_storefield(TosState state)
2886 {
2887   transition(state, vtos);
2888 
2889   ByteSize base = ConstantPoolCache::base_offset();
2890 
2891   jvmti_post_fast_field_mod();
2892 
2893   // access constant pool cache
2894   __ get_cache_and_index_at_bcp(r2, r1, 1);
2895 
2896   // test for volatile with r3
2897   __ ldrw(r3, Address(r2, in_bytes(base +
2898                                    ConstantPoolCacheEntry::flags_offset())));
2899 
2900   // replace index with field offset from cache entry
2901   __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2902 
2903   {
2904     Label notVolatile;
2905     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2906     __ membar(MacroAssembler::StoreStore);
2907     __ bind(notVolatile);
2908   }
2909 
2910   Label notVolatile;
2911 
2912   // Get object from stack
2913   pop_and_check_object(r2);
2914 
2915   // field address
2916   const Address field(r2, r1);
2917 
2918   // access field
2919   switch (bytecode()) {
2920   case Bytecodes::_fast_aputfield:
2921     do_oop_store(_masm, field, r0, _bs->kind(), false);
2922     break;
2923   case Bytecodes::_fast_lputfield:
2924     __ str(r0, field);
2925     break;
2926   case Bytecodes::_fast_iputfield:
2927     __ strw(r0, field);
2928     break;
2929   case Bytecodes::_fast_zputfield:
2930     __ andw(r0, r0, 0x1);  // boolean is true if LSB is 1
2931     // fall through to bputfield
2932   case Bytecodes::_fast_bputfield:
2933     __ strb(r0, field);
2934     break;
2935   case Bytecodes::_fast_sputfield:
2936     // fall through
2937   case Bytecodes::_fast_cputfield:
2938     __ strh(r0, field);
2939     break;
2940   case Bytecodes::_fast_fputfield:
2941     __ strs(v0, field);
2942     break;
2943   case Bytecodes::_fast_dputfield:
2944     __ strd(v0, field);
2945     break;
2946   default:
2947     ShouldNotReachHere();
2948   }
2949 
2950   {
2951     Label notVolatile;
2952     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2953     __ membar(MacroAssembler::StoreLoad);
2954     __ bind(notVolatile);
2955   }
2956 }
2957 
2958 
2959 void TemplateTable::fast_accessfield(TosState state)
2960 {
2961   transition(atos, state);
2962   // Do the JVMTI work here to avoid disturbing the register state below
2963   if (JvmtiExport::can_post_field_access()) {
2964     // Check to see if a field access watch has been set before we
2965     // take the time to call into the VM.
2966     Label L1;
2967     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2968     __ ldrw(r2, Address(rscratch1));
2969     __ cbzw(r2, L1);
2970     // access constant pool cache entry
2971     __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2972     __ verify_oop(r0);
2973     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
2974     __ mov(c_rarg1, r0);
2975     // c_rarg1: object pointer copied above
2976     // c_rarg2: cache entry pointer
2977     __ call_VM(noreg,
2978                CAST_FROM_FN_PTR(address,
2979                                 InterpreterRuntime::post_field_access),
2980                c_rarg1, c_rarg2);
2981     __ pop_ptr(r0); // restore object pointer
2982     __ bind(L1);
2983   }
2984 
2985   // access constant pool cache
2986   __ get_cache_and_index_at_bcp(r2, r1, 1);
2987   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2988                                   ConstantPoolCacheEntry::f2_offset())));
2989   __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2990                                    ConstantPoolCacheEntry::flags_offset())));
2991 
2992   // r0: object
2993   __ verify_oop(r0);
2994   __ null_check(r0);
2995   const Address field(r0, r1);
2996 
2997   // 8179954: We need to make sure that the code generated for
2998   // volatile accesses forms a sequentially-consistent set of
2999   // operations when combined with STLR and LDAR.  Without a leading
3000   // membar it's possible for a simple Dekker test to fail if loads
3001   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3002   // the stores in one method and we interpret the loads in another.
