1 /*
   2  * Copyright (c) 2003, 2022, 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/compilerDefinitions.inline.hpp"
  29 #include "gc/shared/barrierSetAssembler.hpp"
  30 #include "gc/shared/collectedHeap.hpp"
  31 #include "gc/shared/tlab_globals.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "interpreter/interpreterRuntime.hpp"
  34 #include "interpreter/interp_masm.hpp"
  35 #include "interpreter/templateTable.hpp"
  36 #include "memory/universe.hpp"
  37 #include "oops/methodData.hpp"
  38 #include "oops/method.hpp"
  39 #include "oops/objArrayKlass.hpp"
  40 #include "oops/oop.inline.hpp"
  41 #include "prims/jvmtiExport.hpp"
  42 #include "prims/methodHandles.hpp"
  43 #include "runtime/frame.inline.hpp"
  44 #include "runtime/sharedRuntime.hpp"
  45 #include "runtime/stubRoutines.hpp"
  46 #include "runtime/synchronizer.hpp"
  47 #include "utilities/powerOfTwo.hpp"
  48 
  49 #define __ _masm->
  50 
  51 // Address computation: local variables
  52 
  53 static inline Address iaddress(int n) {
  54   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  55 }
  56 
  57 static inline Address laddress(int n) {
  58   return iaddress(n + 1);
  59 }
  60 
  61 static inline Address faddress(int n) {
  62   return iaddress(n);
  63 }
  64 
  65 static inline Address daddress(int n) {
  66   return laddress(n);
  67 }
  68 
  69 static inline Address aaddress(int n) {
  70   return iaddress(n);
  71 }
  72 
  73 static inline Address iaddress(Register r) {
  74   return Address(rlocals, r, Address::lsl(3));
  75 }
  76 
  77 static inline Address laddress(Register r, Register scratch,
  78                                InterpreterMacroAssembler* _masm) {
  79   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  80   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  81 }
  82 
  83 static inline Address faddress(Register r) {
  84   return iaddress(r);
  85 }
  86 
  87 static inline Address daddress(Register r, Register scratch,
  88                                InterpreterMacroAssembler* _masm) {
  89   return laddress(r, scratch, _masm);
  90 }
  91 
  92 static inline Address aaddress(Register r) {
  93   return iaddress(r);
  94 }
  95 
  96 static inline Address at_rsp() {
  97   return Address(esp, 0);
  98 }
  99 
 100 // At top of Java expression stack which may be different than esp().  It
 101 // isn't for category 1 objects.
 102 static inline Address at_tos   () {
 103   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 104 }
 105 
 106 static inline Address at_tos_p1() {
 107   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 108 }
 109 
 110 static inline Address at_tos_p2() {
 111   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 112 }
 113 
 114 static inline Address at_tos_p3() {
 115   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 116 }
 117 
 118 static inline Address at_tos_p4() {
 119   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 120 }
 121 
 122 static inline Address at_tos_p5() {
 123   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 124 }
 125 
 126 // Condition conversion
 127 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 128   switch (cc) {
 129   case TemplateTable::equal        : return Assembler::NE;
 130   case TemplateTable::not_equal    : return Assembler::EQ;
 131   case TemplateTable::less         : return Assembler::GE;
 132   case TemplateTable::less_equal   : return Assembler::GT;
 133   case TemplateTable::greater      : return Assembler::LE;
 134   case TemplateTable::greater_equal: return Assembler::LT;
 135   }
 136   ShouldNotReachHere();
 137   return Assembler::EQ;
 138 }
 139 
 140 
 141 // Miscellaneous helper routines
 142 // Store an oop (or NULL) at the Address described by obj.
 143 // If val == noreg this means store a NULL
 144 static void do_oop_store(InterpreterMacroAssembler* _masm,
 145                          Address dst,
 146                          Register val,
 147                          DecoratorSet decorators) {
 148   assert(val == noreg || val == r0, "parameter is just for looks");
 149   __ store_heap_oop(dst, val, r10, r11, r3, decorators);
 150 }
 151 
 152 static void do_oop_load(InterpreterMacroAssembler* _masm,
 153                         Address src,
 154                         Register dst,
 155                         DecoratorSet decorators) {
 156   __ load_heap_oop(dst, src, r10, r11, decorators);
 157 }
 158 
 159 Address TemplateTable::at_bcp(int offset) {
 160   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 161   return Address(rbcp, offset);
 162 }
 163 
 164 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 165                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 166                                    int byte_no)
 167 {
 168   if (!RewriteBytecodes)  return;
 169   Label L_patch_done;
 170 
 171   switch (bc) {
 172   case Bytecodes::_fast_aputfield:
 173   case Bytecodes::_fast_bputfield:
 174   case Bytecodes::_fast_zputfield:
 175   case Bytecodes::_fast_cputfield:
 176   case Bytecodes::_fast_dputfield:
 177   case Bytecodes::_fast_fputfield:
 178   case Bytecodes::_fast_iputfield:
 179   case Bytecodes::_fast_lputfield:
 180   case Bytecodes::_fast_sputfield:
 181     {
 182       // We skip bytecode quickening for putfield instructions when
 183       // the put_code written to the constant pool cache is zero.
 184       // This is required so that every execution of this instruction
 185       // calls out to InterpreterRuntime::resolve_get_put to do
 186       // additional, required work.
 187       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 188       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 189       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 190       __ movw(bc_reg, bc);
 191       __ cbzw(temp_reg, L_patch_done);  // don't patch
 192     }
 193     break;
 194   default:
 195     assert(byte_no == -1, "sanity");
 196     // the pair bytecodes have already done the load.
 197     if (load_bc_into_bc_reg) {
 198       __ movw(bc_reg, bc);
 199     }
 200   }
 201 
 202   if (JvmtiExport::can_post_breakpoint()) {
 203     Label L_fast_patch;
 204     // if a breakpoint is present we can't rewrite the stream directly
 205     __ load_unsigned_byte(temp_reg, at_bcp(0));
 206     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 207     __ br(Assembler::NE, L_fast_patch);
 208     // Let breakpoint table handling rewrite to quicker bytecode
 209     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 210     __ b(L_patch_done);
 211     __ bind(L_fast_patch);
 212   }
 213 
 214 #ifdef ASSERT
 215   Label L_okay;
 216   __ load_unsigned_byte(temp_reg, at_bcp(0));
 217   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 218   __ br(Assembler::EQ, L_okay);
 219   __ cmpw(temp_reg, bc_reg);
 220   __ br(Assembler::EQ, L_okay);
 221   __ stop("patching the wrong bytecode");
 222   __ bind(L_okay);
 223 #endif
 224 
 225   // patch bytecode
 226   __ strb(bc_reg, at_bcp(0));
 227   __ bind(L_patch_done);
 228 }
 229 
 230 
 231 // Individual instructions
 232 
 233 void TemplateTable::nop() {
 234   transition(vtos, vtos);
 235   // nothing to do
 236 }
 237 
 238 void TemplateTable::shouldnotreachhere() {
 239   transition(vtos, vtos);
 240   __ stop("shouldnotreachhere bytecode");
 241 }
 242 
 243 void TemplateTable::aconst_null()
 244 {
 245   transition(vtos, atos);
 246   __ mov(r0, 0);
 247 }
 248 
 249 void TemplateTable::iconst(int value)
 250 {
 251   transition(vtos, itos);
 252   __ mov(r0, value);
 253 }
 254 
 255 void TemplateTable::lconst(int value)
 256 {
 257   __ mov(r0, value);
 258 }
 259 
 260 void TemplateTable::fconst(int value)
 261 {
 262   transition(vtos, ftos);
 263   switch (value) {
 264   case 0:
 265     __ fmovs(v0, 0.0);
 266     break;
 267   case 1:
 268     __ fmovs(v0, 1.0);
 269     break;
 270   case 2:
 271     __ fmovs(v0, 2.0);
 272     break;
 273   default:
 274     ShouldNotReachHere();
 275     break;
 276   }
 277 }
 278 
 279 void TemplateTable::dconst(int value)
 280 {
 281   transition(vtos, dtos);
 282   switch (value) {
 283   case 0:
 284     __ fmovd(v0, 0.0);
 285     break;
 286   case 1:
 287     __ fmovd(v0, 1.0);
 288     break;
 289   case 2:
 290     __ fmovd(v0, 2.0);
 291     break;
 292   default:
 293     ShouldNotReachHere();
 294     break;
 295   }
 296 }
 297 
 298 void TemplateTable::bipush()
 299 {
 300   transition(vtos, itos);
 301   __ load_signed_byte32(r0, at_bcp(1));
 302 }
 303 
 304 void TemplateTable::sipush()
 305 {
 306   transition(vtos, itos);
 307   __ load_unsigned_short(r0, at_bcp(1));
 308   __ revw(r0, r0);
 309   __ asrw(r0, r0, 16);
 310 }
 311 
 312 void TemplateTable::ldc(bool wide)
 313 {
 314   transition(vtos, vtos);
 315   Label call_ldc, notFloat, notClass, notInt, Done;
 316 
 317   if (wide) {
 318     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 319   } else {
 320     __ load_unsigned_byte(r1, at_bcp(1));
 321   }
 322   __ get_cpool_and_tags(r2, r0);
 323 
 324   const int base_offset = ConstantPool::header_size() * wordSize;
 325   const int tags_offset = Array<u1>::base_offset_in_bytes();
 326 
 327   // get type
 328   __ add(r3, r1, tags_offset);
 329   __ lea(r3, Address(r0, r3));
 330   __ ldarb(r3, r3);
 331 
 332   // unresolved class - get the resolved class
 333   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClass);
 334   __ br(Assembler::EQ, call_ldc);
 335 
 336   // unresolved class in error state - call into runtime to throw the error
 337   // from the first resolution attempt
 338   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClassInError);
 339   __ br(Assembler::EQ, call_ldc);
 340 
 341   // resolved class - need to call vm to get java mirror of the class
 342   __ cmp(r3, (u1)JVM_CONSTANT_Class);
 343   __ br(Assembler::NE, notClass);
 344 
 345   __ bind(call_ldc);
 346   __ mov(c_rarg1, wide);
 347   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 348   __ push_ptr(r0);
 349   __ verify_oop(r0);
 350   __ b(Done);
 351 
 352   __ bind(notClass);
 353   __ cmp(r3, (u1)JVM_CONSTANT_Float);
 354   __ br(Assembler::NE, notFloat);
 355   // ftos
 356   __ adds(r1, r2, r1, Assembler::LSL, 3);
 357   __ ldrs(v0, Address(r1, base_offset));
 358   __ push_f();
 359   __ b(Done);
 360 
 361   __ bind(notFloat);
 362 
 363   __ cmp(r3, (u1)JVM_CONSTANT_Integer);
 364   __ br(Assembler::NE, notInt);
 365 
 366   // itos
 367   __ adds(r1, r2, r1, Assembler::LSL, 3);
 368   __ ldrw(r0, Address(r1, base_offset));
 369   __ push_i(r0);
 370   __ b(Done);
 371 
 372   __ bind(notInt);
 373   condy_helper(Done);
 374 
 375   __ bind(Done);
 376 }
 377 
 378 // Fast path for caching oop constants.
 379 void TemplateTable::fast_aldc(bool wide)
 380 {
 381   transition(vtos, atos);
 382 
 383   Register result = r0;
 384   Register tmp = r1;
 385   Register rarg = r2;
 386 
 387   int index_size = wide ? sizeof(u2) : sizeof(u1);
 388 
 389   Label resolved;
 390 
 391   // We are resolved if the resolved reference cache entry contains a
 392   // non-null object (String, MethodType, etc.)
 393   assert_different_registers(result, tmp);
 394   __ get_cache_index_at_bcp(tmp, 1, index_size);
 395   __ load_resolved_reference_at_index(result, tmp);
 396   __ cbnz(result, resolved);
 397 
 398   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 399 
 400   // first time invocation - must resolve first
 401   __ mov(rarg, (int)bytecode());
 402   __ call_VM(result, entry, rarg);
 403 
 404   __ bind(resolved);
 405 
 406   { // Check for the null sentinel.
 407     // If we just called the VM, it already did the mapping for us,
 408     // but it's harmless to retry.
 409     Label notNull;
 410 
 411     // Stash null_sentinel address to get its value later
 412     __ movptr(rarg, (uintptr_t)Universe::the_null_sentinel_addr());
 413     __ ldr(tmp, Address(rarg));
 414     __ resolve_oop_handle(tmp, r5, rscratch2);
 415     __ cmpoop(result, tmp);
 416     __ br(Assembler::NE, notNull);
 417     __ mov(result, 0);  // NULL object reference
 418     __ bind(notNull);
 419   }
 420 
 421   if (VerifyOops) {
 422     // Safe to call with 0 result
 423     __ verify_oop(result);
 424   }
 425 }
 426 
 427 void TemplateTable::ldc2_w()
 428 {
 429   transition(vtos, vtos);
 430   Label notDouble, notLong, Done;
 431   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 432 
 433   __ get_cpool_and_tags(r1, r2);
 434   const int base_offset = ConstantPool::header_size() * wordSize;
 435   const int tags_offset = Array<u1>::base_offset_in_bytes();
 436 
 437   // get type
 438   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 439   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 440   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 441   __ br(Assembler::NE, notDouble);
 442 
 443   // dtos
 444   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 445   __ ldrd(v0, Address(r2, base_offset));
 446   __ push_d();
 447   __ b(Done);
 448 
 449   __ bind(notDouble);
 450   __ cmpw(r2, (int)JVM_CONSTANT_Long);
 451   __ br(Assembler::NE, notLong);
 452 
 453   // ltos
 454   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 455   __ ldr(r0, Address(r0, base_offset));
 456   __ push_l();
 457   __ b(Done);
 458 
 459   __ bind(notLong);
 460   condy_helper(Done);
 461 
 462   __ bind(Done);
 463 }
 464 
 465 void TemplateTable::condy_helper(Label& Done)
 466 {
 467   Register obj = r0;
 468   Register rarg = r1;
 469   Register flags = r2;
 470   Register off = r3;
 471 
 472   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 473 
 474   __ mov(rarg, (int) bytecode());
 475   __ call_VM(obj, entry, rarg);
 476 
 477   __ get_vm_result_2(flags, rthread);
 478 
 479   // VMr = obj = base address to find primitive value to push
 480   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 481   __ mov(off, flags);
 482   __ andw(off, off, ConstantPoolCacheEntry::field_index_mask);
 483 
 484   const Address field(obj, off);
 485 
 486   // What sort of thing are we loading?
