1 /*
   2  * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "compiler/disassembler.hpp"
  29 #include "compiler/compilerDefinitions.inline.hpp"
  30 #include "gc/shared/barrierSetAssembler.hpp"
  31 #include "gc/shared/collectedHeap.hpp"
  32 #include "gc/shared/tlab_globals.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "interpreter/interpreterRuntime.hpp"
  35 #include "interpreter/interp_masm.hpp"
  36 #include "interpreter/templateTable.hpp"
  37 #include "memory/universe.hpp"
  38 #include "oops/methodData.hpp"
  39 #include "oops/method.inline.hpp"
  40 #include "oops/objArrayKlass.hpp"
  41 #include "oops/oop.inline.hpp"
  42 #include "oops/resolvedFieldEntry.hpp"
  43 #include "oops/resolvedIndyEntry.hpp"
  44 #include "oops/resolvedMethodEntry.hpp"
  45 #include "prims/jvmtiExport.hpp"
  46 #include "prims/methodHandles.hpp"
  47 #include "runtime/frame.inline.hpp"
  48 #include "runtime/sharedRuntime.hpp"
  49 #include "runtime/stubRoutines.hpp"
  50 #include "runtime/synchronizer.hpp"
  51 #include "utilities/powerOfTwo.hpp"
  52 
  53 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
  54 
  55 // Address computation: local variables
  56 
  57 static inline Address iaddress(int n) {
  58   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  59 }
  60 
  61 static inline Address laddress(int n) {
  62   return iaddress(n + 1);
  63 }
  64 
  65 static inline Address faddress(int n) {
  66   return iaddress(n);
  67 }
  68 
  69 static inline Address daddress(int n) {
  70   return laddress(n);
  71 }
  72 
  73 static inline Address aaddress(int n) {
  74   return iaddress(n);
  75 }
  76 
  77 static inline Address iaddress(Register r) {
  78   return Address(rlocals, r, Address::lsl(3));
  79 }
  80 
  81 static inline Address laddress(Register r, Register scratch,
  82                                InterpreterMacroAssembler* _masm) {
  83   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  84   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  85 }
  86 
  87 static inline Address faddress(Register r) {
  88   return iaddress(r);
  89 }
  90 
  91 static inline Address daddress(Register r, Register scratch,
  92                                InterpreterMacroAssembler* _masm) {
  93   return laddress(r, scratch, _masm);
  94 }
  95 
  96 static inline Address aaddress(Register r) {
  97   return iaddress(r);
  98 }
  99 
 100 static inline Address at_rsp() {
 101   return Address(esp, 0);
 102 }
 103 
 104 // At top of Java expression stack which may be different than esp().  It
 105 // isn't for category 1 objects.
 106 static inline Address at_tos   () {
 107   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 108 }
 109 
 110 static inline Address at_tos_p1() {
 111   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 112 }
 113 
 114 static inline Address at_tos_p2() {
 115   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 116 }
 117 
 118 static inline Address at_tos_p3() {
 119   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 120 }
 121 
 122 static inline Address at_tos_p4() {
 123   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 124 }
 125 
 126 static inline Address at_tos_p5() {
 127   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 128 }
 129 
 130 // Condition conversion
 131 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 132   switch (cc) {
 133   case TemplateTable::equal        : return Assembler::NE;
 134   case TemplateTable::not_equal    : return Assembler::EQ;
 135   case TemplateTable::less         : return Assembler::GE;
 136   case TemplateTable::less_equal   : return Assembler::GT;
 137   case TemplateTable::greater      : return Assembler::LE;
 138   case TemplateTable::greater_equal: return Assembler::LT;
 139   }
 140   ShouldNotReachHere();
 141   return Assembler::EQ;
 142 }
 143 
 144 
 145 // Miscellaneous helper routines
 146 // Store an oop (or null) at the Address described by obj.
 147 // If val == noreg this means store a null
 148 static void do_oop_store(InterpreterMacroAssembler* _masm,
 149                          Address dst,
 150                          Register val,
 151                          DecoratorSet decorators) {
 152   assert(val == noreg || val == r0, "parameter is just for looks");
 153   __ store_heap_oop(dst, val, r10, r11, r3, decorators);
 154 }
 155 
 156 static void do_oop_load(InterpreterMacroAssembler* _masm,
 157                         Address src,
 158                         Register dst,
 159                         DecoratorSet decorators) {
 160   __ load_heap_oop(dst, src, r10, r11, decorators);
 161 }
 162 
 163 Address TemplateTable::at_bcp(int offset) {
 164   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 165   return Address(rbcp, offset);
 166 }
 167 
 168 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 169                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 170                                    int byte_no)
 171 {
 172   if (!RewriteBytecodes)  return;
 173   Label L_patch_done;
 174 
 175   switch (bc) {
 176   case Bytecodes::_fast_vputfield:
 177   case Bytecodes::_fast_aputfield:
 178   case Bytecodes::_fast_bputfield:
 179   case Bytecodes::_fast_zputfield:
 180   case Bytecodes::_fast_cputfield:
 181   case Bytecodes::_fast_dputfield:
 182   case Bytecodes::_fast_fputfield:
 183   case Bytecodes::_fast_iputfield:
 184   case Bytecodes::_fast_lputfield:
 185   case Bytecodes::_fast_sputfield:
 186     {
 187       // We skip bytecode quickening for putfield instructions when
 188       // the put_code written to the constant pool cache is zero.
 189       // This is required so that every execution of this instruction
 190       // calls out to InterpreterRuntime::resolve_get_put to do
 191       // additional, required work.
 192       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 193       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 194       __ load_field_entry(temp_reg, bc_reg);
 195       if (byte_no == f1_byte) {
 196         __ lea(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::get_code_offset())));
 197       } else {
 198         __ lea(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::put_code_offset())));
 199       }
 200       // Load-acquire the bytecode to match store-release in ResolvedFieldEntry::fill_in()
 201       __ ldarb(temp_reg, temp_reg);
 202       __ movw(bc_reg, bc);
 203       __ cbzw(temp_reg, L_patch_done);  // don't patch
 204     }
 205     break;
 206   default:
 207     assert(byte_no == -1, "sanity");
 208     // the pair bytecodes have already done the load.
 209     if (load_bc_into_bc_reg) {
 210       __ movw(bc_reg, bc);
 211     }
 212   }
 213 
 214   if (JvmtiExport::can_post_breakpoint()) {
 215     Label L_fast_patch;
 216     // if a breakpoint is present we can't rewrite the stream directly
 217     __ load_unsigned_byte(temp_reg, at_bcp(0));
 218     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 219     __ br(Assembler::NE, L_fast_patch);
 220     // Let breakpoint table handling rewrite to quicker bytecode
 221     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 222     __ b(L_patch_done);
 223     __ bind(L_fast_patch);
 224   }
 225 
 226 #ifdef ASSERT
 227   Label L_okay;
 228   __ load_unsigned_byte(temp_reg, at_bcp(0));
 229   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 230   __ br(Assembler::EQ, L_okay);
 231   __ cmpw(temp_reg, bc_reg);
 232   __ br(Assembler::EQ, L_okay);
 233   __ stop("patching the wrong bytecode");
 234   __ bind(L_okay);
 235 #endif
 236 
 237   // patch bytecode
 238   __ strb(bc_reg, at_bcp(0));
 239   __ bind(L_patch_done);
 240 }
 241 
 242 
 243 // Individual instructions
 244 
 245 void TemplateTable::nop() {
 246   transition(vtos, vtos);
 247   // nothing to do
 248 }
 249 
 250 void TemplateTable::shouldnotreachhere() {
 251   transition(vtos, vtos);
 252   __ stop("shouldnotreachhere bytecode");
 253 }
 254 
 255 void TemplateTable::aconst_null()
 256 {
 257   transition(vtos, atos);
 258   __ mov(r0, 0);
 259 }
 260 
 261 void TemplateTable::iconst(int value)
 262 {
 263   transition(vtos, itos);
 264   __ mov(r0, value);
 265 }
 266 
 267 void TemplateTable::lconst(int value)
 268 {
 269   __ mov(r0, value);
 270 }
 271 
 272 void TemplateTable::fconst(int value)
 273 {
 274   transition(vtos, ftos);
 275   switch (value) {
 276   case 0:
 277     __ fmovs(v0, 0.0);
 278     break;
 279   case 1:
 280     __ fmovs(v0, 1.0);
 281     break;
 282   case 2:
 283     __ fmovs(v0, 2.0);
 284     break;
 285   default:
 286     ShouldNotReachHere();
 287     break;
 288   }
 289 }
 290 
 291 void TemplateTable::dconst(int value)
 292 {
 293   transition(vtos, dtos);
 294   switch (value) {
 295   case 0:
 296     __ fmovd(v0, 0.0);
 297     break;
 298   case 1:
 299     __ fmovd(v0, 1.0);
 300     break;
 301   case 2:
 302     __ fmovd(v0, 2.0);
 303     break;
 304   default:
 305     ShouldNotReachHere();
 306     break;
 307   }
 308 }
 309 
 310 void TemplateTable::bipush()
 311 {
 312   transition(vtos, itos);
 313   __ load_signed_byte32(r0, at_bcp(1));
 314 }
 315 
 316 void TemplateTable::sipush()
 317 {
 318   transition(vtos, itos);
 319   __ load_unsigned_short(r0, at_bcp(1));
 320   __ revw(r0, r0);
 321   __ asrw(r0, r0, 16);
 322 }
 323 
 324 void TemplateTable::ldc(LdcType type)
 325 {
 326   transition(vtos, vtos);
 327   Label call_ldc, notFloat, notClass, notInt, Done;
 328 
 329   if (is_ldc_wide(type)) {
 330     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 331   } else {
 332     __ load_unsigned_byte(r1, at_bcp(1));
 333   }
 334   __ get_cpool_and_tags(r2, r0);
 335 
 336   const int base_offset = ConstantPool::header_size() * wordSize;
 337   const int tags_offset = Array<u1>::base_offset_in_bytes();
 338 
 339   // get type
 340   __ add(r3, r1, tags_offset);
 341   __ lea(r3, Address(r0, r3));
 342   __ ldarb(r3, r3);
 343 
 344   // unresolved class - get the resolved class
 345   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClass);
 346   __ br(Assembler::EQ, call_ldc);
 347 
 348   // unresolved class in error state - call into runtime to throw the error
 349   // from the first resolution attempt
 350   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClassInError);
 351   __ br(Assembler::EQ, call_ldc);
 352 
 353   // resolved class - need to call vm to get java mirror of the class
 354   __ cmp(r3, (u1)JVM_CONSTANT_Class);
 355   __ br(Assembler::NE, notClass);
 356 
 357   __ bind(call_ldc);
 358   __ mov(c_rarg1, is_ldc_wide(type) ? 1 : 0);
 359   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 360   __ push_ptr(r0);
 361   __ verify_oop(r0);
 362   __ b(Done);
 363 
 364   __ bind(notClass);
 365   __ cmp(r3, (u1)JVM_CONSTANT_Float);
 366   __ br(Assembler::NE, notFloat);
 367   // ftos
 368   __ adds(r1, r2, r1, Assembler::LSL, 3);
 369   __ ldrs(v0, Address(r1, base_offset));
 370   __ push_f();
 371   __ b(Done);
 372 
 373   __ bind(notFloat);
 374 
 375   __ cmp(r3, (u1)JVM_CONSTANT_Integer);
 376   __ br(Assembler::NE, notInt);
 377 
 378   // itos
 379   __ adds(r1, r2, r1, Assembler::LSL, 3);
 380   __ ldrw(r0, Address(r1, base_offset));
 381   __ push_i(r0);
 382   __ b(Done);
 383 
 384   __ bind(notInt);
 385   condy_helper(Done);
 386 
 387   __ bind(Done);
 388 }
 389 
 390 // Fast path for caching oop constants.
 391 void TemplateTable::fast_aldc(LdcType type)
 392 {
 393   transition(vtos, atos);
 394 
 395   Register result = r0;
 396   Register tmp = r1;
 397   Register rarg = r2;
 398 
 399   int index_size = is_ldc_wide(type) ? sizeof(u2) : sizeof(u1);
 400 
 401   Label resolved;
 402 
 403   // We are resolved if the resolved reference cache entry contains a
 404   // non-null object (String, MethodType, etc.)
 405   assert_different_registers(result, tmp);
 406   __ get_cache_index_at_bcp(tmp, 1, index_size);
 407   __ load_resolved_reference_at_index(result, tmp);
 408   __ cbnz(result, resolved);
 409 
 410   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 411 
 412   // first time invocation - must resolve first
 413   __ mov(rarg, (int)bytecode());
 414   __ call_VM(result, entry, rarg);
 415 
 416   __ bind(resolved);
 417 
 418   { // Check for the null sentinel.
 419     // If we just called the VM, it already did the mapping for us,
 420     // but it's harmless to retry.
 421     Label notNull;
 422 
 423     // Stash null_sentinel address to get its value later
 424     __ movptr(rarg, (uintptr_t)Universe::the_null_sentinel_addr());
 425     __ ldr(tmp, Address(rarg));
 426     __ resolve_oop_handle(tmp, r5, rscratch2);
 427     __ cmpoop(result, tmp);
 428     __ br(Assembler::NE, notNull);
 429     __ mov(result, 0);  // null object reference
 430     __ bind(notNull);
 431   }
 432 
 433   if (VerifyOops) {
 434     // Safe to call with 0 result
 435     __ verify_oop(result);
 436   }
 437 }
 438 
 439 void TemplateTable::ldc2_w()
 440 {
 441   transition(vtos, vtos);
 442   Label notDouble, notLong, Done;
 443   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 444 
 445   __ get_cpool_and_tags(r1, r2);
 446   const int base_offset = ConstantPool::header_size() * wordSize;
 447   const int tags_offset = Array<u1>::base_offset_in_bytes();
 448 
 449   // get type
 450   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 451   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 452   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 453   __ br(Assembler::NE, notDouble);
 454 
 455   // dtos
 456   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 457   __ ldrd(v0, Address(r2, base_offset));
 458   __ push_d();
 459   __ b(Done);
 460 
 461   __ bind(notDouble);
 462   __ cmpw(r2, (int)JVM_CONSTANT_Long);
 463   __ br(Assembler::NE, notLong);
 464 
 465   // ltos
 466   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 467   __ ldr(r0, Address(r0, base_offset));
 468   __ push_l();
 469   __ b(Done);
 470 
 471   __ bind(notLong);
 472   condy_helper(Done);
 473 
 474   __ bind(Done);
 475 }
 476 
 477 void TemplateTable::condy_helper(Label& Done)
 478 {
 479   Register obj = r0;
 480   Register rarg = r1;
 481   Register flags = r2;
 482   Register off = r3;
 483 
 484   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 485 
 486   __ mov(rarg, (int) bytecode());
 487   __ call_VM(obj, entry, rarg);
 488 
 489   __ get_vm_result_2(flags, rthread);
 490 
 491   // VMr = obj = base address to find primitive value to push
 492   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 493   __ mov(off, flags);
 494   __ andw(off, off, ConstantPoolCache::field_index_mask);
 495 
 496   const Address field(obj, off);
 497 
 498   // What sort of thing are we loading?
 499   // x86 uses a shift and mask or wings it with a shift plus assert
 500   // the mask is not needed. aarch64 just uses bitfield extract
 501   __ ubfxw(flags, flags, ConstantPoolCache::tos_state_shift,
 502            ConstantPoolCache::tos_state_bits);
 503 
 504   switch (bytecode()) {
 505     case Bytecodes::_ldc:
 506     case Bytecodes::_ldc_w:
 507       {
 508         // tos in (itos, ftos, stos, btos, ctos, ztos)
 509         Label notInt, notFloat, notShort, notByte, notChar, notBool;
 510         __ cmpw(flags, itos);
 511         __ br(Assembler::NE, notInt);
 512         // itos
 513         __ ldrw(r0, field);
 514         __ push(itos);
 515         __ b(Done);
 516 
 517         __ bind(notInt);
 518         __ cmpw(flags, ftos);
 519         __ br(Assembler::NE, notFloat);
 520         // ftos
 521         __ load_float(field);
 522         __ push(ftos);
 523         __ b(Done);
 524 
 525         __ bind(notFloat);
 526         __ cmpw(flags, stos);
 527         __ br(Assembler::NE, notShort);
 528         // stos
 529         __ load_signed_short(r0, field);
 530         __ push(stos);
 531         __ b(Done);
 532 
 533         __ bind(notShort);
 534         __ cmpw(flags, btos);
 535         __ br(Assembler::NE, notByte);
 536         // btos
 537         __ load_signed_byte(r0, field);
 538         __ push(btos);
 539         __ b(Done);
 540 
 541         __ bind(notByte);
 542         __ cmpw(flags, ctos);
 543         __ br(Assembler::NE, notChar);
 544         // ctos
 545         __ load_unsigned_short(r0, field);
 546         __ push(ctos);
 547         __ b(Done);
 548 
 549         __ bind(notChar);
 550         __ cmpw(flags, ztos);
 551         __ br(Assembler::NE, notBool);
 552         // ztos
 553         __ load_signed_byte(r0, field);
 554         __ push(ztos);
 555         __ b(Done);
 556 
 557         __ bind(notBool);
 558         break;
 559       }
 560 
 561     case Bytecodes::_ldc2_w:
 562       {
 563         Label notLong, notDouble;
 564         __ cmpw(flags, ltos);
 565         __ br(Assembler::NE, notLong);
 566         // ltos
 567         __ ldr(r0, field);
 568         __ push(ltos);
 569         __ b(Done);
 570 
 571         __ bind(notLong);
 572         __ cmpw(flags, dtos);
 573         __ br(Assembler::NE, notDouble);
 574         // dtos
 575         __ load_double(field);
 576         __ push(dtos);
 577         __ b(Done);
 578 
 579        __ bind(notDouble);
 580         break;
 581       }
 582 
 583     default:
 584       ShouldNotReachHere();
 585     }
 586 
 587     __ stop("bad ldc/condy");
 588 }
 589 
 590 void TemplateTable::locals_index(Register reg, int offset)
 591 {
 592   __ ldrb(reg, at_bcp(offset));
 593   __ neg(reg, reg);
 594 }
 595 
 596 void TemplateTable::iload() {
 597   iload_internal();
 598 }
 599 
 600 void TemplateTable::nofast_iload() {
 601   iload_internal(may_not_rewrite);
 602 }
 603 
 604 void TemplateTable::iload_internal(RewriteControl rc) {
 605   transition(vtos, itos);
 606   if (RewriteFrequentPairs && rc == may_rewrite) {
 607     Label rewrite, done;
 608     Register bc = r4;
 609 
 610     // get next bytecode
 611     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 612 
 613     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 614     // last two iloads in a pair.  Comparing against fast_iload means that
 615     // the next bytecode is neither an iload or a caload, and therefore
 616     // an iload pair.
