1 /*
   2  * Copyright (c) 2013, Red Hat Inc.
   3  * Copyright (c) 2003, 2011, Oracle and/or its affiliates.
   4  * All rights reserved.
   5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6  *
   7  * This code is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License version 2 only, as
   9  * published by the Free Software Foundation.
  10  *
  11  * This code is distributed in the hope that it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14  * version 2 for more details (a copy is included in the LICENSE file that
  15  * accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License version
  18  * 2 along with this work; if not, write to the Free Software Foundation,
  19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  20  *
  21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  22  * or visit www.oracle.com if you need additional information or have any
  23  * questions.
  24  *
  25  */
  26 
  27 #include "precompiled.hpp"
  28 #include "asm/macroAssembler.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.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 #ifndef CC_INTERP
  43 
  44 #define __ _masm->
  45 
  46 // Platform-dependent initialization
  47 
  48 void TemplateTable::pd_initialize() {
  49   // No amd64 specific initialization
  50 }
  51 
  52 // Address computation: local variables
  53 
  54 static inline Address iaddress(int n) {
  55   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  56 }
  57 
  58 static inline Address laddress(int n) {
  59   return iaddress(n + 1);
  60 }
  61 
  62 static inline Address faddress(int n) {
  63   return iaddress(n);
  64 }
  65 
  66 static inline Address daddress(int n) {
  67   return laddress(n);
  68 }
  69 
  70 static inline Address aaddress(int n) {
  71   return iaddress(n);
  72 }
  73 
  74 static inline Address iaddress(Register r) {
  75   return Address(rlocals, r, Address::lsl(3));
  76 }
  77 
  78 static inline Address laddress(Register r, Register scratch,
  79                                InterpreterMacroAssembler* _masm) {
  80   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  81   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  82 }
  83 
  84 static inline Address faddress(Register r) {
  85   return iaddress(r);
  86 }
  87 
  88 static inline Address daddress(Register r, Register scratch,
  89                                InterpreterMacroAssembler* _masm) {
  90   return laddress(r, scratch, _masm);
  91 }
  92 
  93 static inline Address aaddress(Register r) {
  94   return iaddress(r);
  95 }
  96 
  97 static inline Address at_rsp() {
  98   return Address(esp, 0);
  99 }
 100 
 101 // At top of Java expression stack which may be different than esp().  It
 102 // isn't for category 1 objects.
 103 static inline Address at_tos   () {
 104   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 105 }
 106 
 107 static inline Address at_tos_p1() {
 108   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 109 }
 110 
 111 static inline Address at_tos_p2() {
 112   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 113 }
 114 
 115 static inline Address at_tos_p3() {
 116   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 117 }
 118 
 119 static inline Address at_tos_p4() {
 120   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 121 }
 122 
 123 static inline Address at_tos_p5() {
 124   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 125 }
 126 
 127 // Condition conversion
 128 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 129   switch (cc) {
 130   case TemplateTable::equal        : return Assembler::NE;
 131   case TemplateTable::not_equal    : return Assembler::EQ;
 132   case TemplateTable::less         : return Assembler::GE;
 133   case TemplateTable::less_equal   : return Assembler::GT;
 134   case TemplateTable::greater      : return Assembler::LE;
 135   case TemplateTable::greater_equal: return Assembler::LT;
 136   }
 137   ShouldNotReachHere();
 138   return Assembler::EQ;
 139 }
 140 
 141 
 142 // Miscelaneous helper routines
 143 // Store an oop (or NULL) at the Address described by obj.
 144 // If val == noreg this means store a NULL
 145 static void do_oop_store(InterpreterMacroAssembler* _masm,
 146                          Address obj,
 147                          Register val,
 148                          BarrierSet::Name barrier,
 149                          bool precise) {
 150   assert(val == noreg || val == r0, "parameter is just for looks");
 151   switch (barrier) {
 152 #if INCLUDE_ALL_GCS
 153     case BarrierSet::G1SATBCT:
 154     case BarrierSet::G1SATBCTLogging:
 155     case BarrierSet::ShenandoahBarrierSet:
 156       {
 157         // flatten object address if needed
 158         if (obj.index() == noreg && obj.offset() == 0) {
 159           if (obj.base() != r3) {
 160             __ mov(r3, obj.base());
 161           }
 162         } else {
 163           __ lea(r3, obj);
 164         }
 165         __ g1_write_barrier_pre(r3 /* obj */,
 166                                 r1 /* pre_val */,
 167                                 rthread /* thread */,
 168                                 r10  /* tmp */,
 169                                 val != noreg /* tosca_live */,
 170                                 false /* expand_call */);
 171         if (val == noreg) {
 172           __ store_heap_oop_null(Address(r3, 0));
 173         } else {
 174           // G1 barrier needs uncompressed oop for region cross check.
 175           Register new_val = val;
 176           if (UseCompressedOops) {
 177             new_val = rscratch2;
 178             __ mov(new_val, val);
 179           }
 180 
 181           __ store_heap_oop(Address(r3, 0), val);
 182 
 183           __ g1_write_barrier_post(r3 /* store_adr */,
 184                                    new_val /* new_val */,
 185                                    rthread /* thread */,
 186                                    r10 /* tmp */,
 187                                    r1 /* tmp2 */);
 188         }
 189 
 190       }
 191       break;
 192 #endif // INCLUDE_ALL_GCS
 193     case BarrierSet::CardTableModRef:
 194     case BarrierSet::CardTableExtension:
 195       {
 196         if (val == noreg) {
 197           __ store_heap_oop_null(obj);
 198         } else {
 199           __ store_heap_oop(obj, val);
 200           // flatten object address if needed
 201           if (!precise || (obj.index() == noreg && obj.offset() == 0)) {
 202             __ store_check(obj.base());
 203           } else {
 204             __ lea(r3, obj);
 205             __ store_check(r3);
 206           }
 207         }
 208       }
 209       break;
 210     case BarrierSet::ModRef:
 211     case BarrierSet::Other:
 212       if (val == noreg) {
 213         __ store_heap_oop_null(obj);
 214       } else {
 215         __ store_heap_oop(obj, val);
 216       }
 217       break;
 218     default      :
 219       ShouldNotReachHere();
 220 
 221   }
 222 }
 223 
 224 Address TemplateTable::at_bcp(int offset) {
 225   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 226   return Address(rbcp, offset);
 227 }
 228 
 229 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 230                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 231                                    int byte_no)
 232 {
 233   if (!RewriteBytecodes)  return;
 234   Label L_patch_done;
 235 
 236   switch (bc) {
 237   case Bytecodes::_fast_aputfield:
 238   case Bytecodes::_fast_bputfield:
 239   case Bytecodes::_fast_zputfield:
 240   case Bytecodes::_fast_cputfield:
 241   case Bytecodes::_fast_dputfield:
 242   case Bytecodes::_fast_fputfield:
 243   case Bytecodes::_fast_iputfield:
 244   case Bytecodes::_fast_lputfield:
 245   case Bytecodes::_fast_sputfield:
 246     {
 247       // We skip bytecode quickening for putfield instructions when
 248       // the put_code written to the constant pool cache is zero.
 249       // This is required so that every execution of this instruction
 250       // calls out to InterpreterRuntime::resolve_get_put to do
 251       // additional, required work.
 252       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 253       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 254       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 255       __ movw(bc_reg, bc);
 256       __ cmpw(temp_reg, (unsigned) 0);
 257       __ br(Assembler::EQ, L_patch_done);  // don't patch
 258     }
 259     break;
 260   default:
 261     assert(byte_no == -1, "sanity");
 262     // the pair bytecodes have already done the load.
 263     if (load_bc_into_bc_reg) {
 264       __ movw(bc_reg, bc);
 265     }
 266   }
 267 
 268   if (JvmtiExport::can_post_breakpoint()) {
 269     Label L_fast_patch;
 270     // if a breakpoint is present we can't rewrite the stream directly
 271     __ load_unsigned_byte(temp_reg, at_bcp(0));
 272     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 273     __ br(Assembler::NE, L_fast_patch);
 274     // Let breakpoint table handling rewrite to quicker bytecode
 275     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 276     __ b(L_patch_done);
 277     __ bind(L_fast_patch);
 278   }
 279 
 280 #ifdef ASSERT
 281   Label L_okay;
 282   __ load_unsigned_byte(temp_reg, at_bcp(0));
 283   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 284   __ br(Assembler::EQ, L_okay);
 285   __ cmpw(temp_reg, bc_reg);
 286   __ br(Assembler::EQ, L_okay);
 287   __ stop("patching the wrong bytecode");
 288   __ bind(L_okay);
 289 #endif
 290 
 291   // patch bytecode
 292   __ strb(bc_reg, at_bcp(0));
 293   __ bind(L_patch_done);
 294 }
 295 
 296 
 297 // Individual instructions
 298 
 299 void TemplateTable::nop() {
 300   transition(vtos, vtos);
 301   // nothing to do
 302 }
 303 
 304 void TemplateTable::shouldnotreachhere() {
 305   transition(vtos, vtos);
 306   __ stop("shouldnotreachhere bytecode");
 307 }
 308 
 309 void TemplateTable::aconst_null()
 310 {
 311   transition(vtos, atos);
 312   __ mov(r0, 0);
 313 }
 314 
 315 void TemplateTable::iconst(int value)
 316 {
 317   transition(vtos, itos);
 318   __ mov(r0, value);
 319 }
 320 
 321 void TemplateTable::lconst(int value)
 322 {
 323   __ mov(r0, value);
 324 }
 325 
 326 void TemplateTable::fconst(int value)
 327 {
 328   transition(vtos, ftos);
 329   switch (value) {
 330   case 0:
 331     __ fmovs(v0, zr);
 332     break;
 333   case 1:
 334     __ fmovs(v0, 1.0);
 335     break;
 336   case 2:
 337     __ fmovs(v0, 2.0);
 338     break;
 339   default:
 340     ShouldNotReachHere();
 341     break;
 342   }
 343 }
 344 
 345 void TemplateTable::dconst(int value)
 346 {
 347   transition(vtos, dtos);
 348   switch (value) {
 349   case 0:
 350     __ fmovd(v0, zr);
 351     break;
 352   case 1:
 353     __ fmovd(v0, 1.0);
 354     break;
 355   case 2:
 356     __ fmovd(v0, 2.0);
 357     break;
 358   default:
 359     ShouldNotReachHere();
 360     break;
 361   }
 362 }
 363 
 364 void TemplateTable::bipush()
 365 {
 366   transition(vtos, itos);
 367   __ load_signed_byte32(r0, at_bcp(1));
 368 }
 369 
 370 void TemplateTable::sipush()
 371 {
 372   transition(vtos, itos);
 373   __ load_unsigned_short(r0, at_bcp(1));
 374   __ revw(r0, r0);
 375   __ asrw(r0, r0, 16);
 376 }
 377 
 378 void TemplateTable::ldc(bool wide)
 379 {
 380   transition(vtos, vtos);
 381   Label call_ldc, notFloat, notClass, Done;
 382 
 383   if (wide) {
 384     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 385   } else {
 386     __ load_unsigned_byte(r1, at_bcp(1));
 387   }
 388   __ get_cpool_and_tags(r2, r0);
 389 
 390   const int base_offset = ConstantPool::header_size() * wordSize;
 391   const int tags_offset = Array<u1>::base_offset_in_bytes();
 392 
 393   // get type
 394   __ add(r3, r1, tags_offset);
 395   __ lea(r3, Address(r0, r3));
 396   __ ldarb(r3, r3);
 397 
 398   // unresolved class - get the resolved class
 399   __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
 400   __ br(Assembler::EQ, call_ldc);
 401 
 402   // unresolved class in error state - call into runtime to throw the error
 403   // from the first resolution attempt
 404   __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
 405   __ br(Assembler::EQ, call_ldc);
 406 
 407   // resolved class - need to call vm to get java mirror of the class
 408   __ cmp(r3, JVM_CONSTANT_Class);
 409   __ br(Assembler::NE, notClass);
 410 
 411   __ bind(call_ldc);
 412   __ mov(c_rarg1, wide);
 413   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 414   __ push_ptr(r0);
 415   __ verify_oop(r0);
 416   __ b(Done);
 417 
 418   __ bind(notClass);
 419   __ cmp(r3, JVM_CONSTANT_Float);
 420   __ br(Assembler::NE, notFloat);
 421   // ftos
 422   __ adds(r1, r2, r1, Assembler::LSL, 3);
 423   __ ldrs(v0, Address(r1, base_offset));
 424   __ push_f();
 425   __ b(Done);
 426 
 427   __ bind(notFloat);
 428 #ifdef ASSERT
 429   {
 430     Label L;
 431     __ cmp(r3, JVM_CONSTANT_Integer);
 432     __ br(Assembler::EQ, L);
 433     // String and Object are rewritten to fast_aldc
 434     __ stop("unexpected tag type in ldc");
 435     __ bind(L);
 436   }
 437 #endif
 438   // itos JVM_CONSTANT_Integer only
 439   __ adds(r1, r2, r1, Assembler::LSL, 3);
 440   __ ldrw(r0, Address(r1, base_offset));
 441   __ push_i(r0);
 442   __ bind(Done);
 443 }
 444 
 445 // Fast path for caching oop constants.
 446 void TemplateTable::fast_aldc(bool wide)
 447 {
 448   transition(vtos, atos);
 449 
 450   Register result = r0;
 451   Register tmp = r1;
 452   int index_size = wide ? sizeof(u2) : sizeof(u1);
 453 
 454   Label resolved;
 455 
 456   // We are resolved if the resolved reference cache entry contains a
 457   // non-null object (String, MethodType, etc.)
 458   assert_different_registers(result, tmp);
 459   __ get_cache_index_at_bcp(tmp, 1, index_size);
 460   __ load_resolved_reference_at_index(result, tmp);
 461   __ cbnz(result, resolved);
 462 
 463   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 464 
 465   // first time invocation - must resolve first
 466   __ mov(tmp, (int)bytecode());
 467   __ call_VM(result, entry, tmp);
 468 
 469   __ bind(resolved);
 470 
 471   if (VerifyOops) {
 472     __ verify_oop(result);
 473   }
 474 }
 475 
 476 void TemplateTable::ldc2_w()
 477 {
 478   transition(vtos, vtos);
 479   Label Long, Done;
 480   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 481 
 482   __ get_cpool_and_tags(r1, r2);
 483   const int base_offset = ConstantPool::header_size() * wordSize;
 484   const int tags_offset = Array<u1>::base_offset_in_bytes();
 485 
 486   // get type
 487   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 488   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 489   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 490   __ br(Assembler::NE, Long);
 491   // dtos
 492   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 493   __ ldrd(v0, Address(r2, base_offset));
 494   __ push_d();
 495   __ b(Done);
 496 
 497   __ bind(Long);
 498   // ltos
 499   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 500   __ ldr(r0, Address(r0, base_offset));
 501   __ push_l();
 502 
 503   __ bind(Done);
 504 }
 505 
 506 void TemplateTable::locals_index(Register reg, int offset)
 507 {
 508   __ ldrb(reg, at_bcp(offset));
 509   __ neg(reg, reg);
 510 }
 511 
 512 void TemplateTable::iload()
 513 {
 514   transition(vtos, itos);
 515   if (RewriteFrequentPairs) {
 516     Label rewrite, done;
 517     Register bc = r4;
 518 
 519     // get next bytecode
 520     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 521 
 522     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 523     // last two iloads in a pair.  Comparing against fast_iload means that
 524     // the next bytecode is neither an iload or a caload, and therefore
 525     // an iload pair.