 487   // x86 uses a shift and mask or wings it with a shift plus assert
 488   // the mask is not needed. aarch64 just uses bitfield extract
 489   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,
 490            ConstantPoolCacheEntry::tos_state_bits);
 491 
 492   switch (bytecode()) {
 493     case Bytecodes::_ldc:
 494     case Bytecodes::_ldc_w:
 495       {
 496         // tos in (itos, ftos, stos, btos, ctos, ztos)
 497         Label notInt, notFloat, notShort, notByte, notChar, notBool;
 498         __ cmpw(flags, itos);
 499         __ br(Assembler::NE, notInt);
 500         // itos
 501         __ ldrw(r0, field);
 502         __ push(itos);
 503         __ b(Done);
 504 
 505         __ bind(notInt);
 506         __ cmpw(flags, ftos);
 507         __ br(Assembler::NE, notFloat);
 508         // ftos
 509         __ load_float(field);
 510         __ push(ftos);
 511         __ b(Done);
 512 
 513         __ bind(notFloat);
 514         __ cmpw(flags, stos);
 515         __ br(Assembler::NE, notShort);
 516         // stos
 517         __ load_signed_short(r0, field);
 518         __ push(stos);
 519         __ b(Done);
 520 
 521         __ bind(notShort);
 522         __ cmpw(flags, btos);
 523         __ br(Assembler::NE, notByte);
 524         // btos
 525         __ load_signed_byte(r0, field);
 526         __ push(btos);
 527         __ b(Done);
 528 
 529         __ bind(notByte);
 530         __ cmpw(flags, ctos);
 531         __ br(Assembler::NE, notChar);
 532         // ctos
 533         __ load_unsigned_short(r0, field);
 534         __ push(ctos);
 535         __ b(Done);
 536 
 537         __ bind(notChar);
 538         __ cmpw(flags, ztos);
 539         __ br(Assembler::NE, notBool);
 540         // ztos
 541         __ load_signed_byte(r0, field);
 542         __ push(ztos);
 543         __ b(Done);
 544 
 545         __ bind(notBool);
 546         break;
 547       }
 548 
 549     case Bytecodes::_ldc2_w:
 550       {
 551         Label notLong, notDouble;
 552         __ cmpw(flags, ltos);
 553         __ br(Assembler::NE, notLong);
 554         // ltos
 555         __ ldr(r0, field);
 556         __ push(ltos);
 557         __ b(Done);
 558 
 559         __ bind(notLong);
 560         __ cmpw(flags, dtos);
 561         __ br(Assembler::NE, notDouble);
 562         // dtos
 563         __ load_double(field);
 564         __ push(dtos);
 565         __ b(Done);
 566 
 567        __ bind(notDouble);
 568         break;
 569       }
 570 
 571     default:
 572       ShouldNotReachHere();
 573     }
 574 
 575     __ stop("bad ldc/condy");
 576 }
 577 
 578 void TemplateTable::locals_index(Register reg, int offset)
 579 {
 580   __ ldrb(reg, at_bcp(offset));
 581   __ neg(reg, reg);
 582 }
 583 
 584 void TemplateTable::iload() {
 585   iload_internal();
 586 }
 587 
 588 void TemplateTable::nofast_iload() {
 589   iload_internal(may_not_rewrite);
 590 }
 591 
 592 void TemplateTable::iload_internal(RewriteControl rc) {
 593   transition(vtos, itos);
 594   if (RewriteFrequentPairs && rc == may_rewrite) {
 595     Label rewrite, done;
 596     Register bc = r4;
 597 
 598     // get next bytecode
 599     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 600 
 601     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 602     // last two iloads in a pair.  Comparing against fast_iload means that
 603     // the next bytecode is neither an iload or a caload, and therefore
 604     // an iload pair.
 605     __ cmpw(r1, Bytecodes::_iload);
 606     __ br(Assembler::EQ, done);
 607 
 608     // if _fast_iload rewrite to _fast_iload2
 609     __ cmpw(r1, Bytecodes::_fast_iload);
 610     __ movw(bc, Bytecodes::_fast_iload2);
 611     __ br(Assembler::EQ, rewrite);
 612 
 613     // if _caload rewrite to _fast_icaload
 614     __ cmpw(r1, Bytecodes::_caload);
 615     __ movw(bc, Bytecodes::_fast_icaload);
 616     __ br(Assembler::EQ, rewrite);
 617 
 618     // else rewrite to _fast_iload
 619     __ movw(bc, Bytecodes::_fast_iload);
 620 
 621     // rewrite
 622     // bc: new bytecode
 623     __ bind(rewrite);
 624     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 625     __ bind(done);
 626 
 627   }
 628 
 629   // do iload, get the local value into tos
 630   locals_index(r1);
 631   __ ldr(r0, iaddress(r1));
 632 
 633 }
 634 
 635 void TemplateTable::fast_iload2()
 636 {
 637   transition(vtos, itos);
 638   locals_index(r1);
 639   __ ldr(r0, iaddress(r1));
 640   __ push(itos);
 641   locals_index(r1, 3);
 642   __ ldr(r0, iaddress(r1));
 643 }
 644 
 645 void TemplateTable::fast_iload()
 646 {
 647   transition(vtos, itos);
 648   locals_index(r1);
 649   __ ldr(r0, iaddress(r1));
 650 }
 651 
 652 void TemplateTable::lload()
 653 {
 654   transition(vtos, ltos);
 655   __ ldrb(r1, at_bcp(1));
 656   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 657   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 658 }
 659 
 660 void TemplateTable::fload()
 661 {
 662   transition(vtos, ftos);
 663   locals_index(r1);
 664   // n.b. we use ldrd here because this is a 64 bit slot
 665   // this is comparable to the iload case
 666   __ ldrd(v0, faddress(r1));
 667 }
 668 
 669 void TemplateTable::dload()
 670 {
 671   transition(vtos, dtos);
 672   __ ldrb(r1, at_bcp(1));
 673   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 674   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 675 }
 676 
 677 void TemplateTable::aload()
 678 {
 679   transition(vtos, atos);
 680   locals_index(r1);
 681   __ ldr(r0, iaddress(r1));
 682 }
 683 
 684 void TemplateTable::locals_index_wide(Register reg) {
 685   __ ldrh(reg, at_bcp(2));
 686   __ rev16w(reg, reg);
 687   __ neg(reg, reg);
 688 }
 689 
 690 void TemplateTable::wide_iload() {
 691   transition(vtos, itos);
 692   locals_index_wide(r1);
 693   __ ldr(r0, iaddress(r1));
 694 }
 695 
 696 void TemplateTable::wide_lload()
 697 {
 698   transition(vtos, ltos);
 699   __ ldrh(r1, at_bcp(2));
 700   __ rev16w(r1, r1);
 701   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 702   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 703 }
 704 
 705 void TemplateTable::wide_fload()
 706 {
 707   transition(vtos, ftos);
 708   locals_index_wide(r1);
 709   // n.b. we use ldrd here because this is a 64 bit slot
 710   // this is comparable to the iload case
 711   __ ldrd(v0, faddress(r1));
 712 }
 713 
 714 void TemplateTable::wide_dload()
 715 {
 716   transition(vtos, dtos);
 717   __ ldrh(r1, at_bcp(2));
 718   __ rev16w(r1, r1);
 719   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 720   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 721 }
 722 
 723 void TemplateTable::wide_aload()
 724 {
 725   transition(vtos, atos);
 726   locals_index_wide(r1);
 727   __ ldr(r0, aaddress(r1));
 728 }
 729 
 730 void TemplateTable::index_check(Register array, Register index)
 731 {
 732   // destroys r1, rscratch1
 733   // check array
 734   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 735   // sign extend index for use by indexed load
 736   // __ movl2ptr(index, index);
 737   // check index
 738   Register length = rscratch1;
 739   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 740   __ cmpw(index, length);
 741   if (index != r1) {
 742     // ??? convention: move aberrant index into r1 for exception message
 743     assert(r1 != array, "different registers");
 744     __ mov(r1, index);
 745   }
 746   Label ok;
 747   __ br(Assembler::LO, ok);
 748     // ??? convention: move array into r3 for exception message
 749   __ mov(r3, array);
 750   __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 751   __ br(rscratch1);
 752   __ bind(ok);
 753 }
 754 
 755 void TemplateTable::iaload()
 756 {
 757   transition(itos, itos);
 758   __ mov(r1, r0);
 759   __ pop_ptr(r0);
 760   // r0: array
 761   // r1: index
 762   index_check(r0, r1); // leaves index in r1, kills rscratch1
 763   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
 764   __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 765 }
 766 
 767 void TemplateTable::laload()
 768 {
 769   transition(itos, ltos);
 770   __ mov(r1, r0);
 771   __ pop_ptr(r0);
 772   // r0: array
 773   // r1: index
 774   index_check(r0, r1); // leaves index in r1, kills rscratch1
 775   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
 776   __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 777 }
 778 
 779 void TemplateTable::faload()
 780 {
 781   transition(itos, ftos);
 782   __ mov(r1, r0);
 783   __ pop_ptr(r0);
 784   // r0: array
 785   // r1: index
 786   index_check(r0, r1); // leaves index in r1, kills rscratch1
 787   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
 788   __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 789 }
 790 
 791 void TemplateTable::daload()
 792 {
 793   transition(itos, dtos);
 794   __ mov(r1, r0);
 795   __ pop_ptr(r0);
 796   // r0: array
 797   // r1: index
 798   index_check(r0, r1); // leaves index in r1, kills rscratch1
 799   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
 800   __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 801 }
 802 
 803 void TemplateTable::aaload()
 804 {
 805   transition(itos, atos);
 806   __ mov(r1, r0);
 807   __ pop_ptr(r0);
 808   // r0: array
 809   // r1: index
 810   index_check(r0, r1); // leaves index in r1, kills rscratch1
 811   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 812   do_oop_load(_masm,
 813               Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)),
 814               r0,
 815               IS_ARRAY);
 816 }
 817 
 818 void TemplateTable::baload()
 819 {
 820   transition(itos, itos);
 821   __ mov(r1, r0);
 822   __ pop_ptr(r0);
 823   // r0: array
 824   // r1: index
 825   index_check(r0, r1); // leaves index in r1, kills rscratch1
 826   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
 827   __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(0)), noreg, noreg);
 828 }
 829 
 830 void TemplateTable::caload()
 831 {
 832   transition(itos, itos);
 833   __ mov(r1, r0);
 834   __ pop_ptr(r0);
 835   // r0: array
 836   // r1: index
 837   index_check(r0, r1); // leaves index in r1, kills rscratch1
 838   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 839   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 840 }
 841 
 842 // iload followed by caload frequent pair
 843 void TemplateTable::fast_icaload()
 844 {
 845   transition(vtos, itos);
 846   // load index out of locals
 847   locals_index(r2);
 848   __ ldr(r1, iaddress(r2));
 849 
 850   __ pop_ptr(r0);
 851 
 852   // r0: array
 853   // r1: index
 854   index_check(r0, r1); // leaves index in r1, kills rscratch1
 855   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 856   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 857 }
 858 
 859 void TemplateTable::saload()
 860 {
 861   transition(itos, itos);
 862   __ mov(r1, r0);
 863   __ pop_ptr(r0);
 864   // r0: array
 865   // r1: index
 866   index_check(r0, r1); // leaves index in r1, kills rscratch1
 867   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_SHORT) >> 1);
 868   __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 869 }
 870 
 871 void TemplateTable::iload(int n)
 872 {
 873   transition(vtos, itos);
 874   __ ldr(r0, iaddress(n));
 875 }
 876 
 877 void TemplateTable::lload(int n)
 878 {
 879   transition(vtos, ltos);
 880   __ ldr(r0, laddress(n));
 881 }
 882 
 883 void TemplateTable::fload(int n)
 884 {
 885   transition(vtos, ftos);
 886   __ ldrs(v0, faddress(n));
 887 }
 888 
 889 void TemplateTable::dload(int n)
 890 {
 891   transition(vtos, dtos);
 892   __ ldrd(v0, daddress(n));
 893 }
 894 
 895 void TemplateTable::aload(int n)
 896 {
 897   transition(vtos, atos);
 898   __ ldr(r0, iaddress(n));
 899 }
 900 
 901 void TemplateTable::aload_0() {
 902   aload_0_internal();
 903 }
 904 
 905 void TemplateTable::nofast_aload_0() {
 906   aload_0_internal(may_not_rewrite);
 907 }
 908 
 909 void TemplateTable::aload_0_internal(RewriteControl rc) {
 910   // According to bytecode histograms, the pairs:
 911   //
 912   // _aload_0, _fast_igetfield
 913   // _aload_0, _fast_agetfield
 914   // _aload_0, _fast_fgetfield
 915   //
 916   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 917   // _aload_0 bytecode checks if the next bytecode is either
 918   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 919   // rewrites the current bytecode into a pair bytecode; otherwise it
 920   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 921   // the pair check anymore.
 922   //
 923   // Note: If the next bytecode is _getfield, the rewrite must be
 924   //       delayed, otherwise we may miss an opportunity for a pair.
 925   //
 926   // Also rewrite frequent pairs
 927   //   aload_0, aload_1
 928   //   aload_0, iload_1
 929   // These bytecodes with a small amount of code are most profitable
 930   // to rewrite
 931   if (RewriteFrequentPairs && rc == may_rewrite) {
 932     Label rewrite, done;
 933     const Register bc = r4;
 934 
 935     // get next bytecode
 936     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 937 
 938     // if _getfield then wait with rewrite
 939     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 940     __ br(Assembler::EQ, done);
 941 
 942     // if _igetfield then rewrite to _fast_iaccess_0
 943     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 944     __ cmpw(r1, Bytecodes::_fast_igetfield);
 945     __ movw(bc, Bytecodes::_fast_iaccess_0);
 946     __ br(Assembler::EQ, rewrite);
 947 
 948     // if _agetfield then rewrite to _fast_aaccess_0
 949     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 950     __ cmpw(r1, Bytecodes::_fast_agetfield);
 951     __ movw(bc, Bytecodes::_fast_aaccess_0);
 952     __ br(Assembler::EQ, rewrite);
 953 
 954     // if _fgetfield then rewrite to _fast_faccess_0
 955     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 956     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 957     __ movw(bc, Bytecodes::_fast_faccess_0);
 958     __ br(Assembler::EQ, rewrite);
 959 
 960     // else rewrite to _fast_aload0
 961     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 962     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 963 
 964     // rewrite
 965     // bc: new bytecode
 966     __ bind(rewrite);
 967     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 968 
 969     __ bind(done);
 970   }
 971 
 972   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 973   aload(0);
 974 }
 975 
 976 void TemplateTable::istore()
 977 {
 978   transition(itos, vtos);
 979   locals_index(r1);
 980   // FIXME: We're being very pernickerty here storing a jint in a
 981   // local with strw, which costs an extra instruction over what we'd
 982   // be able to do with a simple str.  We should just store the whole
 983   // word.