 617     __ cmpw(r1, Bytecodes::_iload);
 618     __ br(Assembler::EQ, done);
 619 
 620     // if _fast_iload rewrite to _fast_iload2
 621     __ cmpw(r1, Bytecodes::_fast_iload);
 622     __ movw(bc, Bytecodes::_fast_iload2);
 623     __ br(Assembler::EQ, rewrite);
 624 
 625     // if _caload rewrite to _fast_icaload
 626     __ cmpw(r1, Bytecodes::_caload);
 627     __ movw(bc, Bytecodes::_fast_icaload);
 628     __ br(Assembler::EQ, rewrite);
 629 
 630     // else rewrite to _fast_iload
 631     __ movw(bc, Bytecodes::_fast_iload);
 632 
 633     // rewrite
 634     // bc: new bytecode
 635     __ bind(rewrite);
 636     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 637     __ bind(done);
 638 
 639   }
 640 
 641   // do iload, get the local value into tos
 642   locals_index(r1);
 643   __ ldr(r0, iaddress(r1));
 644 
 645 }
 646 
 647 void TemplateTable::fast_iload2()
 648 {
 649   transition(vtos, itos);
 650   locals_index(r1);
 651   __ ldr(r0, iaddress(r1));
 652   __ push(itos);
 653   locals_index(r1, 3);
 654   __ ldr(r0, iaddress(r1));
 655 }
 656 
 657 void TemplateTable::fast_iload()
 658 {
 659   transition(vtos, itos);
 660   locals_index(r1);
 661   __ ldr(r0, iaddress(r1));
 662 }
 663 
 664 void TemplateTable::lload()
 665 {
 666   transition(vtos, ltos);
 667   __ ldrb(r1, at_bcp(1));
 668   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 669   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 670 }
 671 
 672 void TemplateTable::fload()
 673 {
 674   transition(vtos, ftos);
 675   locals_index(r1);
 676   // n.b. we use ldrd here because this is a 64 bit slot
 677   // this is comparable to the iload case
 678   __ ldrd(v0, faddress(r1));
 679 }
 680 
 681 void TemplateTable::dload()
 682 {
 683   transition(vtos, dtos);
 684   __ ldrb(r1, at_bcp(1));
 685   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 686   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 687 }
 688 
 689 void TemplateTable::aload()
 690 {
 691   transition(vtos, atos);
 692   locals_index(r1);
 693   __ ldr(r0, iaddress(r1));
 694 }
 695 
 696 void TemplateTable::locals_index_wide(Register reg) {
 697   __ ldrh(reg, at_bcp(2));
 698   __ rev16w(reg, reg);
 699   __ neg(reg, reg);
 700 }
 701 
 702 void TemplateTable::wide_iload() {
 703   transition(vtos, itos);
 704   locals_index_wide(r1);
 705   __ ldr(r0, iaddress(r1));
 706 }
 707 
 708 void TemplateTable::wide_lload()
 709 {
 710   transition(vtos, ltos);
 711   __ ldrh(r1, at_bcp(2));
 712   __ rev16w(r1, r1);
 713   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 714   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 715 }
 716 
 717 void TemplateTable::wide_fload()
 718 {
 719   transition(vtos, ftos);
 720   locals_index_wide(r1);
 721   // n.b. we use ldrd here because this is a 64 bit slot
 722   // this is comparable to the iload case
 723   __ ldrd(v0, faddress(r1));
 724 }
 725 
 726 void TemplateTable::wide_dload()
 727 {
 728   transition(vtos, dtos);
 729   __ ldrh(r1, at_bcp(2));
 730   __ rev16w(r1, r1);
 731   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 732   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 733 }
 734 
 735 void TemplateTable::wide_aload()
 736 {
 737   transition(vtos, atos);
 738   locals_index_wide(r1);
 739   __ ldr(r0, aaddress(r1));
 740 }
 741 
 742 void TemplateTable::index_check(Register array, Register index)
 743 {
 744   // destroys r1, rscratch1
 745   // sign extend index for use by indexed load
 746   // __ movl2ptr(index, index);
 747   // check index
 748   Register length = rscratch1;
 749   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 750   __ cmpw(index, length);
 751   if (index != r1) {
 752     // ??? convention: move aberrant index into r1 for exception message
 753     assert(r1 != array, "different registers");
 754     __ mov(r1, index);
 755   }
 756   Label ok;
 757   __ br(Assembler::LO, ok);
 758   // ??? convention: move array into r3 for exception message
 759    __ mov(r3, array);
 760    __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 761    __ br(rscratch1);
 762   __ bind(ok);
 763 }
 764 
 765 void TemplateTable::iaload()
 766 {
 767   transition(itos, itos);
 768   __ mov(r1, r0);
 769   __ pop_ptr(r0);
 770   // r0: array
 771   // r1: index
 772   index_check(r0, r1); // leaves index in r1, kills rscratch1
 773   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
 774   __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 775 }
 776 
 777 void TemplateTable::laload()
 778 {
 779   transition(itos, ltos);
 780   __ mov(r1, r0);
 781   __ pop_ptr(r0);
 782   // r0: array
 783   // r1: index
 784   index_check(r0, r1); // leaves index in r1, kills rscratch1
 785   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
 786   __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 787 }
 788 
 789 void TemplateTable::faload()
 790 {
 791   transition(itos, ftos);
 792   __ mov(r1, r0);
 793   __ pop_ptr(r0);
 794   // r0: array
 795   // r1: index
 796   index_check(r0, r1); // leaves index in r1, kills rscratch1
 797   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
 798   __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 799 }
 800 
 801 void TemplateTable::daload()
 802 {
 803   transition(itos, dtos);
 804   __ mov(r1, r0);
 805   __ pop_ptr(r0);
 806   // r0: array
 807   // r1: index
 808   index_check(r0, r1); // leaves index in r1, kills rscratch1
 809   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
 810   __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 811 }
 812 
 813 void TemplateTable::aaload()
 814 {
 815   transition(itos, atos);
 816   __ mov(r1, r0);
 817   __ pop_ptr(r0);
 818   // r0: array
 819   // r1: index
 820   index_check(r0, r1); // leaves index in r1, kills rscratch1
 821   __ profile_array_type<ArrayLoadData>(r2, r0, r4);
 822   if (UseFlatArray) {
 823     Label is_flat_array, done;
 824 
 825     __ test_flat_array_oop(r0, r8 /*temp*/, is_flat_array);
 826     __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 827     do_oop_load(_masm, Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)), r0, IS_ARRAY);
 828 
 829     __ b(done);
 830     __ bind(is_flat_array);
 831     __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), r0, r1);
 832     // Ensure the stores to copy the inline field contents are visible
 833     // before any subsequent store that publishes this reference.
 834     __ membar(Assembler::StoreStore);
 835     __ bind(done);
 836   } else {
 837     __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 838     do_oop_load(_masm, Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)), r0, IS_ARRAY);
 839   }
 840   __ profile_element_type(r2, r0, r4);
 841 }
 842 
 843 void TemplateTable::baload()
 844 {
 845   transition(itos, itos);
 846   __ mov(r1, r0);
 847   __ pop_ptr(r0);
 848   // r0: array
 849   // r1: index
 850   index_check(r0, r1); // leaves index in r1, kills rscratch1
 851   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
 852   __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(0)), noreg, noreg);
 853 }
 854 
 855 void TemplateTable::caload()
 856 {
 857   transition(itos, itos);
 858   __ mov(r1, r0);
 859   __ pop_ptr(r0);
 860   // r0: array
 861   // r1: index
 862   index_check(r0, r1); // leaves index in r1, kills rscratch1
 863   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 864   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 865 }
 866 
 867 // iload followed by caload frequent pair
 868 void TemplateTable::fast_icaload()
 869 {
 870   transition(vtos, itos);
 871   // load index out of locals
 872   locals_index(r2);
 873   __ ldr(r1, iaddress(r2));
 874 
 875   __ pop_ptr(r0);
 876 
 877   // r0: array
 878   // r1: index
 879   index_check(r0, r1); // leaves index in r1, kills rscratch1
 880   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 881   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 882 }
 883 
 884 void TemplateTable::saload()
 885 {
 886   transition(itos, itos);
 887   __ mov(r1, r0);
 888   __ pop_ptr(r0);
 889   // r0: array
 890   // r1: index
 891   index_check(r0, r1); // leaves index in r1, kills rscratch1
 892   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_SHORT) >> 1);
 893   __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 894 }
 895 
 896 void TemplateTable::iload(int n)
 897 {
 898   transition(vtos, itos);
 899   __ ldr(r0, iaddress(n));
 900 }
 901 
 902 void TemplateTable::lload(int n)
 903 {
 904   transition(vtos, ltos);
 905   __ ldr(r0, laddress(n));
 906 }
 907 
 908 void TemplateTable::fload(int n)
 909 {
 910   transition(vtos, ftos);
 911   __ ldrs(v0, faddress(n));
 912 }
 913 
 914 void TemplateTable::dload(int n)
 915 {
 916   transition(vtos, dtos);
 917   __ ldrd(v0, daddress(n));
 918 }
 919 
 920 void TemplateTable::aload(int n)
 921 {
 922   transition(vtos, atos);
 923   __ ldr(r0, iaddress(n));
 924 }
 925 
 926 void TemplateTable::aload_0() {
 927   aload_0_internal();
 928 }
 929 
 930 void TemplateTable::nofast_aload_0() {
 931   aload_0_internal(may_not_rewrite);
 932 }
 933 
 934 void TemplateTable::aload_0_internal(RewriteControl rc) {
 935   // According to bytecode histograms, the pairs:
 936   //
 937   // _aload_0, _fast_igetfield
 938   // _aload_0, _fast_agetfield
 939   // _aload_0, _fast_fgetfield
 940   //
 941   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 942   // _aload_0 bytecode checks if the next bytecode is either
 943   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 944   // rewrites the current bytecode into a pair bytecode; otherwise it
 945   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 946   // the pair check anymore.
 947   //
 948   // Note: If the next bytecode is _getfield, the rewrite must be
 949   //       delayed, otherwise we may miss an opportunity for a pair.
 950   //
 951   // Also rewrite frequent pairs
 952   //   aload_0, aload_1
 953   //   aload_0, iload_1
 954   // These bytecodes with a small amount of code are most profitable
 955   // to rewrite
 956   if (RewriteFrequentPairs && rc == may_rewrite) {
 957     Label rewrite, done;
 958     const Register bc = r4;
 959 
 960     // get next bytecode
 961     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 962 
 963     // if _getfield then wait with rewrite
 964     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 965     __ br(Assembler::EQ, done);
 966 
 967     // if _igetfield then rewrite to _fast_iaccess_0
 968     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 969     __ cmpw(r1, Bytecodes::_fast_igetfield);
 970     __ movw(bc, Bytecodes::_fast_iaccess_0);
 971     __ br(Assembler::EQ, rewrite);
 972 
 973     // if _agetfield then rewrite to _fast_aaccess_0
 974     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 975     __ cmpw(r1, Bytecodes::_fast_agetfield);
 976     __ movw(bc, Bytecodes::_fast_aaccess_0);
 977     __ br(Assembler::EQ, rewrite);
 978 
 979     // if _fgetfield then rewrite to _fast_faccess_0
 980     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 981     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 982     __ movw(bc, Bytecodes::_fast_faccess_0);
 983     __ br(Assembler::EQ, rewrite);
 984 
 985     // else rewrite to _fast_aload0
 986     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 987     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 988 
 989     // rewrite
 990     // bc: new bytecode
 991     __ bind(rewrite);
 992     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 993 
 994     __ bind(done);
 995   }
 996 
 997   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 998   aload(0);
 999 }
1000 
1001 void TemplateTable::istore()
1002 {
1003   transition(itos, vtos);
1004   locals_index(r1);
1005   // FIXME: We're being very pernickerty here storing a jint in a
1006   // local with strw, which costs an extra instruction over what we'd
1007   // be able to do with a simple str.  We should just store the whole
1008   // word.
1009   __ lea(rscratch1, iaddress(r1));
1010   __ strw(r0, Address(rscratch1));
1011 }
1012 
1013 void TemplateTable::lstore()
1014 {
1015   transition(ltos, vtos);
1016   locals_index(r1);
1017   __ str(r0, laddress(r1, rscratch1, _masm));
1018 }
1019 
1020 void TemplateTable::fstore() {
1021   transition(ftos, vtos);
1022   locals_index(r1);
1023   __ lea(rscratch1, iaddress(r1));
1024   __ strs(v0, Address(rscratch1));
1025 }
1026 
1027 void TemplateTable::dstore() {
1028   transition(dtos, vtos);
1029   locals_index(r1);
1030   __ strd(v0, daddress(r1, rscratch1, _masm));
1031 }
1032 
1033 void TemplateTable::astore()
1034 {
1035   transition(vtos, vtos);
1036   __ pop_ptr(r0);
1037   locals_index(r1);
1038   __ str(r0, aaddress(r1));
1039 }
1040 
1041 void TemplateTable::wide_istore() {
1042   transition(vtos, vtos);
1043   __ pop_i();
1044   locals_index_wide(r1);
1045   __ lea(rscratch1, iaddress(r1));
1046   __ strw(r0, Address(rscratch1));
1047 }
1048 
1049 void TemplateTable::wide_lstore() {
1050   transition(vtos, vtos);
1051   __ pop_l();
1052   locals_index_wide(r1);
1053   __ str(r0, laddress(r1, rscratch1, _masm));
1054 }
1055 
1056 void TemplateTable::wide_fstore() {
1057   transition(vtos, vtos);
1058   __ pop_f();
1059   locals_index_wide(r1);
1060   __ lea(rscratch1, faddress(r1));
1061   __ strs(v0, rscratch1);
1062 }
1063 
1064 void TemplateTable::wide_dstore() {
1065   transition(vtos, vtos);
1066   __ pop_d();
1067   locals_index_wide(r1);
1068   __ strd(v0, daddress(r1, rscratch1, _masm));
1069 }
1070 
1071 void TemplateTable::wide_astore() {
1072   transition(vtos, vtos);
1073   __ pop_ptr(r0);
1074   locals_index_wide(r1);
1075   __ str(r0, aaddress(r1));
1076 }
1077 
1078 void TemplateTable::iastore() {
1079   transition(itos, vtos);
1080   __ pop_i(r1);
1081   __ pop_ptr(r3);
1082   // r0: value
1083   // r1: index
1084   // r3: array
1085   index_check(r3, r1); // prefer index in r1
1086   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
1087   __ access_store_at(T_INT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), r0, noreg, noreg, noreg);
1088 }
1089 
1090 void TemplateTable::lastore() {
1091   transition(ltos, vtos);
1092   __ pop_i(r1);
1093   __ pop_ptr(r3);
1094   // r0: value
1095   // r1: index
1096   // r3: array
1097   index_check(r3, r1); // prefer index in r1
1098   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
1099   __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), r0, noreg, noreg, noreg);
1100 }
1101 
1102 void TemplateTable::fastore() {
1103   transition(ftos, vtos);
1104   __ pop_i(r1);
1105   __ pop_ptr(r3);
1106   // v0: value
1107   // r1:  index
1108   // r3:  array
1109   index_check(r3, r1); // prefer index in r1
1110   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
1111   __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), noreg /* ftos */, noreg, noreg, noreg);
1112 }
1113 
1114 void TemplateTable::dastore() {
1115   transition(dtos, vtos);
1116   __ pop_i(r1);
1117   __ pop_ptr(r3);
1118   // v0: value
1119   // r1:  index
1120   // r3:  array
1121   index_check(r3, r1); // prefer index in r1
1122   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
1123   __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), noreg /* dtos */, noreg, noreg, noreg);
1124 }
1125 
1126 void TemplateTable::aastore() {
1127   Label is_null, is_flat_array, ok_is_subtype, done;
1128   transition(vtos, vtos);
1129   // stack: ..., array, index, value
1130   __ ldr(r0, at_tos());    // value
1131   __ ldr(r2, at_tos_p1()); // index
1132   __ ldr(r3, at_tos_p2()); // array
1133 
1134   index_check(r3, r2);     // kills r1
1135 
1136   __ profile_array_type<ArrayStoreData>(r4, r3, r5);
1137   __ profile_multiple_element_types(r4, r0, r5, r6);
1138 
1139   __ add(r4, r2, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
1140   Address element_address(r3, r4, Address::uxtw(LogBytesPerHeapOop));
1141   // Be careful not to clobber r4 below
1142 
1143   // do array store check - check for null value first
1144   __ cbz(r0, is_null);
1145 
1146   // Move array class to r5
1147   __ load_klass(r5, r3);
1148 
1149   if (UseFlatArray) {
1150     __ ldrw(r6, Address(r5, Klass::layout_helper_offset()));
1151     __ test_flat_array_layout(r6, is_flat_array);
1152   }
1153 
1154   // Move subklass into r1
1155   __ load_klass(r1, r0);
1156 
1157   // Move array element superklass into r0
1158   __ ldr(r0, Address(r5, ObjArrayKlass::element_klass_offset()));
1159   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1160 
1161   // Generate subtype check.  Blows r2, r5
1162   // Superklass in r0.  Subklass in r1.