 526     __ cmpw(r1, Bytecodes::_iload);
 527     __ br(Assembler::EQ, done);
 528 
 529     // if _fast_iload rewrite to _fast_iload2
 530     __ cmpw(r1, Bytecodes::_fast_iload);
 531     __ movw(bc, Bytecodes::_fast_iload2);
 532     __ br(Assembler::EQ, rewrite);
 533 
 534     // if _caload rewrite to _fast_icaload
 535     __ cmpw(r1, Bytecodes::_caload);
 536     __ movw(bc, Bytecodes::_fast_icaload);
 537     __ br(Assembler::EQ, rewrite);
 538 
 539     // else rewrite to _fast_iload
 540     __ movw(bc, Bytecodes::_fast_iload);
 541 
 542     // rewrite
 543     // bc: new bytecode
 544     __ bind(rewrite);
 545     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 546     __ bind(done);
 547 
 548   }
 549 
 550   // do iload, get the local value into tos
 551   locals_index(r1);
 552   __ ldr(r0, iaddress(r1));
 553 
 554 }
 555 
 556 void TemplateTable::fast_iload2()
 557 {
 558   transition(vtos, itos);
 559   locals_index(r1);
 560   __ ldr(r0, iaddress(r1));
 561   __ push(itos);
 562   locals_index(r1, 3);
 563   __ ldr(r0, iaddress(r1));
 564 }
 565 
 566 void TemplateTable::fast_iload()
 567 {
 568   transition(vtos, itos);
 569   locals_index(r1);
 570   __ ldr(r0, iaddress(r1));
 571 }
 572 
 573 void TemplateTable::lload()
 574 {
 575   transition(vtos, ltos);
 576   __ ldrb(r1, at_bcp(1));
 577   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 578   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 579 }
 580 
 581 void TemplateTable::fload()
 582 {
 583   transition(vtos, ftos);
 584   locals_index(r1);
 585   // n.b. we use ldrd here because this is a 64 bit slot
 586   // this is comparable to the iload case
 587   __ ldrd(v0, faddress(r1));
 588 }
 589 
 590 void TemplateTable::dload()
 591 {
 592   transition(vtos, dtos);
 593   __ ldrb(r1, at_bcp(1));
 594   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 595   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 596 }
 597 
 598 void TemplateTable::aload()
 599 {
 600   transition(vtos, atos);
 601   locals_index(r1);
 602   __ ldr(r0, iaddress(r1));
 603 }
 604 
 605 void TemplateTable::locals_index_wide(Register reg) {
 606   __ ldrh(reg, at_bcp(2));
 607   __ rev16w(reg, reg);
 608   __ neg(reg, reg);
 609 }
 610 
 611 void TemplateTable::wide_iload() {
 612   transition(vtos, itos);
 613   locals_index_wide(r1);
 614   __ ldr(r0, iaddress(r1));
 615 }
 616 
 617 void TemplateTable::wide_lload()
 618 {
 619   transition(vtos, ltos);
 620   __ ldrh(r1, at_bcp(2));
 621   __ rev16w(r1, r1);
 622   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 623   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 624 }
 625 
 626 void TemplateTable::wide_fload()
 627 {
 628   transition(vtos, ftos);
 629   locals_index_wide(r1);
 630   // n.b. we use ldrd here because this is a 64 bit slot
 631   // this is comparable to the iload case
 632   __ ldrd(v0, faddress(r1));
 633 }
 634 
 635 void TemplateTable::wide_dload()
 636 {
 637   transition(vtos, dtos);
 638   __ ldrh(r1, at_bcp(2));
 639   __ rev16w(r1, r1);
 640   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 641   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 642 }
 643 
 644 void TemplateTable::wide_aload()
 645 {
 646   transition(vtos, atos);
 647   locals_index_wide(r1);
 648   __ ldr(r0, aaddress(r1));
 649 }
 650 
 651 void TemplateTable::index_check(Register array, Register index)
 652 {
 653   // destroys r1, rscratch1
 654   // check array
 655   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 656   // sign extend index for use by indexed load
 657   // __ movl2ptr(index, index);
 658   // check index
 659   Register length = rscratch1;
 660   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 661   __ cmpw(index, length);
 662   if (index != r1) {
 663     // ??? convention: move aberrant index into r1 for exception message
 664     assert(r1 != array, "different registers");
 665     __ mov(r1, index);
 666   }
 667   Label ok;
 668   __ br(Assembler::LO, ok);
 669   __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 670   __ br(rscratch1);
 671   __ bind(ok);
 672 }
 673 
 674 void TemplateTable::iaload()
 675 {
 676   transition(itos, itos);
 677   __ mov(r1, r0);
 678   __ pop_ptr(r0);
 679   // r0: array
 680   // r1: index
 681   index_check(r0, r1); // leaves index in r1, kills rscratch1
 682   __ lea(r1, Address(r0, r1, Address::uxtw(2)));
 683   __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
 684 }
 685 
 686 void TemplateTable::laload()
 687 {
 688   transition(itos, ltos);
 689   __ mov(r1, r0);
 690   __ pop_ptr(r0);
 691   // r0: array
 692   // r1: index
 693   index_check(r0, r1); // leaves index in r1, kills rscratch1
 694   __ lea(r1, Address(r0, r1, Address::uxtw(3)));
 695   __ ldr(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_LONG)));
 696 }
 697 
 698 void TemplateTable::faload()
 699 {
 700   transition(itos, ftos);
 701   __ mov(r1, r0);
 702   __ pop_ptr(r0);
 703   // r0: array
 704   // r1: index
 705   index_check(r0, r1); // leaves index in r1, kills rscratch1
 706   __ lea(r1,  Address(r0, r1, Address::uxtw(2)));
 707   __ ldrs(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 708 }
 709 
 710 void TemplateTable::daload()
 711 {
 712   transition(itos, dtos);
 713   __ mov(r1, r0);
 714   __ pop_ptr(r0);
 715   // r0: array
 716   // r1: index
 717   index_check(r0, r1); // leaves index in r1, kills rscratch1
 718   __ lea(r1,  Address(r0, r1, Address::uxtw(3)));
 719   __ ldrd(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 720 }
 721 
 722 void TemplateTable::aaload()
 723 {
 724   transition(itos, atos);
 725   __ mov(r1, r0);
 726   __ pop_ptr(r0);
 727   // r0: array
 728   // r1: index
 729   index_check(r0, r1); // leaves index in r1, kills rscratch1
 730   int s = (UseCompressedOops ? 2 : 3);
 731   __ lea(r1, Address(r0, r1, Address::uxtw(s)));
 732   __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 733 }
 734 
 735 void TemplateTable::baload()
 736 {
 737   transition(itos, itos);
 738   __ mov(r1, r0);
 739   __ pop_ptr(r0);
 740   // r0: array
 741   // r1: index
 742   index_check(r0, r1); // leaves index in r1, kills rscratch1
 743   __ lea(r1,  Address(r0, r1, Address::uxtw(0)));
 744   __ load_signed_byte(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 745 }
 746 
 747 void TemplateTable::caload()
 748 {
 749   transition(itos, itos);
 750   __ mov(r1, r0);
 751   __ pop_ptr(r0);
 752   // r0: array
 753   // r1: index
 754   index_check(r0, r1); // leaves index in r1, kills rscratch1
 755   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 756   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 757 }
 758 
 759 // iload followed by caload frequent pair
 760 void TemplateTable::fast_icaload()
 761 {
 762   transition(vtos, itos);
 763   // load index out of locals
 764   locals_index(r2);
 765   __ ldr(r1, iaddress(r2));
 766 
 767   __ pop_ptr(r0);
 768 
 769   // r0: array
 770   // r1: index
 771   index_check(r0, r1); // leaves index in r1, kills rscratch1
 772   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 773   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 774 }
 775 
 776 void TemplateTable::saload()
 777 {
 778   transition(itos, itos);
 779   __ mov(r1, r0);
 780   __ pop_ptr(r0);
 781   // r0: array
 782   // r1: index
 783   index_check(r0, r1); // leaves index in r1, kills rscratch1
 784   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 785   __ load_signed_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 786 }
 787 
 788 void TemplateTable::iload(int n)
 789 {
 790   transition(vtos, itos);
 791   __ ldr(r0, iaddress(n));
 792 }
 793 
 794 void TemplateTable::lload(int n)
 795 {
 796   transition(vtos, ltos);
 797   __ ldr(r0, laddress(n));
 798 }
 799 
 800 void TemplateTable::fload(int n)
 801 {
 802   transition(vtos, ftos);
 803   __ ldrs(v0, faddress(n));
 804 }
 805 
 806 void TemplateTable::dload(int n)
 807 {
 808   transition(vtos, dtos);
 809   __ ldrd(v0, daddress(n));
 810 }
 811 
 812 void TemplateTable::aload(int n)
 813 {
 814   transition(vtos, atos);
 815   __ ldr(r0, iaddress(n));
 816 }
 817 
 818 void TemplateTable::aload_0()
 819 {
 820   // According to bytecode histograms, the pairs:
 821   //
 822   // _aload_0, _fast_igetfield
 823   // _aload_0, _fast_agetfield
 824   // _aload_0, _fast_fgetfield
 825   //
 826   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 827   // _aload_0 bytecode checks if the next bytecode is either
 828   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 829   // rewrites the current bytecode into a pair bytecode; otherwise it
 830   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 831   // the pair check anymore.
 832   //
 833   // Note: If the next bytecode is _getfield, the rewrite must be
 834   //       delayed, otherwise we may miss an opportunity for a pair.
 835   //
 836   // Also rewrite frequent pairs
 837   //   aload_0, aload_1
 838   //   aload_0, iload_1
 839   // These bytecodes with a small amount of code are most profitable
 840   // to rewrite
 841   if (RewriteFrequentPairs) {
 842     Label rewrite, done;
 843     const Register bc = r4;
 844 
 845     // get next bytecode
 846     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 847 
 848     // do actual aload_0
 849     aload(0);
 850 
 851     // if _getfield then wait with rewrite
 852     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 853     __ br(Assembler::EQ, done);
 854 
 855     // if _igetfield then reqrite to _fast_iaccess_0
 856     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 857     __ cmpw(r1, Bytecodes::_fast_igetfield);
 858     __ movw(bc, Bytecodes::_fast_iaccess_0);
 859     __ br(Assembler::EQ, rewrite);
 860 
 861     // if _agetfield then reqrite to _fast_aaccess_0
 862     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 863     __ cmpw(r1, Bytecodes::_fast_agetfield);
 864     __ movw(bc, Bytecodes::_fast_aaccess_0);
 865     __ br(Assembler::EQ, rewrite);
 866 
 867     // if _fgetfield then reqrite to _fast_faccess_0
 868     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 869     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 870     __ movw(bc, Bytecodes::_fast_faccess_0);
 871     __ br(Assembler::EQ, rewrite);
 872 
 873     // else rewrite to _fast_aload0
 874     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 875     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 876 
 877     // rewrite
 878     // bc: new bytecode
 879     __ bind(rewrite);
 880     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 881 
 882     __ bind(done);
 883   } else {
 884     aload(0);
 885   }
 886 }
 887 
 888 void TemplateTable::istore()
 889 {
 890   transition(itos, vtos);
 891   locals_index(r1);
 892   // FIXME: We're being very pernickerty here storing a jint in a
 893   // local with strw, which costs an extra instruction over what we'd
 894   // be able to do with a simple str.  We should just store the whole
 895   // word.
 896   __ lea(rscratch1, iaddress(r1));
 897   __ strw(r0, Address(rscratch1));
 898 }
 899 
 900 void TemplateTable::lstore()
 901 {
 902   transition(ltos, vtos);
 903   locals_index(r1);
 904   __ str(r0, laddress(r1, rscratch1, _masm));
 905 }
 906 
 907 void TemplateTable::fstore() {
 908   transition(ftos, vtos);
 909   locals_index(r1);
 910   __ lea(rscratch1, iaddress(r1));
 911   __ strs(v0, Address(rscratch1));
 912 }
 913 
 914 void TemplateTable::dstore() {
 915   transition(dtos, vtos);
 916   locals_index(r1);
 917   __ strd(v0, daddress(r1, rscratch1, _masm));
 918 }
 919 
 920 void TemplateTable::astore()
 921 {
 922   transition(vtos, vtos);
 923   __ pop_ptr(r0);
 924   locals_index(r1);
 925   __ str(r0, aaddress(r1));
 926 }
 927 
 928 void TemplateTable::wide_istore() {
 929   transition(vtos, vtos);
 930   __ pop_i();
 931   locals_index_wide(r1);
 932   __ lea(rscratch1, iaddress(r1));
 933   __ strw(r0, Address(rscratch1));
 934 }
 935 
 936 void TemplateTable::wide_lstore() {
 937   transition(vtos, vtos);
 938   __ pop_l();
 939   locals_index_wide(r1);
 940   __ str(r0, laddress(r1, rscratch1, _masm));
 941 }
 942 
 943 void TemplateTable::wide_fstore() {
 944   transition(vtos, vtos);
 945   __ pop_f();
 946   locals_index_wide(r1);
 947   __ lea(rscratch1, faddress(r1));
 948   __ strs(v0, rscratch1);
 949 }
 950 
 951 void TemplateTable::wide_dstore() {
 952   transition(vtos, vtos);
 953   __ pop_d();
 954   locals_index_wide(r1);
 955   __ strd(v0, daddress(r1, rscratch1, _masm));
 956 }
 957 
 958 void TemplateTable::wide_astore() {
 959   transition(vtos, vtos);
 960   __ pop_ptr(r0);
 961   locals_index_wide(r1);
 962   __ str(r0, aaddress(r1));
 963 }
 964 
 965 void TemplateTable::iastore() {
 966   transition(itos, vtos);
 967   __ pop_i(r1);
 968   __ pop_ptr(r3);
 969   // r0: value
 970   // r1: index
 971   // r3: array
 972   index_check(r3, r1); // prefer index in r1
 973   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
 974   __ strw(r0, Address(rscratch1,
 975                       arrayOopDesc::base_offset_in_bytes(T_INT)));
 976 }
 977 
 978 void TemplateTable::lastore() {
 979   transition(ltos, vtos);
 980   __ pop_i(r1);
 981   __ pop_ptr(r3);
 982   // r0: value
 983   // r1: index
 984   // r3: array
 985   index_check(r3, r1); // prefer index in r1
 986   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
 987   __ str(r0, Address(rscratch1,
 988                       arrayOopDesc::base_offset_in_bytes(T_LONG)));
 989 }
 990 
 991 void TemplateTable::fastore() {
 992   transition(ftos, vtos);
 993   __ pop_i(r1);
 994   __ pop_ptr(r3);
 995   // v0: value
 996   // r1:  index
 997   // r3:  array
 998   index_check(r3, r1); // prefer index in r1
 999   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1000   __ strs(v0, Address(rscratch1,
1001                       arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1002 }
1003 
1004 void TemplateTable::dastore() {
1005   transition(dtos, vtos);
1006   __ pop_i(r1);
1007   __ pop_ptr(r3);
1008   // v0: value
1009   // r1:  index
1010   // r3:  array
1011   index_check(r3, r1); // prefer index in r1
1012   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1013   __ strd(v0, Address(rscratch1,
1014                       arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1015 }
1016 
1017 void TemplateTable::aastore() {
1018   Label is_null, ok_is_subtype, done;
1019   transition(vtos, vtos);
1020   // stack: ..., array, index, value
1021   __ ldr(r0, at_tos());    // value
1022   __ ldr(r2, at_tos_p1()); // index
1023   __ ldr(r3, at_tos_p2()); // array
1024 
1025   Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1026 
1027   index_check(r3, r2);     // kills r1
1028   __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1029 
1030   // do array store check - check for NULL value first
1031   __ cbz(r0, is_null);
1032 
1033   // Move subklass into r1
1034   __ load_klass(r1, r0);
1035   // Move superklass into r0
1036   __ load_klass(r0, r3);
1037   __ ldr(r0, Address(r0,
1038                      ObjArrayKlass::element_klass_offset()));
1039   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1040 
1041   // Generate subtype check.  Blows r2, r5
1042   // Superklass in r0.  Subklass in r1.