 984   __ lea(rscratch1, iaddress(r1));
 985   __ strw(r0, Address(rscratch1));
 986 }
 987 
 988 void TemplateTable::lstore()
 989 {
 990   transition(ltos, vtos);
 991   locals_index(r1);
 992   __ str(r0, laddress(r1, rscratch1, _masm));
 993 }
 994 
 995 void TemplateTable::fstore() {
 996   transition(ftos, vtos);
 997   locals_index(r1);
 998   __ lea(rscratch1, iaddress(r1));
 999   __ strs(v0, Address(rscratch1));
1000 }
1001 
1002 void TemplateTable::dstore() {
1003   transition(dtos, vtos);
1004   locals_index(r1);
1005   __ strd(v0, daddress(r1, rscratch1, _masm));
1006 }
1007 
1008 void TemplateTable::astore()
1009 {
1010   transition(vtos, vtos);
1011   __ pop_ptr(r0);
1012   locals_index(r1);
1013   __ str(r0, aaddress(r1));
1014 }
1015 
1016 void TemplateTable::wide_istore() {
1017   transition(vtos, vtos);
1018   __ pop_i();
1019   locals_index_wide(r1);
1020   __ lea(rscratch1, iaddress(r1));
1021   __ strw(r0, Address(rscratch1));
1022 }
1023 
1024 void TemplateTable::wide_lstore() {
1025   transition(vtos, vtos);
1026   __ pop_l();
1027   locals_index_wide(r1);
1028   __ str(r0, laddress(r1, rscratch1, _masm));
1029 }
1030 
1031 void TemplateTable::wide_fstore() {
1032   transition(vtos, vtos);
1033   __ pop_f();
1034   locals_index_wide(r1);
1035   __ lea(rscratch1, faddress(r1));
1036   __ strs(v0, rscratch1);
1037 }
1038 
1039 void TemplateTable::wide_dstore() {
1040   transition(vtos, vtos);
1041   __ pop_d();
1042   locals_index_wide(r1);
1043   __ strd(v0, daddress(r1, rscratch1, _masm));
1044 }
1045 
1046 void TemplateTable::wide_astore() {
1047   transition(vtos, vtos);
1048   __ pop_ptr(r0);
1049   locals_index_wide(r1);
1050   __ str(r0, aaddress(r1));
1051 }
1052 
1053 void TemplateTable::iastore() {
1054   transition(itos, vtos);
1055   __ pop_i(r1);
1056   __ pop_ptr(r3);
1057   // r0: value
1058   // r1: index
1059   // r3: array
1060   index_check(r3, r1); // prefer index in r1
1061   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
1062   __ access_store_at(T_INT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), r0, noreg, noreg, noreg);
1063 }
1064 
1065 void TemplateTable::lastore() {
1066   transition(ltos, vtos);
1067   __ pop_i(r1);
1068   __ pop_ptr(r3);
1069   // r0: value
1070   // r1: index
1071   // r3: array
1072   index_check(r3, r1); // prefer index in r1
1073   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
1074   __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), r0, noreg, noreg, noreg);
1075 }
1076 
1077 void TemplateTable::fastore() {
1078   transition(ftos, vtos);
1079   __ pop_i(r1);
1080   __ pop_ptr(r3);
1081   // v0: value
1082   // r1:  index
1083   // r3:  array
1084   index_check(r3, r1); // prefer index in r1
1085   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
1086   __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), noreg /* ftos */, noreg, noreg, noreg);
1087 }
1088 
1089 void TemplateTable::dastore() {
1090   transition(dtos, vtos);
1091   __ pop_i(r1);
1092   __ pop_ptr(r3);
1093   // v0: value
1094   // r1:  index
1095   // r3:  array
1096   index_check(r3, r1); // prefer index in r1
1097   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
1098   __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), noreg /* dtos */, noreg, noreg, noreg);
1099 }
1100 
1101 void TemplateTable::aastore() {
1102   Label is_null, ok_is_subtype, done;
1103   transition(vtos, vtos);
1104   // stack: ..., array, index, value
1105   __ ldr(r0, at_tos());    // value
1106   __ ldr(r2, at_tos_p1()); // index
1107   __ ldr(r3, at_tos_p2()); // array
1108 
1109   Address element_address(r3, r4, Address::uxtw(LogBytesPerHeapOop));
1110 
1111   index_check(r3, r2);     // kills r1
1112   __ add(r4, r2, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
1113 
1114   // do array store check - check for NULL value first
1115   __ cbz(r0, is_null);
1116 
1117   // Move subklass into r1
1118   __ load_klass(r1, r0);
1119   // Move superklass into r0
1120   __ load_klass(r0, r3);
1121   __ ldr(r0, Address(r0,
1122                      ObjArrayKlass::element_klass_offset()));
1123   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1124 
1125   // Generate subtype check.  Blows r2, r5
1126   // Superklass in r0.  Subklass in r1.
1127   __ gen_subtype_check(r1, ok_is_subtype);
1128 
1129   // Come here on failure
1130   // object is at TOS
1131   __ b(Interpreter::_throw_ArrayStoreException_entry);
1132 
1133   // Come here on success
1134   __ bind(ok_is_subtype);
1135 
1136   // Get the value we will store
1137   __ ldr(r0, at_tos());
1138   // Now store using the appropriate barrier
1139   do_oop_store(_masm, element_address, r0, IS_ARRAY);
1140   __ b(done);
1141 
1142   // Have a NULL in r0, r3=array, r2=index.  Store NULL at ary[idx]
1143   __ bind(is_null);
1144   __ profile_null_seen(r2);
1145 
1146   // Store a NULL
1147   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1148 
1149   // Pop stack arguments
1150   __ bind(done);
1151   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1152 }
1153 
1154 void TemplateTable::bastore()
1155 {
1156   transition(itos, vtos);
1157   __ pop_i(r1);
1158   __ pop_ptr(r3);
1159   // r0: value
1160   // r1: index
1161   // r3: array
1162   index_check(r3, r1); // prefer index in r1
1163 
1164   // Need to check whether array is boolean or byte
1165   // since both types share the bastore bytecode.
1166   __ load_klass(r2, r3);
1167   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1168   int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1169   Label L_skip;
1170   __ tbz(r2, diffbit_index, L_skip);
1171   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1172   __ bind(L_skip);
1173 
1174   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
1175   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(0)), r0, noreg, noreg, noreg);
1176 }
1177 
1178 void TemplateTable::castore()
1179 {
1180   transition(itos, vtos);
1181   __ pop_i(r1);
1182   __ pop_ptr(r3);
1183   // r0: value
1184   // r1: index
1185   // r3: array
1186   index_check(r3, r1); // prefer index in r1
1187   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
1188   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(1)), r0, noreg, noreg, noreg);
1189 }
1190 
1191 void TemplateTable::sastore()
1192 {
1193   castore();
1194 }
1195 
1196 void TemplateTable::istore(int n)
1197 {
1198   transition(itos, vtos);
1199   __ str(r0, iaddress(n));
1200 }
1201 
1202 void TemplateTable::lstore(int n)
1203 {
1204   transition(ltos, vtos);
1205   __ str(r0, laddress(n));
1206 }
1207 
1208 void TemplateTable::fstore(int n)
1209 {
1210   transition(ftos, vtos);
1211   __ strs(v0, faddress(n));
1212 }
1213 
1214 void TemplateTable::dstore(int n)
1215 {
1216   transition(dtos, vtos);
1217   __ strd(v0, daddress(n));
1218 }
1219 
1220 void TemplateTable::astore(int n)
1221 {
1222   transition(vtos, vtos);
1223   __ pop_ptr(r0);
1224   __ str(r0, iaddress(n));
1225 }
1226 
1227 void TemplateTable::pop()
1228 {
1229   transition(vtos, vtos);
1230   __ add(esp, esp, Interpreter::stackElementSize);
1231 }
1232 
1233 void TemplateTable::pop2()
1234 {
1235   transition(vtos, vtos);
1236   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1237 }
1238 
1239 void TemplateTable::dup()
1240 {
1241   transition(vtos, vtos);
1242   __ ldr(r0, Address(esp, 0));
1243   __ push(r0);
1244   // stack: ..., a, a
1245 }
1246 
1247 void TemplateTable::dup_x1()
1248 {
1249   transition(vtos, vtos);
1250   // stack: ..., a, b
1251   __ ldr(r0, at_tos());  // load b
1252   __ ldr(r2, at_tos_p1());  // load a
1253   __ str(r0, at_tos_p1());  // store b
1254   __ str(r2, at_tos());  // store a
1255   __ push(r0);                  // push b
1256   // stack: ..., b, a, b
1257 }
1258 
1259 void TemplateTable::dup_x2()
1260 {
1261   transition(vtos, vtos);
1262   // stack: ..., a, b, c
1263   __ ldr(r0, at_tos());  // load c
1264   __ ldr(r2, at_tos_p2());  // load a
1265   __ str(r0, at_tos_p2());  // store c in a
1266   __ push(r0);      // push c
1267   // stack: ..., c, b, c, c
1268   __ ldr(r0, at_tos_p2());  // load b
1269   __ str(r2, at_tos_p2());  // store a in b
1270   // stack: ..., c, a, c, c
1271   __ str(r0, at_tos_p1());  // store b in c
1272   // stack: ..., c, a, b, c
1273 }
1274 
1275 void TemplateTable::dup2()
1276 {
1277   transition(vtos, vtos);
1278   // stack: ..., a, b
1279   __ ldr(r0, at_tos_p1());  // load a
1280   __ push(r0);                  // push a
1281   __ ldr(r0, at_tos_p1());  // load b
1282   __ push(r0);                  // push b
1283   // stack: ..., a, b, a, b
1284 }
1285 
1286 void TemplateTable::dup2_x1()
1287 {
1288   transition(vtos, vtos);
1289   // stack: ..., a, b, c
1290   __ ldr(r2, at_tos());  // load c
1291   __ ldr(r0, at_tos_p1());  // load b
1292   __ push(r0);                  // push b
1293   __ push(r2);                  // push c
1294   // stack: ..., a, b, c, b, c
1295   __ str(r2, at_tos_p3());  // store c in b
1296   // stack: ..., a, c, c, b, c
1297   __ ldr(r2, at_tos_p4());  // load a
1298   __ str(r2, at_tos_p2());  // store a in 2nd c
1299   // stack: ..., a, c, a, b, c
1300   __ str(r0, at_tos_p4());  // store b in a
1301   // stack: ..., b, c, a, b, c
1302 }
1303 
1304 void TemplateTable::dup2_x2()
1305 {
1306   transition(vtos, vtos);
1307   // stack: ..., a, b, c, d
1308   __ ldr(r2, at_tos());  // load d
1309   __ ldr(r0, at_tos_p1());  // load c
1310   __ push(r0)            ;      // push c
1311   __ push(r2);                  // push d
1312   // stack: ..., a, b, c, d, c, d
1313   __ ldr(r0, at_tos_p4());  // load b
1314   __ str(r0, at_tos_p2());  // store b in d
1315   __ str(r2, at_tos_p4());  // store d in b
1316   // stack: ..., a, d, c, b, c, d
1317   __ ldr(r2, at_tos_p5());  // load a
1318   __ ldr(r0, at_tos_p3());  // load c
1319   __ str(r2, at_tos_p3());  // store a in c
1320   __ str(r0, at_tos_p5());  // store c in a
1321   // stack: ..., c, d, a, b, c, d
1322 }
1323 
1324 void TemplateTable::swap()
1325 {
1326   transition(vtos, vtos);
1327   // stack: ..., a, b
1328   __ ldr(r2, at_tos_p1());  // load a
1329   __ ldr(r0, at_tos());  // load b
1330   __ str(r2, at_tos());  // store a in b
1331   __ str(r0, at_tos_p1());  // store b in a
1332   // stack: ..., b, a
1333 }
1334 
1335 void TemplateTable::iop2(Operation op)
1336 {
1337   transition(itos, itos);
1338   // r0 <== r1 op r0
1339   __ pop_i(r1);
1340   switch (op) {
1341   case add  : __ addw(r0, r1, r0); break;
1342   case sub  : __ subw(r0, r1, r0); break;
1343   case mul  : __ mulw(r0, r1, r0); break;
1344   case _and : __ andw(r0, r1, r0); break;
1345   case _or  : __ orrw(r0, r1, r0); break;
1346   case _xor : __ eorw(r0, r1, r0); break;
1347   case shl  : __ lslvw(r0, r1, r0); break;
1348   case shr  : __ asrvw(r0, r1, r0); break;
1349   case ushr : __ lsrvw(r0, r1, r0);break;
1350   default   : ShouldNotReachHere();
1351   }
1352 }
1353 
1354 void TemplateTable::lop2(Operation op)
1355 {
1356   transition(ltos, ltos);
1357   // r0 <== r1 op r0
1358   __ pop_l(r1);
1359   switch (op) {
1360   case add  : __ add(r0, r1, r0); break;
1361   case sub  : __ sub(r0, r1, r0); break;
1362   case mul  : __ mul(r0, r1, r0); break;
1363   case _and : __ andr(r0, r1, r0); break;
1364   case _or  : __ orr(r0, r1, r0); break;
1365   case _xor : __ eor(r0, r1, r0); break;
1366   default   : ShouldNotReachHere();
1367   }
1368 }
1369 
1370 void TemplateTable::idiv()
1371 {
1372   transition(itos, itos);
1373   // explicitly check for div0
1374   Label no_div0;
1375   __ cbnzw(r0, no_div0);
1376   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1377   __ br(rscratch1);
1378   __ bind(no_div0);
1379   __ pop_i(r1);
1380   // r0 <== r1 idiv r0
1381   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1382 }
1383 
1384 void TemplateTable::irem()
1385 {
1386   transition(itos, itos);
1387   // explicitly check for div0
1388   Label no_div0;
1389   __ cbnzw(r0, no_div0);
1390   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1391   __ br(rscratch1);
1392   __ bind(no_div0);
1393   __ pop_i(r1);
1394   // r0 <== r1 irem r0
1395   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1396 }
1397 
1398 void TemplateTable::lmul()
1399 {
1400   transition(ltos, ltos);
1401   __ pop_l(r1);
1402   __ mul(r0, r0, r1);
1403 }
1404 
1405 void TemplateTable::ldiv()
1406 {
1407   transition(ltos, ltos);
1408   // explicitly check for div0
1409   Label no_div0;
1410   __ cbnz(r0, no_div0);
1411   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1412   __ br(rscratch1);
1413   __ bind(no_div0);
1414   __ pop_l(r1);
1415   // r0 <== r1 ldiv r0
1416   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1417 }
1418 
1419 void TemplateTable::lrem()
1420 {
1421   transition(ltos, ltos);
1422   // explicitly check for div0
1423   Label no_div0;
1424   __ cbnz(r0, no_div0);
1425   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1426   __ br(rscratch1);
1427   __ bind(no_div0);
1428   __ pop_l(r1);
1429   // r0 <== r1 lrem r0
1430   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1431 }
1432 
1433 void TemplateTable::lshl()
1434 {
1435   transition(itos, ltos);
1436   // shift count is in r0
1437   __ pop_l(r1);
1438   __ lslv(r0, r1, r0);
1439 }
1440 
1441 void TemplateTable::lshr()
1442 {
1443   transition(itos, ltos);
1444   // shift count is in r0
1445   __ pop_l(r1);
1446   __ asrv(r0, r1, r0);
1447 }
1448 
1449 void TemplateTable::lushr()
1450 {
1451   transition(itos, ltos);
1452   // shift count is in r0
1453   __ pop_l(r1);
1454   __ lsrv(r0, r1, r0);
1455 }
1456 
1457 void TemplateTable::fop2(Operation op)
1458 {
1459   transition(ftos, ftos);
1460   switch (op) {
1461   case add:
1462     // n.b. use ldrd because this is a 64 bit slot
1463     __ pop_f(v1);
1464     __ fadds(v0, v1, v0);
1465     break;
1466   case sub:
1467     __ pop_f(v1);
1468     __ fsubs(v0, v1, v0);
1469     break;
1470   case mul:
1471     __ pop_f(v1);
1472     __ fmuls(v0, v1, v0);
1473     break;
1474   case div:
1475     __ pop_f(v1);
1476     __ fdivs(v0, v1, v0);
1477     break;
1478   case rem:
1479     __ fmovs(v1, v0);
1480     __ pop_f(v0);
1481     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1482     break;
1483   default:
1484     ShouldNotReachHere();
1485     break;
1486   }
1487 }
1488 
1489 void TemplateTable::dop2(Operation op)
1490 {
1491   transition(dtos, dtos);
1492   switch (op) {
1493   case add:
1494     // n.b. use ldrd because this is a 64 bit slot
1495     __ pop_d(v1);
1496     __ faddd(v0, v1, v0);
1497     break;
1498   case sub:
1499     __ pop_d(v1);
1500     __ fsubd(v0, v1, v0);
1501     break;
1502   case mul:
1503     __ pop_d(v1);
1504     __ fmuld(v0, v1, v0);
1505     break;
1506   case div:
1507     __ pop_d(v1);
1508     __ fdivd(v0, v1, v0);
1509     break;
1510   case rem:
1511     __ fmovd(v1, v0);
1512     __ pop_d(v0);
1513     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1514     break;
1515   default:
1516     ShouldNotReachHere();
1517     break;
1518   }
1519 }
1520 