3003   if (! UseBarriersForVolatile) {
3004     Label notVolatile;
3005     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3006     __ membar(MacroAssembler::AnyAny);
3007     __ bind(notVolatile);
3008   }
3009 
3010   // access field
3011   switch (bytecode()) {
3012   case Bytecodes::_fast_agetfield:
3013     __ load_heap_oop(r0, field);
3014     __ verify_oop(r0);
3015     break;
3016   case Bytecodes::_fast_lgetfield:
3017     __ ldr(r0, field);
3018     break;
3019   case Bytecodes::_fast_igetfield:
3020     __ ldrw(r0, field);
3021     break;
3022   case Bytecodes::_fast_bgetfield:
3023     __ load_signed_byte(r0, field);
3024     break;
3025   case Bytecodes::_fast_sgetfield:
3026     __ load_signed_short(r0, field);
3027     break;
3028   case Bytecodes::_fast_cgetfield:
3029     __ load_unsigned_short(r0, field);
3030     break;
3031   case Bytecodes::_fast_fgetfield:
3032     __ ldrs(v0, field);
3033     break;
3034   case Bytecodes::_fast_dgetfield:
3035     __ ldrd(v0, field);
3036     break;
3037   default:
3038     ShouldNotReachHere();
3039   }
3040   {
3041     Label notVolatile;
3042     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3043     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3044     __ bind(notVolatile);
3045   }
3046 }
3047 
3048 void TemplateTable::fast_xaccess(TosState state)
3049 {
3050   transition(vtos, state);
3051 
3052   // get receiver
3053   __ ldr(r0, aaddress(0));
3054   // access constant pool cache
3055   __ get_cache_and_index_at_bcp(r2, r3, 2);
3056   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3057                                   ConstantPoolCacheEntry::f2_offset())));
3058 
3059   // 8179954: We need to make sure that the code generated for
3060   // volatile accesses forms a sequentially-consistent set of
3061   // operations when combined with STLR and LDAR.  Without a leading
3062   // membar it's possible for a simple Dekker test to fail if loads
3063   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3064   // the stores in one method and we interpret the loads in another.
3065   if (! UseBarriersForVolatile) {
3066     Label notVolatile;
3067     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3068                                      ConstantPoolCacheEntry::flags_offset())));
3069     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3070     __ membar(MacroAssembler::AnyAny);
3071     __ bind(notVolatile);
3072   }
3073 
3074   // make sure exception is reported in correct bcp range (getfield is
3075   // next instruction)
3076   __ increment(rbcp);
3077   __ null_check(r0);
3078   switch (state) {
3079   case itos:
3080     __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3081     break;
3082   case atos:
3083     __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0)));
3084     __ verify_oop(r0);
3085     break;
3086   case ftos:
3087     __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3088     break;
3089   default:
3090     ShouldNotReachHere();
3091   }
3092 
3093   {
3094     Label notVolatile;
3095     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3096                                      ConstantPoolCacheEntry::flags_offset())));
3097     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3098     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3099     __ bind(notVolatile);
3100   }
3101 
3102   __ decrement(rbcp);
3103 }
3104 
3105 
3106 
3107 //-----------------------------------------------------------------------------
3108 // Calls
3109 
3110 void TemplateTable::count_calls(Register method, Register temp)
3111 {
3112   __ call_Unimplemented();
3113 }
3114 
3115 void TemplateTable::prepare_invoke(int byte_no,
3116                                    Register method, // linked method (or i-klass)
3117                                    Register index,  // itable index, MethodType, etc.
3118                                    Register recv,   // if caller wants to see it
3119                                    Register flags   // if caller wants to test it
3120                                    ) {
3121   // determine flags
3122   Bytecodes::Code code = bytecode();
3123   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3124   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3125   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3126   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3127   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3128   const bool load_receiver       = (recv  != noreg);
3129   const bool save_flags          = (flags != noreg);
3130   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3131   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3132   assert(flags == noreg || flags == r3, "");
3133   assert(recv  == noreg || recv  == r2, "");
3134 
3135   // setup registers & access constant pool cache
3136   if (recv  == noreg)  recv  = r2;
3137   if (flags == noreg)  flags = r3;
3138   assert_different_registers(method, index, recv, flags);
3139 
3140   // save 'interpreter return address'
3141   __ save_bcp();
3142 
3143   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3144 
3145   // maybe push appendix to arguments (just before return address)
3146   if (is_invokedynamic || is_invokehandle) {
3147     Label L_no_push;
3148     __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3149     // Push the appendix as a trailing parameter.