1043   __ gen_subtype_check(r1, ok_is_subtype);
1044 
1045   // Come here on failure
1046   // object is at TOS
1047   __ b(Interpreter::_throw_ArrayStoreException_entry);
1048 
1049   // Come here on success
1050   __ bind(ok_is_subtype);
1051 
1052   // Get the value we will store
1053   __ ldr(r0, at_tos());
1054   // Now store using the appropriate barrier
1055   do_oop_store(_masm, element_address, r0, _bs->kind(), true);
1056   __ b(done);
1057 
1058   // Have a NULL in r0, r3=array, r2=index.  Store NULL at ary[idx]
1059   __ bind(is_null);
1060   __ profile_null_seen(r2);
1061 
1062   // Store a NULL
1063   do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1064 
1065   // Pop stack arguments
1066   __ bind(done);
1067   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1068 }
1069 
1070 void TemplateTable::bastore()
1071 {
1072   transition(itos, vtos);
1073   __ pop_i(r1);
1074   __ pop_ptr(r3);
1075   // r0: value
1076   // r1: index
1077   // r3: array
1078   index_check(r3, r1); // prefer index in r1
1079 
1080   // Need to check whether array is boolean or byte
1081   // since both types share the bastore bytecode.
1082   __ load_klass(r2, r3);
1083   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1084   int diffbit = Klass::layout_helper_boolean_diffbit();
1085   __ andw(rscratch1, r2, diffbit);
1086   Label L_skip;
1087   __ cbzw(rscratch1, L_skip);
1088   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1089   __ bind(L_skip);
1090 
1091   __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1092   __ strb(r0, Address(rscratch1,
1093                       arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1094 }
1095 
1096 void TemplateTable::castore()
1097 {
1098   transition(itos, vtos);
1099   __ pop_i(r1);
1100   __ pop_ptr(r3);
1101   // r0: value
1102   // r1: index
1103   // r3: array
1104   index_check(r3, r1); // prefer index in r1
1105   __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1106   __ strh(r0, Address(rscratch1,
1107                       arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1108 }
1109 
1110 void TemplateTable::sastore()
1111 {
1112   castore();
1113 }
1114 
1115 void TemplateTable::istore(int n)
1116 {
1117   transition(itos, vtos);
1118   __ str(r0, iaddress(n));
1119 }
1120 
1121 void TemplateTable::lstore(int n)
1122 {
1123   transition(ltos, vtos);
1124   __ str(r0, laddress(n));
1125 }
1126 
1127 void TemplateTable::fstore(int n)
1128 {
1129   transition(ftos, vtos);
1130   __ strs(v0, faddress(n));
1131 }
1132 
1133 void TemplateTable::dstore(int n)
1134 {
1135   transition(dtos, vtos);
1136   __ strd(v0, daddress(n));
1137 }
1138 
1139 void TemplateTable::astore(int n)
1140 {
1141   transition(vtos, vtos);
1142   __ pop_ptr(r0);
1143   __ str(r0, iaddress(n));
1144 }
1145 
1146 void TemplateTable::pop()
1147 {
1148   transition(vtos, vtos);
1149   __ add(esp, esp, Interpreter::stackElementSize);
1150 }
1151 
1152 void TemplateTable::pop2()
1153 {
1154   transition(vtos, vtos);
1155   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1156 }
1157 
1158 void TemplateTable::dup()
1159 {
1160   transition(vtos, vtos);
1161   __ ldr(r0, Address(esp, 0));
1162   __ push(r0);
1163   // stack: ..., a, a
1164 }
1165 
1166 void TemplateTable::dup_x1()
1167 {
1168   transition(vtos, vtos);
1169   // stack: ..., a, b
1170   __ ldr(r0, at_tos());  // load b
1171   __ ldr(r2, at_tos_p1());  // load a
1172   __ str(r0, at_tos_p1());  // store b
1173   __ str(r2, at_tos());  // store a
1174   __ push(r0);                  // push b
1175   // stack: ..., b, a, b
1176 }
1177 
1178 void TemplateTable::dup_x2()
1179 {
1180   transition(vtos, vtos);
1181   // stack: ..., a, b, c
1182   __ ldr(r0, at_tos());  // load c
1183   __ ldr(r2, at_tos_p2());  // load a
1184   __ str(r0, at_tos_p2());  // store c in a
1185   __ push(r0);      // push c
1186   // stack: ..., c, b, c, c
1187   __ ldr(r0, at_tos_p2());  // load b
1188   __ str(r2, at_tos_p2());  // store a in b
1189   // stack: ..., c, a, c, c
1190   __ str(r0, at_tos_p1());  // store b in c
1191   // stack: ..., c, a, b, c
1192 }
1193 
1194 void TemplateTable::dup2()
1195 {
1196   transition(vtos, vtos);
1197   // stack: ..., a, b
1198   __ ldr(r0, at_tos_p1());  // load a
1199   __ push(r0);                  // push a
1200   __ ldr(r0, at_tos_p1());  // load b
1201   __ push(r0);                  // push b
1202   // stack: ..., a, b, a, b
1203 }
1204 
1205 void TemplateTable::dup2_x1()
1206 {
1207   transition(vtos, vtos);
1208   // stack: ..., a, b, c
1209   __ ldr(r2, at_tos());  // load c
1210   __ ldr(r0, at_tos_p1());  // load b
1211   __ push(r0);                  // push b
1212   __ push(r2);                  // push c
1213   // stack: ..., a, b, c, b, c
1214   __ str(r2, at_tos_p3());  // store c in b
1215   // stack: ..., a, c, c, b, c
1216   __ ldr(r2, at_tos_p4());  // load a
1217   __ str(r2, at_tos_p2());  // store a in 2nd c
1218   // stack: ..., a, c, a, b, c
1219   __ str(r0, at_tos_p4());  // store b in a
1220   // stack: ..., b, c, a, b, c
1221 }
1222 
1223 void TemplateTable::dup2_x2()
1224 {
1225   transition(vtos, vtos);
1226   // stack: ..., a, b, c, d
1227   __ ldr(r2, at_tos());  // load d
1228   __ ldr(r0, at_tos_p1());  // load c
1229   __ push(r0)            ;      // push c
1230   __ push(r2);                  // push d
1231   // stack: ..., a, b, c, d, c, d
1232   __ ldr(r0, at_tos_p4());  // load b
1233   __ str(r0, at_tos_p2());  // store b in d
1234   __ str(r2, at_tos_p4());  // store d in b
1235   // stack: ..., a, d, c, b, c, d
1236   __ ldr(r2, at_tos_p5());  // load a
1237   __ ldr(r0, at_tos_p3());  // load c
1238   __ str(r2, at_tos_p3());  // store a in c
1239   __ str(r0, at_tos_p5());  // store c in a
1240   // stack: ..., c, d, a, b, c, d
1241 }
1242 
1243 void TemplateTable::swap()
1244 {
1245   transition(vtos, vtos);
1246   // stack: ..., a, b
1247   __ ldr(r2, at_tos_p1());  // load a
1248   __ ldr(r0, at_tos());  // load b
1249   __ str(r2, at_tos());  // store a in b
1250   __ str(r0, at_tos_p1());  // store b in a
1251   // stack: ..., b, a
1252 }
1253 
1254 void TemplateTable::iop2(Operation op)
1255 {
1256   transition(itos, itos);
1257   // r0 <== r1 op r0
1258   __ pop_i(r1);
1259   switch (op) {
1260   case add  : __ addw(r0, r1, r0); break;
1261   case sub  : __ subw(r0, r1, r0); break;
1262   case mul  : __ mulw(r0, r1, r0); break;
1263   case _and : __ andw(r0, r1, r0); break;
1264   case _or  : __ orrw(r0, r1, r0); break;
1265   case _xor : __ eorw(r0, r1, r0); break;
1266   case shl  : __ lslvw(r0, r1, r0); break;
1267   case shr  : __ asrvw(r0, r1, r0); break;
1268   case ushr : __ lsrvw(r0, r1, r0);break;
1269   default   : ShouldNotReachHere();
1270   }
1271 }
1272 
1273 void TemplateTable::lop2(Operation op)
1274 {
1275   transition(ltos, ltos);
1276   // r0 <== r1 op r0
1277   __ pop_l(r1);
1278   switch (op) {
1279   case add  : __ add(r0, r1, r0); break;
1280   case sub  : __ sub(r0, r1, r0); break;
1281   case mul  : __ mul(r0, r1, r0); break;
1282   case _and : __ andr(r0, r1, r0); break;
1283   case _or  : __ orr(r0, r1, r0); break;
1284   case _xor : __ eor(r0, r1, r0); break;
1285   default   : ShouldNotReachHere();
1286   }
1287 }
1288 
1289 void TemplateTable::idiv()
1290 {
1291   transition(itos, itos);
1292   // explicitly check for div0
1293   Label no_div0;
1294   __ cbnzw(r0, no_div0);
1295   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1296   __ br(rscratch1);
1297   __ bind(no_div0);
1298   __ pop_i(r1);
1299   // r0 <== r1 idiv r0
1300   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1301 }
1302 
1303 void TemplateTable::irem()
1304 {
1305   transition(itos, itos);
1306   // explicitly check for div0
1307   Label no_div0;
1308   __ cbnzw(r0, no_div0);
1309   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1310   __ br(rscratch1);
1311   __ bind(no_div0);
1312   __ pop_i(r1);
1313   // r0 <== r1 irem r0
1314   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1315 }
1316 
1317 void TemplateTable::lmul()
1318 {
1319   transition(ltos, ltos);
1320   __ pop_l(r1);
1321   __ mul(r0, r0, r1);
1322 }
1323 
1324 void TemplateTable::ldiv()
1325 {
1326   transition(ltos, ltos);
1327   // explicitly check for div0
1328   Label no_div0;
1329   __ cbnz(r0, no_div0);
1330   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1331   __ br(rscratch1);
1332   __ bind(no_div0);
1333   __ pop_l(r1);
1334   // r0 <== r1 ldiv r0
1335   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1336 }
1337 
1338 void TemplateTable::lrem()
1339 {
1340   transition(ltos, ltos);
1341   // explicitly check for div0
1342   Label no_div0;
1343   __ cbnz(r0, no_div0);
1344   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1345   __ br(rscratch1);
1346   __ bind(no_div0);
1347   __ pop_l(r1);
1348   // r0 <== r1 lrem r0
1349   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1350 }
1351 
1352 void TemplateTable::lshl()
1353 {
1354   transition(itos, ltos);
1355   // shift count is in r0
1356   __ pop_l(r1);
1357   __ lslv(r0, r1, r0);
1358 }
1359 
1360 void TemplateTable::lshr()
1361 {
1362   transition(itos, ltos);
1363   // shift count is in r0
1364   __ pop_l(r1);
1365   __ asrv(r0, r1, r0);
1366 }
1367 
1368 void TemplateTable::lushr()
1369 {
1370   transition(itos, ltos);
1371   // shift count is in r0
1372   __ pop_l(r1);
1373   __ lsrv(r0, r1, r0);
1374 }
1375 
1376 void TemplateTable::fop2(Operation op)
1377 {
1378   transition(ftos, ftos);
1379   switch (op) {
1380   case add:
1381     // n.b. use ldrd because this is a 64 bit slot
1382     __ pop_f(v1);
1383     __ fadds(v0, v1, v0);
1384     break;
1385   case sub:
1386     __ pop_f(v1);
1387     __ fsubs(v0, v1, v0);
1388     break;
1389   case mul:
1390     __ pop_f(v1);
1391     __ fmuls(v0, v1, v0);
1392     break;
1393   case div:
1394     __ pop_f(v1);
1395     __ fdivs(v0, v1, v0);
1396     break;
1397   case rem:
1398     __ fmovs(v1, v0);
1399     __ pop_f(v0);
1400     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1401     break;
1402   default:
1403     ShouldNotReachHere();
1404     break;
1405   }
1406 }
1407 
1408 void TemplateTable::dop2(Operation op)
1409 {
1410   transition(dtos, dtos);
1411   switch (op) {
1412   case add:
1413     // n.b. use ldrd because this is a 64 bit slot
1414     __ pop_d(v1);
1415     __ faddd(v0, v1, v0);
1416     break;
1417   case sub:
1418     __ pop_d(v1);
1419     __ fsubd(v0, v1, v0);
1420     break;
1421   case mul:
1422     __ pop_d(v1);
1423     __ fmuld(v0, v1, v0);
1424     break;
1425   case div:
1426     __ pop_d(v1);
1427     __ fdivd(v0, v1, v0);
1428     break;
1429   case rem:
1430     __ fmovd(v1, v0);
1431     __ pop_d(v0);
1432     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1433     break;
1434   default:
1435     ShouldNotReachHere();
1436     break;
1437   }
1438 }
1439 
1440 void TemplateTable::ineg()
1441 {
1442   transition(itos, itos);
1443   __ negw(r0, r0);
1444 
1445 }
1446 
1447 void TemplateTable::lneg()
1448 {
1449   transition(ltos, ltos);
1450   __ neg(r0, r0);
1451 }
1452 
1453 void TemplateTable::fneg()
1454 {
1455   transition(ftos, ftos);
1456   __ fnegs(v0, v0);
1457 }
1458 
1459 void TemplateTable::dneg()
1460 {
1461   transition(dtos, dtos);
1462   __ fnegd(v0, v0);
1463 }
1464 
1465 void TemplateTable::iinc()
1466 {
1467   transition(vtos, vtos);
1468   __ load_signed_byte(r1, at_bcp(2)); // get constant
1469   locals_index(r2);
1470   __ ldr(r0, iaddress(r2));
1471   __ addw(r0, r0, r1);
1472   __ str(r0, iaddress(r2));
1473 }
1474 
1475 void TemplateTable::wide_iinc()
1476 {
1477   transition(vtos, vtos);
1478   // __ mov(r1, zr);
1479   __ ldrw(r1, at_bcp(2)); // get constant and index
1480   __ rev16(r1, r1);
1481   __ ubfx(r2, r1, 0, 16);
1482   __ neg(r2, r2);
1483   __ sbfx(r1, r1, 16, 16);
1484   __ ldr(r0, iaddress(r2));
1485   __ addw(r0, r0, r1);
1486   __ str(r0, iaddress(r2));
1487 }
1488 
1489 void TemplateTable::convert()
1490 {
1491   // Checking
1492 #ifdef ASSERT
1493   {
1494     TosState tos_in  = ilgl;
1495     TosState tos_out = ilgl;
1496     switch (bytecode()) {
1497     case Bytecodes::_i2l: // fall through
1498     case Bytecodes::_i2f: // fall through
1499     case Bytecodes::_i2d: // fall through
1500     case Bytecodes::_i2b: // fall through
1501     case Bytecodes::_i2c: // fall through
1502     case Bytecodes::_i2s: tos_in = itos; break;
1503     case Bytecodes::_l2i: // fall through
1504     case Bytecodes::_l2f: // fall through
1505     case Bytecodes::_l2d: tos_in = ltos; break;
1506     case Bytecodes::_f2i: // fall through
1507     case Bytecodes::_f2l: // fall through
1508     case Bytecodes::_f2d: tos_in = ftos; break;
1509     case Bytecodes::_d2i: // fall through
1510     case Bytecodes::_d2l: // fall through
1511     case Bytecodes::_d2f: tos_in = dtos; break;
1512     default             : ShouldNotReachHere();
1513     }
1514     switch (bytecode()) {
1515     case Bytecodes::_l2i: // fall through
1516     case Bytecodes::_f2i: // fall through
1517     case Bytecodes::_d2i: // fall through
1518     case Bytecodes::_i2b: // fall through
1519     case Bytecodes::_i2c: // fall through
1520     case Bytecodes::_i2s: tos_out = itos; break;
1521     case Bytecodes::_i2l: // fall through
1522     case Bytecodes::_f2l: // fall through
1523     case Bytecodes::_d2l: tos_out = ltos; break;
1524     case Bytecodes::_i2f: // fall through
1525     case Bytecodes::_l2f: // fall through
1526     case Bytecodes::_d2f: tos_out = ftos; break;
1527     case Bytecodes::_i2d: // fall through
1528     case Bytecodes::_l2d: // fall through
1529     case Bytecodes::_f2d: tos_out = dtos; break;
1530     default             : ShouldNotReachHere();
1531     }
1532     transition(tos_in, tos_out);
1533   }
1534 #endif // ASSERT
1535   // static const int64_t is_nan = 0x8000000000000000L;
1536 
1537   // Conversion
1538   switch (bytecode()) {
1539   case Bytecodes::_i2l:
1540     __ sxtw(r0, r0);
1541     break;
1542   case Bytecodes::_i2f:
1543     __ scvtfws(v0, r0);
1544     break;
1545   case Bytecodes::_i2d:
1546     __ scvtfwd(v0, r0);
1547     break;
1548   case Bytecodes::_i2b:
1549     __ sxtbw(r0, r0);
1550     break;
1551   case Bytecodes::_i2c:
1552     __ uxthw(r0, r0);
1553     break;
1554   case Bytecodes::_i2s:
1555     __ sxthw(r0, r0);
1556     break;
1557   case Bytecodes::_l2i:
1558     __ uxtw(r0, r0);
1559     break;
1560   case Bytecodes::_l2f:
1561     __ scvtfs(v0, r0);
1562     break;
1563   case Bytecodes::_l2d:
1564     __ scvtfd(v0, r0);
1565     break;
1566   case Bytecodes::_f2i:
1567   {
1568     Label L_Okay;
1569     __ clear_fpsr();
1570     __ fcvtzsw(r0, v0);
1571     __ get_fpsr(r1);
1572     __ cbzw(r1, L_Okay);
1573     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i));
1574     __ bind(L_Okay);
1575   }
1576     break;
1577   case Bytecodes::_f2l:
1578   {
1579     Label L_Okay;
1580     __ clear_fpsr();
1581     __ fcvtzs(r0, v0);
1582     __ get_fpsr(r1);
1583     __ cbzw(r1, L_Okay);
1584     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1585     __ bind(L_Okay);
1586   }
1587     break;
1588   case Bytecodes::_f2d:
1589     __ fcvts(v0, v0);
1590     break;
1591   case Bytecodes::_d2i:
1592   {
1593     Label L_Okay;
1594     __ clear_fpsr();
1595     __ fcvtzdw(r0, v0);
1596     __ get_fpsr(r1);
1597     __ cbzw(r1, L_Okay);
1598     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1599     __ bind(L_Okay);
1600   }
1601     break;
1602   case Bytecodes::_d2l:
1603   {
1604     Label L_Okay;
1605     __ clear_fpsr();
1606     __ fcvtzd(r0, v0);
1607     __ get_fpsr(r1);
1608     __ cbzw(r1, L_Okay);
1609     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1610     __ bind(L_Okay);
1611   }
1612     break;
1613   case Bytecodes::_d2f:
1614     __ fcvtd(v0, v0);
1615     break;
1616   default:
1617     ShouldNotReachHere();
1618   }
1619 }
1620 
1621 void TemplateTable::lcmp()
1622 {
1623   transition(ltos, itos);
1624   Label done;
1625   __ pop_l(r1);
1626   __ cmp(r1, r0);
1627   __ mov(r0, (u_int64_t)-1L);
1628   __ br(Assembler::LT, done);
1629   // __ mov(r0, 1UL);
1630   // __ csel(r0, r0, zr, Assembler::NE);
1631   // and here is a faster way
1632   __ csinc(r0, zr, zr, Assembler::EQ);
1633   __ bind(done);
1634 }
1635 
1636 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1637 {
1638   Label done;
1639   if (is_float) {
1640     // XXX get rid of pop here, use ... reg, mem32
1641     __ pop_f(v1);
1642     __ fcmps(v1, v0);
1643   } else {
1644     // XXX get rid of pop here, use ... reg, mem64
1645     __ pop_d(v1);
1646     __ fcmpd(v1, v0);
1647   }
1648   if (unordered_result < 0) {
1649     // we want -1 for unordered or less than, 0 for equal and 1 for
1650     // greater than.