1521 void TemplateTable::ineg()
1522 {
1523   transition(itos, itos);
1524   __ negw(r0, r0);
1525 
1526 }
1527 
1528 void TemplateTable::lneg()
1529 {
1530   transition(ltos, ltos);
1531   __ neg(r0, r0);
1532 }
1533 
1534 void TemplateTable::fneg()
1535 {
1536   transition(ftos, ftos);
1537   __ fnegs(v0, v0);
1538 }
1539 
1540 void TemplateTable::dneg()
1541 {
1542   transition(dtos, dtos);
1543   __ fnegd(v0, v0);
1544 }
1545 
1546 void TemplateTable::iinc()
1547 {
1548   transition(vtos, vtos);
1549   __ load_signed_byte(r1, at_bcp(2)); // get constant
1550   locals_index(r2);
1551   __ ldr(r0, iaddress(r2));
1552   __ addw(r0, r0, r1);
1553   __ str(r0, iaddress(r2));
1554 }
1555 
1556 void TemplateTable::wide_iinc()
1557 {
1558   transition(vtos, vtos);
1559   // __ mov(r1, zr);
1560   __ ldrw(r1, at_bcp(2)); // get constant and index
1561   __ rev16(r1, r1);
1562   __ ubfx(r2, r1, 0, 16);
1563   __ neg(r2, r2);
1564   __ sbfx(r1, r1, 16, 16);
1565   __ ldr(r0, iaddress(r2));
1566   __ addw(r0, r0, r1);
1567   __ str(r0, iaddress(r2));
1568 }
1569 
1570 void TemplateTable::convert()
1571 {
1572   // Checking
1573 #ifdef ASSERT
1574   {
1575     TosState tos_in  = ilgl;
1576     TosState tos_out = ilgl;
1577     switch (bytecode()) {
1578     case Bytecodes::_i2l: // fall through
1579     case Bytecodes::_i2f: // fall through
1580     case Bytecodes::_i2d: // fall through
1581     case Bytecodes::_i2b: // fall through
1582     case Bytecodes::_i2c: // fall through
1583     case Bytecodes::_i2s: tos_in = itos; break;
1584     case Bytecodes::_l2i: // fall through
1585     case Bytecodes::_l2f: // fall through
1586     case Bytecodes::_l2d: tos_in = ltos; break;
1587     case Bytecodes::_f2i: // fall through
1588     case Bytecodes::_f2l: // fall through
1589     case Bytecodes::_f2d: tos_in = ftos; break;
1590     case Bytecodes::_d2i: // fall through
1591     case Bytecodes::_d2l: // fall through
1592     case Bytecodes::_d2f: tos_in = dtos; break;
1593     default             : ShouldNotReachHere();
1594     }
1595     switch (bytecode()) {
1596     case Bytecodes::_l2i: // fall through
1597     case Bytecodes::_f2i: // fall through
1598     case Bytecodes::_d2i: // fall through
1599     case Bytecodes::_i2b: // fall through
1600     case Bytecodes::_i2c: // fall through
1601     case Bytecodes::_i2s: tos_out = itos; break;
1602     case Bytecodes::_i2l: // fall through
1603     case Bytecodes::_f2l: // fall through
1604     case Bytecodes::_d2l: tos_out = ltos; break;
1605     case Bytecodes::_i2f: // fall through
1606     case Bytecodes::_l2f: // fall through
1607     case Bytecodes::_d2f: tos_out = ftos; break;
1608     case Bytecodes::_i2d: // fall through
1609     case Bytecodes::_l2d: // fall through
1610     case Bytecodes::_f2d: tos_out = dtos; break;
1611     default             : ShouldNotReachHere();
1612     }
1613     transition(tos_in, tos_out);
1614   }
1615 #endif // ASSERT
1616   // static const int64_t is_nan = 0x8000000000000000L;
1617 
1618   // Conversion
1619   switch (bytecode()) {
1620   case Bytecodes::_i2l:
1621     __ sxtw(r0, r0);
1622     break;
1623   case Bytecodes::_i2f:
1624     __ scvtfws(v0, r0);
1625     break;
1626   case Bytecodes::_i2d:
1627     __ scvtfwd(v0, r0);
1628     break;
1629   case Bytecodes::_i2b:
1630     __ sxtbw(r0, r0);
1631     break;
1632   case Bytecodes::_i2c:
1633     __ uxthw(r0, r0);
1634     break;
1635   case Bytecodes::_i2s:
1636     __ sxthw(r0, r0);
1637     break;
1638   case Bytecodes::_l2i:
1639     __ uxtw(r0, r0);
1640     break;
1641   case Bytecodes::_l2f:
1642     __ scvtfs(v0, r0);
1643     break;
1644   case Bytecodes::_l2d:
1645     __ scvtfd(v0, r0);
1646     break;
1647   case Bytecodes::_f2i:
1648   {
1649     Label L_Okay;
1650     __ clear_fpsr();
1651     __ fcvtzsw(r0, v0);
1652     __ get_fpsr(r1);
1653     __ cbzw(r1, L_Okay);
1654     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i));
1655     __ bind(L_Okay);
1656   }
1657     break;
1658   case Bytecodes::_f2l:
1659   {
1660     Label L_Okay;
1661     __ clear_fpsr();
1662     __ fcvtzs(r0, v0);
1663     __ get_fpsr(r1);
1664     __ cbzw(r1, L_Okay);
1665     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1666     __ bind(L_Okay);
1667   }
1668     break;
1669   case Bytecodes::_f2d:
1670     __ fcvts(v0, v0);
1671     break;
1672   case Bytecodes::_d2i:
1673   {
1674     Label L_Okay;
1675     __ clear_fpsr();
1676     __ fcvtzdw(r0, v0);
1677     __ get_fpsr(r1);
1678     __ cbzw(r1, L_Okay);
1679     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1680     __ bind(L_Okay);
1681   }
1682     break;
1683   case Bytecodes::_d2l:
1684   {
1685     Label L_Okay;
1686     __ clear_fpsr();
1687     __ fcvtzd(r0, v0);
1688     __ get_fpsr(r1);
1689     __ cbzw(r1, L_Okay);
1690     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1691     __ bind(L_Okay);
1692   }
1693     break;
1694   case Bytecodes::_d2f:
1695     __ fcvtd(v0, v0);
1696     break;
1697   default:
1698     ShouldNotReachHere();
1699   }
1700 }
1701 
1702 void TemplateTable::lcmp()
1703 {
1704   transition(ltos, itos);
1705   Label done;
1706   __ pop_l(r1);
1707   __ cmp(r1, r0);
1708   __ mov(r0, (uint64_t)-1L);
1709   __ br(Assembler::LT, done);
1710   // __ mov(r0, 1UL);
1711   // __ csel(r0, r0, zr, Assembler::NE);
1712   // and here is a faster way
1713   __ csinc(r0, zr, zr, Assembler::EQ);
1714   __ bind(done);
1715 }
1716 
1717 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1718 {
1719   Label done;
1720   if (is_float) {
1721     // XXX get rid of pop here, use ... reg, mem32
1722     __ pop_f(v1);
1723     __ fcmps(v1, v0);
1724   } else {
1725     // XXX get rid of pop here, use ... reg, mem64
1726     __ pop_d(v1);
1727     __ fcmpd(v1, v0);
1728   }
1729   if (unordered_result < 0) {
1730     // we want -1 for unordered or less than, 0 for equal and 1 for
1731     // greater than.
1732     __ mov(r0, (uint64_t)-1L);
1733     // for FP LT tests less than or unordered
1734     __ br(Assembler::LT, done);
1735     // install 0 for EQ otherwise 1
1736     __ csinc(r0, zr, zr, Assembler::EQ);
1737   } else {
1738     // we want -1 for less than, 0 for equal and 1 for unordered or
1739     // greater than.
1740     __ mov(r0, 1L);
1741     // for FP HI tests greater than or unordered
1742     __ br(Assembler::HI, done);
1743     // install 0 for EQ otherwise ~0
1744     __ csinv(r0, zr, zr, Assembler::EQ);
1745 
1746   }
1747   __ bind(done);
1748 }
1749 
1750 void TemplateTable::branch(bool is_jsr, bool is_wide)
1751 {
1752   // We might be moving to a safepoint.  The thread which calls
1753   // Interpreter::notice_safepoints() will effectively flush its cache
1754   // when it makes a system call, but we need to do something to
1755   // ensure that we see the changed dispatch table.
1756   __ membar(MacroAssembler::LoadLoad);
1757 
1758   __ profile_taken_branch(r0, r1);
1759   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1760                              InvocationCounter::counter_offset();
1761   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1762                               InvocationCounter::counter_offset();
1763 
1764   // load branch displacement
1765   if (!is_wide) {
1766     __ ldrh(r2, at_bcp(1));
1767     __ rev16(r2, r2);
1768     // sign extend the 16 bit value in r2
1769     __ sbfm(r2, r2, 0, 15);
1770   } else {
1771     __ ldrw(r2, at_bcp(1));
1772     __ revw(r2, r2);
1773     // sign extend the 32 bit value in r2
1774     __ sbfm(r2, r2, 0, 31);
1775   }
1776 
1777   // Handle all the JSR stuff here, then exit.
1778   // It's much shorter and cleaner than intermingling with the non-JSR
1779   // normal-branch stuff occurring below.
1780 
1781   if (is_jsr) {
1782     // Pre-load the next target bytecode into rscratch1
1783     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1784     // compute return address as bci
1785     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1786     __ add(rscratch2, rscratch2,
1787            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1788     __ sub(r1, rbcp, rscratch2);
1789     __ push_i(r1);
1790     // Adjust the bcp by the 16-bit displacement in r2
1791     __ add(rbcp, rbcp, r2);
1792     __ dispatch_only(vtos, /*generate_poll*/true);
1793     return;
1794   }
1795 
1796   // Normal (non-jsr) branch handling
1797 
1798   // Adjust the bcp by the displacement in r2
1799   __ add(rbcp, rbcp, r2);
1800 
1801   assert(UseLoopCounter || !UseOnStackReplacement,
1802          "on-stack-replacement requires loop counters");
1803   Label backedge_counter_overflow;
1804   Label dispatch;
1805   if (UseLoopCounter) {
1806     // increment backedge counter for backward branches
1807     // r0: MDO
1808     // w1: MDO bumped taken-count
1809     // r2: target offset
1810     __ cmp(r2, zr);
1811     __ br(Assembler::GT, dispatch); // count only if backward branch
1812 
1813     // ECN: FIXME: This code smells
1814     // check if MethodCounters exists
1815     Label has_counters;
1816     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1817     __ cbnz(rscratch1, has_counters);
1818     __ push(r0);
1819     __ push(r1);
1820     __ push(r2);
1821     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1822             InterpreterRuntime::build_method_counters), rmethod);
1823     __ pop(r2);
1824     __ pop(r1);
1825     __ pop(r0);
1826     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1827     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1828     __ bind(has_counters);
1829 
1830     Label no_mdo;
1831     int increment = InvocationCounter::count_increment;
1832     if (ProfileInterpreter) {
1833       // Are we profiling?
1834       __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1835       __ cbz(r1, no_mdo);
1836       // Increment the MDO backedge counter
1837       const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1838                                          in_bytes(InvocationCounter::counter_offset()));
1839       const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1840       __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1841                                  r0, rscratch1, false, Assembler::EQ,
1842                                  UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1843       __ b(dispatch);
1844     }
1845     __ bind(no_mdo);
1846     // Increment backedge counter in MethodCounters*
1847     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1848     const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1849     __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1850                                r0, rscratch2, false, Assembler::EQ,
1851                                UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1852     __ bind(dispatch);
1853   }
1854 
1855   // Pre-load the next target bytecode into rscratch1
1856   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1857 
1858   // continue with the bytecode @ target
1859   // rscratch1: target bytecode
1860   // rbcp: target bcp
1861   __ dispatch_only(vtos, /*generate_poll*/true);
1862 
1863   if (UseLoopCounter && UseOnStackReplacement) {
1864     // invocation counter overflow
1865     __ bind(backedge_counter_overflow);
1866     __ neg(r2, r2);
1867     __ add(r2, r2, rbcp);     // branch bcp
1868     // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1869     __ call_VM(noreg,
1870                CAST_FROM_FN_PTR(address,
1871                                 InterpreterRuntime::frequency_counter_overflow),
1872                r2);
1873     __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1874 
1875     // r0: osr nmethod (osr ok) or NULL (osr not possible)
1876     // w1: target bytecode
1877     // r2: scratch
1878     __ cbz(r0, dispatch);     // test result -- no osr if null
1879     // nmethod may have been invalidated (VM may block upon call_VM return)
1880     __ ldrb(r2, Address(r0, nmethod::state_offset()));
1881     if (nmethod::in_use != 0)
1882       __ sub(r2, r2, nmethod::in_use);
1883     __ cbnz(r2, dispatch);
1884 
1885     // We have the address of an on stack replacement routine in r0
1886     // We need to prepare to execute the OSR method. First we must
1887     // migrate the locals and monitors off of the stack.
1888 
1889     __ mov(r19, r0);                             // save the nmethod
1890 
1891     call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1892 
1893     // r0 is OSR buffer, move it to expected parameter location
1894     __ mov(j_rarg0, r0);
1895 
1896     // remove activation
1897     // get sender esp
1898     __ ldr(esp,
1899         Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1900     // remove frame anchor
1901     __ leave();
1902     // Ensure compiled code always sees stack at proper alignment
1903     __ andr(sp, esp, -16);
1904 
1905     // and begin the OSR nmethod
1906     __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1907     __ br(rscratch1);
1908   }
1909 }
1910 
1911 
1912 void TemplateTable::if_0cmp(Condition cc)
1913 {
1914   transition(itos, vtos);
1915   // assume branch is more often taken than not (loops use backward branches)
1916   Label not_taken;
1917   if (cc == equal)
1918     __ cbnzw(r0, not_taken);
1919   else if (cc == not_equal)
1920     __ cbzw(r0, not_taken);
1921   else {
1922     __ andsw(zr, r0, r0);
1923     __ br(j_not(cc), not_taken);
1924   }
1925 
1926   branch(false, false);
1927   __ bind(not_taken);
1928   __ profile_not_taken_branch(r0);
1929 }
1930 
1931 void TemplateTable::if_icmp(Condition cc)
1932 {
1933   transition(itos, vtos);
1934   // assume branch is more often taken than not (loops use backward branches)
1935   Label not_taken;
1936   __ pop_i(r1);
1937   __ cmpw(r1, r0, Assembler::LSL);
1938   __ br(j_not(cc), not_taken);
1939   branch(false, false);
1940   __ bind(not_taken);
1941   __ profile_not_taken_branch(r0);
1942 }
1943 
1944 void TemplateTable::if_nullcmp(Condition cc)
1945 {
1946   transition(atos, vtos);
1947   // assume branch is more often taken than not (loops use backward branches)
1948   Label not_taken;
1949   if (cc == equal)
1950     __ cbnz(r0, not_taken);
1951   else
1952     __ cbz(r0, not_taken);
1953   branch(false, false);
1954   __ bind(not_taken);
1955   __ profile_not_taken_branch(r0);
1956 }
1957 
1958 void TemplateTable::if_acmp(Condition cc)
1959 {
1960   transition(atos, vtos);
1961   // assume branch is more often taken than not (loops use backward branches)
1962   Label not_taken;
1963   __ pop_ptr(r1);
1964   __ cmpoop(r1, r0);
1965   __ br(j_not(cc), not_taken);
1966   branch(false, false);
1967   __ bind(not_taken);
1968   __ profile_not_taken_branch(r0);
1969 }
1970 
1971 void TemplateTable::ret() {
1972   transition(vtos, vtos);
1973   // We might be moving to a safepoint.  The thread which calls
1974   // Interpreter::notice_safepoints() will effectively flush its cache
1975   // when it makes a system call, but we need to do something to
1976   // ensure that we see the changed dispatch table.