3150     // This must be done before we get the receiver,
3151     // since the parameter_size includes it.
3152     __ push(r19);
3153     __ mov(r19, index);
3154     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3155     __ load_resolved_reference_at_index(index, r19);
3156     __ pop(r19);
3157     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3158     __ bind(L_no_push);
3159   }
3160 
3161   // load receiver if needed (note: no return address pushed yet)
3162   if (load_receiver) {
3163     __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3164     // FIXME -- is this actually correct? looks like it should be 2
3165     // const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3166     // const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3167     // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3168     // __ movptr(recv, recv_addr);
3169     __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3170     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3171     __ verify_oop(recv);
3172   }
3173 
3174   // compute return type
3175   // x86 uses a shift and mask or wings it with a shift plus assert
3176   // the mask is not needed. aarch64 just uses bitfield extract
3177   __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
3178   // load return address
3179   {
3180     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3181     __ mov(rscratch1, table_addr);
3182     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3183   }
3184 }
3185 
3186 
3187 void TemplateTable::invokevirtual_helper(Register index,
3188                                          Register recv,
3189                                          Register flags)
3190 {
3191   // Uses temporary registers r0, r3
3192   assert_different_registers(index, recv, r0, r3);
3193   // Test for an invoke of a final method
3194   Label notFinal;
3195   __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3196 
3197   const Register method = index;  // method must be rmethod
3198   assert(method == rmethod,
3199          "methodOop must be rmethod for interpreter calling convention");
3200 
3201   // do the call - the index is actually the method to call
3202   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3203 
3204   // It's final, need a null check here!
3205   __ null_check(recv);
3206 
3207   // profile this call
3208   __ profile_final_call(r0);
3209   __ profile_arguments_type(r0, method, r4, true);
3210 
3211   __ jump_from_interpreted(method, r0);
3212 
3213   __ bind(notFinal);
3214 
3215   // get receiver klass
3216   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3217   __ load_klass(r0, recv);
3218 
3219   // profile this call
3220   __ profile_virtual_call(r0, rlocals, r3);
3221 
3222   // get target methodOop & entry point
3223   __ lookup_virtual_method(r0, index, method);
3224   __ profile_arguments_type(r3, method, r4, true);
3225   // FIXME -- this looks completely redundant. is it?
3226   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3227   __ jump_from_interpreted(method, r3);
3228 }
3229 
3230 void TemplateTable::invokevirtual(int byte_no)
3231 {
3232   transition(vtos, vtos);
3233   assert(byte_no == f2_byte, "use this argument");
3234 
3235   prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3236 
3237   // rmethod: index (actually a Method*)
3238   // r2: receiver
3239   // r3: flags
3240 
3241   invokevirtual_helper(rmethod, r2, r3);
3242 }
3243 
3244 void TemplateTable::invokespecial(int byte_no)
3245 {
3246   transition(vtos, vtos);
3247   assert(byte_no == f1_byte, "use this argument");
3248 
3249   prepare_invoke(byte_no, rmethod, noreg,  // get f1 Method*
3250                  r2);  // get receiver also for null check
3251   __ verify_oop(r2);
3252   __ null_check(r2);
3253   // do the call
3254   __ profile_call(r0);
3255   __ profile_arguments_type(r0, rmethod, rbcp, false);
3256   __ jump_from_interpreted(rmethod, r0);
3257 }
3258 
3259 void TemplateTable::invokestatic(int byte_no)
3260 {
3261   transition(vtos, vtos);
3262   assert(byte_no == f1_byte, "use this argument");
3263 
3264   prepare_invoke(byte_no, rmethod);  // get f1 Method*
3265   // do the call
3266   __ profile_call(r0);
3267   __ profile_arguments_type(r0, rmethod, r4, false);
3268   __ jump_from_interpreted(rmethod, r0);
3269 }
3270 
3271 void TemplateTable::fast_invokevfinal(int byte_no)
3272 {
3273   __ call_Unimplemented();
3274 }
3275 
3276 void TemplateTable::invokeinterface(int byte_no) {
3277   transition(vtos, vtos);
3278   assert(byte_no == f1_byte, "use this argument");
3279 
3280   prepare_invoke(byte_no, r0, rmethod,  // get f1 Klass*, f2 itable index
3281                  r2, r3); // recv, flags
3282 
3283   // r0: interface klass (from f1)
3284   // rmethod: itable index (from f2)
3285   // r2: receiver
3286   // r3: flags
3287 
3288   // Special case of invokeinterface called for virtual method of
3289   // java.lang.Object.  See cpCacheOop.cpp for details.