1651     __ mov(r0, (u_int64_t)-1L);
1652     // for FP LT tests less than or unordered
1653     __ br(Assembler::LT, done);
1654     // install 0 for EQ otherwise 1
1655     __ csinc(r0, zr, zr, Assembler::EQ);
1656   } else {
1657     // we want -1 for less than, 0 for equal and 1 for unordered or
1658     // greater than.
1659     __ mov(r0, 1L);
1660     // for FP HI tests greater than or unordered
1661     __ br(Assembler::HI, done);
1662     // install 0 for EQ otherwise ~0
1663     __ csinv(r0, zr, zr, Assembler::EQ);
1664 
1665   }
1666   __ bind(done);
1667 }
1668 
1669 void TemplateTable::branch(bool is_jsr, bool is_wide)
1670 {
1671   // We might be moving to a safepoint.  The thread which calls
1672   // Interpreter::notice_safepoints() will effectively flush its cache
1673   // when it makes a system call, but we need to do something to
1674   // ensure that we see the changed dispatch table.
1675   __ membar(MacroAssembler::LoadLoad);
1676 
1677   __ profile_taken_branch(r0, r1);
1678   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1679                              InvocationCounter::counter_offset();
1680   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1681                               InvocationCounter::counter_offset();
1682 
1683   // load branch displacement
1684   if (!is_wide) {
1685     __ ldrh(r2, at_bcp(1));
1686     __ rev16(r2, r2);
1687     // sign extend the 16 bit value in r2
1688     __ sbfm(r2, r2, 0, 15);
1689   } else {
1690     __ ldrw(r2, at_bcp(1));
1691     __ revw(r2, r2);
1692     // sign extend the 32 bit value in r2
1693     __ sbfm(r2, r2, 0, 31);
1694   }
1695 
1696   // Handle all the JSR stuff here, then exit.
1697   // It's much shorter and cleaner than intermingling with the non-JSR
1698   // normal-branch stuff occurring below.
1699 
1700   if (is_jsr) {
1701     // Pre-load the next target bytecode into rscratch1
1702     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1703     // compute return address as bci
1704     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1705     __ add(rscratch2, rscratch2,
1706            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1707     __ sub(r1, rbcp, rscratch2);
1708     __ push_i(r1);
1709     // Adjust the bcp by the 16-bit displacement in r2
1710     __ add(rbcp, rbcp, r2);
1711     __ dispatch_only(vtos);
1712     return;
1713   }
1714 
1715   // Normal (non-jsr) branch handling
1716 
1717   // Adjust the bcp by the displacement in r2
1718   __ add(rbcp, rbcp, r2);
1719 
1720   assert(UseLoopCounter || !UseOnStackReplacement,
1721          "on-stack-replacement requires loop counters");
1722   Label backedge_counter_overflow;
1723   Label profile_method;
1724   Label dispatch;
1725   if (UseLoopCounter) {
1726     // increment backedge counter for backward branches
1727     // r0: MDO
1728     // w1: MDO bumped taken-count
1729     // r2: target offset
1730     __ cmp(r2, zr);
1731     __ br(Assembler::GT, dispatch); // count only if backward branch
1732 
1733     // ECN: FIXME: This code smells
1734     // check if MethodCounters exists
1735     Label has_counters;
1736     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1737     __ cbnz(rscratch1, has_counters);
1738     __ push(r0);
1739     __ push(r1);
1740     __ push(r2);
1741     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1742             InterpreterRuntime::build_method_counters), rmethod);
1743     __ pop(r2);
1744     __ pop(r1);
1745     __ pop(r0);
1746     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1747     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1748     __ bind(has_counters);
1749 
1750     if (TieredCompilation) {
1751       Label no_mdo;
1752       int increment = InvocationCounter::count_increment;
1753       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1754       if (ProfileInterpreter) {
1755         // Are we profiling?
1756         __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1757         __ cbz(r1, no_mdo);
1758         // Increment the MDO backedge counter
1759         const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1760                                            in_bytes(InvocationCounter::counter_offset()));
1761         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1762                                    r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1763         __ b(dispatch);
1764       }
1765       __ bind(no_mdo);
1766       // Increment backedge counter in MethodCounters*
1767       __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1768       __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1769                                  r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1770     } else {
1771       // increment counter
1772       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1773       __ ldrw(r0, Address(rscratch2, be_offset));        // load backedge counter
1774       __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1775       __ strw(rscratch1, Address(rscratch2, be_offset));        // store counter
1776 
1777       __ ldrw(r0, Address(rscratch2, inv_offset));    // load invocation counter
1778       __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1779       __ addw(r0, r0, rscratch1);        // add both counters
1780 
1781       if (ProfileInterpreter) {
1782         // Test to see if we should create a method data oop
1783         __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1784         __ ldrw(rscratch1, rscratch1);
1785         __ cmpw(r0, rscratch1);
1786         __ br(Assembler::LT, dispatch);
1787 
1788         // if no method data exists, go to profile method
1789         __ test_method_data_pointer(r0, profile_method);
1790 
1791         if (UseOnStackReplacement) {
1792           // check for overflow against w1 which is the MDO taken count
1793           __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1794           __ ldrw(rscratch1, rscratch1);
1795           __ cmpw(r1, rscratch1);
1796           __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1797 
1798           // When ProfileInterpreter is on, the backedge_count comes
1799           // from the MethodData*, which value does not get reset on
1800           // the call to frequency_counter_overflow().  To avoid
1801           // excessive calls to the overflow routine while the method is
1802           // being compiled, add a second test to make sure the overflow
1803           // function is called only once every overflow_frequency.
1804           const int overflow_frequency = 1024;
1805           __ andsw(r1, r1, overflow_frequency - 1);
1806           __ br(Assembler::EQ, backedge_counter_overflow);
1807 
1808         }
1809       } else {
1810         if (UseOnStackReplacement) {
1811           // check for overflow against w0, which is the sum of the
1812           // counters
1813           __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1814           __ ldrw(rscratch1, rscratch1);
1815           __ cmpw(r0, rscratch1);
1816           __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1817         }
1818       }
1819     }
1820   }
1821   __ bind(dispatch);
1822 
1823   // Pre-load the next target bytecode into rscratch1
1824   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1825 
1826   // continue with the bytecode @ target
1827   // rscratch1: target bytecode
1828   // rbcp: target bcp
1829   __ dispatch_only(vtos);
1830 
1831   if (UseLoopCounter) {
1832     if (ProfileInterpreter) {
1833       // Out-of-line code to allocate method data oop.
1834       __ bind(profile_method);
1835       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1836       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1837       __ set_method_data_pointer_for_bcp();
1838       __ b(dispatch);
1839     }
1840 
1841     if (TieredCompilation || UseOnStackReplacement) {
1842       // invocation counter overflow
1843       __ bind(backedge_counter_overflow);
1844       __ neg(r2, r2);
1845       __ add(r2, r2, rbcp);     // branch bcp
1846       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1847       __ call_VM(noreg,
1848                  CAST_FROM_FN_PTR(address,
1849                                   InterpreterRuntime::frequency_counter_overflow),
1850                  r2);
1851       if (!UseOnStackReplacement)
1852         __ b(dispatch);
1853     }
1854 
1855     if (UseOnStackReplacement) {
1856       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1857 
1858       // r0: osr nmethod (osr ok) or NULL (osr not possible)
1859       // w1: target bytecode
1860       // r2: scratch
1861       __ cbz(r0, dispatch);     // test result -- no osr if null
1862       // nmethod may have been invalidated (VM may block upon call_VM return)
1863       __ ldrw(r2, Address(r0, nmethod::entry_bci_offset()));
1864       // InvalidOSREntryBci == -2 which overflows cmpw as unsigned
1865       // use cmnw against -InvalidOSREntryBci which does the same thing
1866       __ cmn(r2, -InvalidOSREntryBci);
1867       __ br(Assembler::EQ, dispatch);
1868 
1869       // We have the address of an on stack replacement routine in r0
1870       // We need to prepare to execute the OSR method. First we must
1871       // migrate the locals and monitors off of the stack.
1872 
1873       __ mov(r19, r0);                             // save the nmethod
1874 
1875       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1876 
1877       // r0 is OSR buffer, move it to expected parameter location
1878       __ mov(j_rarg0, r0);
1879 
1880       // remove activation
1881       // get sender esp
1882       __ ldr(esp,
1883           Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1884       // remove frame anchor
1885       __ leave();
1886       // Ensure compiled code always sees stack at proper alignment
1887       __ andr(sp, esp, -16);
1888 
1889       // and begin the OSR nmethod
1890       __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1891       __ br(rscratch1);
1892     }
1893   }
1894 }
1895 
1896 
1897 void TemplateTable::if_0cmp(Condition cc)
1898 {
1899   transition(itos, vtos);
1900   // assume branch is more often taken than not (loops use backward branches)
1901   Label not_taken;
1902   if (cc == equal)
1903     __ cbnzw(r0, not_taken);
1904   else if (cc == not_equal)
1905     __ cbzw(r0, not_taken);
1906   else {
1907     __ andsw(zr, r0, r0);
1908     __ br(j_not(cc), not_taken);
1909   }
1910 
1911   branch(false, false);
1912   __ bind(not_taken);
1913   __ profile_not_taken_branch(r0);
1914 }
1915 
1916 void TemplateTable::if_icmp(Condition cc)
1917 {
1918   transition(itos, vtos);
1919   // assume branch is more often taken than not (loops use backward branches)
1920   Label not_taken;
1921   __ pop_i(r1);
1922   __ cmpw(r1, r0, Assembler::LSL);
1923   __ br(j_not(cc), not_taken);
1924   branch(false, false);
1925   __ bind(not_taken);
1926   __ profile_not_taken_branch(r0);
1927 }
1928 
1929 void TemplateTable::if_nullcmp(Condition cc)
1930 {
1931   transition(atos, vtos);
1932   // assume branch is more often taken than not (loops use backward branches)
1933   Label not_taken;
1934   if (cc == equal)
1935     __ cbnz(r0, not_taken);
1936   else
1937     __ cbz(r0, not_taken);
1938   branch(false, false);
1939   __ bind(not_taken);
1940   __ profile_not_taken_branch(r0);
1941 }
1942 
1943 void TemplateTable::if_acmp(Condition cc)
1944 {
1945   transition(atos, vtos);
1946   // assume branch is more often taken than not (loops use backward branches)
1947   Label not_taken;
1948   __ pop_ptr(r1);
1949   __ cmp(r1, r0);
1950   __ br(j_not(cc), not_taken);
1951   branch(false, false);
1952   __ bind(not_taken);
1953   __ profile_not_taken_branch(r0);
1954 }
1955 
1956 void TemplateTable::ret() {
1957   transition(vtos, vtos);
1958   // We might be moving to a safepoint.  The thread which calls
1959   // Interpreter::notice_safepoints() will effectively flush its cache
1960   // when it makes a system call, but we need to do something to
1961   // ensure that we see the changed dispatch table.