1977   __ membar(MacroAssembler::LoadLoad);
1978 
1979   locals_index(r1);
1980   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1981   __ profile_ret(r1, r2);
1982   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1983   __ lea(rbcp, Address(rbcp, r1));
1984   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1985   __ dispatch_next(vtos, 0, /*generate_poll*/true);
1986 }
1987 
1988 void TemplateTable::wide_ret() {
1989   transition(vtos, vtos);
1990   locals_index_wide(r1);
1991   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1992   __ profile_ret(r1, r2);
1993   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1994   __ lea(rbcp, Address(rbcp, r1));
1995   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1996   __ dispatch_next(vtos, 0, /*generate_poll*/true);
1997 }
1998 
1999 
2000 void TemplateTable::tableswitch() {
2001   Label default_case, continue_execution;
2002   transition(itos, vtos);
2003   // align rbcp
2004   __ lea(r1, at_bcp(BytesPerInt));
2005   __ andr(r1, r1, -BytesPerInt);
2006   // load lo & hi
2007   __ ldrw(r2, Address(r1, BytesPerInt));
2008   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2009   __ rev32(r2, r2);
2010   __ rev32(r3, r3);
2011   // check against lo & hi
2012   __ cmpw(r0, r2);
2013   __ br(Assembler::LT, default_case);
2014   __ cmpw(r0, r3);
2015   __ br(Assembler::GT, default_case);
2016   // lookup dispatch offset
2017   __ subw(r0, r0, r2);
2018   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2019   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2020   __ profile_switch_case(r0, r1, r2);
2021   // continue execution
2022   __ bind(continue_execution);
2023   __ rev32(r3, r3);
2024   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2025   __ add(rbcp, rbcp, r3, ext::sxtw);
2026   __ dispatch_only(vtos, /*generate_poll*/true);
2027   // handle default
2028   __ bind(default_case);
2029   __ profile_switch_default(r0);
2030   __ ldrw(r3, Address(r1, 0));
2031   __ b(continue_execution);
2032 }
2033 
2034 void TemplateTable::lookupswitch() {
2035   transition(itos, itos);
2036   __ stop("lookupswitch bytecode should have been rewritten");
2037 }
2038 
2039 void TemplateTable::fast_linearswitch() {
2040   transition(itos, vtos);
2041   Label loop_entry, loop, found, continue_execution;
2042   // bswap r0 so we can avoid bswapping the table entries
2043   __ rev32(r0, r0);
2044   // align rbcp
2045   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2046                                     // this instruction (change offsets
2047                                     // below)
2048   __ andr(r19, r19, -BytesPerInt);
2049   // set counter
2050   __ ldrw(r1, Address(r19, BytesPerInt));
2051   __ rev32(r1, r1);
2052   __ b(loop_entry);
2053   // table search
2054   __ bind(loop);
2055   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2056   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2057   __ cmpw(r0, rscratch1);
2058   __ br(Assembler::EQ, found);
2059   __ bind(loop_entry);
2060   __ subs(r1, r1, 1);
2061   __ br(Assembler::PL, loop);
2062   // default case
2063   __ profile_switch_default(r0);
2064   __ ldrw(r3, Address(r19, 0));
2065   __ b(continue_execution);
2066   // entry found -> get offset
2067   __ bind(found);
2068   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2069   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2070   __ profile_switch_case(r1, r0, r19);
2071   // continue execution
2072   __ bind(continue_execution);
2073   __ rev32(r3, r3);
2074   __ add(rbcp, rbcp, r3, ext::sxtw);
2075   __ ldrb(rscratch1, Address(rbcp, 0));
2076   __ dispatch_only(vtos, /*generate_poll*/true);
2077 }
2078 
2079 void TemplateTable::fast_binaryswitch() {
2080   transition(itos, vtos);
2081   // Implementation using the following core algorithm:
2082   //
2083   // int binary_search(int key, LookupswitchPair* array, int n) {
2084   //   // Binary search according to "Methodik des Programmierens" by
2085   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2086   //   int i = 0;
2087   //   int j = n;
2088   //   while (i+1 < j) {
2089   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2090   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2091   //     // where a stands for the array and assuming that the (inexisting)
2092   //     // element a[n] is infinitely big.
2093   //     int h = (i + j) >> 1;
2094   //     // i < h < j
2095   //     if (key < array[h].fast_match()) {
2096   //       j = h;
2097   //     } else {
2098   //       i = h;
2099   //     }
2100   //   }
2101   //   // R: a[i] <= key < a[i+1] or Q
2102   //   // (i.e., if key is within array, i is the correct index)
2103   //   return i;
2104   // }
2105 
2106   // Register allocation
2107   const Register key   = r0; // already set (tosca)
2108   const Register array = r1;
2109   const Register i     = r2;
2110   const Register j     = r3;
2111   const Register h     = rscratch1;
2112   const Register temp  = rscratch2;
2113 
2114   // Find array start
2115   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2116                                           // get rid of this
2117                                           // instruction (change
2118                                           // offsets below)
2119   __ andr(array, array, -BytesPerInt);
2120 
2121   // Initialize i & j
2122   __ mov(i, 0);                            // i = 0;
2123   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2124 
2125   // Convert j into native byteordering
2126   __ rev32(j, j);
2127 
2128   // And start
2129   Label entry;
2130   __ b(entry);
2131 
2132   // binary search loop
2133   {
2134     Label loop;
2135     __ bind(loop);
2136     // int h = (i + j) >> 1;
2137     __ addw(h, i, j);                           // h = i + j;
2138     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2139     // if (key < array[h].fast_match()) {
2140     //   j = h;
2141     // } else {
2142     //   i = h;
2143     // }
2144     // Convert array[h].match to native byte-ordering before compare
2145     __ ldr(temp, Address(array, h, Address::lsl(3)));
2146     __ rev32(temp, temp);
2147     __ cmpw(key, temp);
2148     // j = h if (key <  array[h].fast_match())
2149     __ csel(j, h, j, Assembler::LT);
2150     // i = h if (key >= array[h].fast_match())
2151     __ csel(i, h, i, Assembler::GE);
2152     // while (i+1 < j)
2153     __ bind(entry);
2154     __ addw(h, i, 1);          // i+1
2155     __ cmpw(h, j);             // i+1 < j
2156     __ br(Assembler::LT, loop);
2157   }
2158 
2159   // end of binary search, result index is i (must check again!)
2160   Label default_case;
2161   // Convert array[i].match to native byte-ordering before compare
2162   __ ldr(temp, Address(array, i, Address::lsl(3)));
2163   __ rev32(temp, temp);
2164   __ cmpw(key, temp);
2165   __ br(Assembler::NE, default_case);
2166 
2167   // entry found -> j = offset
2168   __ add(j, array, i, ext::uxtx, 3);
2169   __ ldrw(j, Address(j, BytesPerInt));
2170   __ profile_switch_case(i, key, array);
2171   __ rev32(j, j);
2172   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2173   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2174   __ dispatch_only(vtos, /*generate_poll*/true);
2175 
2176   // default case -> j = default offset
2177   __ bind(default_case);
2178   __ profile_switch_default(i);
2179   __ ldrw(j, Address(array, -2 * BytesPerInt));
2180   __ rev32(j, j);
2181   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2182   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2183   __ dispatch_only(vtos, /*generate_poll*/true);
2184 }
2185 
2186 
2187 void TemplateTable::_return(TosState state)
2188 {
2189   transition(state, state);
2190   assert(_desc->calls_vm(),
2191          "inconsistent calls_vm information"); // call in remove_activation
2192 
2193   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2194     assert(state == vtos, "only valid state");
2195 
2196     __ ldr(c_rarg1, aaddress(0));
2197     __ load_klass(r3, c_rarg1);
2198     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2199     Label skip_register_finalizer;
2200     __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2201 
2202     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2203 
2204     __ bind(skip_register_finalizer);
2205   }
2206 
2207   // Issue a StoreStore barrier after all stores but before return
2208   // from any constructor for any class with a final field.  We don't
2209   // know if this is a finalizer, so we always do so.
2210   if (_desc->bytecode() == Bytecodes::_return)
2211     __ membar(MacroAssembler::StoreStore);
2212 
2213   // Narrow result if state is itos but result type is smaller.
2214   // Need to narrow in the return bytecode rather than in generate_return_entry
2215   // since compiled code callers expect the result to already be narrowed.
2216   if (state == itos) {
2217     __ narrow(r0);
2218   }
2219 
2220   __ remove_activation(state);
2221   __ ret(lr);
2222 }
2223 
2224 // ----------------------------------------------------------------------------
2225 // Volatile variables demand their effects be made known to all CPU's
2226 // in order.  Store buffers on most chips allow reads & writes to
2227 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2228 // without some kind of memory barrier (i.e., it's not sufficient that
2229 // the interpreter does not reorder volatile references, the hardware
2230 // also must not reorder them).
2231 //
2232 // According to the new Java Memory Model (JMM):
2233 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2234 //     writes act as acquire & release, so:
2235 // (2) A read cannot let unrelated NON-volatile memory refs that
2236 //     happen after the read float up to before the read.  It's OK for
2237 //     non-volatile memory refs that happen before the volatile read to
2238 //     float down below it.
2239 // (3) Similar a volatile write cannot let unrelated NON-volatile
2240 //     memory refs that happen BEFORE the write float down to after the
2241 //     write.  It's OK for non-volatile memory refs that happen after the
2242 //     volatile write to float up before it.
2243 //
2244 // We only put in barriers around volatile refs (they are expensive),
2245 // not _between_ memory refs (that would require us to track the
2246 // flavor of the previous memory refs).  Requirements (2) and (3)
2247 // require some barriers before volatile stores and after volatile
2248 // loads.  These nearly cover requirement (1) but miss the
2249 // volatile-store-volatile-load case.  This final case is placed after
2250 // volatile-stores although it could just as well go before
2251 // volatile-loads.
2252 
2253 void TemplateTable::resolve_cache_and_index(int byte_no,
2254                                             Register Rcache,
2255                                             Register index,
2256                                             size_t index_size) {
2257   const Register temp = r19;
2258   assert_different_registers(Rcache, index, temp);
2259 
2260   Label resolved, clinit_barrier_slow;
2261 
2262   Bytecodes::Code code = bytecode();
2263   switch (code) {
2264   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2265   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2266   default: break;
2267   }
2268 
2269   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2270   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2271   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2272   __ br(Assembler::EQ, resolved);
2273 
2274   // resolve first time through
2275   // Class initialization barrier slow path lands here as well.
2276   __ bind(clinit_barrier_slow);
2277   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2278   __ mov(temp, (int) code);
2279   __ call_VM(noreg, entry, temp);
2280 
2281   // Update registers with resolved info
2282   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2283   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2284   // so all clients ofthis method must be modified accordingly
2285   __ bind(resolved);
2286 
2287   // Class initialization barrier for static methods
2288   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2289     __ load_resolved_method_at_index(byte_no, temp, Rcache);
2290     __ load_method_holder(temp, temp);
2291     __ clinit_barrier(temp, rscratch1, NULL, &clinit_barrier_slow);
2292   }
2293 }
2294 
2295 // The Rcache and index registers must be set before call
2296 // n.b unlike x86 cache already includes the index offset
2297 void TemplateTable::load_field_cp_cache_entry(Register obj,
2298                                               Register cache,
2299                                               Register index,
2300                                               Register off,
2301                                               Register flags,
2302                                               bool is_static = false) {
2303   assert_different_registers(cache, index, flags, off);
2304 
2305   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2306   // Field offset
2307   __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2308                                           ConstantPoolCacheEntry::f2_offset())));
2309   // Flags
2310   __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2311                                            ConstantPoolCacheEntry::flags_offset())));
2312 
2313   // klass overwrite register
2314   if (is_static) {
2315     __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2316                                         ConstantPoolCacheEntry::f1_offset())));
2317     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2318     __ ldr(obj, Address(obj, mirror_offset));
2319     __ resolve_oop_handle(obj, r5, rscratch2);
2320   }
2321 }
2322 
2323 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2324                                                Register method,
2325                                                Register itable_index,
2326                                                Register flags,
2327                                                bool is_invokevirtual,
2328                                                bool is_invokevfinal, /*unused*/
2329                                                bool is_invokedynamic) {
2330   // setup registers
2331   const Register cache = rscratch2;
2332   const Register index = r4;
2333   assert_different_registers(method, flags);
2334   assert_different_registers(method, cache, index);
2335   assert_different_registers(itable_index, flags);
2336   assert_different_registers(itable_index, cache, index);
2337   // determine constant pool cache field offsets
2338   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2339   const int method_offset = in_bytes(
2340     ConstantPoolCache::base_offset() +
2341       (is_invokevirtual
2342        ? ConstantPoolCacheEntry::f2_offset()
2343        : ConstantPoolCacheEntry::f1_offset()));
2344   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2345                                     ConstantPoolCacheEntry::flags_offset());
2346   // access constant pool cache fields
2347   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2348                                     ConstantPoolCacheEntry::f2_offset());
2349 
2350   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2351   resolve_cache_and_index(byte_no, cache, index, index_size);
2352   __ ldr(method, Address(cache, method_offset));
2353 
2354   if (itable_index != noreg) {
2355     __ ldr(itable_index, Address(cache, index_offset));
2356   }
2357   __ ldrw(flags, Address(cache, flags_offset));
2358 }
2359 
2360 
2361 // The registers cache and index expected to be set before call.
2362 // Correct values of the cache and index registers are preserved.
2363 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2364                                             bool is_static, bool has_tos) {
2365   // do the JVMTI work here to avoid disturbing the register state below
2366   // We use c_rarg registers here because we want to use the register used in
2367   // the call to the VM
2368   if (JvmtiExport::can_post_field_access()) {
2369     // Check to see if a field access watch has been set before we
2370     // take the time to call into the VM.
2371     Label L1;
2372     assert_different_registers(cache, index, r0);
2373     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2374     __ ldrw(r0, Address(rscratch1));
2375     __ cbzw(r0, L1);
2376 
2377     __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2378     __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2379 
2380     if (is_static) {
2381       __ mov(c_rarg1, zr); // NULL object reference
2382     } else {
2383       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2384       __ verify_oop(c_rarg1);
2385     }
2386     // c_rarg1: object pointer or NULL
2387     // c_rarg2: cache entry pointer
2388     // c_rarg3: jvalue object on the stack
2389     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2390                                        InterpreterRuntime::post_field_access),
2391                c_rarg1, c_rarg2, c_rarg3);
2392     __ get_cache_and_index_at_bcp(cache, index, 1);
2393     __ bind(L1);
2394   }
2395 }
2396 
2397 void TemplateTable::pop_and_check_object(Register r)
2398 {
2399   __ pop_ptr(r);
2400   __ null_check(r);  // for field access must check obj.
2401   __ verify_oop(r);
2402 }
2403 
2404 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2405 {
2406   const Register cache = r2;
2407   const Register index = r3;
2408   const Register obj   = r4;
2409   const Register off   = r19;
2410   const Register flags = r0;
2411   const Register raw_flags = r6;
2412   const Register bc    = r4; // uses same reg as obj, so don't mix them
2413 
2414   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2415   jvmti_post_field_access(cache, index, is_static, false);
2416   load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2417 
2418   if (!is_static) {
2419     // obj is on the stack
2420     pop_and_check_object(obj);
2421   }
2422 
2423   // 8179954: We need to make sure that the code generated for
2424   // volatile accesses forms a sequentially-consistent set of
2425   // operations when combined with STLR and LDAR.  Without a leading
2426   // membar it's possible for a simple Dekker test to fail if loads
2427   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2428   // the stores in one method and we interpret the loads in another.