1163 
1164   // is "r1 <: r0" ? (value subclass <: array element superclass)
1165   __ gen_subtype_check(r1, ok_is_subtype, false);
1166 
1167   // Come here on failure
1168   // object is at TOS
1169   __ b(Interpreter::_throw_ArrayStoreException_entry);
1170 
1171   // Come here on success
1172   __ bind(ok_is_subtype);
1173 
1174   // Get the value we will store
1175   __ ldr(r0, at_tos());
1176   // Now store using the appropriate barrier
1177   do_oop_store(_masm, element_address, r0, IS_ARRAY);
1178   __ b(done);
1179 
1180   // Have a null in r0, r3=array, r2=index.  Store null at ary[idx]
1181   __ bind(is_null);
1182   if (EnableValhalla) {
1183     Label is_null_into_value_array_npe, store_null;
1184 
1185     // No way to store null in flat null-free array
1186     __ test_null_free_array_oop(r3, r8, is_null_into_value_array_npe);
1187     __ b(store_null);
1188 
1189     __ bind(is_null_into_value_array_npe);
1190     __ b(ExternalAddress(Interpreter::_throw_NullPointerException_entry));
1191 
1192     __ bind(store_null);
1193   }
1194 
1195   // Store a null
1196   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1197   __ b(done);
1198 
1199   if (UseFlatArray) {
1200      Label is_type_ok;
1201     __ bind(is_flat_array); // Store non-null value to flat
1202 
1203     // Simplistic type check...
1204     // r0 - value, r2 - index, r3 - array.
1205 
1206     // Profile the not-null value's klass.
1207     // Load value class
1208      __ load_klass(r1, r0);
1209 
1210     // Move element klass into r7
1211      __ ldr(r7, Address(r5, ArrayKlass::element_klass_offset()));
1212 
1213     // flat value array needs exact type match
1214     // is "r1 == r7" (value subclass == array element superclass)
1215 
1216      __ cmp(r7, r1);
1217      __ br(Assembler::EQ, is_type_ok);
1218 
1219      __ b(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1220 
1221      __ bind(is_type_ok);
1222     // r1: value's klass
1223     // r3: array
1224     // r5: array klass
1225     __ test_klass_is_empty_inline_type(r1, r7, done);
1226 
1227     // calc dst for copy
1228     __ ldrw(r7, at_tos_p1()); // index
1229     __ data_for_value_array_index(r3, r5, r7, r7);
1230 
1231     // ...and src for copy
1232     __ ldr(r6, at_tos());  // value
1233     __ data_for_oop(r6, r6, r1);
1234 
1235     __ mov(r4, r1);  // Shuffle arguments to avoid conflict with c_rarg1
1236     __ access_value_copy(IN_HEAP, r6, r7, r4);
1237   }
1238 
1239   // Pop stack arguments
1240   __ bind(done);
1241   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1242 }
1243 
1244 void TemplateTable::bastore()
1245 {
1246   transition(itos, vtos);
1247   __ pop_i(r1);
1248   __ pop_ptr(r3);
1249   // r0: value
1250   // r1: index
1251   // r3: array
1252   index_check(r3, r1); // prefer index in r1
1253 
1254   // Need to check whether array is boolean or byte
1255   // since both types share the bastore bytecode.
1256   __ load_klass(r2, r3);
1257   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1258   int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1259   Label L_skip;
1260   __ tbz(r2, diffbit_index, L_skip);
1261   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1262   __ bind(L_skip);
1263 
1264   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
1265   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(0)), r0, noreg, noreg, noreg);
1266 }
1267 
1268 void TemplateTable::castore()
1269 {
1270   transition(itos, vtos);
1271   __ pop_i(r1);
1272   __ pop_ptr(r3);
1273   // r0: value
1274   // r1: index
1275   // r3: array
1276   index_check(r3, r1); // prefer index in r1
1277   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
1278   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(1)), r0, noreg, noreg, noreg);
1279 }
1280 
1281 void TemplateTable::sastore()
1282 {
1283   castore();
1284 }
1285 
1286 void TemplateTable::istore(int n)
1287 {
1288   transition(itos, vtos);
1289   __ str(r0, iaddress(n));
1290 }
1291 
1292 void TemplateTable::lstore(int n)
1293 {
1294   transition(ltos, vtos);
1295   __ str(r0, laddress(n));
1296 }
1297 
1298 void TemplateTable::fstore(int n)
1299 {
1300   transition(ftos, vtos);
1301   __ strs(v0, faddress(n));
1302 }
1303 
1304 void TemplateTable::dstore(int n)
1305 {
1306   transition(dtos, vtos);
1307   __ strd(v0, daddress(n));
1308 }
1309 
1310 void TemplateTable::astore(int n)
1311 {
1312   transition(vtos, vtos);
1313   __ pop_ptr(r0);
1314   __ str(r0, iaddress(n));
1315 }
1316 
1317 void TemplateTable::pop()
1318 {
1319   transition(vtos, vtos);
1320   __ add(esp, esp, Interpreter::stackElementSize);
1321 }
1322 
1323 void TemplateTable::pop2()
1324 {
1325   transition(vtos, vtos);
1326   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1327 }
1328 
1329 void TemplateTable::dup()
1330 {
1331   transition(vtos, vtos);
1332   __ ldr(r0, Address(esp, 0));
1333   __ push(r0);
1334   // stack: ..., a, a
1335 }
1336 
1337 void TemplateTable::dup_x1()
1338 {
1339   transition(vtos, vtos);
1340   // stack: ..., a, b
1341   __ ldr(r0, at_tos());  // load b
1342   __ ldr(r2, at_tos_p1());  // load a
1343   __ str(r0, at_tos_p1());  // store b
1344   __ str(r2, at_tos());  // store a
1345   __ push(r0);                  // push b
1346   // stack: ..., b, a, b
1347 }
1348 
1349 void TemplateTable::dup_x2()
1350 {
1351   transition(vtos, vtos);
1352   // stack: ..., a, b, c
1353   __ ldr(r0, at_tos());  // load c
1354   __ ldr(r2, at_tos_p2());  // load a
1355   __ str(r0, at_tos_p2());  // store c in a
1356   __ push(r0);      // push c
1357   // stack: ..., c, b, c, c
1358   __ ldr(r0, at_tos_p2());  // load b
1359   __ str(r2, at_tos_p2());  // store a in b
1360   // stack: ..., c, a, c, c
1361   __ str(r0, at_tos_p1());  // store b in c
1362   // stack: ..., c, a, b, c
1363 }
1364 
1365 void TemplateTable::dup2()
1366 {
1367   transition(vtos, vtos);
1368   // stack: ..., a, b
1369   __ ldr(r0, at_tos_p1());  // load a
1370   __ push(r0);                  // push a
1371   __ ldr(r0, at_tos_p1());  // load b
1372   __ push(r0);                  // push b
1373   // stack: ..., a, b, a, b
1374 }
1375 
1376 void TemplateTable::dup2_x1()
1377 {
1378   transition(vtos, vtos);
1379   // stack: ..., a, b, c
1380   __ ldr(r2, at_tos());  // load c
1381   __ ldr(r0, at_tos_p1());  // load b
1382   __ push(r0);                  // push b
1383   __ push(r2);                  // push c
1384   // stack: ..., a, b, c, b, c
1385   __ str(r2, at_tos_p3());  // store c in b
1386   // stack: ..., a, c, c, b, c
1387   __ ldr(r2, at_tos_p4());  // load a
1388   __ str(r2, at_tos_p2());  // store a in 2nd c
1389   // stack: ..., a, c, a, b, c
1390   __ str(r0, at_tos_p4());  // store b in a
1391   // stack: ..., b, c, a, b, c
1392 }
1393 
1394 void TemplateTable::dup2_x2()
1395 {
1396   transition(vtos, vtos);
1397   // stack: ..., a, b, c, d
1398   __ ldr(r2, at_tos());  // load d
1399   __ ldr(r0, at_tos_p1());  // load c
1400   __ push(r0)            ;      // push c
1401   __ push(r2);                  // push d
1402   // stack: ..., a, b, c, d, c, d
1403   __ ldr(r0, at_tos_p4());  // load b
1404   __ str(r0, at_tos_p2());  // store b in d
1405   __ str(r2, at_tos_p4());  // store d in b
1406   // stack: ..., a, d, c, b, c, d
1407   __ ldr(r2, at_tos_p5());  // load a
1408   __ ldr(r0, at_tos_p3());  // load c
1409   __ str(r2, at_tos_p3());  // store a in c
1410   __ str(r0, at_tos_p5());  // store c in a
1411   // stack: ..., c, d, a, b, c, d
1412 }
1413 
1414 void TemplateTable::swap()
1415 {
1416   transition(vtos, vtos);
1417   // stack: ..., a, b
1418   __ ldr(r2, at_tos_p1());  // load a
1419   __ ldr(r0, at_tos());  // load b
1420   __ str(r2, at_tos());  // store a in b
1421   __ str(r0, at_tos_p1());  // store b in a
1422   // stack: ..., b, a
1423 }
1424 
1425 void TemplateTable::iop2(Operation op)
1426 {
1427   transition(itos, itos);
1428   // r0 <== r1 op r0
1429   __ pop_i(r1);
1430   switch (op) {
1431   case add  : __ addw(r0, r1, r0); break;
1432   case sub  : __ subw(r0, r1, r0); break;
1433   case mul  : __ mulw(r0, r1, r0); break;
1434   case _and : __ andw(r0, r1, r0); break;
1435   case _or  : __ orrw(r0, r1, r0); break;
1436   case _xor : __ eorw(r0, r1, r0); break;
1437   case shl  : __ lslvw(r0, r1, r0); break;
1438   case shr  : __ asrvw(r0, r1, r0); break;
1439   case ushr : __ lsrvw(r0, r1, r0);break;
1440   default   : ShouldNotReachHere();
1441   }
1442 }
1443 
1444 void TemplateTable::lop2(Operation op)
1445 {
1446   transition(ltos, ltos);
1447   // r0 <== r1 op r0
1448   __ pop_l(r1);
1449   switch (op) {
1450   case add  : __ add(r0, r1, r0); break;
1451   case sub  : __ sub(r0, r1, r0); break;
1452   case mul  : __ mul(r0, r1, r0); break;
1453   case _and : __ andr(r0, r1, r0); break;
1454   case _or  : __ orr(r0, r1, r0); break;
1455   case _xor : __ eor(r0, r1, r0); break;
1456   default   : ShouldNotReachHere();
1457   }
1458 }
1459 
1460 void TemplateTable::idiv()
1461 {
1462   transition(itos, itos);
1463   // explicitly check for div0
1464   Label no_div0;
1465   __ cbnzw(r0, no_div0);
1466   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1467   __ br(rscratch1);
1468   __ bind(no_div0);
1469   __ pop_i(r1);
1470   // r0 <== r1 idiv r0
1471   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1472 }
1473 
1474 void TemplateTable::irem()
1475 {
1476   transition(itos, itos);
1477   // explicitly check for div0
1478   Label no_div0;
1479   __ cbnzw(r0, no_div0);
1480   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1481   __ br(rscratch1);
1482   __ bind(no_div0);
1483   __ pop_i(r1);
1484   // r0 <== r1 irem r0
1485   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1486 }
1487 
1488 void TemplateTable::lmul()
1489 {
1490   transition(ltos, ltos);
1491   __ pop_l(r1);
1492   __ mul(r0, r0, r1);
1493 }
1494 
1495 void TemplateTable::ldiv()
1496 {
1497   transition(ltos, ltos);
1498   // explicitly check for div0
1499   Label no_div0;
1500   __ cbnz(r0, no_div0);
1501   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1502   __ br(rscratch1);
1503   __ bind(no_div0);
1504   __ pop_l(r1);
1505   // r0 <== r1 ldiv r0
1506   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1507 }
1508 
1509 void TemplateTable::lrem()
1510 {
1511   transition(ltos, ltos);
1512   // explicitly check for div0
1513   Label no_div0;
1514   __ cbnz(r0, no_div0);
1515   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1516   __ br(rscratch1);
1517   __ bind(no_div0);
1518   __ pop_l(r1);
1519   // r0 <== r1 lrem r0
1520   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1521 }
1522 
1523 void TemplateTable::lshl()
1524 {
1525   transition(itos, ltos);
1526   // shift count is in r0
1527   __ pop_l(r1);
1528   __ lslv(r0, r1, r0);
1529 }
1530 
1531 void TemplateTable::lshr()
1532 {
1533   transition(itos, ltos);
1534   // shift count is in r0
1535   __ pop_l(r1);
1536   __ asrv(r0, r1, r0);
1537 }
1538 
1539 void TemplateTable::lushr()
1540 {
1541   transition(itos, ltos);
1542   // shift count is in r0
1543   __ pop_l(r1);
1544   __ lsrv(r0, r1, r0);
1545 }
1546 
1547 void TemplateTable::fop2(Operation op)
1548 {
1549   transition(ftos, ftos);
1550   switch (op) {
1551   case add:
1552     // n.b. use ldrd because this is a 64 bit slot
1553     __ pop_f(v1);
1554     __ fadds(v0, v1, v0);
1555     break;
1556   case sub:
1557     __ pop_f(v1);
1558     __ fsubs(v0, v1, v0);
1559     break;
1560   case mul:
1561     __ pop_f(v1);
1562     __ fmuls(v0, v1, v0);
1563     break;
1564   case div:
1565     __ pop_f(v1);
1566     __ fdivs(v0, v1, v0);
1567     break;
1568   case rem:
1569     __ fmovs(v1, v0);
1570     __ pop_f(v0);
1571     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1572     break;
1573   default:
1574     ShouldNotReachHere();
1575     break;
1576   }
1577 }
1578 
1579 void TemplateTable::dop2(Operation op)
1580 {
1581   transition(dtos, dtos);
1582   switch (op) {
1583   case add:
1584     // n.b. use ldrd because this is a 64 bit slot
1585     __ pop_d(v1);
1586     __ faddd(v0, v1, v0);
1587     break;
1588   case sub:
1589     __ pop_d(v1);
1590     __ fsubd(v0, v1, v0);
1591     break;
1592   case mul:
1593     __ pop_d(v1);
1594     __ fmuld(v0, v1, v0);
1595     break;
1596   case div:
1597     __ pop_d(v1);
1598     __ fdivd(v0, v1, v0);
1599     break;
1600   case rem:
1601     __ fmovd(v1, v0);
1602     __ pop_d(v0);
1603     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1604     break;
1605   default:
1606     ShouldNotReachHere();
1607     break;
1608   }
1609 }
1610 
1611 void TemplateTable::ineg()
1612 {
1613   transition(itos, itos);
1614   __ negw(r0, r0);
1615 
1616 }
1617 
1618 void TemplateTable::lneg()
1619 {
1620   transition(ltos, ltos);
1621   __ neg(r0, r0);
1622 }
1623 
1624 void TemplateTable::fneg()
1625 {
1626   transition(ftos, ftos);
1627   __ fnegs(v0, v0);
1628 }
1629 
1630 void TemplateTable::dneg()
1631 {
1632   transition(dtos, dtos);
1633   __ fnegd(v0, v0);
1634 }
1635 
1636 void TemplateTable::iinc()
1637 {
1638   transition(vtos, vtos);
1639   __ load_signed_byte(r1, at_bcp(2)); // get constant
1640   locals_index(r2);
1641   __ ldr(r0, iaddress(r2));
1642   __ addw(r0, r0, r1);
1643   __ str(r0, iaddress(r2));
1644 }
1645 
1646 void TemplateTable::wide_iinc()
1647 {
1648   transition(vtos, vtos);
1649   // __ mov(r1, zr);
1650   __ ldrw(r1, at_bcp(2)); // get constant and index
1651   __ rev16(r1, r1);
1652   __ ubfx(r2, r1, 0, 16);
1653   __ neg(r2, r2);
1654   __ sbfx(r1, r1, 16, 16);
1655   __ ldr(r0, iaddress(r2));
1656   __ addw(r0, r0, r1);
1657   __ str(r0, iaddress(r2));
1658 }
1659 
1660 void TemplateTable::convert()
1661 {
1662   // Checking
1663 #ifdef ASSERT
1664   {
1665     TosState tos_in  = ilgl;
1666     TosState tos_out = ilgl;
1667     switch (bytecode()) {
1668     case Bytecodes::_i2l: // fall through
1669     case Bytecodes::_i2f: // fall through
1670     case Bytecodes::_i2d: // fall through
1671     case Bytecodes::_i2b: // fall through
1672     case Bytecodes::_i2c: // fall through
1673     case Bytecodes::_i2s: tos_in = itos; break;
1674     case Bytecodes::_l2i: // fall through
1675     case Bytecodes::_l2f: // fall through
1676     case Bytecodes::_l2d: tos_in = ltos; break;
1677     case Bytecodes::_f2i: // fall through
1678     case Bytecodes::_f2l: // fall through
1679     case Bytecodes::_f2d: tos_in = ftos; break;
1680     case Bytecodes::_d2i: // fall through
1681     case Bytecodes::_d2l: // fall through
1682     case Bytecodes::_d2f: tos_in = dtos; break;
1683     default             : ShouldNotReachHere();
1684     }
1685     switch (bytecode()) {
1686     case Bytecodes::_l2i: // fall through
1687     case Bytecodes::_f2i: // fall through
1688     case Bytecodes::_d2i: // fall through
1689     case Bytecodes::_i2b: // fall through
1690     case Bytecodes::_i2c: // fall through
1691     case Bytecodes::_i2s: tos_out = itos; break;
1692     case Bytecodes::_i2l: // fall through
1693     case Bytecodes::_f2l: // fall through
1694     case Bytecodes::_d2l: tos_out = ltos; break;
1695     case Bytecodes::_i2f: // fall through
1696     case Bytecodes::_l2f: // fall through
1697     case Bytecodes::_d2f: tos_out = ftos; break;
1698     case Bytecodes::_i2d: // fall through
1699     case Bytecodes::_l2d: // fall through
1700     case Bytecodes::_f2d: tos_out = dtos; break;
1701     default             : ShouldNotReachHere();
1702     }
1703     transition(tos_in, tos_out);
1704   }
1705 #endif // ASSERT
1706   // static const int64_t is_nan = 0x8000000000000000L;
1707 
1708   // Conversion
1709   switch (bytecode()) {
1710   case Bytecodes::_i2l:
1711     __ sxtw(r0, r0);
1712     break;
1713   case Bytecodes::_i2f:
1714     __ scvtfws(v0, r0);
1715     break;
1716   case Bytecodes::_i2d:
1717     __ scvtfwd(v0, r0);
1718     break;
1719   case Bytecodes::_i2b:
1720     __ sxtbw(r0, r0);
1721     break;
1722   case Bytecodes::_i2c:
1723     __ uxthw(r0, r0);
1724     break;
1725   case Bytecodes::_i2s:
1726     __ sxthw(r0, r0);
1727     break;
1728   case Bytecodes::_l2i:
1729     __ uxtw(r0, r0);
1730     break;
1731   case Bytecodes::_l2f:
1732     __ scvtfs(v0, r0);
1733     break;
1734   case Bytecodes::_l2d:
1735     __ scvtfd(v0, r0);
1736     break;
1737   case Bytecodes::_f2i:
1738   {
1739     Label L_Okay;
1740     __ clear_fpsr();
1741     __ fcvtzsw(r0, v0);
1742     __ get_fpsr(r1);
1743     __ cbzw(r1, L_Okay);
1744     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i));
1745     __ bind(L_Okay);
1746   }
1747     break;
1748   case Bytecodes::_f2l:
1749   {
1750     Label L_Okay;
1751     __ clear_fpsr();
1752     __ fcvtzs(r0, v0);
1753     __ get_fpsr(r1);
1754     __ cbzw(r1, L_Okay);
1755     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1756     __ bind(L_Okay);
1757   }
1758     break;
1759   case Bytecodes::_f2d:
1760     __ fcvts(v0, v0);
1761     break;
1762   case Bytecodes::_d2i:
1763   {
1764     Label L_Okay;
1765     __ clear_fpsr();
1766     __ fcvtzdw(r0, v0);
1767     __ get_fpsr(r1);
1768     __ cbzw(r1, L_Okay);
1769     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1770     __ bind(L_Okay);
1771   }
1772     break;
1773   case Bytecodes::_d2l:
1774   {
1775     Label L_Okay;
1776     __ clear_fpsr();
1777     __ fcvtzd(r0, v0);
1778     __ get_fpsr(r1);
1779     __ cbzw(r1, L_Okay);
1780     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1781     __ bind(L_Okay);
1782   }
1783     break;
1784   case Bytecodes::_d2f:
1785     __ fcvtd(v0, v0);
1786     break;
1787   default:
1788     ShouldNotReachHere();
1789   }
1790 }
1791 
1792 void TemplateTable::lcmp()
1793 {
1794   transition(ltos, itos);
1795   Label done;
1796   __ pop_l(r1);
1797   __ cmp(r1, r0);
1798   __ mov(r0, (uint64_t)-1L);
1799   __ br(Assembler::LT, done);
1800   // __ mov(r0, 1UL);
1801   // __ csel(r0, r0, zr, Assembler::NE);
1802   // and here is a faster way
1803   __ csinc(r0, zr, zr, Assembler::EQ);
1804   __ bind(done);
1805 }
1806 
1807 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1808 {
1809   Label done;
1810   if (is_float) {
1811     // XXX get rid of pop here, use ... reg, mem32
1812     __ pop_f(v1);
1813     __ fcmps(v1, v0);
1814   } else {
1815     // XXX get rid of pop here, use ... reg, mem64
1816     __ pop_d(v1);
1817     __ fcmpd(v1, v0);
1818   }
1819   if (unordered_result < 0) {
1820     // we want -1 for unordered or less than, 0 for equal and 1 for
1821     // greater than.