3290   // This code isn't produced by javac, but could be produced by
3291   // another compliant java compiler.
3292   Label notMethod;
3293   __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3294 
3295   invokevirtual_helper(rmethod, r2, r3);
3296   __ bind(notMethod);
3297 
3298   // Get receiver klass into r3 - also a null check
3299   __ restore_locals();
3300   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3301   __ load_klass(r3, r2);
3302 
3303   // profile this call
3304   __ profile_virtual_call(r3, r13, r19);
3305 
3306   Label no_such_interface, no_such_method;
3307 
3308   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3309                              r3, r0, rmethod,
3310                              // outputs: method, scan temp. reg
3311                              rmethod, r13,
3312                              no_such_interface);
3313 
3314   // rmethod,: methodOop to call
3315   // r2: receiver
3316   // Check for abstract method error
3317   // Note: This should be done more efficiently via a throw_abstract_method_error
3318   //       interpreter entry point and a conditional jump to it in case of a null
3319   //       method.
3320   __ cbz(rmethod, no_such_method);
3321 
3322   __ profile_arguments_type(r3, rmethod, r13, true);
3323 
3324   // do the call
3325   // r2: receiver
3326   // rmethod,: methodOop
3327   __ jump_from_interpreted(rmethod, r3);
3328   __ should_not_reach_here();
3329 
3330   // exception handling code follows...
3331   // note: must restore interpreter registers to canonical
3332   //       state for exception handling to work correctly!
3333 
3334   __ bind(no_such_method);
3335   // throw exception
3336   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3337   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3338   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3339   // the call_VM checks for exception, so we should never return here.
3340   __ should_not_reach_here();
3341 
3342   __ bind(no_such_interface);
3343   // throw exception
3344   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3345   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3346   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3347                    InterpreterRuntime::throw_IncompatibleClassChangeError));
3348   // the call_VM checks for exception, so we should never return here.
3349   __ should_not_reach_here();
3350   return;
3351 }
3352 
3353 void TemplateTable::invokehandle(int byte_no) {
3354   transition(vtos, vtos);
3355   assert(byte_no == f1_byte, "use this argument");
3356 
3357   prepare_invoke(byte_no, rmethod, r0, r2);
3358   __ verify_method_ptr(r2);
3359   __ verify_oop(r2);
3360   __ null_check(r2);
3361 
3362   // FIXME: profile the LambdaForm also
3363 
3364   // r13 is safe to use here as a scratch reg because it is about to
3365   // be clobbered by jump_from_interpreted().