2429   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()){
2430     Label notVolatile;
2431     __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2432     __ membar(MacroAssembler::AnyAny);
2433     __ bind(notVolatile);
2434   }
2435 
2436   const Address field(obj, off);
2437 
2438   Label Done, notByte, notBool, notInt, notShort, notChar,
2439               notLong, notFloat, notObj, notDouble;
2440 
2441   // x86 uses a shift and mask or wings it with a shift plus assert
2442   // the mask is not needed. aarch64 just uses bitfield extract
2443   __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2444            ConstantPoolCacheEntry::tos_state_bits);
2445 
2446   assert(btos == 0, "change code, btos != 0");
2447   __ cbnz(flags, notByte);
2448 
2449   // Don't rewrite getstatic, only getfield
2450   if (is_static) rc = may_not_rewrite;
2451 
2452   // btos
2453   __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
2454   __ push(btos);
2455   // Rewrite bytecode to be faster
2456   if (rc == may_rewrite) {
2457     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2458   }
2459   __ b(Done);
2460 
2461   __ bind(notByte);
2462   __ cmp(flags, (u1)ztos);
2463   __ br(Assembler::NE, notBool);
2464 
2465   // ztos (same code as btos)
2466   __ access_load_at(T_BOOLEAN, IN_HEAP, r0, field, noreg, noreg);
2467   __ push(ztos);
2468   // Rewrite bytecode to be faster
2469   if (rc == may_rewrite) {
2470     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2471     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2472   }
2473   __ b(Done);
2474 
2475   __ bind(notBool);
2476   __ cmp(flags, (u1)atos);
2477   __ br(Assembler::NE, notObj);
2478   // atos
2479   do_oop_load(_masm, field, r0, IN_HEAP);
2480   __ push(atos);
2481   if (rc == may_rewrite) {
2482     patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2483   }
2484   __ b(Done);
2485 
2486   __ bind(notObj);
2487   __ cmp(flags, (u1)itos);
2488   __ br(Assembler::NE, notInt);
2489   // itos
2490   __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
2491   __ push(itos);
2492   // Rewrite bytecode to be faster
2493   if (rc == may_rewrite) {
2494     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2495   }
2496   __ b(Done);
2497 
2498   __ bind(notInt);
2499   __ cmp(flags, (u1)ctos);
2500   __ br(Assembler::NE, notChar);
2501   // ctos
2502   __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
2503   __ push(ctos);
2504   // Rewrite bytecode to be faster
2505   if (rc == may_rewrite) {
2506     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2507   }
2508   __ b(Done);
2509 
2510   __ bind(notChar);
2511   __ cmp(flags, (u1)stos);
2512   __ br(Assembler::NE, notShort);
2513   // stos
2514   __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
2515   __ push(stos);
2516   // Rewrite bytecode to be faster
2517   if (rc == may_rewrite) {
2518     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2519   }
2520   __ b(Done);
2521 
2522   __ bind(notShort);
2523   __ cmp(flags, (u1)ltos);
2524   __ br(Assembler::NE, notLong);
2525   // ltos
2526   __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
2527   __ push(ltos);
2528   // Rewrite bytecode to be faster
2529   if (rc == may_rewrite) {
2530     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2531   }
2532   __ b(Done);
2533 
2534   __ bind(notLong);
2535   __ cmp(flags, (u1)ftos);
2536   __ br(Assembler::NE, notFloat);
2537   // ftos
2538   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2539   __ push(ftos);
2540   // Rewrite bytecode to be faster
2541   if (rc == may_rewrite) {
2542     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2543   }
2544   __ b(Done);
2545 
2546   __ bind(notFloat);
2547 #ifdef ASSERT
2548   __ cmp(flags, (u1)dtos);
2549   __ br(Assembler::NE, notDouble);
2550 #endif
2551   // dtos
2552   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2553   __ push(dtos);
2554   // Rewrite bytecode to be faster
2555   if (rc == may_rewrite) {
2556     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2557   }
2558 #ifdef ASSERT
2559   __ b(Done);
2560 
2561   __ bind(notDouble);
2562   __ stop("Bad state");
2563 #endif
2564 
2565   __ bind(Done);
2566 
2567   Label notVolatile;
2568   __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2569   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2570   __ bind(notVolatile);
2571 }
2572 
2573 
2574 void TemplateTable::getfield(int byte_no)
2575 {
2576   getfield_or_static(byte_no, false);
2577 }
2578 
2579 void TemplateTable::nofast_getfield(int byte_no) {
2580   getfield_or_static(byte_no, false, may_not_rewrite);
2581 }
2582 
2583 void TemplateTable::getstatic(int byte_no)
2584 {
2585   getfield_or_static(byte_no, true);
2586 }
2587 
2588 // The registers cache and index expected to be set before call.
2589 // The function may destroy various registers, just not the cache and index registers.
2590 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2591   transition(vtos, vtos);
2592 
2593   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2594 
2595   if (JvmtiExport::can_post_field_modification()) {
2596     // Check to see if a field modification watch has been set before
2597     // we take the time to call into the VM.
2598     Label L1;
2599     assert_different_registers(cache, index, r0);
2600     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2601     __ ldrw(r0, Address(rscratch1));
2602     __ cbz(r0, L1);
2603 
2604     __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2605 
2606     if (is_static) {
2607       // Life is simple.  Null out the object pointer.
2608       __ mov(c_rarg1, zr);
2609     } else {
2610       // Life is harder. The stack holds the value on top, followed by
2611       // the object.  We don't know the size of the value, though; it
2612       // could be one or two words depending on its type. As a result,
2613       // we must find the type to determine where the object is.
2614       __ ldrw(c_rarg3, Address(c_rarg2,
2615                                in_bytes(cp_base_offset +
2616                                         ConstantPoolCacheEntry::flags_offset())));
2617       __ lsr(c_rarg3, c_rarg3,
2618              ConstantPoolCacheEntry::tos_state_shift);
2619       ConstantPoolCacheEntry::verify_tos_state_shift();
2620       Label nope2, done, ok;
2621       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2622       __ cmpw(c_rarg3, ltos);
2623       __ br(Assembler::EQ, ok);
2624       __ cmpw(c_rarg3, dtos);
2625       __ br(Assembler::NE, nope2);
2626       __ bind(ok);
2627       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2628       __ bind(nope2);
2629     }
2630     // cache entry pointer
2631     __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2632     // object (tos)
2633     __ mov(c_rarg3, esp);
2634     // c_rarg1: object pointer set up above (NULL if static)
2635     // c_rarg2: cache entry pointer
2636     // c_rarg3: jvalue object on the stack
2637     __ call_VM(noreg,
2638                CAST_FROM_FN_PTR(address,
2639                                 InterpreterRuntime::post_field_modification),
2640                c_rarg1, c_rarg2, c_rarg3);
2641     __ get_cache_and_index_at_bcp(cache, index, 1);
2642     __ bind(L1);
2643   }
2644 }
2645 
2646 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2647   transition(vtos, vtos);
2648 
2649   const Register cache = r2;
2650   const Register index = r3;
2651   const Register obj   = r2;
2652   const Register off   = r19;
2653   const Register flags = r0;
2654   const Register bc    = r4;
2655 
2656   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2657   jvmti_post_field_mod(cache, index, is_static);
2658   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2659 
2660   Label Done;
2661   __ mov(r5, flags);
2662 
2663   {
2664     Label notVolatile;
2665     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2666     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2667     __ bind(notVolatile);
2668   }
2669 
2670   // field address
2671   const Address field(obj, off);
2672 
2673   Label notByte, notBool, notInt, notShort, notChar,
2674         notLong, notFloat, notObj, notDouble;
2675 
2676   // x86 uses a shift and mask or wings it with a shift plus assert
2677   // the mask is not needed. aarch64 just uses bitfield extract
2678   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2679 
2680   assert(btos == 0, "change code, btos != 0");
2681   __ cbnz(flags, notByte);
2682 
2683   // Don't rewrite putstatic, only putfield
2684   if (is_static) rc = may_not_rewrite;
2685 
2686   // btos
2687   {
2688     __ pop(btos);
2689     if (!is_static) pop_and_check_object(obj);
2690     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
2691     if (rc == may_rewrite) {
2692       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2693     }
2694     __ b(Done);
2695   }
2696 
2697   __ bind(notByte);
2698   __ cmp(flags, (u1)ztos);
2699   __ br(Assembler::NE, notBool);
2700 
2701   // ztos
2702   {
2703     __ pop(ztos);
2704     if (!is_static) pop_and_check_object(obj);
2705     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
2706     if (rc == may_rewrite) {
2707       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2708     }
2709     __ b(Done);
2710   }
2711 
2712   __ bind(notBool);
2713   __ cmp(flags, (u1)atos);
2714   __ br(Assembler::NE, notObj);
2715 
2716   // atos
2717   {
2718     __ pop(atos);
2719     if (!is_static) pop_and_check_object(obj);
2720     // Store into the field
2721     do_oop_store(_masm, field, r0, IN_HEAP);
2722     if (rc == may_rewrite) {
2723       patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2724     }
2725     __ b(Done);
2726   }
2727 
2728   __ bind(notObj);
2729   __ cmp(flags, (u1)itos);
2730   __ br(Assembler::NE, notInt);
2731 
2732   // itos
2733   {
2734     __ pop(itos);
2735     if (!is_static) pop_and_check_object(obj);
2736     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
2737     if (rc == may_rewrite) {
2738       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2739     }
2740     __ b(Done);
2741   }
2742 
2743   __ bind(notInt);
2744   __ cmp(flags, (u1)ctos);
2745   __ br(Assembler::NE, notChar);
2746 
2747   // ctos
2748   {
2749     __ pop(ctos);
2750     if (!is_static) pop_and_check_object(obj);
2751     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
2752     if (rc == may_rewrite) {
2753       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2754     }
2755     __ b(Done);
2756   }
2757 
2758   __ bind(notChar);
2759   __ cmp(flags, (u1)stos);
2760   __ br(Assembler::NE, notShort);
2761 
2762   // stos
2763   {
2764     __ pop(stos);
2765     if (!is_static) pop_and_check_object(obj);
2766     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
2767     if (rc == may_rewrite) {
2768       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2769     }
2770     __ b(Done);
2771   }
2772 
2773   __ bind(notShort);
2774   __ cmp(flags, (u1)ltos);
2775   __ br(Assembler::NE, notLong);
2776 
2777   // ltos
2778   {
2779     __ pop(ltos);
2780     if (!is_static) pop_and_check_object(obj);
2781     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
2782     if (rc == may_rewrite) {
2783       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2784     }
2785     __ b(Done);
2786   }
2787 
2788   __ bind(notLong);
2789   __ cmp(flags, (u1)ftos);
2790   __ br(Assembler::NE, notFloat);
2791 
2792   // ftos
2793   {
2794     __ pop(ftos);
2795     if (!is_static) pop_and_check_object(obj);
2796     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
2797     if (rc == may_rewrite) {
2798       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2799     }
2800     __ b(Done);
2801   }
2802 
2803   __ bind(notFloat);
2804 #ifdef ASSERT
2805   __ cmp(flags, (u1)dtos);
2806   __ br(Assembler::NE, notDouble);
2807 #endif
2808 
2809   // dtos
2810   {
2811     __ pop(dtos);
2812     if (!is_static) pop_and_check_object(obj);
2813     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
2814     if (rc == may_rewrite) {
2815       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2816     }
2817   }
2818 
2819 #ifdef ASSERT
2820   __ b(Done);
2821 
2822   __ bind(notDouble);
2823   __ stop("Bad state");
2824 #endif
2825 
2826   __ bind(Done);
2827 
2828   {
2829     Label notVolatile;
2830     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2831     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2832     __ bind(notVolatile);
2833   }
2834 }
2835 
2836 void TemplateTable::putfield(int byte_no)
2837 {
2838   putfield_or_static(byte_no, false);
2839 }
2840 
2841 void TemplateTable::nofast_putfield(int byte_no) {
2842   putfield_or_static(byte_no, false, may_not_rewrite);
2843 }
2844 
2845 void TemplateTable::putstatic(int byte_no) {
2846   putfield_or_static(byte_no, true);
2847 }
2848 
2849 void TemplateTable::jvmti_post_fast_field_mod()
2850 {
2851   if (JvmtiExport::can_post_field_modification()) {
2852     // Check to see if a field modification watch has been set before
2853     // we take the time to call into the VM.
2854     Label L2;
2855     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2856     __ ldrw(c_rarg3, Address(rscratch1));
2857     __ cbzw(c_rarg3, L2);
2858     __ pop_ptr(r19);                  // copy the object pointer from tos
2859     __ verify_oop(r19);
2860     __ push_ptr(r19);                 // put the object pointer back on tos
2861     // Save tos values before call_VM() clobbers them. Since we have
2862     // to do it for every data type, we use the saved values as the
2863     // jvalue object.
2864     switch (bytecode()) {          // load values into the jvalue object
2865     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2866     case Bytecodes::_fast_bputfield: // fall through
2867     case Bytecodes::_fast_zputfield: // fall through
2868     case Bytecodes::_fast_sputfield: // fall through
2869     case Bytecodes::_fast_cputfield: // fall through
2870     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2871     case Bytecodes::_fast_dputfield: __ push_d(); break;
2872     case Bytecodes::_fast_fputfield: __ push_f(); break;
2873     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2874 
2875     default:
2876       ShouldNotReachHere();
2877     }
2878     __ mov(c_rarg3, esp);             // points to jvalue on the stack
2879     // access constant pool cache entry
2880     __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2881     __ verify_oop(r19);
2882     // r19: object pointer copied above
2883     // c_rarg2: cache entry pointer
2884     // c_rarg3: jvalue object on the stack
2885     __ call_VM(noreg,
2886                CAST_FROM_FN_PTR(address,
2887                                 InterpreterRuntime::post_field_modification),
2888                r19, c_rarg2, c_rarg3);
2889 
2890     switch (bytecode()) {             // restore tos values
2891     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2892     case Bytecodes::_fast_bputfield: // fall through
2893     case Bytecodes::_fast_zputfield: // fall through
2894     case Bytecodes::_fast_sputfield: // fall through
2895     case Bytecodes::_fast_cputfield: // fall through
2896     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2897     case Bytecodes::_fast_dputfield: __ pop_d(); break;
2898     case Bytecodes::_fast_fputfield: __ pop_f(); break;
2899     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2900     default: break;
2901     }
2902     __ bind(L2);
2903   }
2904 }
2905 
2906 void TemplateTable::fast_storefield(TosState state)
2907 {
2908   transition(state, vtos);
2909 
2910   ByteSize base = ConstantPoolCache::base_offset();
2911 
2912   jvmti_post_fast_field_mod();
2913 
2914   // access constant pool cache
2915   __ get_cache_and_index_at_bcp(r2, r1, 1);
2916 
2917   // Must prevent reordering of the following cp cache loads with bytecode load
2918   __ membar(MacroAssembler::LoadLoad);
2919 
2920   // test for volatile with r3
2921   __ ldrw(r3, Address(r2, in_bytes(base +
2922                                    ConstantPoolCacheEntry::flags_offset())));
2923 
2924   // replace index with field offset from cache entry
2925   __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2926 
2927   {
2928     Label notVolatile;
2929     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2930     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2931     __ bind(notVolatile);
2932   }
2933 
2934   Label notVolatile;
2935 
2936   // Get object from stack
2937   pop_and_check_object(r2);
2938 
2939   // field address
2940   const Address field(r2, r1);
2941 
2942   // access field
2943   switch (bytecode()) {
2944   case Bytecodes::_fast_aputfield:
2945     do_oop_store(_masm, field, r0, IN_HEAP);
2946     break;
2947   case Bytecodes::_fast_lputfield:
2948     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
2949     break;
2950   case Bytecodes::_fast_iputfield:
2951     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
2952     break;
2953   case Bytecodes::_fast_zputfield:
2954     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
2955     break;
2956   case Bytecodes::_fast_bputfield:
2957     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
2958     break;
2959   case Bytecodes::_fast_sputfield:
2960     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
2961     break;
2962   case Bytecodes::_fast_cputfield:
2963     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
2964     break;
2965   case Bytecodes::_fast_fputfield:
2966     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
2967     break;
2968   case Bytecodes::_fast_dputfield:
2969     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
2970     break;
2971   default:
2972     ShouldNotReachHere();
2973   }
2974 
2975   {
2976     Label notVolatile;
2977     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2978     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2979     __ bind(notVolatile);
2980   }
2981 }
2982 
2983 
2984 void TemplateTable::fast_accessfield(TosState state)
2985 {
2986   transition(atos, state);
2987   // Do the JVMTI work here to avoid disturbing the register state below
2988   if (JvmtiExport::can_post_field_access()) {
2989     // Check to see if a field access watch has been set before we
2990     // take the time to call into the VM.