1962   __ membar(MacroAssembler::LoadLoad);
1963 
1964   locals_index(r1);
1965   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1966   __ profile_ret(r1, r2);
1967   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1968   __ lea(rbcp, Address(rbcp, r1));
1969   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1970   __ dispatch_next(vtos);
1971 }
1972 
1973 void TemplateTable::wide_ret() {
1974   transition(vtos, vtos);
1975   locals_index_wide(r1);
1976   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1977   __ profile_ret(r1, r2);
1978   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1979   __ lea(rbcp, Address(rbcp, r1));
1980   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1981   __ dispatch_next(vtos);
1982 }
1983 
1984 
1985 void TemplateTable::tableswitch() {
1986   Label default_case, continue_execution;
1987   transition(itos, vtos);
1988   // align rbcp
1989   __ lea(r1, at_bcp(BytesPerInt));
1990   __ andr(r1, r1, -BytesPerInt);
1991   // load lo & hi
1992   __ ldrw(r2, Address(r1, BytesPerInt));
1993   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
1994   __ rev32(r2, r2);
1995   __ rev32(r3, r3);
1996   // check against lo & hi
1997   __ cmpw(r0, r2);
1998   __ br(Assembler::LT, default_case);
1999   __ cmpw(r0, r3);
2000   __ br(Assembler::GT, default_case);
2001   // lookup dispatch offset
2002   __ subw(r0, r0, r2);
2003   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2004   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2005   __ profile_switch_case(r0, r1, r2);
2006   // continue execution
2007   __ bind(continue_execution);
2008   __ rev32(r3, r3);
2009   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2010   __ add(rbcp, rbcp, r3, ext::sxtw);
2011   __ dispatch_only(vtos);
2012   // handle default
2013   __ bind(default_case);
2014   __ profile_switch_default(r0);
2015   __ ldrw(r3, Address(r1, 0));
2016   __ b(continue_execution);
2017 }
2018 
2019 void TemplateTable::lookupswitch() {
2020   transition(itos, itos);
2021   __ stop("lookupswitch bytecode should have been rewritten");
2022 }
2023 
2024 void TemplateTable::fast_linearswitch() {
2025   transition(itos, vtos);
2026   Label loop_entry, loop, found, continue_execution;
2027   // bswap r0 so we can avoid bswapping the table entries
2028   __ rev32(r0, r0);
2029   // align rbcp
2030   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2031                                     // this instruction (change offsets
2032                                     // below)
2033   __ andr(r19, r19, -BytesPerInt);
2034   // set counter
2035   __ ldrw(r1, Address(r19, BytesPerInt));
2036   __ rev32(r1, r1);
2037   __ b(loop_entry);
2038   // table search
2039   __ bind(loop);
2040   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2041   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2042   __ cmpw(r0, rscratch1);
2043   __ br(Assembler::EQ, found);
2044   __ bind(loop_entry);
2045   __ subs(r1, r1, 1);
2046   __ br(Assembler::PL, loop);
2047   // default case
2048   __ profile_switch_default(r0);
2049   __ ldrw(r3, Address(r19, 0));
2050   __ b(continue_execution);
2051   // entry found -> get offset
2052   __ bind(found);
2053   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2054   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2055   __ profile_switch_case(r1, r0, r19);
2056   // continue execution
2057   __ bind(continue_execution);
2058   __ rev32(r3, r3);
2059   __ add(rbcp, rbcp, r3, ext::sxtw);
2060   __ ldrb(rscratch1, Address(rbcp, 0));
2061   __ dispatch_only(vtos);
2062 }
2063 
2064 void TemplateTable::fast_binaryswitch() {
2065   transition(itos, vtos);
2066   // Implementation using the following core algorithm:
2067   //
2068   // int binary_search(int key, LookupswitchPair* array, int n) {
2069   //   // Binary search according to "Methodik des Programmierens" by
2070   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2071   //   int i = 0;
2072   //   int j = n;
2073   //   while (i+1 < j) {
2074   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2075   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2076   //     // where a stands for the array and assuming that the (inexisting)
2077   //     // element a[n] is infinitely big.
2078   //     int h = (i + j) >> 1;
2079   //     // i < h < j
2080   //     if (key < array[h].fast_match()) {
2081   //       j = h;
2082   //     } else {
2083   //       i = h;
2084   //     }
2085   //   }
2086   //   // R: a[i] <= key < a[i+1] or Q
2087   //   // (i.e., if key is within array, i is the correct index)
2088   //   return i;
2089   // }
2090 
2091   // Register allocation
2092   const Register key   = r0; // already set (tosca)
2093   const Register array = r1;
2094   const Register i     = r2;
2095   const Register j     = r3;
2096   const Register h     = rscratch1;
2097   const Register temp  = rscratch2;
2098 
2099   // Find array start
2100   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2101                                           // get rid of this
2102                                           // instruction (change
2103                                           // offsets below)
2104   __ andr(array, array, -BytesPerInt);
2105 
2106   // Initialize i & j
2107   __ mov(i, 0);                            // i = 0;
2108   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2109 
2110   // Convert j into native byteordering
2111   __ rev32(j, j);
2112 
2113   // And start
2114   Label entry;
2115   __ b(entry);
2116 
2117   // binary search loop
2118   {
2119     Label loop;
2120     __ bind(loop);
2121     // int h = (i + j) >> 1;
2122     __ addw(h, i, j);                           // h = i + j;
2123     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2124     // if (key < array[h].fast_match()) {
2125     //   j = h;
2126     // } else {
2127     //   i = h;
2128     // }
2129     // Convert array[h].match to native byte-ordering before compare
2130     __ ldr(temp, Address(array, h, Address::lsl(3)));
2131     __ rev32(temp, temp);
2132     __ cmpw(key, temp);
2133     // j = h if (key <  array[h].fast_match())
2134     __ csel(j, h, j, Assembler::LT);
2135     // i = h if (key >= array[h].fast_match())
2136     __ csel(i, h, i, Assembler::GE);
2137     // while (i+1 < j)
2138     __ bind(entry);
2139     __ addw(h, i, 1);          // i+1
2140     __ cmpw(h, j);             // i+1 < j
2141     __ br(Assembler::LT, loop);
2142   }
2143 
2144   // end of binary search, result index is i (must check again!)
2145   Label default_case;
2146   // Convert array[i].match to native byte-ordering before compare
2147   __ ldr(temp, Address(array, i, Address::lsl(3)));
2148   __ rev32(temp, temp);
2149   __ cmpw(key, temp);
2150   __ br(Assembler::NE, default_case);
2151 
2152   // entry found -> j = offset
2153   __ add(j, array, i, ext::uxtx, 3);
2154   __ ldrw(j, Address(j, BytesPerInt));
2155   __ profile_switch_case(i, key, array);
2156   __ rev32(j, j);
2157   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2158   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2159   __ dispatch_only(vtos);
2160 
2161   // default case -> j = default offset
2162   __ bind(default_case);
2163   __ profile_switch_default(i);
2164   __ ldrw(j, Address(array, -2 * BytesPerInt));
2165   __ rev32(j, j);
2166   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2167   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2168   __ dispatch_only(vtos);
2169 }
2170 
2171 
2172 void TemplateTable::_return(TosState state)
2173 {
2174   transition(state, state);
2175   assert(_desc->calls_vm(),
2176          "inconsistent calls_vm information"); // call in remove_activation
2177 
2178   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2179     assert(state == vtos, "only valid state");
2180 
2181     __ ldr(c_rarg1, aaddress(0));
2182     __ load_klass(r3, c_rarg1);
2183     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2184     __ tst(r3, JVM_ACC_HAS_FINALIZER);
2185     Label skip_register_finalizer;
2186     __ br(Assembler::EQ, skip_register_finalizer);
2187 
2188     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2189 
2190     __ bind(skip_register_finalizer);
2191   }
2192 
2193   // Issue a StoreStore barrier after all stores but before return
2194   // from any constructor for any class with a final field.  We don't
2195   // know if this is a finalizer, so we always do so.
2196   if (_desc->bytecode() == Bytecodes::_return)
2197     __ membar(MacroAssembler::StoreStore);
2198 
2199   // Narrow result if state is itos but result type is smaller.
2200   // Need to narrow in the return bytecode rather than in generate_return_entry
2201   // since compiled code callers expect the result to already be narrowed.
2202   if (state == itos) {
2203     __ narrow(r0);
2204   }
2205 
2206   __ remove_activation(state);
2207   __ ret(lr);
2208 }
2209 
2210 // ----------------------------------------------------------------------------
2211 // Volatile variables demand their effects be made known to all CPU's
2212 // in order.  Store buffers on most chips allow reads & writes to
2213 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2214 // without some kind of memory barrier (i.e., it's not sufficient that
2215 // the interpreter does not reorder volatile references, the hardware
2216 // also must not reorder them).
2217 //
2218 // According to the new Java Memory Model (JMM):
2219 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2220 //     writes act as aquire & release, so:
2221 // (2) A read cannot let unrelated NON-volatile memory refs that
2222 //     happen after the read float up to before the read.  It's OK for
2223 //     non-volatile memory refs that happen before the volatile read to
2224 //     float down below it.
2225 // (3) Similar a volatile write cannot let unrelated NON-volatile
2226 //     memory refs that happen BEFORE the write float down to after the
2227 //     write.  It's OK for non-volatile memory refs that happen after the
2228 //     volatile write to float up before it.
2229 //
2230 // We only put in barriers around volatile refs (they are expensive),
2231 // not _between_ memory refs (that would require us to track the
2232 // flavor of the previous memory refs).  Requirements (2) and (3)
2233 // require some barriers before volatile stores and after volatile
2234 // loads.  These nearly cover requirement (1) but miss the
2235 // volatile-store-volatile-load case.  This final case is placed after
2236 // volatile-stores although it could just as well go before
2237 // volatile-loads.
2238 
2239 void TemplateTable::resolve_cache_and_index(int byte_no,
2240                                             Register Rcache,
2241                                             Register index,
2242                                             size_t index_size) {
2243   const Register temp = r19;
2244   assert_different_registers(Rcache, index, temp);
2245 
2246   Label resolved;
2247   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2248   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2249   __ cmp(temp, (int) bytecode());  // have we resolved this bytecode?
2250   __ br(Assembler::EQ, resolved);
2251 
2252   // resolve first time through
2253   address entry;
2254   switch (bytecode()) {
2255   case Bytecodes::_getstatic:
2256   case Bytecodes::_putstatic:
2257   case Bytecodes::_getfield:
2258   case Bytecodes::_putfield:
2259     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2260     break;
2261   case Bytecodes::_invokevirtual:
2262   case Bytecodes::_invokespecial:
2263   case Bytecodes::_invokestatic:
2264   case Bytecodes::_invokeinterface:
2265     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2266     break;
2267   case Bytecodes::_invokehandle:
2268     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2269     break;
2270   case Bytecodes::_invokedynamic:
2271     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2272     break;
2273   default:
2274     fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2275     break;
2276   }
2277   __ mov(temp, (int) bytecode());
2278   __ call_VM(noreg, entry, temp);
2279 
2280   // Update registers with resolved info
2281   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2282   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2283   // so all clients ofthis method must be modified accordingly
2284   __ bind(resolved);
2285 }
2286 
2287 // The Rcache and index registers must be set before call
2288 // n.b unlike x86 cache already includes the index offset
2289 void TemplateTable::load_field_cp_cache_entry(Register obj,
2290                                               Register cache,
2291                                               Register index,
2292                                               Register off,
2293                                               Register flags,
2294                                               bool is_static = false) {
2295   assert_different_registers(cache, index, flags, off);
2296 
2297   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2298   // Field offset
2299   __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2300                                           ConstantPoolCacheEntry::f2_offset())));
2301   // Flags
2302   __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2303                                            ConstantPoolCacheEntry::flags_offset())));
2304 
2305   // klass overwrite register
2306   if (is_static) {
2307     __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2308                                         ConstantPoolCacheEntry::f1_offset())));
2309     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2310     __ ldr(obj, Address(obj, mirror_offset));
2311   }
2312 }
2313 
2314 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2315                                                Register method,
2316                                                Register itable_index,
2317                                                Register flags,
2318                                                bool is_invokevirtual,
2319                                                bool is_invokevfinal, /*unused*/
2320                                                bool is_invokedynamic) {
2321   // setup registers
2322   const Register cache = rscratch2;
2323   const Register index = r4;
2324   assert_different_registers(method, flags);
2325   assert_different_registers(method, cache, index);
2326   assert_different_registers(itable_index, flags);
2327   assert_different_registers(itable_index, cache, index);
2328   // determine constant pool cache field offsets
2329   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2330   const int method_offset = in_bytes(
2331     ConstantPoolCache::base_offset() +
2332       (is_invokevirtual
2333        ? ConstantPoolCacheEntry::f2_offset()
2334        : ConstantPoolCacheEntry::f1_offset()));
2335   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2336                                     ConstantPoolCacheEntry::flags_offset());
2337   // access constant pool cache fields
2338   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2339                                     ConstantPoolCacheEntry::f2_offset());
2340 
2341   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2342   resolve_cache_and_index(byte_no, cache, index, index_size);
2343   __ ldr(method, Address(cache, method_offset));
2344 
2345   if (itable_index != noreg) {
2346     __ ldr(itable_index, Address(cache, index_offset));
2347   }
2348   __ ldrw(flags, Address(cache, flags_offset));
2349 }
2350 
2351 
2352 // The registers cache and index expected to be set before call.
2353 // Correct values of the cache and index registers are preserved.
2354 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2355                                             bool is_static, bool has_tos) {
2356   // do the JVMTI work here to avoid disturbing the register state below
2357   // We use c_rarg registers here because we want to use the register used in
2358   // the call to the VM
2359   if (JvmtiExport::can_post_field_access()) {
2360     // Check to see if a field access watch has been set before we
2361     // take the time to call into the VM.
2362     Label L1;
2363     assert_different_registers(cache, index, r0);
2364     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2365     __ ldrw(r0, Address(rscratch1));
2366     __ cbzw(r0, L1);
2367 
2368     __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2369     __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2370 
2371     if (is_static) {
2372       __ mov(c_rarg1, zr); // NULL object reference
2373     } else {
2374       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2375       __ verify_oop(c_rarg1);
2376     }
2377     // c_rarg1: object pointer or NULL
2378     // c_rarg2: cache entry pointer
2379     // c_rarg3: jvalue object on the stack
2380     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2381                                        InterpreterRuntime::post_field_access),
2382                c_rarg1, c_rarg2, c_rarg3);
2383     __ get_cache_and_index_at_bcp(cache, index, 1);
2384     __ bind(L1);
2385   }
2386 }
2387 
2388 void TemplateTable::pop_and_check_object(Register r)
2389 {
2390   __ pop_ptr(r);
2391   __ null_check(r);  // for field access must check obj.
2392   __ verify_oop(r);
2393 }
2394 
2395 void TemplateTable::getfield_or_static(int byte_no, bool is_static)
2396 {
2397   const Register cache = r2;
2398   const Register index = r3;
2399   const Register obj   = r4;
2400   const Register off   = r19;
2401   const Register flags = r0;
2402   const Register raw_flags = r6;
2403   const Register bc    = r4; // uses same reg as obj, so don't mix them
2404 
2405   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2406   jvmti_post_field_access(cache, index, is_static, false);
2407   load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2408 
2409   if (!is_static) {
2410     // obj is on the stack
2411     pop_and_check_object(obj);
2412   }
2413 
2414   // 8179954: We need to make sure that the code generated for
2415   // volatile accesses forms a sequentially-consistent set of
2416   // operations when combined with STLR and LDAR.  Without a leading
2417   // membar it's possible for a simple Dekker test to fail if loads
2418   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2419   // the stores in one method and we interpret the loads in another.