1822     __ mov(r0, (uint64_t)-1L);
1823     // for FP LT tests less than or unordered
1824     __ br(Assembler::LT, done);
1825     // install 0 for EQ otherwise 1
1826     __ csinc(r0, zr, zr, Assembler::EQ);
1827   } else {
1828     // we want -1 for less than, 0 for equal and 1 for unordered or
1829     // greater than.
1830     __ mov(r0, 1L);
1831     // for FP HI tests greater than or unordered
1832     __ br(Assembler::HI, done);
1833     // install 0 for EQ otherwise ~0
1834     __ csinv(r0, zr, zr, Assembler::EQ);
1835 
1836   }
1837   __ bind(done);
1838 }
1839 
1840 void TemplateTable::branch(bool is_jsr, bool is_wide)
1841 {
1842   __ profile_taken_branch(r0, r1);
1843   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1844                              InvocationCounter::counter_offset();
1845   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1846                               InvocationCounter::counter_offset();
1847 
1848   // load branch displacement
1849   if (!is_wide) {
1850     __ ldrh(r2, at_bcp(1));
1851     __ rev16(r2, r2);
1852     // sign extend the 16 bit value in r2
1853     __ sbfm(r2, r2, 0, 15);
1854   } else {
1855     __ ldrw(r2, at_bcp(1));
1856     __ revw(r2, r2);
1857     // sign extend the 32 bit value in r2
1858     __ sbfm(r2, r2, 0, 31);
1859   }
1860 
1861   // Handle all the JSR stuff here, then exit.
1862   // It's much shorter and cleaner than intermingling with the non-JSR
1863   // normal-branch stuff occurring below.
1864 
1865   if (is_jsr) {
1866     // Pre-load the next target bytecode into rscratch1
1867     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1868     // compute return address as bci
1869     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1870     __ add(rscratch2, rscratch2,
1871            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1872     __ sub(r1, rbcp, rscratch2);
1873     __ push_i(r1);
1874     // Adjust the bcp by the 16-bit displacement in r2
1875     __ add(rbcp, rbcp, r2);
1876     __ dispatch_only(vtos, /*generate_poll*/true);
1877     return;
1878   }
1879 
1880   // Normal (non-jsr) branch handling
1881 
1882   // Adjust the bcp by the displacement in r2
1883   __ add(rbcp, rbcp, r2);
1884 
1885   assert(UseLoopCounter || !UseOnStackReplacement,
1886          "on-stack-replacement requires loop counters");
1887   Label backedge_counter_overflow;
1888   Label dispatch;
1889   if (UseLoopCounter) {
1890     // increment backedge counter for backward branches
1891     // r0: MDO
1892     // w1: MDO bumped taken-count
1893     // r2: target offset
1894     __ cmp(r2, zr);
1895     __ br(Assembler::GT, dispatch); // count only if backward branch
1896 
1897     // ECN: FIXME: This code smells
1898     // check if MethodCounters exists
1899     Label has_counters;
1900     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1901     __ cbnz(rscratch1, has_counters);
1902     __ push(r0);
1903     __ push(r1);
1904     __ push(r2);
1905     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1906             InterpreterRuntime::build_method_counters), rmethod);
1907     __ pop(r2);
1908     __ pop(r1);
1909     __ pop(r0);
1910     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1911     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1912     __ bind(has_counters);
1913 
1914     Label no_mdo;
1915     int increment = InvocationCounter::count_increment;
1916     if (ProfileInterpreter) {
1917       // Are we profiling?
1918       __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1919       __ cbz(r1, no_mdo);
1920       // Increment the MDO backedge counter
1921       const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1922                                          in_bytes(InvocationCounter::counter_offset()));
1923       const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1924       __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1925                                  r0, rscratch1, false, Assembler::EQ,
1926                                  UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1927       __ b(dispatch);
1928     }
1929     __ bind(no_mdo);
1930     // Increment backedge counter in MethodCounters*
1931     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1932     const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1933     __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1934                                r0, rscratch2, false, Assembler::EQ,
1935                                UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1936     __ bind(dispatch);
1937   }
1938 
1939   // Pre-load the next target bytecode into rscratch1
1940   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1941 
1942   // continue with the bytecode @ target
1943   // rscratch1: target bytecode
1944   // rbcp: target bcp
1945   __ dispatch_only(vtos, /*generate_poll*/true);
1946 
1947   if (UseLoopCounter && UseOnStackReplacement) {
1948     // invocation counter overflow
1949     __ bind(backedge_counter_overflow);
1950     __ neg(r2, r2);
1951     __ add(r2, r2, rbcp);     // branch bcp
1952     // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1953     __ call_VM(noreg,
1954                CAST_FROM_FN_PTR(address,
1955                                 InterpreterRuntime::frequency_counter_overflow),
1956                r2);
1957     __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1958 
1959     // r0: osr nmethod (osr ok) or null (osr not possible)
1960     // w1: target bytecode
1961     // r2: scratch
1962     __ cbz(r0, dispatch);     // test result -- no osr if null
1963     // nmethod may have been invalidated (VM may block upon call_VM return)
1964     __ ldrb(r2, Address(r0, nmethod::state_offset()));
1965     if (nmethod::in_use != 0)
1966       __ sub(r2, r2, nmethod::in_use);
1967     __ cbnz(r2, dispatch);
1968 
1969     // We have the address of an on stack replacement routine in r0
1970     // We need to prepare to execute the OSR method. First we must
1971     // migrate the locals and monitors off of the stack.
1972 
1973     __ mov(r19, r0);                             // save the nmethod
1974 
1975     call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1976 
1977     // r0 is OSR buffer, move it to expected parameter location
1978     __ mov(j_rarg0, r0);
1979 
1980     // remove activation
1981     // get sender esp
1982     __ ldr(esp,
1983         Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1984     // remove frame anchor
1985     __ leave();
1986     // Ensure compiled code always sees stack at proper alignment
1987     __ andr(sp, esp, -16);
1988 
1989     // and begin the OSR nmethod
1990     __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1991     __ br(rscratch1);
1992   }
1993 }
1994 
1995 
1996 void TemplateTable::if_0cmp(Condition cc)
1997 {
1998   transition(itos, vtos);
1999   // assume branch is more often taken than not (loops use backward branches)
2000   Label not_taken;
2001   if (cc == equal)
2002     __ cbnzw(r0, not_taken);
2003   else if (cc == not_equal)
2004     __ cbzw(r0, not_taken);
2005   else {
2006     __ andsw(zr, r0, r0);
2007     __ br(j_not(cc), not_taken);
2008   }
2009 
2010   branch(false, false);
2011   __ bind(not_taken);
2012   __ profile_not_taken_branch(r0);
2013 }
2014 
2015 void TemplateTable::if_icmp(Condition cc)
2016 {
2017   transition(itos, vtos);
2018   // assume branch is more often taken than not (loops use backward branches)
2019   Label not_taken;
2020   __ pop_i(r1);
2021   __ cmpw(r1, r0, Assembler::LSL);
2022   __ br(j_not(cc), not_taken);
2023   branch(false, false);
2024   __ bind(not_taken);
2025   __ profile_not_taken_branch(r0);
2026 }
2027 
2028 void TemplateTable::if_nullcmp(Condition cc)
2029 {
2030   transition(atos, vtos);
2031   // assume branch is more often taken than not (loops use backward branches)
2032   Label not_taken;
2033   if (cc == equal)
2034     __ cbnz(r0, not_taken);
2035   else
2036     __ cbz(r0, not_taken);
2037   branch(false, false);
2038   __ bind(not_taken);
2039   __ profile_not_taken_branch(r0);
2040 }
2041 
2042 void TemplateTable::if_acmp(Condition cc) {
2043   transition(atos, vtos);
2044   // assume branch is more often taken than not (loops use backward branches)
2045   Label taken, not_taken;
2046   __ pop_ptr(r1);
2047 
2048   __ profile_acmp(r2, r1, r0, r4);
2049 
2050   Register is_inline_type_mask = rscratch1;
2051   __ mov(is_inline_type_mask, markWord::inline_type_pattern);
2052 
2053   if (EnableValhalla) {
2054     __ cmp(r1, r0);
2055     __ br(Assembler::EQ, (cc == equal) ? taken : not_taken);
2056 
2057     // might be substitutable, test if either r0 or r1 is null
2058     __ andr(r2, r0, r1);
2059     __ cbz(r2, (cc == equal) ? not_taken : taken);
2060 
2061     // and both are values ?
2062     __ ldr(r2, Address(r1, oopDesc::mark_offset_in_bytes()));
2063     __ andr(r2, r2, is_inline_type_mask);
2064     __ ldr(r4, Address(r0, oopDesc::mark_offset_in_bytes()));
2065     __ andr(r4, r4, is_inline_type_mask);
2066     __ andr(r2, r2, r4);
2067     __ cmp(r2,  is_inline_type_mask);
2068     __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2069 
2070     // same value klass ?
2071     __ load_metadata(r2, r1);
2072     __ load_metadata(r4, r0);
2073     __ cmp(r2, r4);
2074     __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2075 
2076     // Know both are the same type, let's test for substitutability...
2077     if (cc == equal) {
2078       invoke_is_substitutable(r0, r1, taken, not_taken);
2079     } else {
2080       invoke_is_substitutable(r0, r1, not_taken, taken);
2081     }
2082     __ stop("Not reachable");
2083   }
2084 
2085   __ cmpoop(r1, r0);
2086   __ br(j_not(cc), not_taken);
2087   __ bind(taken);
2088   branch(false, false);
2089   __ bind(not_taken);
2090   __ profile_not_taken_branch(r0, true);
2091 }
2092 
2093 void TemplateTable::invoke_is_substitutable(Register aobj, Register bobj,
2094                                             Label& is_subst, Label& not_subst) {
2095 
2096   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::is_substitutable), aobj, bobj);
2097   // Restored... r0 answer, jmp to outcome...
2098   __ cbz(r0, not_subst);
2099   __ b(is_subst);
2100 }
2101 
2102 
2103 void TemplateTable::ret() {
2104   transition(vtos, vtos);
2105   locals_index(r1);
2106   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2107   __ profile_ret(r1, r2);
2108   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2109   __ lea(rbcp, Address(rbcp, r1));
2110   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2111   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2112 }
2113 
2114 void TemplateTable::wide_ret() {
2115   transition(vtos, vtos);
2116   locals_index_wide(r1);
2117   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2118   __ profile_ret(r1, r2);
2119   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2120   __ lea(rbcp, Address(rbcp, r1));
2121   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2122   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2123 }
2124 
2125 
2126 void TemplateTable::tableswitch() {
2127   Label default_case, continue_execution;
2128   transition(itos, vtos);
2129   // align rbcp
2130   __ lea(r1, at_bcp(BytesPerInt));
2131   __ andr(r1, r1, -BytesPerInt);
2132   // load lo & hi
2133   __ ldrw(r2, Address(r1, BytesPerInt));
2134   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2135   __ rev32(r2, r2);
2136   __ rev32(r3, r3);
2137   // check against lo & hi
2138   __ cmpw(r0, r2);
2139   __ br(Assembler::LT, default_case);
2140   __ cmpw(r0, r3);
2141   __ br(Assembler::GT, default_case);
2142   // lookup dispatch offset
2143   __ subw(r0, r0, r2);
2144   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2145   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2146   __ profile_switch_case(r0, r1, r2);
2147   // continue execution
2148   __ bind(continue_execution);
2149   __ rev32(r3, r3);
2150   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2151   __ add(rbcp, rbcp, r3, ext::sxtw);
2152   __ dispatch_only(vtos, /*generate_poll*/true);
2153   // handle default
2154   __ bind(default_case);
2155   __ profile_switch_default(r0);
2156   __ ldrw(r3, Address(r1, 0));
2157   __ b(continue_execution);
2158 }
2159 
2160 void TemplateTable::lookupswitch() {
2161   transition(itos, itos);
2162   __ stop("lookupswitch bytecode should have been rewritten");
2163 }
2164 
2165 void TemplateTable::fast_linearswitch() {
2166   transition(itos, vtos);
2167   Label loop_entry, loop, found, continue_execution;
2168   // bswap r0 so we can avoid bswapping the table entries
2169   __ rev32(r0, r0);
2170   // align rbcp
2171   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2172                                     // this instruction (change offsets
2173                                     // below)
2174   __ andr(r19, r19, -BytesPerInt);
2175   // set counter
2176   __ ldrw(r1, Address(r19, BytesPerInt));
2177   __ rev32(r1, r1);
2178   __ b(loop_entry);
2179   // table search
2180   __ bind(loop);
2181   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2182   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2183   __ cmpw(r0, rscratch1);
2184   __ br(Assembler::EQ, found);
2185   __ bind(loop_entry);
2186   __ subs(r1, r1, 1);
2187   __ br(Assembler::PL, loop);
2188   // default case
2189   __ profile_switch_default(r0);
2190   __ ldrw(r3, Address(r19, 0));
2191   __ b(continue_execution);
2192   // entry found -> get offset
2193   __ bind(found);
2194   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2195   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2196   __ profile_switch_case(r1, r0, r19);
2197   // continue execution
2198   __ bind(continue_execution);
2199   __ rev32(r3, r3);
2200   __ add(rbcp, rbcp, r3, ext::sxtw);
2201   __ ldrb(rscratch1, Address(rbcp, 0));
2202   __ dispatch_only(vtos, /*generate_poll*/true);
2203 }
2204 
2205 void TemplateTable::fast_binaryswitch() {
2206   transition(itos, vtos);
2207   // Implementation using the following core algorithm:
2208   //
2209   // int binary_search(int key, LookupswitchPair* array, int n) {
2210   //   // Binary search according to "Methodik des Programmierens" by
2211   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2212   //   int i = 0;
2213   //   int j = n;
2214   //   while (i+1 < j) {
2215   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2216   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2217   //     // where a stands for the array and assuming that the (inexisting)
2218   //     // element a[n] is infinitely big.