2991     Label L1;
2992     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2993     __ ldrw(r2, Address(rscratch1));
2994     __ cbzw(r2, L1);
2995     // access constant pool cache entry
2996     __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2997     __ verify_oop(r0);
2998     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
2999     __ mov(c_rarg1, r0);
3000     // c_rarg1: object pointer copied above
3001     // c_rarg2: cache entry pointer
3002     __ call_VM(noreg,
3003                CAST_FROM_FN_PTR(address,
3004                                 InterpreterRuntime::post_field_access),
3005                c_rarg1, c_rarg2);
3006     __ pop_ptr(r0); // restore object pointer
3007     __ bind(L1);
3008   }
3009 
3010   // access constant pool cache
3011   __ get_cache_and_index_at_bcp(r2, r1, 1);
3012 
3013   // Must prevent reordering of the following cp cache loads with bytecode load
3014   __ membar(MacroAssembler::LoadLoad);
3015 
3016   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3017                                   ConstantPoolCacheEntry::f2_offset())));
3018   __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3019                                    ConstantPoolCacheEntry::flags_offset())));
3020 
3021   // r0: object
3022   __ verify_oop(r0);
3023   __ null_check(r0);
3024   const Address field(r0, r1);
3025 
3026   // 8179954: We need to make sure that the code generated for
3027   // volatile accesses forms a sequentially-consistent set of
3028   // operations when combined with STLR and LDAR.  Without a leading
3029   // membar it's possible for a simple Dekker test to fail if loads
3030   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3031   // the stores in one method and we interpret the loads in another.
3032   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3033     Label notVolatile;
3034     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3035     __ membar(MacroAssembler::AnyAny);
3036     __ bind(notVolatile);
3037   }
3038 
3039   // access field
3040   switch (bytecode()) {
3041   case Bytecodes::_fast_agetfield:
3042     do_oop_load(_masm, field, r0, IN_HEAP);
3043     __ verify_oop(r0);
3044     break;
3045   case Bytecodes::_fast_lgetfield:
3046     __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
3047     break;
3048   case Bytecodes::_fast_igetfield:
3049     __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
3050     break;
3051   case Bytecodes::_fast_bgetfield:
3052     __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
3053     break;
3054   case Bytecodes::_fast_sgetfield:
3055     __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
3056     break;
3057   case Bytecodes::_fast_cgetfield:
3058     __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
3059     break;
3060   case Bytecodes::_fast_fgetfield:
3061     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3062     break;
3063   case Bytecodes::_fast_dgetfield:
3064     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3065     break;
3066   default:
3067     ShouldNotReachHere();
3068   }
3069   {
3070     Label notVolatile;
3071     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3072     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3073     __ bind(notVolatile);
3074   }
3075 }
3076 
3077 void TemplateTable::fast_xaccess(TosState state)
3078 {
3079   transition(vtos, state);
3080 
3081   // get receiver
3082   __ ldr(r0, aaddress(0));
3083   // access constant pool cache
3084   __ get_cache_and_index_at_bcp(r2, r3, 2);
3085   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3086                                   ConstantPoolCacheEntry::f2_offset())));
3087 
3088   // 8179954: We need to make sure that the code generated for
3089   // volatile accesses forms a sequentially-consistent set of
3090   // operations when combined with STLR and LDAR.  Without a leading
3091   // membar it's possible for a simple Dekker test to fail if loads
3092   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3093   // the stores in one method and we interpret the loads in another.
3094   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3095     Label notVolatile;
3096     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3097                                      ConstantPoolCacheEntry::flags_offset())));
3098     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3099     __ membar(MacroAssembler::AnyAny);
3100     __ bind(notVolatile);
3101   }
3102 
3103   // make sure exception is reported in correct bcp range (getfield is
3104   // next instruction)
3105   __ increment(rbcp);
3106   __ null_check(r0);
3107   switch (state) {
3108   case itos:
3109     __ access_load_at(T_INT, IN_HEAP, r0, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3110     break;
3111   case atos:
3112     do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3113     __ verify_oop(r0);
3114     break;
3115   case ftos:
3116     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3117     break;
3118   default:
3119     ShouldNotReachHere();
3120   }
3121 
3122   {
3123     Label notVolatile;
3124     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3125                                      ConstantPoolCacheEntry::flags_offset())));
3126     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3127     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3128     __ bind(notVolatile);
3129   }
3130 
3131   __ decrement(rbcp);
3132 }
3133 
3134 
3135 
3136 //-----------------------------------------------------------------------------
3137 // Calls
3138 
3139 void TemplateTable::prepare_invoke(int byte_no,
3140                                    Register method, // linked method (or i-klass)
3141                                    Register index,  // itable index, MethodType, etc.
3142                                    Register recv,   // if caller wants to see it
3143                                    Register flags   // if caller wants to test it
3144                                    ) {
3145   // determine flags
3146   Bytecodes::Code code = bytecode();
3147   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3148   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3149   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3150   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3151   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3152   const bool load_receiver       = (recv  != noreg);
3153   const bool save_flags          = (flags != noreg);
3154   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3155   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3156   assert(flags == noreg || flags == r3, "");
3157   assert(recv  == noreg || recv  == r2, "");
3158 
3159   // setup registers & access constant pool cache
3160   if (recv  == noreg)  recv  = r2;
3161   if (flags == noreg)  flags = r3;
3162   assert_different_registers(method, index, recv, flags);
3163 
3164   // save 'interpreter return address'
3165   __ save_bcp();
3166 
3167   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3168 
3169   // maybe push appendix to arguments (just before return address)
3170   if (is_invokedynamic || is_invokehandle) {
3171     Label L_no_push;
3172     __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3173     // Push the appendix as a trailing parameter.
3174     // This must be done before we get the receiver,
3175     // since the parameter_size includes it.
3176     __ push(r19);
3177     __ mov(r19, index);
3178     __ load_resolved_reference_at_index(index, r19);
3179     __ pop(r19);
3180     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3181     __ bind(L_no_push);
3182   }
3183 
3184   // load receiver if needed (note: no return address pushed yet)
3185   if (load_receiver) {
3186     __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3187     // FIXME -- is this actually correct? looks like it should be 2
3188     // const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3189     // const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3190     // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3191     // __ movptr(recv, recv_addr);
3192     __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3193     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3194     __ verify_oop(recv);
3195   }
3196 
3197   // compute return type
3198   // x86 uses a shift and mask or wings it with a shift plus assert
3199   // the mask is not needed. aarch64 just uses bitfield extract
3200   __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
3201   // load return address
3202   {
3203     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3204     __ mov(rscratch1, table_addr);
3205     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3206   }
3207 }
3208 
3209 
3210 void TemplateTable::invokevirtual_helper(Register index,
3211                                          Register recv,
3212                                          Register flags)
3213 {
3214   // Uses temporary registers r0, r3
3215   assert_different_registers(index, recv, r0, r3);
3216   // Test for an invoke of a final method
3217   Label notFinal;
3218   __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3219 
3220   const Register method = index;  // method must be rmethod
3221   assert(method == rmethod,
3222          "Method must be rmethod for interpreter calling convention");
3223 
3224   // do the call - the index is actually the method to call
3225   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3226 
3227   // It's final, need a null check here!
3228   __ null_check(recv);
3229 
3230   // profile this call
3231   __ profile_final_call(r0);
3232   __ profile_arguments_type(r0, method, r4, true);
3233 
3234   __ jump_from_interpreted(method, r0);
3235 
3236   __ bind(notFinal);
3237 
3238   // get receiver klass
3239   __ null_check(recv, oopDesc::mark_offset_in_bytes());
3240   __ load_klass(r0, recv);
3241 
3242   // profile this call
3243   __ profile_virtual_call(r0, rlocals, r3);
3244 
3245   // get target Method & entry point
3246   __ lookup_virtual_method(r0, index, method);
3247   __ profile_arguments_type(r3, method, r4, true);
3248   // FIXME -- this looks completely redundant. is it?
3249   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3250   __ jump_from_interpreted(method, r3);
3251 }
3252 
3253 void TemplateTable::invokevirtual(int byte_no)
3254 {
3255   transition(vtos, vtos);
3256   assert(byte_no == f2_byte, "use this argument");
3257 
3258   prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3259 
3260   // rmethod: index (actually a Method*)
3261   // r2: receiver
3262   // r3: flags
3263 
3264   invokevirtual_helper(rmethod, r2, r3);
3265 }
3266 
3267 void TemplateTable::invokespecial(int byte_no)
3268 {
3269   transition(vtos, vtos);
3270   assert(byte_no == f1_byte, "use this argument");
3271 
3272   prepare_invoke(byte_no, rmethod, noreg,  // get f1 Method*
3273                  r2);  // get receiver also for null check
3274   __ verify_oop(r2);
3275   __ null_check(r2);
3276   // do the call
3277   __ profile_call(r0);
3278   __ profile_arguments_type(r0, rmethod, rbcp, false);
3279   __ jump_from_interpreted(rmethod, r0);
3280 }
3281 
3282 void TemplateTable::invokestatic(int byte_no)
3283 {
3284   transition(vtos, vtos);
3285   assert(byte_no == f1_byte, "use this argument");
3286 
3287   prepare_invoke(byte_no, rmethod);  // get f1 Method*
3288   // do the call
3289   __ profile_call(r0);
3290   __ profile_arguments_type(r0, rmethod, r4, false);
3291   __ jump_from_interpreted(rmethod, r0);
3292 }
3293 
3294 void TemplateTable::fast_invokevfinal(int byte_no)
3295 {
3296   __ call_Unimplemented();
3297 }
3298 
3299 void TemplateTable::invokeinterface(int byte_no) {
3300   transition(vtos, vtos);
3301   assert(byte_no == f1_byte, "use this argument");
3302 
3303   prepare_invoke(byte_no, r0, rmethod,  // get f1 Klass*, f2 Method*
3304                  r2, r3); // recv, flags
3305 
3306   // r0: interface klass (from f1)
3307   // rmethod: method (from f2)
3308   // r2: receiver
3309   // r3: flags
3310 
3311   // First check for Object case, then private interface method,
3312   // then regular interface method.
3313 
3314   // Special case of invokeinterface called for virtual method of
3315   // java.lang.Object.  See cpCache.cpp for details.
3316   Label notObjectMethod;
3317   __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notObjectMethod);
3318 
3319   invokevirtual_helper(rmethod, r2, r3);
3320   __ bind(notObjectMethod);
3321 
3322   Label no_such_interface;
3323 
3324   // Check for private method invocation - indicated by vfinal
3325   Label notVFinal;
3326   __ tbz(r3, ConstantPoolCacheEntry::is_vfinal_shift, notVFinal);
3327 
3328   // Get receiver klass into r3 - also a null check
3329   __ null_check(r2, oopDesc::mark_offset_in_bytes());
3330   __ load_klass(r3, r2);
3331 
3332   Label subtype;
3333   __ check_klass_subtype(r3, r0, r4, subtype);
3334   // If we get here the typecheck failed
3335   __ b(no_such_interface);
3336   __ bind(subtype);
3337 
3338   __ profile_final_call(r0);
3339   __ profile_arguments_type(r0, rmethod, r4, true);
3340   __ jump_from_interpreted(rmethod, r0);
3341 
3342   __ bind(notVFinal);
3343 
3344   // Get receiver klass into r3 - also a null check
3345   __ restore_locals();
3346   __ null_check(r2, oopDesc::mark_offset_in_bytes());
3347   __ load_klass(r3, r2);
3348 
3349   Label no_such_method;
3350 
3351   // Preserve method for throw_AbstractMethodErrorVerbose.
3352   __ mov(r16, rmethod);
3353   // Receiver subtype check against REFC.
3354   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3355   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3356                              r3, r0, noreg,
3357                              // outputs: scan temp. reg, scan temp. reg
3358                              rscratch2, r13,
3359                              no_such_interface,
3360                              /*return_method=*/false);
3361 
3362   // profile this call
3363   __ profile_virtual_call(r3, r13, r19);
3364 
3365   // Get declaring interface class from method, and itable index
3366 
3367   __ load_method_holder(r0, rmethod);
3368   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3369   __ subw(rmethod, rmethod, Method::itable_index_max);
3370   __ negw(rmethod, rmethod);
3371 
3372   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3373   __ mov(rlocals, r3);
3374   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3375                              rlocals, r0, rmethod,
3376                              // outputs: method, scan temp. reg
3377                              rmethod, r13,
3378                              no_such_interface);
3379 
3380   // rmethod,: Method to call
3381   // r2: receiver
3382   // Check for abstract method error
3383   // Note: This should be done more efficiently via a throw_abstract_method_error
3384   //       interpreter entry point and a conditional jump to it in case of a null
3385   //       method.
3386   __ cbz(rmethod, no_such_method);
3387 
3388   __ profile_arguments_type(r3, rmethod, r13, true);
3389 
3390   // do the call
3391   // r2: receiver
3392   // rmethod,: Method
3393   __ jump_from_interpreted(rmethod, r3);
3394   __ should_not_reach_here();
3395 
3396   // exception handling code follows...
3397   // note: must restore interpreter registers to canonical
3398   //       state for exception handling to work correctly!
3399 
3400   __ bind(no_such_method);
3401   // throw exception
3402   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3403   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3404   // Pass arguments for generating a verbose error message.
3405   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3406   // the call_VM checks for exception, so we should never return here.
3407   __ should_not_reach_here();
3408 
3409   __ bind(no_such_interface);
3410   // throw exception
3411   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3412   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3413   // Pass arguments for generating a verbose error message.
3414   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3415                    InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3416   // the call_VM checks for exception, so we should never return here.
3417   __ should_not_reach_here();
3418   return;
3419 }
3420 
3421 void TemplateTable::invokehandle(int byte_no) {
3422   transition(vtos, vtos);
3423   assert(byte_no == f1_byte, "use this argument");
3424 
3425   prepare_invoke(byte_no, rmethod, r0, r2);
3426   __ verify_method_ptr(r2);
3427   __ verify_oop(r2);
3428   __ null_check(r2);
3429 
3430   // FIXME: profile the LambdaForm also
3431 
3432   // r13 is safe to use here as a scratch reg because it is about to
3433   // be clobbered by jump_from_interpreted().