2420   if (! UseBarriersForVolatile) {
2421     Label notVolatile;
2422     __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2423     __ membar(MacroAssembler::AnyAny);
2424     __ bind(notVolatile);
2425   }
2426 
2427   const Address field(obj, off);
2428 
2429   Label Done, notByte, notBool, notInt, notShort, notChar,
2430               notLong, notFloat, notObj, notDouble;
2431 
2432   // x86 uses a shift and mask or wings it with a shift plus assert
2433   // the mask is not needed. aarch64 just uses bitfield extract
2434   __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2435            ConstantPoolCacheEntry::tos_state_bits);
2436 
2437   assert(btos == 0, "change code, btos != 0");
2438   __ cbnz(flags, notByte);
2439 
2440   // btos
2441   __ load_signed_byte(r0, field);
2442   __ push(btos);
2443   // Rewrite bytecode to be faster
2444   if (!is_static) {
2445     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2446   }
2447   __ b(Done);
2448 
2449   __ bind(notByte);
2450   __ cmp(flags, ztos);
2451   __ br(Assembler::NE, notBool);
2452 
2453   // ztos (same code as btos)
2454   __ ldrsb(r0, field);
2455   __ push(ztos);
2456   // Rewrite bytecode to be faster
2457   if (!is_static) {
2458     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2459     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2460   }
2461   __ b(Done);
2462 
2463   __ bind(notBool);
2464   __ cmp(flags, atos);
2465   __ br(Assembler::NE, notObj);
2466   // atos
2467   __ load_heap_oop(r0, field);
2468   __ push(atos);
2469   if (!is_static) {
2470     patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2471   }
2472   __ b(Done);
2473 
2474   __ bind(notObj);
2475   __ cmp(flags, itos);
2476   __ br(Assembler::NE, notInt);
2477   // itos
2478   __ ldrw(r0, field);
2479   __ push(itos);
2480   // Rewrite bytecode to be faster
2481   if (!is_static) {
2482     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2483   }
2484   __ b(Done);
2485 
2486   __ bind(notInt);
2487   __ cmp(flags, ctos);
2488   __ br(Assembler::NE, notChar);
2489   // ctos
2490   __ load_unsigned_short(r0, field);
2491   __ push(ctos);
2492   // Rewrite bytecode to be faster
2493   if (!is_static) {
2494     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2495   }
2496   __ b(Done);
2497 
2498   __ bind(notChar);
2499   __ cmp(flags, stos);
2500   __ br(Assembler::NE, notShort);
2501   // stos
2502   __ load_signed_short(r0, field);
2503   __ push(stos);
2504   // Rewrite bytecode to be faster
2505   if (!is_static) {
2506     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2507   }
2508   __ b(Done);
2509 
2510   __ bind(notShort);
2511   __ cmp(flags, ltos);
2512   __ br(Assembler::NE, notLong);
2513   // ltos
2514   __ ldr(r0, field);
2515   __ push(ltos);
2516   // Rewrite bytecode to be faster
2517   if (!is_static) {
2518     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2519   }
2520   __ b(Done);
2521 
2522   __ bind(notLong);
2523   __ cmp(flags, ftos);
2524   __ br(Assembler::NE, notFloat);
2525   // ftos
2526   __ ldrs(v0, field);
2527   __ push(ftos);
2528   // Rewrite bytecode to be faster
2529   if (!is_static) {
2530     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2531   }
2532   __ b(Done);
2533 
2534   __ bind(notFloat);
2535 #ifdef ASSERT
2536   __ cmp(flags, dtos);
2537   __ br(Assembler::NE, notDouble);
2538 #endif
2539   // dtos
2540   __ ldrd(v0, field);
2541   __ push(dtos);
2542   // Rewrite bytecode to be faster
2543   if (!is_static) {
2544     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2545   }
2546 #ifdef ASSERT
2547   __ b(Done);
2548 
2549   __ bind(notDouble);
2550   __ stop("Bad state");
2551 #endif
2552 
2553   __ bind(Done);
2554 
2555   Label notVolatile;
2556   __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2557   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2558   __ bind(notVolatile);
2559 }
2560 
2561 
2562 void TemplateTable::getfield(int byte_no)
2563 {
2564   getfield_or_static(byte_no, false);
2565 }
2566 
2567 void TemplateTable::getstatic(int byte_no)
2568 {
2569   getfield_or_static(byte_no, true);
2570 }
2571 
2572 // The registers cache and index expected to be set before call.
2573 // The function may destroy various registers, just not the cache and index registers.
2574 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2575   transition(vtos, vtos);
2576 
2577   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2578 
2579   if (JvmtiExport::can_post_field_modification()) {
2580     // Check to see if a field modification watch has been set before
2581     // we take the time to call into the VM.
2582     Label L1;
2583     assert_different_registers(cache, index, r0);
2584     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2585     __ ldrw(r0, Address(rscratch1));
2586     __ cbz(r0, L1);
2587 
2588     __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2589 
2590     if (is_static) {
2591       // Life is simple.  Null out the object pointer.
2592       __ mov(c_rarg1, zr);
2593     } else {
2594       // Life is harder. The stack holds the value on top, followed by
2595       // the object.  We don't know the size of the value, though; it
2596       // could be one or two words depending on its type. As a result,
2597       // we must find the type to determine where the object is.
2598       __ ldrw(c_rarg3, Address(c_rarg2,
2599                                in_bytes(cp_base_offset +
2600                                         ConstantPoolCacheEntry::flags_offset())));
2601       __ lsr(c_rarg3, c_rarg3,
2602              ConstantPoolCacheEntry::tos_state_shift);
2603       ConstantPoolCacheEntry::verify_tos_state_shift();
2604       Label nope2, done, ok;
2605       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2606       __ cmpw(c_rarg3, ltos);
2607       __ br(Assembler::EQ, ok);
2608       __ cmpw(c_rarg3, dtos);
2609       __ br(Assembler::NE, nope2);
2610       __ bind(ok);
2611       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2612       __ bind(nope2);
2613     }
2614     // cache entry pointer
2615     __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2616     // object (tos)
2617     __ mov(c_rarg3, esp);
2618     // c_rarg1: object pointer set up above (NULL if static)
2619     // c_rarg2: cache entry pointer
2620     // c_rarg3: jvalue object on the stack
2621     __ call_VM(noreg,
2622                CAST_FROM_FN_PTR(address,
2623                                 InterpreterRuntime::post_field_modification),
2624                c_rarg1, c_rarg2, c_rarg3);
2625     __ get_cache_and_index_at_bcp(cache, index, 1);
2626     __ bind(L1);
2627   }
2628 }
2629 
2630 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2631   transition(vtos, vtos);
2632 
2633   const Register cache = r2;
2634   const Register index = r3;
2635   const Register obj   = r2;
2636   const Register off   = r19;
2637   const Register flags = r0;
2638   const Register bc    = r4;
2639 
2640   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2641   jvmti_post_field_mod(cache, index, is_static);
2642   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2643 
2644   Label Done;
2645   __ mov(r5, flags);
2646 
2647   {
2648     Label notVolatile;
2649     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2650     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2651     __ bind(notVolatile);
2652   }
2653 
2654   // field address
2655   const Address field(obj, off);
2656 
2657   Label notByte, notBool, notInt, notShort, notChar,
2658         notLong, notFloat, notObj, notDouble;
2659 
2660   // x86 uses a shift and mask or wings it with a shift plus assert
2661   // the mask is not needed. aarch64 just uses bitfield extract
2662   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2663 
2664   assert(btos == 0, "change code, btos != 0");
2665   __ cbnz(flags, notByte);
2666 
2667   // btos
2668   {
2669     __ pop(btos);
2670     if (!is_static) pop_and_check_object(obj);
2671     __ strb(r0, field);
2672     if (!is_static) {
2673       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2674     }
2675     __ b(Done);
2676   }
2677 
2678   __ bind(notByte);
2679   __ cmp(flags, ztos);
2680   __ br(Assembler::NE, notBool);
2681 
2682   // ztos
2683   {
2684     __ pop(ztos);
2685     if (!is_static) pop_and_check_object(obj);
2686     __ andw(r0, r0, 0x1);
2687     __ strb(r0, field);
2688     if (!is_static) {
2689       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2690     }
2691     __ b(Done);
2692   }
2693 
2694   __ bind(notBool);
2695   __ cmp(flags, atos);
2696   __ br(Assembler::NE, notObj);
2697 
2698   // atos
2699   {
2700     __ pop(atos);
2701     if (!is_static) pop_and_check_object(obj);
2702     // Store into the field
2703     do_oop_store(_masm, field, r0, _bs->kind(), false);
2704     if (!is_static) {
2705       patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2706     }
2707     __ b(Done);
2708   }
2709 
2710   __ bind(notObj);
2711   __ cmp(flags, itos);
2712   __ br(Assembler::NE, notInt);
2713 
2714   // itos
2715   {
2716     __ pop(itos);
2717     if (!is_static) pop_and_check_object(obj);
2718     __ strw(r0, field);
2719     if (!is_static) {
2720       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2721     }
2722     __ b(Done);
2723   }
2724 
2725   __ bind(notInt);
2726   __ cmp(flags, ctos);
2727   __ br(Assembler::NE, notChar);
2728 
2729   // ctos
2730   {
2731     __ pop(ctos);
2732     if (!is_static) pop_and_check_object(obj);
2733     __ strh(r0, field);
2734     if (!is_static) {
2735       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2736     }
2737     __ b(Done);
2738   }
2739 
2740   __ bind(notChar);
2741   __ cmp(flags, stos);
2742   __ br(Assembler::NE, notShort);
2743 
2744   // stos
2745   {
2746     __ pop(stos);
2747     if (!is_static) pop_and_check_object(obj);
2748     __ strh(r0, field);
2749     if (!is_static) {
2750       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2751     }
2752     __ b(Done);
2753   }
2754 
2755   __ bind(notShort);
2756   __ cmp(flags, ltos);
2757   __ br(Assembler::NE, notLong);
2758 
2759   // ltos
2760   {
2761     __ pop(ltos);
2762     if (!is_static) pop_and_check_object(obj);
2763     __ str(r0, field);
2764     if (!is_static) {
2765       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2766     }
2767     __ b(Done);
2768   }
2769 
2770   __ bind(notLong);
2771   __ cmp(flags, ftos);
2772   __ br(Assembler::NE, notFloat);
2773 
2774   // ftos
2775   {
2776     __ pop(ftos);
2777     if (!is_static) pop_and_check_object(obj);
2778     __ strs(v0, field);
2779     if (!is_static) {
2780       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2781     }
2782     __ b(Done);
2783   }
2784 
2785   __ bind(notFloat);
2786 #ifdef ASSERT
2787   __ cmp(flags, dtos);
2788   __ br(Assembler::NE, notDouble);
2789 #endif
2790 
2791   // dtos
2792   {
2793     __ pop(dtos);
2794     if (!is_static) pop_and_check_object(obj);
2795     __ strd(v0, field);
2796     if (!is_static) {
2797       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2798     }
2799   }
2800 
2801 #ifdef ASSERT
2802   __ b(Done);
2803 
2804   __ bind(notDouble);
2805   __ stop("Bad state");
2806 #endif
2807 
2808   __ bind(Done);
2809 
2810   {
2811     Label notVolatile;
2812     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2813     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2814     __ bind(notVolatile);
2815   }
2816 }
2817 
2818 void TemplateTable::putfield(int byte_no)
2819 {
2820   putfield_or_static(byte_no, false);
2821 }
2822 
2823 void TemplateTable::putstatic(int byte_no) {
2824   putfield_or_static(byte_no, true);
2825 }
2826 
2827 void TemplateTable::jvmti_post_fast_field_mod()
2828 {
2829   if (JvmtiExport::can_post_field_modification()) {
2830     // Check to see if a field modification watch has been set before
2831     // we take the time to call into the VM.
2832     Label L2;
2833     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2834     __ ldrw(c_rarg3, Address(rscratch1));
2835     __ cbzw(c_rarg3, L2);
2836     __ pop_ptr(r19);                  // copy the object pointer from tos
2837     __ verify_oop(r19);
2838     __ push_ptr(r19);                 // put the object pointer back on tos
2839     // Save tos values before call_VM() clobbers them. Since we have
2840     // to do it for every data type, we use the saved values as the
2841     // jvalue object.
2842     switch (bytecode()) {          // load values into the jvalue object
2843     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2844     case Bytecodes::_fast_bputfield: // fall through
2845     case Bytecodes::_fast_zputfield: // fall through
2846     case Bytecodes::_fast_sputfield: // fall through
2847     case Bytecodes::_fast_cputfield: // fall through
2848     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2849     case Bytecodes::_fast_dputfield: __ push_d(); break;
2850     case Bytecodes::_fast_fputfield: __ push_f(); break;
2851     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2852 
2853     default:
2854       ShouldNotReachHere();
2855     }
2856     __ mov(c_rarg3, esp);             // points to jvalue on the stack
2857     // access constant pool cache entry
2858     __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2859     __ verify_oop(r19);
2860     // r19: object pointer copied above
2861     // c_rarg2: cache entry pointer
2862     // c_rarg3: jvalue object on the stack
2863     __ call_VM(noreg,
2864                CAST_FROM_FN_PTR(address,
2865                                 InterpreterRuntime::post_field_modification),
2866                r19, c_rarg2, c_rarg3);
2867 
2868     switch (bytecode()) {             // restore tos values
2869     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2870     case Bytecodes::_fast_bputfield: // fall through
2871     case Bytecodes::_fast_zputfield: // fall through
2872     case Bytecodes::_fast_sputfield: // fall through
2873     case Bytecodes::_fast_cputfield: // fall through
2874     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2875     case Bytecodes::_fast_dputfield: __ pop_d(); break;
2876     case Bytecodes::_fast_fputfield: __ pop_f(); break;
2877     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2878     }
2879     __ bind(L2);
2880   }
2881 }
2882 
2883 void TemplateTable::fast_storefield(TosState state)
2884 {
2885   transition(state, vtos);
2886 
2887   ByteSize base = ConstantPoolCache::base_offset();
2888 
2889   jvmti_post_fast_field_mod();
2890 
2891   // access constant pool cache
2892   __ get_cache_and_index_at_bcp(r2, r1, 1);
2893 
2894   // test for volatile with r3
2895   __ ldrw(r3, Address(r2, in_bytes(base +
2896                                    ConstantPoolCacheEntry::flags_offset())));
2897 
2898   // replace index with field offset from cache entry
2899   __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2900 
2901   {
2902     Label notVolatile;
2903     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2904     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2905     __ bind(notVolatile);
2906   }
2907 
2908   Label notVolatile;
2909 
2910   // Get object from stack
2911   pop_and_check_object(r2);
2912 
2913   // field address
2914   const Address field(r2, r1);
2915 
2916   // access field
2917   switch (bytecode()) {
2918   case Bytecodes::_fast_aputfield:
2919     do_oop_store(_masm, field, r0, _bs->kind(), false);
2920     break;
2921   case Bytecodes::_fast_lputfield:
2922     __ str(r0, field);
2923     break;
2924   case Bytecodes::_fast_iputfield:
2925     __ strw(r0, field);
2926     break;
2927   case Bytecodes::_fast_zputfield:
2928     __ andw(r0, r0, 0x1);  // boolean is true if LSB is 1
2929     // fall through to bputfield
2930   case Bytecodes::_fast_bputfield:
2931     __ strb(r0, field);
2932     break;
2933   case Bytecodes::_fast_sputfield:
2934     // fall through
2935   case Bytecodes::_fast_cputfield:
2936     __ strh(r0, field);
2937     break;
2938   case Bytecodes::_fast_fputfield:
2939     __ strs(v0, field);
2940     break;
2941   case Bytecodes::_fast_dputfield:
2942     __ strd(v0, field);
2943     break;
2944   default:
2945     ShouldNotReachHere();
2946   }
2947 
2948   {
2949     Label notVolatile;
2950     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2951     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2952     __ bind(notVolatile);
2953   }
2954 }
2955 
2956 
2957 void TemplateTable::fast_accessfield(TosState state)
2958 {
2959   transition(atos, state);
2960   // Do the JVMTI work here to avoid disturbing the register state below
2961   if (JvmtiExport::can_post_field_access()) {
2962     // Check to see if a field access watch has been set before we
2963     // take the time to call into the VM.