3366   __ profile_final_call(r13);
3367   __ profile_arguments_type(r13, rmethod, r4, true);
3368 
3369   __ jump_from_interpreted(rmethod, r0);
3370 }
3371 
3372 void TemplateTable::invokedynamic(int byte_no) {
3373   transition(vtos, vtos);
3374   assert(byte_no == f1_byte, "use this argument");
3375 
3376   prepare_invoke(byte_no, rmethod, r0);
3377 
3378   // r0: CallSite object (from cpool->resolved_references[])
3379   // rmethod: MH.linkToCallSite method (from f2)
3380 
3381   // Note:  r0_callsite is already pushed by prepare_invoke
3382 
3383   // %%% should make a type profile for any invokedynamic that takes a ref argument
3384   // profile this call
3385   __ profile_call(rbcp);
3386   __ profile_arguments_type(r3, rmethod, r13, false);
3387 
3388   __ verify_oop(r0);
3389 
3390   __ jump_from_interpreted(rmethod, r0);
3391 }
3392 
3393 
3394 //-----------------------------------------------------------------------------
3395 // Allocation
3396 
3397 void TemplateTable::_new() {
3398   transition(vtos, atos);
3399 
3400   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3401   Label slow_case;
3402   Label done;
3403   Label initialize_header;
3404   Label initialize_object; // including clearing the fields
3405   Label allocate_shared;
3406 
3407   __ get_cpool_and_tags(r4, r0);
3408   // Make sure the class we're about to instantiate has been resolved.
3409   // This is done before loading InstanceKlass to be consistent with the order
3410   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3411   const int tags_offset = Array<u1>::base_offset_in_bytes();
3412   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3413   __ lea(rscratch1, Address(rscratch1, tags_offset));
3414   __ ldarb(rscratch1, rscratch1);
3415   __ cmp(rscratch1, JVM_CONSTANT_Class);
3416   __ br(Assembler::NE, slow_case);
3417 
3418   // get InstanceKlass
3419   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3420 
3421   // make sure klass is initialized & doesn't have finalizer
3422   // make sure klass is fully initialized
3423   __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3424   __ cmp(rscratch1, InstanceKlass::fully_initialized);
3425   __ br(Assembler::NE, slow_case);
3426 
3427   // get instance_size in InstanceKlass (scaled to a count of bytes)
3428   __ ldrw(r3,
3429           Address(r4,
3430                   Klass::layout_helper_offset()));
3431   // test to see if it has a finalizer or is malformed in some way
3432   __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3433 
3434   // Allocate the instance
3435   // 1) Try to allocate in the TLAB
3436   // 2) if fail and the object is large allocate in the shared Eden
3437   // 3) if the above fails (or is not applicable), go to a slow case
3438   // (creates a new TLAB, etc.)
3439 
3440   const bool allow_shared_alloc =
3441     Universe::heap()->supports_inline_contig_alloc();
3442 
3443   if (UseTLAB) {
3444     __ tlab_allocate(r0, r3, 0, noreg, r1,
3445                      allow_shared_alloc ? allocate_shared : slow_case);
3446 
3447     if (ZeroTLAB) {
3448       // the fields have been already cleared
3449       __ b(initialize_header);
3450     } else {
3451       // initialize both the header and fields
3452       __ b(initialize_object);
3453     }
3454   }
3455 
3456   // Allocation in the shared Eden, if allowed.
3457   //
3458   // r3: instance size in bytes
3459   if (allow_shared_alloc) {
3460     __ bind(allocate_shared);
3461 
3462     __ eden_allocate(r0, r3, 0, r10, slow_case);
3463     __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3464   }
3465 
3466   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3467     // The object is initialized before the header.  If the object size is
3468     // zero, go directly to the header initialization.
3469     __ bind(initialize_object);
3470     __ sub(r3, r3, sizeof(oopDesc));
3471     __ cbz(r3, initialize_header);
3472 
3473     // Initialize object fields
3474     {
3475       __ add(r2, r0, sizeof(oopDesc));
3476       Label loop;
3477       __ bind(loop);
3478       __ str(zr, Address(__ post(r2, BytesPerLong)));
3479       __ sub(r3, r3, BytesPerLong);
3480       __ cbnz(r3, loop);
3481     }
3482 
3483     // initialize object header only.