2219   //     int h = (i + j) >> 1;
2220   //     // i < h < j
2221   //     if (key < array[h].fast_match()) {
2222   //       j = h;
2223   //     } else {
2224   //       i = h;
2225   //     }
2226   //   }
2227   //   // R: a[i] <= key < a[i+1] or Q
2228   //   // (i.e., if key is within array, i is the correct index)
2229   //   return i;
2230   // }
2231 
2232   // Register allocation
2233   const Register key   = r0; // already set (tosca)
2234   const Register array = r1;
2235   const Register i     = r2;
2236   const Register j     = r3;
2237   const Register h     = rscratch1;
2238   const Register temp  = rscratch2;
2239 
2240   // Find array start
2241   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2242                                           // get rid of this
2243                                           // instruction (change
2244                                           // offsets below)
2245   __ andr(array, array, -BytesPerInt);
2246 
2247   // Initialize i & j
2248   __ mov(i, 0);                            // i = 0;
2249   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2250 
2251   // Convert j into native byteordering
2252   __ rev32(j, j);
2253 
2254   // And start
2255   Label entry;
2256   __ b(entry);
2257 
2258   // binary search loop
2259   {
2260     Label loop;
2261     __ bind(loop);
2262     // int h = (i + j) >> 1;
2263     __ addw(h, i, j);                           // h = i + j;
2264     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2265     // if (key < array[h].fast_match()) {
2266     //   j = h;
2267     // } else {
2268     //   i = h;
2269     // }
2270     // Convert array[h].match to native byte-ordering before compare
2271     __ ldr(temp, Address(array, h, Address::lsl(3)));
2272     __ rev32(temp, temp);
2273     __ cmpw(key, temp);
2274     // j = h if (key <  array[h].fast_match())
2275     __ csel(j, h, j, Assembler::LT);
2276     // i = h if (key >= array[h].fast_match())
2277     __ csel(i, h, i, Assembler::GE);
2278     // while (i+1 < j)
2279     __ bind(entry);
2280     __ addw(h, i, 1);          // i+1
2281     __ cmpw(h, j);             // i+1 < j
2282     __ br(Assembler::LT, loop);
2283   }
2284 
2285   // end of binary search, result index is i (must check again!)
2286   Label default_case;
2287   // Convert array[i].match to native byte-ordering before compare
2288   __ ldr(temp, Address(array, i, Address::lsl(3)));
2289   __ rev32(temp, temp);
2290   __ cmpw(key, temp);
2291   __ br(Assembler::NE, default_case);
2292 
2293   // entry found -> j = offset
2294   __ add(j, array, i, ext::uxtx, 3);
2295   __ ldrw(j, Address(j, BytesPerInt));
2296   __ profile_switch_case(i, key, array);
2297   __ rev32(j, j);
2298   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2299   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2300   __ dispatch_only(vtos, /*generate_poll*/true);
2301 
2302   // default case -> j = default offset
2303   __ bind(default_case);
2304   __ profile_switch_default(i);
2305   __ ldrw(j, Address(array, -2 * BytesPerInt));
2306   __ rev32(j, j);
2307   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2308   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2309   __ dispatch_only(vtos, /*generate_poll*/true);
2310 }
2311 
2312 
2313 void TemplateTable::_return(TosState state)
2314 {
2315   transition(state, state);
2316   assert(_desc->calls_vm(),
2317          "inconsistent calls_vm information"); // call in remove_activation
2318 
2319   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2320     assert(state == vtos, "only valid state");
2321 
2322     __ ldr(c_rarg1, aaddress(0));
2323     __ load_klass(r3, c_rarg1);
2324     __ ldrb(r3, Address(r3, Klass::misc_flags_offset()));
2325     Label skip_register_finalizer;
2326     __ tbz(r3, exact_log2(KlassFlags::_misc_has_finalizer), skip_register_finalizer);
2327 
2328     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2329 
2330     __ bind(skip_register_finalizer);
2331   }
2332 
2333   // Issue a StoreStore barrier after all stores but before return
2334   // from any constructor for any class with a final field.  We don't
2335   // know if this is a finalizer, so we always do so.
2336   if (_desc->bytecode() == Bytecodes::_return)
2337     __ membar(MacroAssembler::StoreStore);
2338 
2339   if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2340     Label no_safepoint;
2341     __ ldr(rscratch1, Address(rthread, JavaThread::polling_word_offset()));
2342     __ tbz(rscratch1, log2i_exact(SafepointMechanism::poll_bit()), no_safepoint);
2343     __ push(state);
2344     __ push_cont_fastpath(rthread);
2345     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint));
2346     __ pop_cont_fastpath(rthread);
2347     __ pop(state);
2348     __ bind(no_safepoint);
2349   }
2350 
2351   // Narrow result if state is itos but result type is smaller.
2352   // Need to narrow in the return bytecode rather than in generate_return_entry
2353   // since compiled code callers expect the result to already be narrowed.
2354   if (state == itos) {
2355     __ narrow(r0);
2356   }
2357 
2358   __ remove_activation(state);
2359   __ ret(lr);
2360 }
2361 
2362 // ----------------------------------------------------------------------------
2363 // Volatile variables demand their effects be made known to all CPU's
2364 // in order.  Store buffers on most chips allow reads & writes to
2365 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2366 // without some kind of memory barrier (i.e., it's not sufficient that
2367 // the interpreter does not reorder volatile references, the hardware
2368 // also must not reorder them).
2369 //
2370 // According to the new Java Memory Model (JMM):
2371 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2372 //     writes act as acquire & release, so:
2373 // (2) A read cannot let unrelated NON-volatile memory refs that
2374 //     happen after the read float up to before the read.  It's OK for
2375 //     non-volatile memory refs that happen before the volatile read to
2376 //     float down below it.
2377 // (3) Similar a volatile write cannot let unrelated NON-volatile
2378 //     memory refs that happen BEFORE the write float down to after the
2379 //     write.  It's OK for non-volatile memory refs that happen after the
2380 //     volatile write to float up before it.
2381 //
2382 // We only put in barriers around volatile refs (they are expensive),
2383 // not _between_ memory refs (that would require us to track the
2384 // flavor of the previous memory refs).  Requirements (2) and (3)
2385 // require some barriers before volatile stores and after volatile
2386 // loads.  These nearly cover requirement (1) but miss the
2387 // volatile-store-volatile-load case.  This final case is placed after
2388 // volatile-stores although it could just as well go before
2389 // volatile-loads.
2390 
2391 void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
2392                                             Register Rcache,
2393                                             Register index) {
2394   const Register temp = r19;
2395   assert_different_registers(Rcache, index, temp);
2396   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2397 
2398   Label resolved, clinit_barrier_slow;
2399 
2400   Bytecodes::Code code = bytecode();
2401   __ load_method_entry(Rcache, index);
2402   switch(byte_no) {
2403     case f1_byte:
2404       __ lea(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::bytecode1_offset())));
2405       break;
2406     case f2_byte:
2407       __ lea(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::bytecode2_offset())));
2408       break;
2409   }
2410   // Load-acquire the bytecode to match store-release in InterpreterRuntime
2411   __ ldarb(temp, temp);
2412   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2413   __ br(Assembler::EQ, resolved);
2414 
2415   // resolve first time through
2416   // Class initialization barrier slow path lands here as well.
2417   __ bind(clinit_barrier_slow);
2418   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2419   __ mov(temp, (int) code);
2420   __ call_VM(noreg, entry, temp);
2421 
2422   // Update registers with resolved info
2423   __ load_method_entry(Rcache, index);
2424   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2425   // so all clients ofthis method must be modified accordingly
2426   __ bind(resolved);
2427 
2428   // Class initialization barrier for static methods
2429   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2430     __ ldr(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::method_offset())));
2431     __ load_method_holder(temp, temp);
2432     __ clinit_barrier(temp, rscratch1, nullptr, &clinit_barrier_slow);
2433   }
2434 }
2435 
2436 void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
2437                                             Register Rcache,
2438                                             Register index) {
2439   const Register temp = r19;
2440   assert_different_registers(Rcache, index, temp);
2441 
2442   Label resolved;
2443 
2444   Bytecodes::Code code = bytecode();
2445   switch (code) {
2446   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2447   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2448   default: break;
2449   }
2450 
2451   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2452   __ load_field_entry(Rcache, index);
2453   if (byte_no == f1_byte) {
2454     __ lea(temp, Address(Rcache, in_bytes(ResolvedFieldEntry::get_code_offset())));
2455   } else {
2456     __ lea(temp, Address(Rcache, in_bytes(ResolvedFieldEntry::put_code_offset())));
2457   }
2458   // Load-acquire the bytecode to match store-release in ResolvedFieldEntry::fill_in()
2459   __ ldarb(temp, temp);
2460   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2461   __ br(Assembler::EQ, resolved);
2462 
2463   // resolve first time through
2464   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2465   __ mov(temp, (int) code);
2466   __ call_VM(noreg, entry, temp);
2467 
2468   // Update registers with resolved info
2469   __ load_field_entry(Rcache, index);
2470   __ bind(resolved);
2471 }
2472 
2473 void TemplateTable::load_resolved_field_entry(Register obj,
2474                                               Register cache,
2475                                               Register tos_state,
2476                                               Register offset,
2477                                               Register flags,
2478                                               bool is_static = false) {
2479   assert_different_registers(cache, tos_state, flags, offset);
2480 
2481   // Field offset
2482   __ load_sized_value(offset, Address(cache, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
2483 
2484   // Flags
2485   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedFieldEntry::flags_offset())));
2486 
2487   // TOS state
2488   if (tos_state != noreg) {
2489     __ load_unsigned_byte(tos_state, Address(cache, in_bytes(ResolvedFieldEntry::type_offset())));
2490   }
2491 
2492   // Klass overwrite register
2493   if (is_static) {
2494     __ ldr(obj, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2495     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2496     __ ldr(obj, Address(obj, mirror_offset));
2497     __ resolve_oop_handle(obj, r5, rscratch2);
2498   }
2499 }
2500 
2501 void TemplateTable::load_resolved_method_entry_special_or_static(Register cache,
2502                                                                  Register method,
2503                                                                  Register flags) {
2504 
2505   // setup registers
2506   const Register index = flags;
2507   assert_different_registers(method, cache, flags);
2508 
2509   // determine constant pool cache field offsets
2510   resolve_cache_and_index_for_method(f1_byte, cache, index);
2511   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2512   __ ldr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2513 }
2514 
2515 void TemplateTable::load_resolved_method_entry_handle(Register cache,
2516                                                       Register method,
2517                                                       Register ref_index,
2518                                                       Register flags) {
2519   // setup registers
2520   const Register index = ref_index;
2521   assert_different_registers(method, flags);
2522   assert_different_registers(method, cache, index);
2523 
2524   // determine constant pool cache field offsets
2525   resolve_cache_and_index_for_method(f1_byte, cache, index);
2526   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2527 
2528   // maybe push appendix to arguments (just before return address)
2529   Label L_no_push;
2530   __ tbz(flags, ResolvedMethodEntry::has_appendix_shift, L_no_push);
2531   // invokehandle uses an index into the resolved references array
2532   __ load_unsigned_short(ref_index, Address(cache, in_bytes(ResolvedMethodEntry::resolved_references_index_offset())));
2533   // Push the appendix as a trailing parameter.
2534   // This must be done before we get the receiver,
2535   // since the parameter_size includes it.
2536   Register appendix = method;
2537   __ load_resolved_reference_at_index(appendix, ref_index);
2538   __ push(appendix);  // push appendix (MethodType, CallSite, etc.)
2539   __ bind(L_no_push);
2540 
2541   __ ldr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2542 }
2543 
2544 void TemplateTable::load_resolved_method_entry_interface(Register cache,
2545                                                          Register klass,
2546                                                          Register method_or_table_index,
2547                                                          Register flags) {
2548   // setup registers
2549   const Register index = method_or_table_index;
2550   assert_different_registers(method_or_table_index, cache, flags);
2551 
2552   // determine constant pool cache field offsets
2553   resolve_cache_and_index_for_method(f1_byte, cache, index);
2554   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2555 
2556   // Invokeinterface can behave in different ways:
2557   // If calling a method from java.lang.Object, the forced virtual flag is true so the invocation will
2558   // behave like an invokevirtual call. The state of the virtual final flag will determine whether a method or
2559   // vtable index is placed in the register.
2560   // Otherwise, the registers will be populated with the klass and method.
2561 
2562   Label NotVirtual; Label NotVFinal; Label Done;
2563   __ tbz(flags, ResolvedMethodEntry::is_forced_virtual_shift, NotVirtual);
2564   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, NotVFinal);
2565   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2566   __ b(Done);
2567 
2568   __ bind(NotVFinal);
2569   __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2570   __ b(Done);
2571 
2572   __ bind(NotVirtual);
2573   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2574   __ ldr(klass, Address(cache, in_bytes(ResolvedMethodEntry::klass_offset())));
2575   __ bind(Done);
2576 }
2577 
2578 void TemplateTable::load_resolved_method_entry_virtual(Register cache,
2579                                                        Register method_or_table_index,
2580                                                        Register flags) {
2581   // setup registers
2582   const Register index = flags;
2583   assert_different_registers(method_or_table_index, cache, flags);
2584 
2585   // determine constant pool cache field offsets
2586   resolve_cache_and_index_for_method(f2_byte, cache, index);
2587   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2588 
2589   // method_or_table_index can either be an itable index or a method depending on the virtual final flag
2590   Label NotVFinal; Label Done;
2591   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, NotVFinal);
2592   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2593   __ b(Done);
2594 
2595   __ bind(NotVFinal);
2596   __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2597   __ bind(Done);
2598 }
2599 
2600 // The rmethod register is input and overwritten to be the adapter method for the
2601 // indy call. Link Register (lr) is set to the return address for the adapter and
2602 // an appendix may be pushed to the stack. Registers r0-r3 are clobbered
2603 void TemplateTable::load_invokedynamic_entry(Register method) {
2604   // setup registers
2605   const Register appendix = r0;
2606   const Register cache = r2;
2607   const Register index = r3;
2608   assert_different_registers(method, appendix, cache, index, rcpool);
2609 
2610   __ save_bcp();
2611 
2612   Label resolved;
2613 
2614   __ load_resolved_indy_entry(cache, index);
2615   // Load-acquire the adapter method to match store-release in ResolvedIndyEntry::fill_in()
2616   __ lea(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2617   __ ldar(method, method);
2618 
2619   // Compare the method to zero
2620   __ cbnz(method, resolved);
2621 
2622   Bytecodes::Code code = bytecode();
2623 
2624   // Call to the interpreter runtime to resolve invokedynamic
2625   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2626   __ mov(method, code); // this is essentially Bytecodes::_invokedynamic
2627   __ call_VM(noreg, entry, method);
2628   // Update registers with resolved info
2629   __ load_resolved_indy_entry(cache, index);
2630   // Load-acquire the adapter method to match store-release in ResolvedIndyEntry::fill_in()
2631   __ lea(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2632   __ ldar(method, method);
2633 
2634 #ifdef ASSERT
2635   __ cbnz(method, resolved);
2636   __ stop("Should be resolved by now");
2637 #endif // ASSERT
2638   __ bind(resolved);
2639 
2640   Label L_no_push;
2641   // Check if there is an appendix
2642   __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::flags_offset())));
2643   __ tbz(index, ResolvedIndyEntry::has_appendix_shift, L_no_push);
2644 
2645   // Get appendix
2646   __ load_unsigned_short(index, Address(cache, in_bytes(ResolvedIndyEntry::resolved_references_index_offset())));
2647   // Push the appendix as a trailing parameter
2648   // since the parameter_size includes it.
2649   __ push(method);
2650   __ mov(method, index);
2651   __ load_resolved_reference_at_index(appendix, method);
2652   __ verify_oop(appendix);
2653   __ pop(method);
2654   __ push(appendix);  // push appendix (MethodType, CallSite, etc.)
2655   __ bind(L_no_push);
2656 
2657   // compute return type
2658   __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::result_type_offset())));
2659   // load return address
2660   // Return address is loaded into link register(lr) and not pushed to the stack
2661   // like x86
2662   {
2663     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2664     __ mov(rscratch1, table_addr);
2665     __ ldr(lr, Address(rscratch1, index, Address::lsl(3)));
2666   }
2667 }
2668 
2669 // The registers cache and index expected to be set before call.
2670 // Correct values of the cache and index registers are preserved.
2671 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2672                                             bool is_static, bool has_tos) {
2673   // do the JVMTI work here to avoid disturbing the register state below
2674   // We use c_rarg registers here because we want to use the register used in
2675   // the call to the VM
2676   if (JvmtiExport::can_post_field_access()) {
2677     // Check to see if a field access watch has been set before we
2678     // take the time to call into the VM.
2679     Label L1;
2680     assert_different_registers(cache, index, r0);
2681     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2682     __ ldrw(r0, Address(rscratch1));
2683     __ cbzw(r0, L1);
2684 
2685     __ load_field_entry(c_rarg2, index);
2686 
2687     if (is_static) {
2688       __ mov(c_rarg1, zr); // null object reference
2689     } else {
2690       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2691       __ verify_oop(c_rarg1);
2692     }
2693     // c_rarg1: object pointer or null
2694     // c_rarg2: cache entry pointer
2695     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2696                                        InterpreterRuntime::post_field_access),
2697                c_rarg1, c_rarg2);
2698     __ load_field_entry(cache, index);
2699     __ bind(L1);
2700   }
2701 }
2702 
2703 void TemplateTable::pop_and_check_object(Register r)
2704 {
2705   __ pop_ptr(r);
2706   __ null_check(r);  // for field access must check obj.