2964     Label L1;
2965     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2966     __ ldrw(r2, Address(rscratch1));
2967     __ cbzw(r2, L1);
2968     // access constant pool cache entry
2969     __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2970     __ verify_oop(r0);
2971     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
2972     __ mov(c_rarg1, r0);
2973     // c_rarg1: object pointer copied above
2974     // c_rarg2: cache entry pointer
2975     __ call_VM(noreg,
2976                CAST_FROM_FN_PTR(address,
2977                                 InterpreterRuntime::post_field_access),
2978                c_rarg1, c_rarg2);
2979     __ pop_ptr(r0); // restore object pointer
2980     __ bind(L1);
2981   }
2982 
2983   // access constant pool cache
2984   __ get_cache_and_index_at_bcp(r2, r1, 1);
2985   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2986                                   ConstantPoolCacheEntry::f2_offset())));
2987   __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2988                                    ConstantPoolCacheEntry::flags_offset())));
2989 
2990   // r0: object
2991   __ verify_oop(r0);
2992   __ null_check(r0);
2993   const Address field(r0, r1);
2994 
2995   // 8179954: We need to make sure that the code generated for
2996   // volatile accesses forms a sequentially-consistent set of
2997   // operations when combined with STLR and LDAR.  Without a leading
2998   // membar it's possible for a simple Dekker test to fail if loads
2999   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3000   // the stores in one method and we interpret the loads in another.
3001   if (! UseBarriersForVolatile) {
3002     Label notVolatile;
3003     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3004     __ membar(MacroAssembler::AnyAny);
3005     __ bind(notVolatile);
3006   }
3007 
3008   // access field
3009   switch (bytecode()) {
3010   case Bytecodes::_fast_agetfield:
3011     __ load_heap_oop(r0, field);
3012     __ verify_oop(r0);
3013     break;
3014   case Bytecodes::_fast_lgetfield:
3015     __ ldr(r0, field);
3016     break;
3017   case Bytecodes::_fast_igetfield:
3018     __ ldrw(r0, field);
3019     break;
3020   case Bytecodes::_fast_bgetfield:
3021     __ load_signed_byte(r0, field);
3022     break;
3023   case Bytecodes::_fast_sgetfield:
3024     __ load_signed_short(r0, field);
3025     break;
3026   case Bytecodes::_fast_cgetfield:
3027     __ load_unsigned_short(r0, field);
3028     break;
3029   case Bytecodes::_fast_fgetfield:
3030     __ ldrs(v0, field);
3031     break;
3032   case Bytecodes::_fast_dgetfield:
3033     __ ldrd(v0, field);
3034     break;
3035   default:
3036     ShouldNotReachHere();
3037   }
3038   {
3039     Label notVolatile;
3040     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3041     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3042     __ bind(notVolatile);
3043   }
3044 }
3045 
3046 void TemplateTable::fast_xaccess(TosState state)
3047 {
3048   transition(vtos, state);
3049 
3050   // get receiver
3051   __ ldr(r0, aaddress(0));
3052   // access constant pool cache
3053   __ get_cache_and_index_at_bcp(r2, r3, 2);
3054   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3055                                   ConstantPoolCacheEntry::f2_offset())));
3056 
3057   // 8179954: We need to make sure that the code generated for
3058   // volatile accesses forms a sequentially-consistent set of
3059   // operations when combined with STLR and LDAR.  Without a leading
3060   // membar it's possible for a simple Dekker test to fail if loads
3061   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3062   // the stores in one method and we interpret the loads in another.
3063   if (! UseBarriersForVolatile) {
3064     Label notVolatile;
3065     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3066                                      ConstantPoolCacheEntry::flags_offset())));
3067     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3068     __ membar(MacroAssembler::AnyAny);
3069     __ bind(notVolatile);
3070   }
3071 
3072   // make sure exception is reported in correct bcp range (getfield is
3073   // next instruction)
3074   __ increment(rbcp);
3075   __ null_check(r0);
3076   switch (state) {
3077   case itos:
3078     __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3079     break;
3080   case atos:
3081     __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0)));
3082     __ verify_oop(r0);
3083     break;
3084   case ftos:
3085     __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3086     break;
3087   default:
3088     ShouldNotReachHere();
3089   }
3090 
3091   {
3092     Label notVolatile;
3093     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3094                                      ConstantPoolCacheEntry::flags_offset())));
3095     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3096     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3097     __ bind(notVolatile);
3098   }
3099 
3100   __ decrement(rbcp);
3101 }
3102 
3103 
3104 
3105 //-----------------------------------------------------------------------------
3106 // Calls
3107 
3108 void TemplateTable::count_calls(Register method, Register temp)
3109 {
3110   __ call_Unimplemented();
3111 }
3112 
3113 void TemplateTable::prepare_invoke(int byte_no,
3114                                    Register method, // linked method (or i-klass)
3115                                    Register index,  // itable index, MethodType, etc.
3116                                    Register recv,   // if caller wants to see it
3117                                    Register flags   // if caller wants to test it
3118                                    ) {
3119   // determine flags
3120   Bytecodes::Code code = bytecode();
3121   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3122   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3123   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3124   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3125   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3126   const bool load_receiver       = (recv  != noreg);
3127   const bool save_flags          = (flags != noreg);
3128   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3129   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3130   assert(flags == noreg || flags == r3, "");
3131   assert(recv  == noreg || recv  == r2, "");
3132 
3133   // setup registers & access constant pool cache
3134   if (recv  == noreg)  recv  = r2;
3135   if (flags == noreg)  flags = r3;
3136   assert_different_registers(method, index, recv, flags);
3137 
3138   // save 'interpreter return address'
3139   __ save_bcp();
3140 
3141   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3142 
3143   // maybe push appendix to arguments (just before return address)
3144   if (is_invokedynamic || is_invokehandle) {
3145     Label L_no_push;
3146     __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3147     // Push the appendix as a trailing parameter.
3148     // This must be done before we get the receiver,
3149     // since the parameter_size includes it.
3150     __ push(r19);
3151     __ mov(r19, index);
3152     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3153     __ load_resolved_reference_at_index(index, r19);
3154     __ pop(r19);
3155     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3156     __ bind(L_no_push);
3157   }
3158 
3159   // load receiver if needed (note: no return address pushed yet)
3160   if (load_receiver) {
3161     __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3162     // FIXME -- is this actually correct? looks like it should be 2
3163     // const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3164     // const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3165     // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3166     // __ movptr(recv, recv_addr);
3167     __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3168     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3169     __ verify_oop(recv);
3170   }
3171 
3172   // compute return type
3173   // x86 uses a shift and mask or wings it with a shift plus assert
3174   // the mask is not needed. aarch64 just uses bitfield extract
3175   __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
3176   // load return address
3177   {
3178     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3179     __ mov(rscratch1, table_addr);
3180     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3181   }
3182 }
3183 
3184 
3185 void TemplateTable::invokevirtual_helper(Register index,
3186                                          Register recv,
3187                                          Register flags)
3188 {
3189   // Uses temporary registers r0, r3
3190   assert_different_registers(index, recv, r0, r3);
3191   // Test for an invoke of a final method
3192   Label notFinal;
3193   __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3194 
3195   const Register method = index;  // method must be rmethod
3196   assert(method == rmethod,
3197          "methodOop must be rmethod for interpreter calling convention");
3198 
3199   // do the call - the index is actually the method to call
3200   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3201 
3202   // It's final, need a null check here!
3203   __ null_check(recv);
3204 
3205   // profile this call
3206   __ profile_final_call(r0);
3207   __ profile_arguments_type(r0, method, r4, true);
3208 
3209   __ jump_from_interpreted(method, r0);
3210 
3211   __ bind(notFinal);
3212 
3213   // get receiver klass
3214   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3215   __ load_klass(r0, recv);
3216 
3217   // profile this call
3218   __ profile_virtual_call(r0, rlocals, r3);
3219 
3220   // get target methodOop & entry point
3221   __ lookup_virtual_method(r0, index, method);
3222   __ profile_arguments_type(r3, method, r4, true);
3223   // FIXME -- this looks completely redundant. is it?
3224   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3225   __ jump_from_interpreted(method, r3);
3226 }
3227 
3228 void TemplateTable::invokevirtual(int byte_no)
3229 {
3230   transition(vtos, vtos);
3231   assert(byte_no == f2_byte, "use this argument");
3232 
3233   prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3234 
3235   // rmethod: index (actually a Method*)
3236   // r2: receiver
3237   // r3: flags
3238 
3239   invokevirtual_helper(rmethod, r2, r3);
3240 }
3241 
3242 void TemplateTable::invokespecial(int byte_no)
3243 {
3244   transition(vtos, vtos);
3245   assert(byte_no == f1_byte, "use this argument");
3246 
3247   prepare_invoke(byte_no, rmethod, noreg,  // get f1 Method*
3248                  r2);  // get receiver also for null check
3249   __ verify_oop(r2);
3250   __ null_check(r2);
3251   // do the call
3252   __ profile_call(r0);
3253   __ profile_arguments_type(r0, rmethod, rbcp, false);
3254   __ jump_from_interpreted(rmethod, r0);
3255 }
3256 
3257 void TemplateTable::invokestatic(int byte_no)
3258 {
3259   transition(vtos, vtos);
3260   assert(byte_no == f1_byte, "use this argument");
3261 
3262   prepare_invoke(byte_no, rmethod);  // get f1 Method*
3263   // do the call
3264   __ profile_call(r0);
3265   __ profile_arguments_type(r0, rmethod, r4, false);
3266   __ jump_from_interpreted(rmethod, r0);
3267 }
3268 
3269 void TemplateTable::fast_invokevfinal(int byte_no)
3270 {
3271   __ call_Unimplemented();
3272 }
3273 
3274 void TemplateTable::invokeinterface(int byte_no) {
3275   transition(vtos, vtos);
3276   assert(byte_no == f1_byte, "use this argument");
3277 
3278   prepare_invoke(byte_no, r0, rmethod,  // get f1 Klass*, f2 Method*
3279                  r2, r3); // recv, flags
3280 
3281   // r0: interface klass (from f1)
3282   // rmethod: method (from f2)
3283   // r2: receiver
3284   // r3: flags
3285 
3286   // Special case of invokeinterface called for virtual method of
3287   // java.lang.Object.  See cpCacheOop.cpp for details.
3288   // This code isn't produced by javac, but could be produced by
3289   // another compliant java compiler.
3290   Label notMethod;
3291   __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3292 
3293   invokevirtual_helper(rmethod, r2, r3);
3294   __ bind(notMethod);
3295 
3296   // Get receiver klass into r3 - also a null check
3297   __ restore_locals();
3298   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3299   __ load_klass(r3, r2);
3300 
3301   Label no_such_interface, no_such_method;
3302 
3303   // Receiver subtype check against REFC.
3304   // Superklass in r0. Subklass in r3. Blows rscratch2, r13.
3305   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3306                              r3, r0, noreg,
3307                              // outputs: scan temp. reg, scan temp. reg
3308                              rscratch2, r13,
3309                              no_such_interface,
3310                              /*return_method=*/false);
3311 
3312   // profile this call
3313   __ profile_virtual_call(r3, r13, r19);
3314 
3315   // Get declaring interface class from method, and itable index
3316   __ ldr(r0, Address(rmethod, Method::const_offset()));
3317   __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
3318   __ ldr(r0, Address(r0, ConstantPool::pool_holder_offset_in_bytes()));
3319   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3320   __ subw(rmethod, rmethod, Method::itable_index_max);
3321   __ negw(rmethod, rmethod);
3322 
3323   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3324                              r3, r0, rmethod,
3325                              // outputs: method, scan temp. reg
3326                              rmethod, r13,
3327                              no_such_interface);
3328 
3329   // rmethod,: methodOop to call
3330   // r2: receiver
3331   // Check for abstract method error
3332   // Note: This should be done more efficiently via a throw_abstract_method_error
3333   //       interpreter entry point and a conditional jump to it in case of a null
3334   //       method.
3335   __ cbz(rmethod, no_such_method);
3336 
3337   __ profile_arguments_type(r3, rmethod, r13, true);
3338 
3339   // do the call
3340   // r2: receiver
3341   // rmethod,: methodOop
3342   __ jump_from_interpreted(rmethod, r3);
3343   __ should_not_reach_here();
3344 
3345   // exception handling code follows...
3346   // note: must restore interpreter registers to canonical
3347   //       state for exception handling to work correctly!
3348 
3349   __ bind(no_such_method);
3350   // throw exception
3351   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3352   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3353   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3354   // the call_VM checks for exception, so we should never return here.
3355   __ should_not_reach_here();
3356 
3357   __ bind(no_such_interface);
3358   // throw exception
3359   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3360   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3361   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3362                    InterpreterRuntime::throw_IncompatibleClassChangeError));
3363   // the call_VM checks for exception, so we should never return here.
3364   __ should_not_reach_here();
3365   return;
3366 }
3367 
3368 void TemplateTable::invokehandle(int byte_no) {
3369   transition(vtos, vtos);
3370   assert(byte_no == f1_byte, "use this argument");
3371 
3372   if (!EnableInvokeDynamic) {
3373     // rewriter does not generate this bytecode
3374     __ should_not_reach_here();
3375     return;
3376   }
3377 
3378   prepare_invoke(byte_no, rmethod, r0, r2);
3379   __ verify_method_ptr(r2);
3380   __ verify_oop(r2);
3381   __ null_check(r2);
3382 
3383   // FIXME: profile the LambdaForm also
3384 
3385   // r13 is safe to use here as a scratch reg because it is about to
3386   // be clobbered by jump_from_interpreted().
3387   __ profile_final_call(r13);
3388   __ profile_arguments_type(r13, rmethod, r4, true);
3389 
3390   __ jump_from_interpreted(rmethod, r0);
3391 }
3392 
3393 void TemplateTable::invokedynamic(int byte_no) {
3394   transition(vtos, vtos);
3395   assert(byte_no == f1_byte, "use this argument");
3396 
3397   if (!EnableInvokeDynamic) {
3398     // We should not encounter this bytecode if !EnableInvokeDynamic.
3399     // The verifier will stop it.  However, if we get past the verifier,
3400     // this will stop the thread in a reasonable way, without crashing the JVM.
3401     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3402                      InterpreterRuntime::throw_IncompatibleClassChangeError));
3403     // the call_VM checks for exception, so we should never return here.
3404     __ should_not_reach_here();
3405     return;
3406   }
3407 
3408   prepare_invoke(byte_no, rmethod, r0);
3409 
3410   // r0: CallSite object (from cpool->resolved_references[])
3411   // rmethod: MH.linkToCallSite method (from f2)
3412 
3413   // Note:  r0_callsite is already pushed by prepare_invoke
3414 
3415   // %%% should make a type profile for any invokedynamic that takes a ref argument
3416   // profile this call
3417   __ profile_call(rbcp);
3418   __ profile_arguments_type(r3, rmethod, r13, false);
3419 
3420   __ verify_oop(r0);
3421 
3422   __ jump_from_interpreted(rmethod, r0);
3423 }
3424 
3425 
3426 //-----------------------------------------------------------------------------
3427 // Allocation
3428 
3429 void TemplateTable::_new() {
3430   transition(vtos, atos);
3431 
3432   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3433   Label slow_case;
3434   Label done;
3435   Label initialize_header;
3436   Label initialize_object; // including clearing the fields
3437   Label allocate_shared;
3438 
3439   __ get_cpool_and_tags(r4, r0);
3440   // Make sure the class we're about to instantiate has been resolved.