3484     __ bind(initialize_header);
3485     if (UseBiasedLocking) {
3486       __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3487     } else {
3488       __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3489     }
3490     __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3491     __ store_klass_gap(r0, zr);  // zero klass gap for compressed oops
3492     __ store_klass(r0, r4);      // store klass last
3493 
3494     {
3495       SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3496       // Trigger dtrace event for fastpath
3497       __ push(atos); // save the return value
3498       __ call_VM_leaf(
3499            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3500       __ pop(atos); // restore the return value
3501 
3502     }
3503     __ b(done);
3504   }
3505 
3506   // slow case
3507   __ bind(slow_case);
3508   __ get_constant_pool(c_rarg1);
3509   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3510   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3511   __ verify_oop(r0);
3512 
3513   // continue
3514   __ bind(done);
3515   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3516   __ membar(Assembler::StoreStore);
3517 }
3518 
3519 void TemplateTable::newarray() {
3520   transition(itos, atos);
3521   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3522   __ mov(c_rarg2, r0);
3523   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3524           c_rarg1, c_rarg2);
3525   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3526   __ membar(Assembler::StoreStore);
3527 }
3528 
3529 void TemplateTable::anewarray() {
3530   transition(itos, atos);
3531   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3532   __ get_constant_pool(c_rarg1);
3533   __ mov(c_rarg3, r0);
3534   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3535           c_rarg1, c_rarg2, c_rarg3);
3536   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3537   __ membar(Assembler::StoreStore);
3538 }
3539 
3540 void TemplateTable::arraylength() {
3541   transition(atos, itos);
3542   __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3543   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3544 }
3545 
3546 void TemplateTable::checkcast()
3547 {
3548   transition(atos, atos);
3549   Label done, is_null, ok_is_subtype, quicked, resolved;
3550   __ cbz(r0, is_null);
3551 
3552   // Get cpool & tags index
3553   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3554   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3555   // See if bytecode has already been quicked
3556   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3557   __ lea(r1, Address(rscratch1, r19));
3558   __ ldarb(r1, r1);
3559   __ cmp(r1, JVM_CONSTANT_Class);
3560   __ br(Assembler::EQ, quicked);
3561 
3562   __ push(atos); // save receiver for result, and for GC
3563   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3564   // vm_result_2 has metadata result
3565   __ get_vm_result_2(r0, rthread);
3566   __ pop(r3); // restore receiver
3567   __ b(resolved);
3568 
3569   // Get superklass in r0 and subklass in r3
3570   __ bind(quicked);
3571   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3572   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3573 
3574   __ bind(resolved);
3575   __ load_klass(r19, r3);
3576 
3577   // Generate subtype check.  Blows r2, r5.  Object in r3.
3578   // Superklass in r0.  Subklass in r19.
3579   __ gen_subtype_check(r19, ok_is_subtype);
3580 
3581   // Come here on failure
3582   __ push(r3);
3583   // object is at TOS
3584   __ b(Interpreter::_throw_ClassCastException_entry);
3585 
3586   // Come here on success
3587   __ bind(ok_is_subtype);
3588   __ mov(r0, r3); // Restore object in r3
3589 
3590   // Collect counts on whether this test sees NULLs a lot or not.
3591   if (ProfileInterpreter) {
3592     __ b(done);
3593     __ bind(is_null);
3594     __ profile_null_seen(r2);
3595   } else {
3596     __ bind(is_null);   // same as 'done'
3597   }
3598   __ bind(done);
3599 }
3600 
3601 void TemplateTable::instanceof() {
3602   transition(atos, itos);
3603   Label done, is_null, ok_is_subtype, quicked, resolved;
3604   __ cbz(r0, is_null);
3605 
3606   // Get cpool & tags index
3607   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3608   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3609   // See if bytecode has already been quicked
3610   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3611   __ lea(r1, Address(rscratch1, r19));
3612   __ ldarb(r1, r1);
3613   __ cmp(r1, JVM_CONSTANT_Class);
3614   __ br(Assembler::EQ, quicked);
3615 
3616   __ push(atos); // save receiver for result, and for GC
3617   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3618   // vm_result_2 has metadata result
3619   __ get_vm_result_2(r0, rthread);
3620   __ pop(r3); // restore receiver
3621   __ verify_oop(r3);
3622   __ load_klass(r3, r3);
3623   __ b(resolved);
3624 
3625   // Get superklass in r0 and subklass in r3
3626   __ bind(quicked);
3627   __ load_klass(r3, r0);
3628   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3629 
3630   __ bind(resolved);
3631 
3632   // Generate subtype check.  Blows r2, r5
3633   // Superklass in r0.  Subklass in r3.