2707   __ verify_oop(r);
2708 }
2709 
2710 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2711 {
2712   const Register cache     = r2;
2713   const Register obj       = r4;
2714   const Register klass     = r5;
2715   const Register inline_klass = r7;
2716   const Register field_index = r23;
2717   const Register index     = r3;
2718   const Register tos_state = r3;
2719   const Register off       = r19;
2720   const Register flags     = r6;
2721   const Register bc        = r4; // uses same reg as obj, so don't mix them
2722 
2723   resolve_cache_and_index_for_field(byte_no, cache, index);
2724   jvmti_post_field_access(cache, index, is_static, false);
2725 
2726   // Valhalla extras
2727   __ load_unsigned_short(field_index, Address(cache, in_bytes(ResolvedFieldEntry::field_index_offset())));
2728   __ ldr(klass, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2729 
2730   load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
2731 
2732   if (!is_static) {
2733     // obj is on the stack
2734     pop_and_check_object(obj);
2735   }
2736 
2737   // 8179954: We need to make sure that the code generated for
2738   // volatile accesses forms a sequentially-consistent set of
2739   // operations when combined with STLR and LDAR.  Without a leading
2740   // membar it's possible for a simple Dekker test to fail if loads
2741   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2742   // the stores in one method and we interpret the loads in another.
2743   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()){
2744     Label notVolatile;
2745     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
2746     __ membar(MacroAssembler::AnyAny);
2747     __ bind(notVolatile);
2748   }
2749 
2750   const Address field(obj, off);
2751 
2752   Label Done, notByte, notBool, notInt, notShort, notChar,
2753               notLong, notFloat, notObj, notDouble;
2754 
2755   assert(btos == 0, "change code, btos != 0");
2756   __ cbnz(tos_state, notByte);
2757 
2758   // Don't rewrite getstatic, only getfield
2759   if (is_static) rc = may_not_rewrite;
2760 
2761   // btos
2762   __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
2763   __ push(btos);
2764   // Rewrite bytecode to be faster
2765   if (rc == may_rewrite) {
2766     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2767   }
2768   __ b(Done);
2769 
2770   __ bind(notByte);
2771   __ cmp(tos_state, (u1)ztos);
2772   __ br(Assembler::NE, notBool);
2773 
2774   // ztos (same code as btos)
2775   __ access_load_at(T_BOOLEAN, IN_HEAP, r0, field, noreg, noreg);
2776   __ push(ztos);
2777   // Rewrite bytecode to be faster
2778   if (rc == may_rewrite) {
2779     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2780     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2781   }
2782   __ b(Done);
2783 
2784   __ bind(notBool);
2785   __ cmp(tos_state, (u1)atos);
2786   __ br(Assembler::NE, notObj);
2787   // atos
2788   if (!EnableValhalla) {
2789     do_oop_load(_masm, field, r0, IN_HEAP);
2790     __ push(atos);
2791     if (rc == may_rewrite) {
2792       patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2793     }
2794     __ b(Done);
2795   } else { // Valhalla
2796     if (is_static) {
2797       __ load_heap_oop(r0, field, rscratch1, rscratch2);
2798       Label is_null_free_inline_type, uninitialized;
2799       // Issue below if the static field has not been initialized yet
2800       __ test_field_is_null_free_inline_type(flags, noreg /*temp*/, is_null_free_inline_type);
2801         // field is not a null free inline type
2802         __ push(atos);
2803         __ b(Done);
2804       // field is a null free inline type, must not return null even if uninitialized
2805       __ bind(is_null_free_inline_type);
2806         __ cbz(r0, uninitialized);
2807           __ push(atos);
2808           __ b(Done);
2809         __ bind(uninitialized);
2810           Label slow_case, finish;
2811           __ ldrb(rscratch1, Address(klass, InstanceKlass::init_state_offset()));
2812           __ cmp(rscratch1, (u1)InstanceKlass::fully_initialized);
2813           __ br(Assembler::NE, slow_case);
2814           __ get_default_value_oop(klass, off /* temp */, r0);
2815         __ b(finish);
2816         __ bind(slow_case);
2817           __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_inline_type_field), obj, cache);
2818           __ bind(finish);
2819           __ verify_oop(r0);
2820           __ push(atos);
2821           __ b(Done);
2822     } else {
2823       Label is_flat, nonnull, is_inline_type, has_null_marker, rewrite_inline;
2824       __ test_field_is_null_free_inline_type(flags, noreg /*temp*/, is_inline_type);
2825       __ test_field_has_null_marker(flags, noreg /*temp*/, has_null_marker);
2826         // Non-inline field case
2827         __ load_heap_oop(r0, field, rscratch1, rscratch2);
2828         __ push(atos);
2829         if (rc == may_rewrite) {
2830           patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2831         }
2832         __ b(Done);
2833       __ bind(is_inline_type);
2834         __ test_field_is_flat(flags, noreg /* temp */, is_flat);
2835          // field is not flat
2836           __ load_heap_oop(r0, field, rscratch1, rscratch2);
2837           __ cbnz(r0, nonnull);
2838             __ get_inline_type_field_klass(klass, field_index, inline_klass);
2839             __ get_default_value_oop(inline_klass, klass /* temp */, r0);
2840           __ bind(nonnull);
2841           __ verify_oop(r0);
2842           __ push(atos);
2843           __ b(rewrite_inline);
2844         __ bind(is_flat);
2845         // field is flat
2846           __ mov(r0, obj);
2847           __ read_flat_field(cache, field_index, off, inline_klass /* temp */, r0);
2848           __ verify_oop(r0);
2849           __ push(atos);
2850           __ b(rewrite_inline);
2851         __ bind(has_null_marker);
2852           call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), obj, cache);
2853           __ verify_oop(r0);
2854           __ push(atos);
2855       __ bind(rewrite_inline);
2856       if (rc == may_rewrite) {
2857         patch_bytecode(Bytecodes::_fast_vgetfield, bc, r1);
2858       }
2859       __ b(Done);
2860     }
2861   }
2862 
2863   __ bind(notObj);
2864   __ cmp(tos_state, (u1)itos);
2865   __ br(Assembler::NE, notInt);
2866   // itos
2867   __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
2868   __ push(itos);
2869   // Rewrite bytecode to be faster
2870   if (rc == may_rewrite) {
2871     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2872   }
2873   __ b(Done);
2874 
2875   __ bind(notInt);
2876   __ cmp(tos_state, (u1)ctos);
2877   __ br(Assembler::NE, notChar);
2878   // ctos
2879   __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
2880   __ push(ctos);
2881   // Rewrite bytecode to be faster
2882   if (rc == may_rewrite) {
2883     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2884   }
2885   __ b(Done);
2886 
2887   __ bind(notChar);
2888   __ cmp(tos_state, (u1)stos);
2889   __ br(Assembler::NE, notShort);
2890   // stos
2891   __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
2892   __ push(stos);
2893   // Rewrite bytecode to be faster
2894   if (rc == may_rewrite) {
2895     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2896   }
2897   __ b(Done);
2898 
2899   __ bind(notShort);
2900   __ cmp(tos_state, (u1)ltos);
2901   __ br(Assembler::NE, notLong);
2902   // ltos
2903   __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
2904   __ push(ltos);
2905   // Rewrite bytecode to be faster
2906   if (rc == may_rewrite) {
2907     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2908   }
2909   __ b(Done);
2910 
2911   __ bind(notLong);
2912   __ cmp(tos_state, (u1)ftos);
2913   __ br(Assembler::NE, notFloat);
2914   // ftos
2915   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2916   __ push(ftos);
2917   // Rewrite bytecode to be faster
2918   if (rc == may_rewrite) {
2919     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2920   }
2921   __ b(Done);
2922 
2923   __ bind(notFloat);
2924 #ifdef ASSERT
2925   __ cmp(tos_state, (u1)dtos);
2926   __ br(Assembler::NE, notDouble);
2927 #endif
2928   // dtos
2929   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2930   __ push(dtos);
2931   // Rewrite bytecode to be faster
2932   if (rc == may_rewrite) {
2933     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2934   }
2935 #ifdef ASSERT
2936   __ b(Done);
2937 
2938   __ bind(notDouble);
2939   __ stop("Bad state");
2940 #endif
2941 
2942   __ bind(Done);
2943 
2944   Label notVolatile;
2945   __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
2946   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2947   __ bind(notVolatile);
2948 }
2949 
2950 
2951 void TemplateTable::getfield(int byte_no)
2952 {
2953   getfield_or_static(byte_no, false);
2954 }
2955 
2956 void TemplateTable::nofast_getfield(int byte_no) {
2957   getfield_or_static(byte_no, false, may_not_rewrite);
2958 }
2959 
2960 void TemplateTable::getstatic(int byte_no)
2961 {
2962   getfield_or_static(byte_no, true);
2963 }
2964 
2965 // The registers cache and index expected to be set before call.
2966 // The function may destroy various registers, just not the cache and index registers.
2967 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2968   transition(vtos, vtos);
2969 
2970   if (JvmtiExport::can_post_field_modification()) {
2971     // Check to see if a field modification watch has been set before
2972     // we take the time to call into the VM.
2973     Label L1;
2974     assert_different_registers(cache, index, r0);
2975     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2976     __ ldrw(r0, Address(rscratch1));
2977     __ cbz(r0, L1);
2978 
2979     __ mov(c_rarg2, cache);
2980 
2981     if (is_static) {
2982       // Life is simple.  Null out the object pointer.
2983       __ mov(c_rarg1, zr);
2984     } else {
2985       // Life is harder. The stack holds the value on top, followed by
2986       // the object.  We don't know the size of the value, though; it
2987       // could be one or two words depending on its type. As a result,
2988       // we must find the type to determine where the object is.
2989       __ load_unsigned_byte(c_rarg3, Address(c_rarg2, in_bytes(ResolvedFieldEntry::type_offset())));
2990       Label nope2, done, ok;
2991       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2992       __ cmpw(c_rarg3, ltos);
2993       __ br(Assembler::EQ, ok);
2994       __ cmpw(c_rarg3, dtos);
2995       __ br(Assembler::NE, nope2);
2996       __ bind(ok);
2997       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2998       __ bind(nope2);
2999     }
3000     // object (tos)
3001     __ mov(c_rarg3, esp);
3002     // c_rarg1: object pointer set up above (null if static)
3003     // c_rarg2: cache entry pointer
3004     // c_rarg3: jvalue object on the stack
3005     __ call_VM(noreg,
3006                CAST_FROM_FN_PTR(address,
3007                                 InterpreterRuntime::post_field_modification),
3008                c_rarg1, c_rarg2, c_rarg3);
3009     __ load_field_entry(cache, index);
3010     __ bind(L1);
3011   }
3012 }
3013 
3014 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3015   transition(vtos, vtos);
3016 
3017   const Register cache     = r2;
3018   const Register index     = r3;
3019   const Register tos_state = r3;
3020   const Register obj       = r2;
3021   const Register off       = r19;
3022   const Register flags     = r6;
3023   const Register bc        = r4;
3024   const Register inline_klass = r5;
3025 
3026   resolve_cache_and_index_for_field(byte_no, cache, index);
3027   jvmti_post_field_mod(cache, index, is_static);
3028   load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3029 
3030   Label Done;
3031   {
3032     Label notVolatile;
3033     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3034     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
3035     __ bind(notVolatile);
3036   }
3037 
3038   // field address
3039   const Address field(obj, off);
3040 
3041   Label notByte, notBool, notInt, notShort, notChar,
3042         notLong, notFloat, notObj, notDouble;
3043 
3044   assert(btos == 0, "change code, btos != 0");
3045   __ cbnz(tos_state, notByte);
3046 
3047   // Don't rewrite putstatic, only putfield
3048   if (is_static) rc = may_not_rewrite;
3049 
3050   // btos
3051   {
3052     __ pop(btos);
3053     if (!is_static) pop_and_check_object(obj);
3054     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
3055     if (rc == may_rewrite) {
3056       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
3057     }
3058     __ b(Done);
3059   }
3060 
3061   __ bind(notByte);
3062   __ cmp(tos_state, (u1)ztos);
3063   __ br(Assembler::NE, notBool);
3064 
3065   // ztos
3066   {
3067     __ pop(ztos);
3068     if (!is_static) pop_and_check_object(obj);
3069     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
3070     if (rc == may_rewrite) {
3071       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
3072     }
3073     __ b(Done);
3074   }
3075 
3076   __ bind(notBool);
3077   __ cmp(tos_state, (u1)atos);
3078   __ br(Assembler::NE, notObj);
3079 
3080   // atos
3081   {
3082      if (!EnableValhalla) {
3083       __ pop(atos);
3084       if (!is_static) pop_and_check_object(obj);
3085       // Store into the field
3086       do_oop_store(_masm, field, r0, IN_HEAP);
3087       if (rc == may_rewrite) {
3088         patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
3089       }
3090       __ b(Done);
3091      } else { // Valhalla
3092       __ pop(atos);
3093       if (is_static) {
3094         Label is_inline_type;
3095          __ test_field_is_not_null_free_inline_type(flags, noreg /* temp */, is_inline_type);
3096          __ null_check(r0);
3097          __ bind(is_inline_type);
3098          do_oop_store(_masm, field, r0, IN_HEAP);
3099          __ b(Done);
3100       } else {
3101         Label is_inline_type, is_flat, has_null_marker, rewrite_not_inline, rewrite_inline;
3102         __ test_field_is_null_free_inline_type(flags, noreg /*temp*/, is_inline_type);
3103         __ test_field_has_null_marker(flags, noreg /*temp*/, has_null_marker);
3104         // Not an inline type
3105         pop_and_check_object(obj);
3106         // Store into the field
3107         do_oop_store(_masm, field, r0, IN_HEAP);
3108         __ bind(rewrite_not_inline);
3109         if (rc == may_rewrite) {
3110           patch_bytecode(Bytecodes::_fast_aputfield, bc, r19, true, byte_no);
3111         }
3112         __ b(Done);
3113         // Implementation of the inline type semantic
3114         __ bind(is_inline_type);
3115         __ null_check(r0);
3116         __ test_field_is_flat(flags, noreg /*temp*/, is_flat);
3117         // field is not flat
3118         pop_and_check_object(obj);
3119         // Store into the field
3120         do_oop_store(_masm, field, r0, IN_HEAP);
3121         __ b(rewrite_inline);
3122         __ bind(is_flat);
3123         __ load_field_entry(cache, index); // reload field entry (cache) because it was erased by tos_state
3124         __ load_unsigned_short(index, Address(cache, in_bytes(ResolvedFieldEntry::field_index_offset())));
3125         __ ldr(r2, Address(cache, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3126         __ inline_layout_info(r2, index, r6);
3127         pop_and_check_object(obj);
3128         __ load_klass(inline_klass, r0);
3129         __ data_for_oop(r0, r0, inline_klass);
3130         __ add(obj, obj, off);
3131         // because we use InlineLayoutInfo, we need special value access code specialized for fields (arrays will need a different API)
3132         __ flat_field_copy(IN_HEAP, r0, obj, r6);
3133         __ b(rewrite_inline);
3134         __ bind(has_null_marker);
3135         assert_different_registers(r0, cache, r19);
3136         pop_and_check_object(r19);
3137         __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), r19, r0, cache);
3138         __ bind(rewrite_inline);
3139         if (rc == may_rewrite) {
3140           patch_bytecode(Bytecodes::_fast_vputfield, bc, r19, true, byte_no);
3141         }
3142         __ b(Done);
3143       }
3144      }  // Valhalla
3145   }
3146 
3147   __ bind(notObj);
3148   __ cmp(tos_state, (u1)itos);
3149   __ br(Assembler::NE, notInt);
3150 
3151   // itos
3152   {
3153     __ pop(itos);
3154     if (!is_static) pop_and_check_object(obj);
3155     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
3156     if (rc == may_rewrite) {
3157       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
3158     }
3159     __ b(Done);
3160   }
3161 
3162   __ bind(notInt);
3163   __ cmp(tos_state, (u1)ctos);
3164   __ br(Assembler::NE, notChar);
3165 
3166   // ctos
3167   {
3168     __ pop(ctos);
3169     if (!is_static) pop_and_check_object(obj);
3170     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
3171     if (rc == may_rewrite) {
3172       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
3173     }
3174     __ b(Done);
3175   }
3176 
3177   __ bind(notChar);
3178   __ cmp(tos_state, (u1)stos);
3179   __ br(Assembler::NE, notShort);
3180 
3181   // stos
3182   {
3183     __ pop(stos);
3184     if (!is_static) pop_and_check_object(obj);
3185     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
3186     if (rc == may_rewrite) {
3187       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
3188     }
3189     __ b(Done);
3190   }
3191 
3192   __ bind(notShort);
3193   __ cmp(tos_state, (u1)ltos);
3194   __ br(Assembler::NE, notLong);
3195 
3196   // ltos
3197   {
3198     __ pop(ltos);
3199     if (!is_static) pop_and_check_object(obj);
3200     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
3201     if (rc == may_rewrite) {
3202       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
3203     }
3204     __ b(Done);
3205   }
3206 
3207   __ bind(notLong);
3208   __ cmp(tos_state, (u1)ftos);
3209   __ br(Assembler::NE, notFloat);
3210 
3211   // ftos
3212   {
3213     __ pop(ftos);
3214     if (!is_static) pop_and_check_object(obj);
3215     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3216     if (rc == may_rewrite) {
3217       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
3218     }
3219     __ b(Done);
3220   }
3221 
3222   __ bind(notFloat);
3223 #ifdef ASSERT
3224   __ cmp(tos_state, (u1)dtos);
3225   __ br(Assembler::NE, notDouble);
3226 #endif
3227 
3228   // dtos
3229   {
3230     __ pop(dtos);
3231     if (!is_static) pop_and_check_object(obj);
3232     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
3233     if (rc == may_rewrite) {
3234       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
3235     }
3236   }
3237 
3238 #ifdef ASSERT
3239   __ b(Done);
3240 
3241   __ bind(notDouble);
3242   __ stop("Bad state");
3243 #endif
3244 
3245   __ bind(Done);
3246 
3247   {
3248     Label notVolatile;
3249     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3250     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3251     __ bind(notVolatile);
3252   }
3253 }
3254 
3255 void TemplateTable::putfield(int byte_no)
3256 {
3257   putfield_or_static(byte_no, false);
3258 }
3259 
3260 void TemplateTable::nofast_putfield(int byte_no) {
3261   putfield_or_static(byte_no, false, may_not_rewrite);
3262 }
3263 
3264 void TemplateTable::putstatic(int byte_no) {
3265   putfield_or_static(byte_no, true);
3266 }
3267 
3268 void TemplateTable::jvmti_post_fast_field_mod() {
3269   if (JvmtiExport::can_post_field_modification()) {
3270     // Check to see if a field modification watch has been set before
3271     // we take the time to call into the VM.