3441   // This is done before loading InstanceKlass to be consistent with the order
3442   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3443   const int tags_offset = Array<u1>::base_offset_in_bytes();
3444   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3445   __ lea(rscratch1, Address(rscratch1, tags_offset));
3446   __ ldarb(rscratch1, rscratch1);
3447   __ cmp(rscratch1, JVM_CONSTANT_Class);
3448   __ br(Assembler::NE, slow_case);
3449 
3450   // get InstanceKlass
3451   __ lea(r4, Address(r4, r3, Address::lsl(3)));
3452   __ ldr(r4, Address(r4, sizeof(ConstantPool)));
3453 
3454   // make sure klass is initialized & doesn't have finalizer
3455   // make sure klass is fully initialized
3456   __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3457   __ cmp(rscratch1, InstanceKlass::fully_initialized);
3458   __ br(Assembler::NE, slow_case);
3459 
3460   // get instance_size in InstanceKlass (scaled to a count of bytes)
3461   __ ldrw(r3,
3462           Address(r4,
3463                   Klass::layout_helper_offset()));
3464   // test to see if it has a finalizer or is malformed in some way
3465   __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3466 
3467   // Allocate the instance
3468   // 1) Try to allocate in the TLAB
3469   // 2) if fail and the object is large allocate in the shared Eden
3470   // 3) if the above fails (or is not applicable), go to a slow case
3471   // (creates a new TLAB, etc.)
3472 
3473   const bool allow_shared_alloc =
3474     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3475 
3476   if (UseTLAB) {
3477     __ tlab_allocate(r0, r3, 0, noreg, r1,
3478                      allow_shared_alloc ? allocate_shared : slow_case);
3479 
3480     if (ZeroTLAB) {
3481       // the fields have been already cleared
3482       __ b(initialize_header);
3483     } else {
3484       // initialize both the header and fields
3485       __ b(initialize_object);
3486     }
3487   }
3488 
3489   // Allocation in the shared Eden, if allowed.
3490   //
3491   // r3: instance size in bytes
3492   if (allow_shared_alloc) {
3493     __ bind(allocate_shared);
3494 
3495     __ eden_allocate(r0, r3, 0, r10, slow_case);
3496     __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3497   }
3498 
3499   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3500     // The object is initialized before the header.  If the object size is
3501     // zero, go directly to the header initialization.
3502     __ bind(initialize_object);
3503     __ sub(r3, r3, sizeof(oopDesc));
3504     __ cbz(r3, initialize_header);
3505 
3506     // Initialize object fields
3507     {
3508       __ add(r2, r0, sizeof(oopDesc));
3509       Label loop;
3510       __ bind(loop);
3511       __ str(zr, Address(__ post(r2, BytesPerLong)));
3512       __ sub(r3, r3, BytesPerLong);
3513       __ cbnz(r3, loop);
3514     }
3515 
3516     // initialize object header only.
3517     __ bind(initialize_header);
3518     if (UseBiasedLocking) {
3519       __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3520     } else {
3521       __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3522     }
3523     __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3524     __ store_klass_gap(r0, zr);  // zero klass gap for compressed oops
3525     __ store_klass(r0, r4);      // store klass last
3526 
3527     {
3528       SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3529       // Trigger dtrace event for fastpath
3530       __ push(atos); // save the return value
3531       __ call_VM_leaf(
3532            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3533       __ pop(atos); // restore the return value
3534 
3535     }
3536     __ b(done);
3537   }
3538 
3539   // slow case
3540   __ bind(slow_case);
3541   __ get_constant_pool(c_rarg1);
3542   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3543   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3544   __ verify_oop(r0);
3545 
3546   // continue
3547   __ bind(done);
3548   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3549   __ membar(Assembler::StoreStore);
3550 }
3551 
3552 void TemplateTable::newarray() {
3553   transition(itos, atos);
3554   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3555   __ mov(c_rarg2, r0);
3556   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3557           c_rarg1, c_rarg2);
3558   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3559   __ membar(Assembler::StoreStore);
3560 }
3561 
3562 void TemplateTable::anewarray() {
3563   transition(itos, atos);
3564   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3565   __ get_constant_pool(c_rarg1);
3566   __ mov(c_rarg3, r0);
3567   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3568           c_rarg1, c_rarg2, c_rarg3);
3569   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3570   __ membar(Assembler::StoreStore);
3571 }
3572 
3573 void TemplateTable::arraylength() {
3574   transition(atos, itos);
3575   __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3576   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3577 }
3578 
3579 void TemplateTable::checkcast()
3580 {
3581   transition(atos, atos);
3582   Label done, is_null, ok_is_subtype, quicked, resolved;
3583   __ cbz(r0, is_null);
3584 
3585   // Get cpool & tags index
3586   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3587   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3588   // See if bytecode has already been quicked
3589   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3590   __ lea(r1, Address(rscratch1, r19));
3591   __ ldarb(r1, r1);
3592   __ cmp(r1, JVM_CONSTANT_Class);
3593   __ br(Assembler::EQ, quicked);
3594 
3595   __ push(atos); // save receiver for result, and for GC
3596   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3597   // vm_result_2 has metadata result
3598   __ get_vm_result_2(r0, rthread);
3599   __ pop(r3); // restore receiver
3600   __ b(resolved);
3601 
3602   // Get superklass in r0 and subklass in r3
3603   __ bind(quicked);
3604   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3605   __ lea(r0, Address(r2, r19, Address::lsl(3)));
3606   __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3607 
3608   __ bind(resolved);
3609   __ load_klass(r19, r3);
3610 
3611   // Generate subtype check.  Blows r2, r5.  Object in r3.
3612   // Superklass in r0.  Subklass in r19.
3613   __ gen_subtype_check(r19, ok_is_subtype);
3614 
3615   // Come here on failure
3616   __ push(r3);
3617   // object is at TOS
3618   __ b(Interpreter::_throw_ClassCastException_entry);
3619 
3620   // Come here on success
3621   __ bind(ok_is_subtype);
3622   __ mov(r0, r3); // Restore object in r3
3623 
3624   // Collect counts on whether this test sees NULLs a lot or not.
3625   if (ProfileInterpreter) {
3626     __ b(done);
3627     __ bind(is_null);
3628     __ profile_null_seen(r2);
3629   } else {
3630     __ bind(is_null);   // same as 'done'
3631   }
3632   __ bind(done);
3633 }
3634 
3635 void TemplateTable::instanceof() {
3636   transition(atos, itos);
3637   Label done, is_null, ok_is_subtype, quicked, resolved;
3638   __ cbz(r0, is_null);
3639 
3640   // Get cpool & tags index
3641   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3642   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3643   // See if bytecode has already been quicked
3644   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3645   __ lea(r1, Address(rscratch1, r19));
3646   __ ldarb(r1, r1);
3647   __ cmp(r1, JVM_CONSTANT_Class);
3648   __ br(Assembler::EQ, quicked);
3649 
3650   __ push(atos); // save receiver for result, and for GC
3651   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3652   // vm_result_2 has metadata result
3653   __ get_vm_result_2(r0, rthread);
3654   __ pop(r3); // restore receiver
3655   __ verify_oop(r3);
3656   __ load_klass(r3, r3);
3657   __ b(resolved);
3658 
3659   // Get superklass in r0 and subklass in r3
3660   __ bind(quicked);
3661   __ load_klass(r3, r0);
3662   __ lea(r0, Address(r2, r19, Address::lsl(3)));
3663   __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3664 
3665   __ bind(resolved);
3666 
3667   // Generate subtype check.  Blows r2, r5
3668   // Superklass in r0.  Subklass in r3.
3669   __ gen_subtype_check(r3, ok_is_subtype);
3670 
3671   // Come here on failure
3672   __ mov(r0, 0);
3673   __ b(done);
3674   // Come here on success
3675   __ bind(ok_is_subtype);
3676   __ mov(r0, 1);
3677 
3678   // Collect counts on whether this test sees NULLs a lot or not.
3679   if (ProfileInterpreter) {
3680     __ b(done);
3681     __ bind(is_null);
3682     __ profile_null_seen(r2);
3683   } else {
3684     __ bind(is_null);   // same as 'done'
3685   }
3686   __ bind(done);
3687   // r0 = 0: obj == NULL or  obj is not an instanceof the specified klass
3688   // r0 = 1: obj != NULL and obj is     an instanceof the specified klass
3689 }
3690 
3691 //-----------------------------------------------------------------------------
3692 // Breakpoints
3693 void TemplateTable::_breakpoint() {
3694   // Note: We get here even if we are single stepping..
3695   // jbug inists on setting breakpoints at every bytecode
3696   // even if we are in single step mode.
3697 
3698   transition(vtos, vtos);
3699 
3700   // get the unpatched byte code
3701   __ get_method(c_rarg1);
3702   __ call_VM(noreg,
3703              CAST_FROM_FN_PTR(address,
3704                               InterpreterRuntime::get_original_bytecode_at),
3705              c_rarg1, rbcp);
3706   __ mov(r19, r0);
3707 
3708   // post the breakpoint event
3709   __ call_VM(noreg,
3710              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3711              rmethod, rbcp);
3712 
3713   // complete the execution of original bytecode
3714   __ mov(rscratch1, r19);
3715   __ dispatch_only_normal(vtos);
3716 }
3717 
3718 //-----------------------------------------------------------------------------
3719 // Exceptions
3720 
3721 void TemplateTable::athrow() {
3722   transition(atos, vtos);
3723   __ null_check(r0);
3724   __ b(Interpreter::throw_exception_entry());
3725 }
3726 
3727 //-----------------------------------------------------------------------------
3728 // Synchronization
3729 //
3730 // Note: monitorenter & exit are symmetric routines; which is reflected
3731 //       in the assembly code structure as well
3732 //
3733 // Stack layout:
3734 //
3735 // [expressions  ] <--- esp               = expression stack top
3736 // ..
3737 // [expressions  ]
3738 // [monitor entry] <--- monitor block top = expression stack bot
3739 // ..
3740 // [monitor entry]
3741 // [frame data   ] <--- monitor block bot
3742 // ...
3743 // [saved rbp    ] <--- rbp
3744 void TemplateTable::monitorenter()
3745 {
3746   transition(atos, vtos);
3747 
3748   // check for NULL object
3749   __ null_check(r0);
3750 
3751   const Address monitor_block_top(
3752         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3753   const Address monitor_block_bot(
3754         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3755   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3756 
3757   Label allocated;
3758 
3759   // initialize entry pointer
3760   __ mov(c_rarg1, zr); // points to free slot or NULL
3761 
3762   // find a free slot in the monitor block (result in c_rarg1)
3763   {
3764     Label entry, loop, exit;
3765     __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3766                                         // starting with top-most entry
3767     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3768 
3769     __ b(entry);
3770 
3771     __ bind(loop);
3772     // check if current entry is used
3773     // if not used then remember entry in c_rarg1
3774     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3775     __ cmp(zr, rscratch1);
3776     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3777     // check if current entry is for same object
3778     __ cmp(r0, rscratch1);
3779     // if same object then stop searching
3780     __ br(Assembler::EQ, exit);
3781     // otherwise advance to next entry
3782     __ add(c_rarg3, c_rarg3, entry_size);
3783     __ bind(entry);
3784     // check if bottom reached
3785     __ cmp(c_rarg3, c_rarg2);
3786     // if not at bottom then check this entry
3787     __ br(Assembler::NE, loop);
3788     __ bind(exit);
3789   }
3790 
3791   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3792                             // if found, continue with that on
3793 
3794   // allocate one if there's no free slot
3795   {
3796     Label entry, loop;
3797     // 1. compute new pointers            // rsp: old expression stack top
3798     __ ldr(c_rarg1, monitor_block_bot);   // c_rarg1: old expression stack bottom
3799     __ sub(esp, esp, entry_size);         // move expression stack top
3800     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3801     __ mov(c_rarg3, esp);                 // set start value for copy loop
3802     __ str(c_rarg1, monitor_block_bot);   // set new monitor block bottom
3803 
3804     __ sub(sp, sp, entry_size);           // make room for the monitor
3805 
3806     __ b(entry);
3807     // 2. move expression stack contents
3808     __ bind(loop);
3809     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3810                                                    // word from old location
3811     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
3812     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
3813     __ bind(entry);
3814     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
3815     __ br(Assembler::NE, loop);      // if not at bottom then
3816                                      // copy next word
3817   }
3818 
3819   // call run-time routine
3820   // c_rarg1: points to monitor entry
3821   __ bind(allocated);
3822 
3823   // Increment bcp to point to the next bytecode, so exception
3824   // handling for async. exceptions work correctly.
3825   // The object has already been poped from the stack, so the
3826   // expression stack looks correct.
3827   __ increment(rbcp);
3828 
3829   // store object
3830   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3831   __ lock_object(c_rarg1);
3832 
3833   // check to make sure this monitor doesn't cause stack overflow after locking
3834   __ save_bcp();  // in case of exception
3835   __ generate_stack_overflow_check(0);
3836 
3837   // The bcp has already been incremented. Just need to dispatch to
3838   // next instruction.
3839   __ dispatch_next(vtos);
3840 }
3841 
3842 
3843 void TemplateTable::monitorexit()
3844 {
3845   transition(atos, vtos);
3846 
3847   // check for NULL object
3848   __ null_check(r0);
3849 
3850   const Address monitor_block_top(
3851         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3852   const Address monitor_block_bot(
3853         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3854   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3855 
3856   Label found;
3857 
3858   // find matching slot
3859   {
3860     Label entry, loop;
3861     __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3862                                         // starting with top-most entry
3863     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3864                                         // of monitor block
3865     __ b(entry);
3866 
3867     __ bind(loop);
3868     // check if current entry is for same object
3869     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3870     __ cmp(r0, rscratch1);
3871     // if same object then stop searching
3872     __ br(Assembler::EQ, found);
3873     // otherwise advance to next entry
3874     __ add(c_rarg1, c_rarg1, entry_size);
3875     __ bind(entry);
3876     // check if bottom reached
3877     __ cmp(c_rarg1, c_rarg2);
3878     // if not at bottom then check this entry
3879     __ br(Assembler::NE, loop);
3880   }
3881 
3882   // error handling. Unlocking was not block-structured
3883   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3884                    InterpreterRuntime::throw_illegal_monitor_state_exception));
3885   __ should_not_reach_here();
3886 
3887   // call run-time routine
3888   __ bind(found);
3889   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3890   __ unlock_object(c_rarg1);
3891   __ pop_ptr(r0); // discard object
3892 }
3893 
3894 
3895 // Wide instructions
3896 void TemplateTable::wide()
3897 {
3898   __ load_unsigned_byte(r19, at_bcp(1));
3899   __ mov(rscratch1, (address)Interpreter::_wentry_point);
3900   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3901   __ br(rscratch1);
3902 }
3903 
3904 
3905 // Multi arrays
3906 void TemplateTable::multianewarray() {
3907   transition(vtos, atos);
3908   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3909   // last dim is on top of stack; we want address of first one:
3910   // first_addr = last_addr + (ndims - 1) * wordSize
3911   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3912   __ sub(c_rarg1, c_rarg1, wordSize);
3913   call_VM(r0,
3914           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3915           c_rarg1);
3916   __ load_unsigned_byte(r1, at_bcp(3));
3917   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3918 }
3919 #endif // !CC_INTERP