3634   __ gen_subtype_check(r3, ok_is_subtype);
3635 
3636   // Come here on failure
3637   __ mov(r0, 0);
3638   __ b(done);
3639   // Come here on success
3640   __ bind(ok_is_subtype);
3641   __ mov(r0, 1);
3642 
3643   // Collect counts on whether this test sees NULLs a lot or not.
3644   if (ProfileInterpreter) {
3645     __ b(done);
3646     __ bind(is_null);
3647     __ profile_null_seen(r2);
3648   } else {
3649     __ bind(is_null);   // same as 'done'
3650   }
3651   __ bind(done);
3652   // r0 = 0: obj == NULL or  obj is not an instanceof the specified klass
3653   // r0 = 1: obj != NULL and obj is     an instanceof the specified klass
3654 }
3655 
3656 //-----------------------------------------------------------------------------
3657 // Breakpoints
3658 void TemplateTable::_breakpoint() {
3659   // Note: We get here even if we are single stepping..
3660   // jbug inists on setting breakpoints at every bytecode
3661   // even if we are in single step mode.
3662 
3663   transition(vtos, vtos);
3664 
3665   // get the unpatched byte code
3666   __ get_method(c_rarg1);
3667   __ call_VM(noreg,
3668              CAST_FROM_FN_PTR(address,
3669                               InterpreterRuntime::get_original_bytecode_at),
3670              c_rarg1, rbcp);
3671   __ mov(r19, r0);
3672 
3673   // post the breakpoint event
3674   __ call_VM(noreg,
3675              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3676              rmethod, rbcp);
3677 
3678   // complete the execution of original bytecode
3679   __ mov(rscratch1, r19);
3680   __ dispatch_only_normal(vtos);
3681 }
3682 
3683 //-----------------------------------------------------------------------------
3684 // Exceptions
3685 
3686 void TemplateTable::athrow() {
3687   transition(atos, vtos);
3688   __ null_check(r0);
3689   __ b(Interpreter::throw_exception_entry());
3690 }
3691 
3692 //-----------------------------------------------------------------------------
3693 // Synchronization
3694 //
3695 // Note: monitorenter & exit are symmetric routines; which is reflected
3696 //       in the assembly code structure as well
3697 //
3698 // Stack layout:
3699 //
3700 // [expressions  ] <--- esp               = expression stack top
3701 // ..
3702 // [expressions  ]
3703 // [monitor entry] <--- monitor block top = expression stack bot
3704 // ..
3705 // [monitor entry]
3706 // [frame data   ] <--- monitor block bot
3707 // ...
3708 // [saved rbp    ] <--- rbp
3709 void TemplateTable::monitorenter()
3710 {
3711   transition(atos, vtos);
3712 
3713   // check for NULL object
3714   __ null_check(r0);
3715 
3716   const Address monitor_block_top(
3717         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3718   const Address monitor_block_bot(
3719         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3720   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3721 
3722   Label allocated;
3723 
3724   // initialize entry pointer
3725   __ mov(c_rarg1, zr); // points to free slot or NULL
3726 
3727   // find a free slot in the monitor block (result in c_rarg1)
3728   {
3729     Label entry, loop, exit;
3730     __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3731                                         // starting with top-most entry
3732     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3733 
3734     __ b(entry);
3735 
3736     __ bind(loop);
3737     // check if current entry is used
3738     // if not used then remember entry in c_rarg1
3739     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3740     __ cmp(zr, rscratch1);
3741     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3742     // check if current entry is for same object
3743     __ cmp(r0, rscratch1);
3744     // if same object then stop searching
3745     __ br(Assembler::EQ, exit);
3746     // otherwise advance to next entry
3747     __ add(c_rarg3, c_rarg3, entry_size);
3748     __ bind(entry);
3749     // check if bottom reached
3750     __ cmp(c_rarg3, c_rarg2);
3751     // if not at bottom then check this entry
3752     __ br(Assembler::NE, loop);
3753     __ bind(exit);
3754   }
3755 
3756   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3757                             // if found, continue with that on
3758 
3759   // allocate one if there's no free slot
3760   {
3761     Label entry, loop;
3762     // 1. compute new pointers            // rsp: old expression stack top
3763     __ ldr(c_rarg1, monitor_block_bot);   // c_rarg1: old expression stack bottom