3272     Label L2;
3273     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3274     __ ldrw(c_rarg3, Address(rscratch1));
3275     __ cbzw(c_rarg3, L2);
3276     __ pop_ptr(r19);                  // copy the object pointer from tos
3277     __ verify_oop(r19);
3278     __ push_ptr(r19);                 // put the object pointer back on tos
3279     // Save tos values before call_VM() clobbers them. Since we have
3280     // to do it for every data type, we use the saved values as the
3281     // jvalue object.
3282     switch (bytecode()) {          // load values into the jvalue object
3283     case Bytecodes::_fast_vputfield: //fall through
3284     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
3285     case Bytecodes::_fast_bputfield: // fall through
3286     case Bytecodes::_fast_zputfield: // fall through
3287     case Bytecodes::_fast_sputfield: // fall through
3288     case Bytecodes::_fast_cputfield: // fall through
3289     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
3290     case Bytecodes::_fast_dputfield: __ push_d(); break;
3291     case Bytecodes::_fast_fputfield: __ push_f(); break;
3292     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
3293 
3294     default:
3295       ShouldNotReachHere();
3296     }
3297     __ mov(c_rarg3, esp);             // points to jvalue on the stack
3298     // access constant pool cache entry
3299     __ load_field_entry(c_rarg2, r0);
3300     __ verify_oop(r19);
3301     // r19: object pointer copied above
3302     // c_rarg2: cache entry pointer
3303     // c_rarg3: jvalue object on the stack
3304     __ call_VM(noreg,
3305                CAST_FROM_FN_PTR(address,
3306                                 InterpreterRuntime::post_field_modification),
3307                r19, c_rarg2, c_rarg3);
3308 
3309     switch (bytecode()) {             // restore tos values
3310     case Bytecodes::_fast_vputfield: //fall through
3311     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
3312     case Bytecodes::_fast_bputfield: // fall through
3313     case Bytecodes::_fast_zputfield: // fall through
3314     case Bytecodes::_fast_sputfield: // fall through
3315     case Bytecodes::_fast_cputfield: // fall through
3316     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
3317     case Bytecodes::_fast_dputfield: __ pop_d(); break;
3318     case Bytecodes::_fast_fputfield: __ pop_f(); break;
3319     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
3320     default: break;
3321     }
3322     __ bind(L2);
3323   }
3324 }
3325 
3326 void TemplateTable::fast_storefield(TosState state)
3327 {
3328   transition(state, vtos);
3329 
3330   ByteSize base = ConstantPoolCache::base_offset();
3331 
3332   jvmti_post_fast_field_mod();
3333 
3334   // access constant pool cache
3335   __ load_field_entry(r2, r1);
3336 
3337   // R1: field offset, R2: field holder, R3: flags
3338   load_resolved_field_entry(r2, r2, noreg, r1, r3);
3339 
3340   {
3341     Label notVolatile;
3342     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3343     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
3344     __ bind(notVolatile);
3345   }
3346 
3347   Label notVolatile;
3348 
3349   // Get object from stack
3350   pop_and_check_object(r2);
3351 
3352   // field address
3353   const Address field(r2, r1);
3354 
3355   // access field
3356   switch (bytecode()) {
3357   case Bytecodes::_fast_vputfield:
3358    {
3359       Label is_flat, has_null_marker, done;
3360       __ test_field_has_null_marker(r3, noreg /* temp */, has_null_marker);
3361       __ null_check(r0);
3362       __ test_field_is_flat(r3, noreg /* temp */, is_flat);
3363       // field is not flat
3364       do_oop_store(_masm, field, r0, IN_HEAP);
3365       __ b(done);
3366       __ bind(is_flat);
3367       // field is flat
3368       __ load_field_entry(r4, r3);
3369       __ load_unsigned_short(r3, Address(r4, in_bytes(ResolvedFieldEntry::field_index_offset())));
3370       __ ldr(r4, Address(r4, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3371       __ inline_layout_info(r4, r3, r5);
3372       __ load_klass(r4, r0);
3373       __ data_for_oop(r0, r0, r4);
3374       __ lea(rscratch1, field);
3375       __ flat_field_copy(IN_HEAP, r0, rscratch1, r5);
3376       __ b(done);
3377       __ bind(has_null_marker);
3378       __ load_field_entry(r4, r1);
3379       __ mov(r1, r2);
3380       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), r1, r0, r4);
3381       __ bind(done);
3382     }
3383     break;
3384   case Bytecodes::_fast_aputfield:
3385     do_oop_store(_masm, field, r0, IN_HEAP);
3386     break;
3387   case Bytecodes::_fast_lputfield:
3388     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
3389     break;
3390   case Bytecodes::_fast_iputfield:
3391     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
3392     break;
3393   case Bytecodes::_fast_zputfield:
3394     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
3395     break;
3396   case Bytecodes::_fast_bputfield:
3397     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
3398     break;
3399   case Bytecodes::_fast_sputfield:
3400     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
3401     break;
3402   case Bytecodes::_fast_cputfield:
3403     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
3404     break;
3405   case Bytecodes::_fast_fputfield:
3406     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3407     break;
3408   case Bytecodes::_fast_dputfield:
3409     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
3410     break;
3411   default:
3412     ShouldNotReachHere();
3413   }
3414 
3415   {
3416     Label notVolatile;
3417     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3418     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3419     __ bind(notVolatile);
3420   }
3421 }
3422 
3423 
3424 void TemplateTable::fast_accessfield(TosState state)
3425 {
3426   transition(atos, state);
3427   // Do the JVMTI work here to avoid disturbing the register state below
3428   if (JvmtiExport::can_post_field_access()) {
3429     // Check to see if a field access watch has been set before we
3430     // take the time to call into the VM.
3431     Label L1;
3432     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3433     __ ldrw(r2, Address(rscratch1));
3434     __ cbzw(r2, L1);
3435     // access constant pool cache entry
3436     __ load_field_entry(c_rarg2, rscratch2);
3437     __ verify_oop(r0);
3438     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
3439     __ mov(c_rarg1, r0);
3440     // c_rarg1: object pointer copied above
3441     // c_rarg2: cache entry pointer
3442     __ call_VM(noreg,
3443                CAST_FROM_FN_PTR(address,
3444                                 InterpreterRuntime::post_field_access),
3445                c_rarg1, c_rarg2);
3446     __ pop_ptr(r0); // restore object pointer
3447     __ bind(L1);
3448   }
3449 
3450   // access constant pool cache
3451   __ load_field_entry(r2, r1);
3452 
3453   __ load_sized_value(r1, Address(r2, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3454   __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3455 
3456   // r0: object
3457   __ verify_oop(r0);
3458   __ null_check(r0);
3459   const Address field(r0, r1);
3460 
3461   // 8179954: We need to make sure that the code generated for
3462   // volatile accesses forms a sequentially-consistent set of
3463   // operations when combined with STLR and LDAR.  Without a leading
3464   // membar it's possible for a simple Dekker test to fail if loads
3465   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3466   // the stores in one method and we interpret the loads in another.
3467   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3468     Label notVolatile;
3469     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3470     __ membar(MacroAssembler::AnyAny);
3471     __ bind(notVolatile);
3472   }
3473 
3474   // access field
3475   switch (bytecode()) {
3476   case Bytecodes::_fast_vgetfield:
3477     {
3478       Register index = r4, klass = r5, inline_klass = r6, tmp = r7;
3479       Label is_flat, has_null_marker, nonnull, Done;
3480       __ test_field_has_null_marker(r3, noreg /*temp*/, has_null_marker);
3481       __ test_field_is_flat(r3, noreg /* temp */, is_flat);
3482         // field is not flat
3483         __ load_heap_oop(r0, field, rscratch1, rscratch2);
3484         __ cbnz(r0, nonnull);
3485           __ load_unsigned_short(index, Address(r2, in_bytes(ResolvedFieldEntry::field_index_offset())));
3486           __ ldr(klass, Address(r2, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3487           __ get_inline_type_field_klass(klass, index, inline_klass);
3488           __ get_default_value_oop(inline_klass, tmp /* temp */, r0);
3489         __ bind(nonnull);
3490         __ verify_oop(r0);
3491         __ b(Done);
3492       __ bind(is_flat);
3493       // field is flat
3494         __ load_unsigned_short(index, Address(r2, in_bytes(ResolvedFieldEntry::field_index_offset())));
3495         __ read_flat_field(r2, index, r1, tmp /* temp */, r0);
3496         __ verify_oop(r0);
3497         __ b(Done);
3498       __ bind(has_null_marker);
3499         call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), r0, r2);
3500         __ verify_oop(r0);
3501       __ bind(Done);
3502     }
3503     break;
3504   case Bytecodes::_fast_agetfield:
3505     do_oop_load(_masm, field, r0, IN_HEAP);
3506     __ verify_oop(r0);
3507     break;
3508   case Bytecodes::_fast_lgetfield:
3509     __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
3510     break;
3511   case Bytecodes::_fast_igetfield:
3512     __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
3513     break;
3514   case Bytecodes::_fast_bgetfield:
3515     __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
3516     break;
3517   case Bytecodes::_fast_sgetfield:
3518     __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
3519     break;
3520   case Bytecodes::_fast_cgetfield:
3521     __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
3522     break;
3523   case Bytecodes::_fast_fgetfield:
3524     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3525     break;
3526   case Bytecodes::_fast_dgetfield:
3527     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3528     break;
3529   default:
3530     ShouldNotReachHere();
3531   }
3532   {
3533     Label notVolatile;
3534     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3535     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3536     __ bind(notVolatile);
3537   }
3538 }
3539 
3540 void TemplateTable::fast_xaccess(TosState state)
3541 {
3542   transition(vtos, state);
3543 
3544   // get receiver
3545   __ ldr(r0, aaddress(0));
3546   // access constant pool cache
3547   __ load_field_entry(r2, r3, 2);
3548   __ load_sized_value(r1, Address(r2, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3549 
3550   // 8179954: We need to make sure that the code generated for
3551   // volatile accesses forms a sequentially-consistent set of
3552   // operations when combined with STLR and LDAR.  Without a leading
3553   // membar it's possible for a simple Dekker test to fail if loads
3554   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3555   // the stores in one method and we interpret the loads in another.
3556   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3557     Label notVolatile;
3558     __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3559     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3560     __ membar(MacroAssembler::AnyAny);
3561     __ bind(notVolatile);
3562   }
3563 
3564   // make sure exception is reported in correct bcp range (getfield is
3565   // next instruction)
3566   __ increment(rbcp);
3567   __ null_check(r0);
3568   switch (state) {
3569   case itos:
3570     __ access_load_at(T_INT, IN_HEAP, r0, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3571     break;
3572   case atos:
3573     do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3574     __ verify_oop(r0);
3575     break;
3576   case ftos:
3577     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3578     break;
3579   default:
3580     ShouldNotReachHere();
3581   }
3582 
3583   {
3584     Label notVolatile;
3585     __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3586     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3587     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3588     __ bind(notVolatile);
3589   }
3590 
3591   __ decrement(rbcp);
3592 }
3593 
3594 
3595 
3596 //-----------------------------------------------------------------------------
3597 // Calls
3598 
3599 void TemplateTable::prepare_invoke(Register cache, Register recv) {
3600 
3601   Bytecodes::Code code = bytecode();
3602   const bool load_receiver       = (code != Bytecodes::_invokestatic) && (code != Bytecodes::_invokedynamic);
3603 
3604   // save 'interpreter return address'
3605   __ save_bcp();
3606 
3607   // Load TOS state for later
3608   __ load_unsigned_byte(rscratch2, Address(cache, in_bytes(ResolvedMethodEntry::type_offset())));
3609 
3610   // load receiver if needed (note: no return address pushed yet)
3611   if (load_receiver) {
3612     __ load_unsigned_short(recv, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
3613     __ add(rscratch1, esp, recv, ext::uxtx, 3);
3614     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3615     __ verify_oop(recv);
3616   }
3617 
3618   // load return address
3619   {
3620     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3621     __ mov(rscratch1, table_addr);
3622     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3623   }
3624 }
3625 
3626 
3627 void TemplateTable::invokevirtual_helper(Register index,
3628                                          Register recv,
3629                                          Register flags)
3630 {
3631   // Uses temporary registers r0, r3
3632   assert_different_registers(index, recv, r0, r3);
3633   // Test for an invoke of a final method
3634   Label notFinal;
3635   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, notFinal);
3636 
3637   const Register method = index;  // method must be rmethod
3638   assert(method == rmethod,
3639          "Method must be rmethod for interpreter calling convention");
3640 
3641   // do the call - the index is actually the method to call
3642   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3643 
3644   // It's final, need a null check here!
3645   __ null_check(recv);
3646 
3647   // profile this call
3648   __ profile_final_call(r0);
3649   __ profile_arguments_type(r0, method, r4, true);
3650 
3651   __ jump_from_interpreted(method, r0);
3652 
3653   __ bind(notFinal);
3654 
3655   // get receiver klass
3656   __ load_klass(r0, recv);
3657 
3658   // profile this call
3659   __ profile_virtual_call(r0, rlocals, r3);
3660 
3661   // get target Method & entry point
3662   __ lookup_virtual_method(r0, index, method);
3663   __ profile_arguments_type(r3, method, r4, true);
3664   // FIXME -- this looks completely redundant. is it?
3665   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3666   __ jump_from_interpreted(method, r3);
3667 }
3668 
3669 void TemplateTable::invokevirtual(int byte_no)
3670 {
3671   transition(vtos, vtos);
3672   assert(byte_no == f2_byte, "use this argument");
3673 
3674   load_resolved_method_entry_virtual(r2,      // ResolvedMethodEntry*
3675                                      rmethod, // Method* or itable index
3676                                      r3);     // flags
3677   prepare_invoke(r2, r2); // recv
3678 
3679   // rmethod: index (actually a Method*)
3680   // r2: receiver
3681   // r3: flags
3682 
3683   invokevirtual_helper(rmethod, r2, r3);
3684 }
3685 
3686 void TemplateTable::invokespecial(int byte_no)
3687 {
3688   transition(vtos, vtos);
3689   assert(byte_no == f1_byte, "use this argument");
3690 
3691   load_resolved_method_entry_special_or_static(r2,      // ResolvedMethodEntry*
3692                                                rmethod, // Method*
3693                                                r3);     // flags
3694   prepare_invoke(r2, r2);  // get receiver also for null check
3695   __ verify_oop(r2);
3696   __ null_check(r2);
3697   // do the call
3698   __ profile_call(r0);
3699   __ profile_arguments_type(r0, rmethod, rbcp, false);
3700   __ jump_from_interpreted(rmethod, r0);
3701 }
3702 
3703 void TemplateTable::invokestatic(int byte_no)
3704 {
3705   transition(vtos, vtos);
3706   assert(byte_no == f1_byte, "use this argument");
3707 
3708   load_resolved_method_entry_special_or_static(r2,      // ResolvedMethodEntry*
3709                                                rmethod, // Method*
3710                                                r3);     // flags
3711   prepare_invoke(r2, r2);  // get receiver also for null check
3712 
3713   // do the call
3714   __ profile_call(r0);
3715   __ profile_arguments_type(r0, rmethod, r4, false);
3716   __ jump_from_interpreted(rmethod, r0);
3717 }
3718 
3719 void TemplateTable::fast_invokevfinal(int byte_no)
3720 {
3721   __ call_Unimplemented();
3722 }
3723 
3724 void TemplateTable::invokeinterface(int byte_no) {
3725   transition(vtos, vtos);
3726   assert(byte_no == f1_byte, "use this argument");
3727 
3728   load_resolved_method_entry_interface(r2,      // ResolvedMethodEntry*
3729                                        r0,      // Klass*
3730                                        rmethod, // Method* or itable/vtable index
3731                                        r3);     // flags
3732   prepare_invoke(r2, r2); // receiver
3733 
3734   // r0: interface klass (from f1)
3735   // rmethod: method (from f2)
3736   // r2: receiver
3737   // r3: flags
3738 
3739   // First check for Object case, then private interface method,
3740   // then regular interface method.
3741 
3742   // Special case of invokeinterface called for virtual method of
3743   // java.lang.Object.  See cpCache.cpp for details.
3744   Label notObjectMethod;
3745   __ tbz(r3, ResolvedMethodEntry::is_forced_virtual_shift, notObjectMethod);
3746 
3747   invokevirtual_helper(rmethod, r2, r3);
3748   __ bind(notObjectMethod);
3749 
3750   Label no_such_interface;
3751 
3752   // Check for private method invocation - indicated by vfinal
3753   Label notVFinal;
3754   __ tbz(r3, ResolvedMethodEntry::is_vfinal_shift, notVFinal);
3755 
3756   // Get receiver klass into r3
3757   __ load_klass(r3, r2);
3758 
3759   Label subtype;
3760   __ check_klass_subtype(r3, r0, r4, subtype);
3761   // If we get here the typecheck failed
3762   __ b(no_such_interface);
3763   __ bind(subtype);
3764 
3765   __ profile_final_call(r0);
3766   __ profile_arguments_type(r0, rmethod, r4, true);
3767   __ jump_from_interpreted(rmethod, r0);
3768 
3769   __ bind(notVFinal);
3770 
3771   // Get receiver klass into r3
3772   __ restore_locals();
3773   __ load_klass(r3, r2);
3774 
3775   Label no_such_method;
3776 
3777   // Preserve method for throw_AbstractMethodErrorVerbose.
3778   __ mov(r16, rmethod);
3779   // Receiver subtype check against REFC.