3434   __ profile_final_call(r13);
3435   __ profile_arguments_type(r13, rmethod, r4, true);
3436 
3437   __ jump_from_interpreted(rmethod, r0);
3438 }
3439 
3440 void TemplateTable::invokedynamic(int byte_no) {
3441   transition(vtos, vtos);
3442   assert(byte_no == f1_byte, "use this argument");
3443 
3444   prepare_invoke(byte_no, rmethod, r0);
3445 
3446   // r0: CallSite object (from cpool->resolved_references[])
3447   // rmethod: MH.linkToCallSite method (from f2)
3448 
3449   // Note:  r0_callsite is already pushed by prepare_invoke
3450 
3451   // %%% should make a type profile for any invokedynamic that takes a ref argument
3452   // profile this call
3453   __ profile_call(rbcp);
3454   __ profile_arguments_type(r3, rmethod, r13, false);
3455 
3456   __ verify_oop(r0);
3457 
3458   __ jump_from_interpreted(rmethod, r0);
3459 }
3460 
3461 
3462 //-----------------------------------------------------------------------------
3463 // Allocation
3464 
3465 void TemplateTable::_new() {
3466   transition(vtos, atos);
3467 
3468   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3469   Label slow_case;
3470   Label done;
3471   Label initialize_header;
3472 
3473   __ get_cpool_and_tags(r4, r0);
3474   // Make sure the class we're about to instantiate has been resolved.
3475   // This is done before loading InstanceKlass to be consistent with the order
3476   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3477   const int tags_offset = Array<u1>::base_offset_in_bytes();
3478   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3479   __ lea(rscratch1, Address(rscratch1, tags_offset));
3480   __ ldarb(rscratch1, rscratch1);
3481   __ cmp(rscratch1, (u1)JVM_CONSTANT_Class);
3482   __ br(Assembler::NE, slow_case);
3483 
3484   // get InstanceKlass
3485   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3486 
3487   // make sure klass is initialized & doesn't have finalizer
3488   // make sure klass is fully initialized
3489   __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3490   __ cmp(rscratch1, (u1)InstanceKlass::fully_initialized);
3491   __ br(Assembler::NE, slow_case);
3492 
3493   // get instance_size in InstanceKlass (scaled to a count of bytes)
3494   __ ldrw(r3,
3495           Address(r4,
3496                   Klass::layout_helper_offset()));
3497   // test to see if it has a finalizer or is malformed in some way
3498   __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3499 
3500   // Allocate the instance:
3501   //  If TLAB is enabled:
3502   //    Try to allocate in the TLAB.
3503   //    If fails, go to the slow path.
3504   //    Initialize the allocation.
3505   //    Exit.
3506   //
3507   //  Go to slow path.
3508 
3509   if (UseTLAB) {
3510     __ tlab_allocate(r0, r3, 0, noreg, r1, slow_case);
3511 
3512     if (ZeroTLAB) {
3513       // the fields have been already cleared
3514       __ b(initialize_header);
3515     }
3516 
3517     // The object is initialized before the header.  If the object size is
3518     // zero, go directly to the header initialization.
3519     __ sub(r3, r3, sizeof(oopDesc));
3520     __ cbz(r3, initialize_header);
3521 
3522     // Initialize object fields
3523     {
3524       __ add(r2, r0, sizeof(oopDesc));
3525       Label loop;
3526       __ bind(loop);
3527       __ str(zr, Address(__ post(r2, BytesPerLong)));
3528       __ sub(r3, r3, BytesPerLong);
3529       __ cbnz(r3, loop);
3530     }
3531 
3532     // initialize object header only.
3533     __ bind(initialize_header);
3534     __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3535     __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3536 
3537     {
3538       SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3539       // Trigger dtrace event for fastpath
3540       __ push(atos); // save the return value
3541       __ call_VM_leaf(
3542            CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), r0);
3543       __ pop(atos); // restore the return value
3544 
3545     }
3546     __ b(done);
3547   }
3548 
3549   // slow case
3550   __ bind(slow_case);
3551   __ get_constant_pool(c_rarg1);
3552   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3553   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3554   __ verify_oop(r0);
3555 
3556   // continue
3557   __ bind(done);
3558   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3559   __ membar(Assembler::StoreStore);
3560 }
3561 
3562 void TemplateTable::newarray() {
3563   transition(itos, atos);
3564   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3565   __ mov(c_rarg2, r0);
3566   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3567           c_rarg1, c_rarg2);
3568   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3569   __ membar(Assembler::StoreStore);
3570 }
3571 
3572 void TemplateTable::anewarray() {
3573   transition(itos, atos);
3574   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3575   __ get_constant_pool(c_rarg1);
3576   __ mov(c_rarg3, r0);
3577   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3578           c_rarg1, c_rarg2, c_rarg3);
3579   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3580   __ membar(Assembler::StoreStore);
3581 }
3582 
3583 void TemplateTable::arraylength() {
3584   transition(atos, itos);
3585   __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3586   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3587 }
3588 
3589 void TemplateTable::checkcast()
3590 {
3591   transition(atos, atos);
3592   Label done, is_null, ok_is_subtype, quicked, resolved;
3593   __ cbz(r0, is_null);
3594 
3595   // Get cpool & tags index
3596   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3597   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3598   // See if bytecode has already been quicked
3599   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3600   __ lea(r1, Address(rscratch1, r19));
3601   __ ldarb(r1, r1);
3602   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3603   __ br(Assembler::EQ, quicked);
3604 
3605   __ push(atos); // save receiver for result, and for GC
3606   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3607   // vm_result_2 has metadata result
3608   __ get_vm_result_2(r0, rthread);
3609   __ pop(r3); // restore receiver
3610   __ b(resolved);
3611 
3612   // Get superklass in r0 and subklass in r3
3613   __ bind(quicked);
3614   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3615   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3616 
3617   __ bind(resolved);
3618   __ load_klass(r19, r3);
3619 
3620   // Generate subtype check.  Blows r2, r5.  Object in r3.
3621   // Superklass in r0.  Subklass in r19.
3622   __ gen_subtype_check(r19, ok_is_subtype);
3623 
3624   // Come here on failure
3625   __ push(r3);
3626   // object is at TOS
3627   __ b(Interpreter::_throw_ClassCastException_entry);
3628 
3629   // Come here on success
3630   __ bind(ok_is_subtype);
3631   __ mov(r0, r3); // Restore object in r3
3632 
3633   // Collect counts on whether this test sees NULLs a lot or not.
3634   if (ProfileInterpreter) {
3635     __ b(done);
3636     __ bind(is_null);
3637     __ profile_null_seen(r2);
3638   } else {
3639     __ bind(is_null);   // same as 'done'
3640   }
3641   __ bind(done);
3642 }
3643 
3644 void TemplateTable::instanceof() {
3645   transition(atos, itos);
3646   Label done, is_null, ok_is_subtype, quicked, resolved;
3647   __ cbz(r0, is_null);
3648 
3649   // Get cpool & tags index
3650   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3651   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3652   // See if bytecode has already been quicked
3653   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3654   __ lea(r1, Address(rscratch1, r19));
3655   __ ldarb(r1, r1);
3656   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3657   __ br(Assembler::EQ, quicked);
3658 
3659   __ push(atos); // save receiver for result, and for GC
3660   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3661   // vm_result_2 has metadata result
3662   __ get_vm_result_2(r0, rthread);
3663   __ pop(r3); // restore receiver
3664   __ verify_oop(r3);
3665   __ load_klass(rscratch1, r3);
3666   __ mov(r3, rscratch1);
3667   __ b(resolved);
3668 
3669   // Get superklass in r0 and subklass in r3
3670   __ bind(quicked);
3671   __ load_klass(r3, r0);
3672   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3673 
3674   __ bind(resolved);
3675 
3676   // Generate subtype check.  Blows r2, r5
3677   // Superklass in r0.  Subklass in r3.
3678   __ gen_subtype_check(r3, ok_is_subtype);
3679 
3680   // Come here on failure
3681   __ mov(r0, 0);
3682   __ b(done);
3683   // Come here on success
3684   __ bind(ok_is_subtype);
3685   __ mov(r0, 1);
3686 
3687   // Collect counts on whether this test sees NULLs a lot or not.
3688   if (ProfileInterpreter) {
3689     __ b(done);
3690     __ bind(is_null);
3691     __ profile_null_seen(r2);
3692   } else {
3693     __ bind(is_null);   // same as 'done'
3694   }
3695   __ bind(done);
3696   // r0 = 0: obj == NULL or  obj is not an instanceof the specified klass
3697   // r0 = 1: obj != NULL and obj is     an instanceof the specified klass
3698 }
3699 
3700 //-----------------------------------------------------------------------------
3701 // Breakpoints
3702 void TemplateTable::_breakpoint() {
3703   // Note: We get here even if we are single stepping..
3704   // jbug inists on setting breakpoints at every bytecode
3705   // even if we are in single step mode.
3706 
3707   transition(vtos, vtos);
3708 
3709   // get the unpatched byte code
3710   __ get_method(c_rarg1);
3711   __ call_VM(noreg,
3712              CAST_FROM_FN_PTR(address,
3713                               InterpreterRuntime::get_original_bytecode_at),
3714              c_rarg1, rbcp);
3715   __ mov(r19, r0);
3716 
3717   // post the breakpoint event
3718   __ call_VM(noreg,
3719              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3720              rmethod, rbcp);
3721 
3722   // complete the execution of original bytecode
3723   __ mov(rscratch1, r19);
3724   __ dispatch_only_normal(vtos);
3725 }
3726 
3727 //-----------------------------------------------------------------------------
3728 // Exceptions
3729 
3730 void TemplateTable::athrow() {
3731   transition(atos, vtos);
3732   __ null_check(r0);
3733   __ b(Interpreter::throw_exception_entry());
3734 }
3735 
3736 //-----------------------------------------------------------------------------
3737 // Synchronization
3738 //
3739 // Note: monitorenter & exit are symmetric routines; which is reflected
3740 //       in the assembly code structure as well
3741 //
3742 // Stack layout:
3743 //
3744 // [expressions  ] <--- esp               = expression stack top
3745 // ..
3746 // [expressions  ]
3747 // [monitor entry] <--- monitor block top = expression stack bot
3748 // ..
3749 // [monitor entry]
3750 // [frame data   ] <--- monitor block bot
3751 // ...
3752 // [saved rfp    ] <--- rfp
3753 void TemplateTable::monitorenter()
3754 {
3755   transition(atos, vtos);
3756 
3757   // check for NULL object
3758   __ null_check(r0);
3759 
3760   const Address monitor_block_top(
3761         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3762   const Address monitor_block_bot(
3763         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3764   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3765 
3766   Label allocated;
3767 
3768   // initialize entry pointer
3769   __ mov(c_rarg1, zr); // points to free slot or NULL
3770 
3771   // find a free slot in the monitor block (result in c_rarg1)
3772   {
3773     Label entry, loop, exit;
3774     __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3775                                         // starting with top-most entry
3776     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3777 
3778     __ b(entry);
3779 
3780     __ bind(loop);
3781     // check if current entry is used
3782     // if not used then remember entry in c_rarg1
3783     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3784     __ cmp(zr, rscratch1);
3785     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3786     // check if current entry is for same object
3787     __ cmp(r0, rscratch1);
3788     // if same object then stop searching
3789     __ br(Assembler::EQ, exit);
3790     // otherwise advance to next entry
3791     __ add(c_rarg3, c_rarg3, entry_size);
3792     __ bind(entry);
3793     // check if bottom reached
3794     __ cmp(c_rarg3, c_rarg2);
3795     // if not at bottom then check this entry
3796     __ br(Assembler::NE, loop);
3797     __ bind(exit);
3798   }
3799 
3800   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3801                             // if found, continue with that on
3802 
3803   // allocate one if there's no free slot
3804   {
3805     Label entry, loop;
3806     // 1. compute new pointers            // rsp: old expression stack top
3807 
3808     __ check_extended_sp();
3809     __ sub(sp, sp, entry_size);           // make room for the monitor
3810     __ mov(rscratch1, sp);
3811     __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
3812 
3813     __ ldr(c_rarg1, monitor_block_bot);   // c_rarg1: old expression stack bottom
3814     __ sub(esp, esp, entry_size);         // move expression stack top
3815     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3816     __ mov(c_rarg3, esp);                 // set start value for copy loop
3817     __ str(c_rarg1, monitor_block_bot);   // set new monitor block bottom
3818 
3819     __ b(entry);
3820     // 2. move expression stack contents
3821     __ bind(loop);
3822     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3823                                                    // word from old location
3824     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
3825     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
3826     __ bind(entry);
3827     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
3828     __ br(Assembler::NE, loop);      // if not at bottom then
3829                                      // copy next word
3830   }
3831 
3832   // call run-time routine
3833   // c_rarg1: points to monitor entry
3834   __ bind(allocated);
3835 
3836   // Increment bcp to point to the next bytecode, so exception
3837   // handling for async. exceptions work correctly.
3838   // The object has already been popped from the stack, so the
3839   // expression stack looks correct.
3840   __ increment(rbcp);
3841 
3842   // store object
3843   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3844   __ lock_object(c_rarg1);
3845 
3846   // check to make sure this monitor doesn't cause stack overflow after locking
3847   __ save_bcp();  // in case of exception
3848   __ generate_stack_overflow_check(0);
3849 
3850   // The bcp has already been incremented. Just need to dispatch to
3851   // next instruction.
3852   __ dispatch_next(vtos);
3853 }
3854 
3855 
3856 void TemplateTable::monitorexit()
3857 {
3858   transition(atos, vtos);
3859 
3860   // check for NULL object
3861   __ null_check(r0);
3862 
3863   const Address monitor_block_top(
3864         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3865   const Address monitor_block_bot(
3866         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3867   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3868 
3869   Label found;
3870 
3871   // find matching slot
3872   {
3873     Label entry, loop;
3874     __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3875                                         // starting with top-most entry
3876     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3877                                         // of monitor block
3878     __ b(entry);
3879 
3880     __ bind(loop);
3881     // check if current entry is for same object
3882     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3883     __ cmp(r0, rscratch1);
3884     // if same object then stop searching
3885     __ br(Assembler::EQ, found);
3886     // otherwise advance to next entry
3887     __ add(c_rarg1, c_rarg1, entry_size);
3888     __ bind(entry);
3889     // check if bottom reached
3890     __ cmp(c_rarg1, c_rarg2);
3891     // if not at bottom then check this entry
3892     __ br(Assembler::NE, loop);
3893   }
3894 
3895   // error handling. Unlocking was not block-structured
3896   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3897                    InterpreterRuntime::throw_illegal_monitor_state_exception));
3898   __ should_not_reach_here();
3899 
3900   // call run-time routine
3901   __ bind(found);
3902   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3903   __ unlock_object(c_rarg1);
3904   __ pop_ptr(r0); // discard object
3905 }
3906 
3907 
3908 // Wide instructions
3909 void TemplateTable::wide()
3910 {
3911   __ load_unsigned_byte(r19, at_bcp(1));
3912   __ mov(rscratch1, (address)Interpreter::_wentry_point);
3913   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3914   __ br(rscratch1);
3915 }
3916 
3917 
3918 // Multi arrays
3919 void TemplateTable::multianewarray() {
3920   transition(vtos, atos);
3921   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3922   // last dim is on top of stack; we want address of first one:
3923   // first_addr = last_addr + (ndims - 1) * wordSize
3924   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3925   __ sub(c_rarg1, c_rarg1, wordSize);
3926   call_VM(r0,
3927           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3928           c_rarg1);
3929   __ load_unsigned_byte(r1, at_bcp(3));
3930   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3931 }