3764     __ sub(esp, esp, entry_size);         // move expression stack top
3765     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3766     __ mov(c_rarg3, esp);                 // set start value for copy loop
3767     __ str(c_rarg1, monitor_block_bot);   // set new monitor block bottom
3768 
3769     __ sub(sp, sp, entry_size);           // make room for the monitor
3770 
3771     __ b(entry);
3772     // 2. move expression stack contents
3773     __ bind(loop);
3774     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3775                                                    // word from old location
3776     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
3777     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
3778     __ bind(entry);
3779     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
3780     __ br(Assembler::NE, loop);      // if not at bottom then
3781                                      // copy next word
3782   }
3783 
3784   // call run-time routine
3785   // c_rarg1: points to monitor entry
3786   __ bind(allocated);
3787 
3788   // Increment bcp to point to the next bytecode, so exception
3789   // handling for async. exceptions work correctly.
3790   // The object has already been poped from the stack, so the
3791   // expression stack looks correct.
3792   __ increment(rbcp);
3793 
3794   // store object
3795   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3796   __ lock_object(c_rarg1);
3797 
3798   // check to make sure this monitor doesn't cause stack overflow after locking
3799   __ save_bcp();  // in case of exception
3800   __ generate_stack_overflow_check(0);
3801 
3802   // The bcp has already been incremented. Just need to dispatch to
3803   // next instruction.
3804   __ dispatch_next(vtos);
3805 }
3806 
3807 
3808 void TemplateTable::monitorexit()
3809 {
3810   transition(atos, vtos);
3811 
3812   // check for NULL object
3813   __ null_check(r0);
3814 
3815   const Address monitor_block_top(
3816         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3817   const Address monitor_block_bot(
3818         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3819   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3820 
3821   Label found;
3822 
3823   // find matching slot
3824   {
3825     Label entry, loop;
3826     __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3827                                         // starting with top-most entry
3828     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3829                                         // of monitor block
3830     __ b(entry);
3831 
3832     __ bind(loop);
3833     // check if current entry is for same object
3834     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3835     __ cmp(r0, rscratch1);
3836     // if same object then stop searching
3837     __ br(Assembler::EQ, found);
3838     // otherwise advance to next entry
3839     __ add(c_rarg1, c_rarg1, entry_size);
3840     __ bind(entry);
3841     // check if bottom reached
3842     __ cmp(c_rarg1, c_rarg2);
3843     // if not at bottom then check this entry
3844     __ br(Assembler::NE, loop);
3845   }
3846 
3847   // error handling. Unlocking was not block-structured
3848   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3849                    InterpreterRuntime::throw_illegal_monitor_state_exception));
3850   __ should_not_reach_here();
3851 
3852   // call run-time routine
3853   __ bind(found);
3854   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3855   __ unlock_object(c_rarg1);
3856   __ pop_ptr(r0); // discard object
3857 }
3858 
3859 
3860 // Wide instructions
3861 void TemplateTable::wide()
3862 {
3863   __ load_unsigned_byte(r19, at_bcp(1));
3864   __ mov(rscratch1, (address)Interpreter::_wentry_point);
3865   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3866   __ br(rscratch1);
3867 }
3868 
3869 
3870 // Multi arrays
3871 void TemplateTable::multianewarray() {
3872   transition(vtos, atos);
3873   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3874   // last dim is on top of stack; we want address of first one:
3875   // first_addr = last_addr + (ndims - 1) * wordSize
3876   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3877   __ sub(c_rarg1, c_rarg1, wordSize);
3878   call_VM(r0,
3879           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3880           c_rarg1);
3881   __ load_unsigned_byte(r1, at_bcp(3));
3882   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3883 }