3780   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3781   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3782                              r3, r0, noreg,
3783                              // outputs: scan temp. reg, scan temp. reg
3784                              rscratch2, r13,
3785                              no_such_interface,
3786                              /*return_method=*/false);
3787 
3788   // profile this call
3789   __ profile_virtual_call(r3, r13, r19);
3790 
3791   // Get declaring interface class from method, and itable index
3792 
3793   __ load_method_holder(r0, rmethod);
3794   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3795   __ subw(rmethod, rmethod, Method::itable_index_max);
3796   __ negw(rmethod, rmethod);
3797 
3798   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3799   __ mov(rlocals, r3);
3800   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3801                              rlocals, r0, rmethod,
3802                              // outputs: method, scan temp. reg
3803                              rmethod, r13,
3804                              no_such_interface);
3805 
3806   // rmethod,: Method to call
3807   // r2: receiver
3808   // Check for abstract method error
3809   // Note: This should be done more efficiently via a throw_abstract_method_error
3810   //       interpreter entry point and a conditional jump to it in case of a null
3811   //       method.
3812   __ cbz(rmethod, no_such_method);
3813 
3814   __ profile_arguments_type(r3, rmethod, r13, true);
3815 
3816   // do the call
3817   // r2: receiver
3818   // rmethod,: Method
3819   __ jump_from_interpreted(rmethod, r3);
3820   __ should_not_reach_here();
3821 
3822   // exception handling code follows...
3823   // note: must restore interpreter registers to canonical
3824   //       state for exception handling to work correctly!
3825 
3826   __ bind(no_such_method);
3827   // throw exception
3828   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3829   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3830   // Pass arguments for generating a verbose error message.
3831   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3832   // the call_VM checks for exception, so we should never return here.
3833   __ should_not_reach_here();
3834 
3835   __ bind(no_such_interface);
3836   // throw exception
3837   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3838   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3839   // Pass arguments for generating a verbose error message.
3840   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3841                    InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3842   // the call_VM checks for exception, so we should never return here.
3843   __ should_not_reach_here();
3844   return;
3845 }
3846 
3847 void TemplateTable::invokehandle(int byte_no) {
3848   transition(vtos, vtos);
3849   assert(byte_no == f1_byte, "use this argument");
3850 
3851   load_resolved_method_entry_handle(r2,      // ResolvedMethodEntry*
3852                                     rmethod, // Method*
3853                                     r0,      // Resolved reference
3854                                     r3);     // flags
3855   prepare_invoke(r2, r2);
3856 
3857   __ verify_method_ptr(r2);
3858   __ verify_oop(r2);
3859   __ null_check(r2);
3860 
3861   // FIXME: profile the LambdaForm also
3862 
3863   // r13 is safe to use here as a scratch reg because it is about to
3864   // be clobbered by jump_from_interpreted().
3865   __ profile_final_call(r13);
3866   __ profile_arguments_type(r13, rmethod, r4, true);
3867 
3868   __ jump_from_interpreted(rmethod, r0);
3869 }
3870 
3871 void TemplateTable::invokedynamic(int byte_no) {
3872   transition(vtos, vtos);
3873   assert(byte_no == f1_byte, "use this argument");
3874 
3875   load_invokedynamic_entry(rmethod);
3876 
3877   // r0: CallSite object (from cpool->resolved_references[])
3878   // rmethod: MH.linkToCallSite method
3879 
3880   // Note:  r0_callsite is already pushed
3881 
3882   // %%% should make a type profile for any invokedynamic that takes a ref argument
3883   // profile this call
3884   __ profile_call(rbcp);
3885   __ profile_arguments_type(r3, rmethod, r13, false);
3886 
3887   __ verify_oop(r0);
3888 
3889   __ jump_from_interpreted(rmethod, r0);
3890 }
3891 
3892 
3893 //-----------------------------------------------------------------------------
3894 // Allocation
3895 
3896 void TemplateTable::_new() {
3897   transition(vtos, atos);
3898 
3899   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3900   Label slow_case;
3901   Label done;
3902   Label initialize_header;
3903 
3904   __ get_cpool_and_tags(r4, r0);
3905   // Make sure the class we're about to instantiate has been resolved.
3906   // This is done before loading InstanceKlass to be consistent with the order
3907   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3908   const int tags_offset = Array<u1>::base_offset_in_bytes();
3909   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3910   __ lea(rscratch1, Address(rscratch1, tags_offset));
3911   __ ldarb(rscratch1, rscratch1);
3912   __ cmp(rscratch1, (u1)JVM_CONSTANT_Class);
3913   __ br(Assembler::NE, slow_case);
3914 
3915   // get InstanceKlass
3916   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3917 
3918   // make sure klass is initialized
3919   assert(VM_Version::supports_fast_class_init_checks(), "Optimization requires support for fast class initialization checks");
3920   __ clinit_barrier(r4, rscratch1, nullptr /*L_fast_path*/, &slow_case);
3921 
3922   __ allocate_instance(r4, r0, r3, r1, true, slow_case);
3923     if (DTraceAllocProbes) {
3924       // Trigger dtrace event for fastpath
3925       __ push(atos); // save the return value
3926       __ call_VM_leaf(
3927            CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), r0);
3928       __ pop(atos); // restore the return value
3929 
3930     }
3931   __ b(done);
3932 
3933   // slow case
3934   __ bind(slow_case);
3935   __ get_constant_pool(c_rarg1);
3936   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3937   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3938   __ verify_oop(r0);
3939 
3940   // continue
3941   __ bind(done);
3942   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3943   __ membar(Assembler::StoreStore);
3944 }
3945 
3946 void TemplateTable::newarray() {
3947   transition(itos, atos);
3948   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3949   __ mov(c_rarg2, r0);
3950   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3951           c_rarg1, c_rarg2);
3952   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3953   __ membar(Assembler::StoreStore);
3954 }
3955 
3956 void TemplateTable::anewarray() {
3957   transition(itos, atos);
3958   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3959   __ get_constant_pool(c_rarg1);
3960   __ mov(c_rarg3, r0);
3961   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3962           c_rarg1, c_rarg2, c_rarg3);
3963   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3964   __ membar(Assembler::StoreStore);
3965 }
3966 
3967 void TemplateTable::arraylength() {
3968   transition(atos, itos);
3969   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3970 }
3971 
3972 void TemplateTable::checkcast()
3973 {
3974   transition(atos, atos);
3975   Label done, is_null, ok_is_subtype, quicked, resolved;
3976   __ cbz(r0, is_null);
3977 
3978   // Get cpool & tags index
3979   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3980   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3981   // See if bytecode has already been quicked
3982   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3983   __ lea(r1, Address(rscratch1, r19));
3984   __ ldarb(r1, r1);
3985   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3986   __ br(Assembler::EQ, quicked);
3987 
3988   __ push(atos); // save receiver for result, and for GC
3989   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3990   // vm_result_2 has metadata result
3991   __ get_vm_result_2(r0, rthread);
3992   __ pop(r3); // restore receiver
3993   __ b(resolved);
3994 
3995   // Get superklass in r0 and subklass in r3
3996   __ bind(quicked);
3997   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3998   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3999 
4000   __ bind(resolved);
4001   __ load_klass(r19, r3);
4002 
4003   // Generate subtype check.  Blows r2, r5.  Object in r3.
4004   // Superklass in r0.  Subklass in r19.
4005   __ gen_subtype_check(r19, ok_is_subtype);
4006 
4007   // Come here on failure
4008   __ push(r3);
4009   // object is at TOS
4010   __ b(Interpreter::_throw_ClassCastException_entry);
4011 
4012   // Come here on success
4013   __ bind(ok_is_subtype);
4014   __ mov(r0, r3); // Restore object in r3
4015 
4016   __ b(done);
4017   __ bind(is_null);
4018 
4019   // Collect counts on whether this test sees nulls a lot or not.
4020   if (ProfileInterpreter) {
4021     __ profile_null_seen(r2);
4022   }
4023 
4024   __ bind(done);
4025 }
4026 
4027 void TemplateTable::instanceof() {
4028   transition(atos, itos);
4029   Label done, is_null, ok_is_subtype, quicked, resolved;
4030   __ cbz(r0, is_null);
4031 
4032   // Get cpool & tags index
4033   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
4034   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
4035   // See if bytecode has already been quicked
4036   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
4037   __ lea(r1, Address(rscratch1, r19));
4038   __ ldarb(r1, r1);
4039   __ cmp(r1, (u1)JVM_CONSTANT_Class);
4040   __ br(Assembler::EQ, quicked);
4041 
4042   __ push(atos); // save receiver for result, and for GC
4043   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4044   // vm_result_2 has metadata result
4045   __ get_vm_result_2(r0, rthread);
4046   __ pop(r3); // restore receiver
4047   __ verify_oop(r3);
4048   __ load_klass(r3, r3);
4049   __ b(resolved);
4050 
4051   // Get superklass in r0 and subklass in r3
4052   __ bind(quicked);
4053   __ load_klass(r3, r0);
4054   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
4055 
4056   __ bind(resolved);
4057 
4058   // Generate subtype check.  Blows r2, r5
4059   // Superklass in r0.  Subklass in r3.
4060   __ gen_subtype_check(r3, ok_is_subtype);
4061 
4062   // Come here on failure
4063   __ mov(r0, 0);
4064   __ b(done);
4065   // Come here on success
4066   __ bind(ok_is_subtype);
4067   __ mov(r0, 1);
4068 
4069   // Collect counts on whether this test sees nulls a lot or not.
4070   if (ProfileInterpreter) {
4071     __ b(done);
4072     __ bind(is_null);
4073     __ profile_null_seen(r2);
4074   } else {
4075     __ bind(is_null);   // same as 'done'
4076   }
4077   __ bind(done);
4078   // r0 = 0: obj == nullptr or  obj is not an instanceof the specified klass
4079   // r0 = 1: obj != nullptr and obj is     an instanceof the specified klass
4080 }
4081 
4082 //-----------------------------------------------------------------------------
4083 // Breakpoints
4084 void TemplateTable::_breakpoint() {
4085   // Note: We get here even if we are single stepping..
4086   // jbug inists on setting breakpoints at every bytecode
4087   // even if we are in single step mode.
4088 
4089   transition(vtos, vtos);
4090 
4091   // get the unpatched byte code
4092   __ get_method(c_rarg1);
4093   __ call_VM(noreg,
4094              CAST_FROM_FN_PTR(address,
4095                               InterpreterRuntime::get_original_bytecode_at),
4096              c_rarg1, rbcp);
4097   __ mov(r19, r0);
4098 
4099   // post the breakpoint event
4100   __ call_VM(noreg,
4101              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4102              rmethod, rbcp);
4103 
4104   // complete the execution of original bytecode
4105   __ mov(rscratch1, r19);
4106   __ dispatch_only_normal(vtos);
4107 }
4108 
4109 //-----------------------------------------------------------------------------
4110 // Exceptions
4111 
4112 void TemplateTable::athrow() {
4113   transition(atos, vtos);
4114   __ null_check(r0);
4115   __ b(Interpreter::throw_exception_entry());
4116 }
4117 
4118 //-----------------------------------------------------------------------------
4119 // Synchronization
4120 //
4121 // Note: monitorenter & exit are symmetric routines; which is reflected
4122 //       in the assembly code structure as well
4123 //
4124 // Stack layout:
4125 //
4126 // [expressions  ] <--- esp               = expression stack top
4127 // ..
4128 // [expressions  ]
4129 // [monitor entry] <--- monitor block top = expression stack bot
4130 // ..
4131 // [monitor entry]
4132 // [frame data   ] <--- monitor block bot
4133 // ...
4134 // [saved rfp    ] <--- rfp
4135 void TemplateTable::monitorenter()
4136 {
4137   transition(atos, vtos);
4138 
4139   // check for null object
4140   __ null_check(r0);
4141 
4142   Label is_inline_type;
4143   __ ldr(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
4144   __ test_markword_is_inline_type(rscratch1, is_inline_type);
4145 
4146   const Address monitor_block_top(
4147         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4148   const Address monitor_block_bot(
4149         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4150   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4151 
4152   Label allocated;
4153 
4154   // initialize entry pointer
4155   __ mov(c_rarg1, zr); // points to free slot or null
4156 
4157   // find a free slot in the monitor block (result in c_rarg1)
4158   {
4159     Label entry, loop, exit;
4160     __ ldr(c_rarg3, monitor_block_top); // derelativize pointer
4161     __ lea(c_rarg3, Address(rfp, c_rarg3, Address::lsl(Interpreter::logStackElementSize)));
4162     // c_rarg3 points to current entry, starting with top-most entry
4163 
4164     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4165 
4166     __ b(entry);
4167 
4168     __ bind(loop);
4169     // check if current entry is used
4170     // if not used then remember entry in c_rarg1
4171     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset()));
4172     __ cmp(zr, rscratch1);
4173     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
4174     // check if current entry is for same object
4175     __ cmp(r0, rscratch1);
4176     // if same object then stop searching
4177     __ br(Assembler::EQ, exit);
4178     // otherwise advance to next entry
4179     __ add(c_rarg3, c_rarg3, entry_size);
4180     __ bind(entry);
4181     // check if bottom reached
4182     __ cmp(c_rarg3, c_rarg2);
4183     // if not at bottom then check this entry
4184     __ br(Assembler::NE, loop);
4185     __ bind(exit);
4186   }
4187 
4188   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
4189                             // if found, continue with that on
4190 
4191   // allocate one if there's no free slot
4192   {
4193     Label entry, loop;
4194     // 1. compute new pointers            // rsp: old expression stack top
4195 
4196     __ check_extended_sp();
4197     __ sub(sp, sp, entry_size);           // make room for the monitor
4198     __ sub(rscratch1, sp, rfp);
4199     __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
4200     __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
4201 
4202     __ ldr(c_rarg1, monitor_block_bot);   // derelativize pointer
4203     __ lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
4204     // c_rarg1 points to the old expression stack bottom
4205 
4206     __ sub(esp, esp, entry_size);         // move expression stack top
4207     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
4208     __ mov(c_rarg3, esp);                 // set start value for copy loop
4209     __ sub(rscratch1, c_rarg1, rfp);      // relativize pointer
4210     __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
4211     __ str(rscratch1, monitor_block_bot);  // set new monitor block bottom
4212 
4213     __ b(entry);
4214     // 2. move expression stack contents
4215     __ bind(loop);
4216     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
4217                                                    // word from old location
4218     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
4219     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
4220     __ bind(entry);
4221     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
4222     __ br(Assembler::NE, loop);      // if not at bottom then
4223                                      // copy next word
4224   }
4225 
4226   // call run-time routine
4227   // c_rarg1: points to monitor entry
4228   __ bind(allocated);
4229 
4230   // Increment bcp to point to the next bytecode, so exception
4231   // handling for async. exceptions work correctly.
4232   // The object has already been popped from the stack, so the
4233   // expression stack looks correct.
4234   __ increment(rbcp);
4235 
4236   // store object
4237   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
4238   __ lock_object(c_rarg1);
4239 
4240   // check to make sure this monitor doesn't cause stack overflow after locking
4241   __ save_bcp();  // in case of exception
4242   __ generate_stack_overflow_check(0);
4243 
4244   // The bcp has already been incremented. Just need to dispatch to
4245   // next instruction.
4246   __ dispatch_next(vtos);
4247 
4248   __ bind(is_inline_type);
4249   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4250                     InterpreterRuntime::throw_identity_exception), r0);
4251   __ should_not_reach_here();
4252 }
4253 
4254 
4255 void TemplateTable::monitorexit()
4256 {
4257   transition(atos, vtos);
4258 
4259   // check for null object
4260   __ null_check(r0);
4261 
4262   const int is_inline_type_mask = markWord::inline_type_pattern;
4263   Label has_identity;
4264   __ ldr(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
4265   __ mov(rscratch2, is_inline_type_mask);
4266   __ andr(rscratch1, rscratch1, rscratch2);
4267   __ cmp(rscratch1, rscratch2);
4268   __ br(Assembler::NE, has_identity);
4269   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4270                      InterpreterRuntime::throw_illegal_monitor_state_exception));
4271   __ should_not_reach_here();
4272   __ bind(has_identity);
4273 
4274   const Address monitor_block_top(
4275         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4276   const Address monitor_block_bot(
4277         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4278   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4279 
4280   Label found;
4281 
4282   // find matching slot
4283   {
4284     Label entry, loop;
4285     __ ldr(c_rarg1, monitor_block_top); // derelativize pointer
4286     __ lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
4287     // c_rarg1 points to current entry, starting with top-most entry
4288 
4289     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4290                                         // of monitor block
4291     __ b(entry);
4292 
4293     __ bind(loop);
4294     // check if current entry is for same object
4295     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
4296     __ cmp(r0, rscratch1);
4297     // if same object then stop searching
4298     __ br(Assembler::EQ, found);
4299     // otherwise advance to next entry
4300     __ add(c_rarg1, c_rarg1, entry_size);
4301     __ bind(entry);
4302     // check if bottom reached
4303     __ cmp(c_rarg1, c_rarg2);
4304     // if not at bottom then check this entry
4305     __ br(Assembler::NE, loop);
4306   }
4307 
4308   // error handling. Unlocking was not block-structured
4309   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4310                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4311   __ should_not_reach_here();
4312 
4313   // call run-time routine
4314   __ bind(found);
4315   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
4316   __ unlock_object(c_rarg1);
4317   __ pop_ptr(r0); // discard object
4318 }
4319 
4320 
4321 // Wide instructions
4322 void TemplateTable::wide()
4323 {
4324   __ load_unsigned_byte(r19, at_bcp(1));
4325   __ mov(rscratch1, (address)Interpreter::_wentry_point);
4326   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
4327   __ br(rscratch1);
4328 }
4329 
4330 
4331 // Multi arrays
4332 void TemplateTable::multianewarray() {
4333   transition(vtos, atos);
4334   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
4335   // last dim is on top of stack; we want address of first one:
4336   // first_addr = last_addr + (ndims - 1) * wordSize
4337   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
4338   __ sub(c_rarg1, c_rarg1, wordSize);
4339   call_VM(r0,
4340           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
4341           c_rarg1);
4342   __ load_unsigned_byte(r1, at_bcp(3));
4343   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
4344 }