1 /*
   2  * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2018, 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 "gc/shared/barrierSetAssembler.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/interp_masm.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "memory/universe.hpp"
  34 #include "oops/methodData.hpp"
  35 #include "oops/method.hpp"
  36 #include "oops/objArrayKlass.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "prims/methodHandles.hpp"
  39 #include "runtime/frame.inline.hpp"
  40 #include "runtime/sharedRuntime.hpp"
  41 #include "runtime/stubRoutines.hpp"
  42 #include "runtime/synchronizer.hpp"
  43 
  44 #define __ _masm->
  45 
  46 // Platform-dependent initialization
  47 
  48 void TemplateTable::pd_initialize() {
  49   // No aarch64 specific initialization
  50 }
  51 
  52 // Address computation: local variables
  53 
  54 static inline Address iaddress(int n) {
  55   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  56 }
  57 
  58 static inline Address laddress(int n) {
  59   return iaddress(n + 1);
  60 }
  61 
  62 static inline Address faddress(int n) {
  63   return iaddress(n);
  64 }
  65 
  66 static inline Address daddress(int n) {
  67   return laddress(n);
  68 }
  69 
  70 static inline Address aaddress(int n) {
  71   return iaddress(n);
  72 }
  73 
  74 static inline Address iaddress(Register r) {
  75   return Address(rlocals, r, Address::lsl(3));
  76 }
  77 
  78 static inline Address laddress(Register r, Register scratch,
  79                                InterpreterMacroAssembler* _masm) {
  80   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  81   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  82 }
  83 
  84 static inline Address faddress(Register r) {
  85   return iaddress(r);
  86 }
  87 
  88 static inline Address daddress(Register r, Register scratch,
  89                                InterpreterMacroAssembler* _masm) {
  90   return laddress(r, scratch, _masm);
  91 }
  92 
  93 static inline Address aaddress(Register r) {
  94   return iaddress(r);
  95 }
  96 
  97 static inline Address at_rsp() {
  98   return Address(esp, 0);
  99 }
 100 
 101 // At top of Java expression stack which may be different than esp().  It
 102 // isn't for category 1 objects.
 103 static inline Address at_tos   () {
 104   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 105 }
 106 
 107 static inline Address at_tos_p1() {
 108   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 109 }
 110 
 111 static inline Address at_tos_p2() {
 112   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 113 }
 114 
 115 static inline Address at_tos_p3() {
 116   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 117 }
 118 
 119 static inline Address at_tos_p4() {
 120   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 121 }
 122 
 123 static inline Address at_tos_p5() {
 124   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 125 }
 126 
 127 // Condition conversion
 128 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 129   switch (cc) {
 130   case TemplateTable::equal        : return Assembler::NE;
 131   case TemplateTable::not_equal    : return Assembler::EQ;
 132   case TemplateTable::less         : return Assembler::GE;
 133   case TemplateTable::less_equal   : return Assembler::GT;
 134   case TemplateTable::greater      : return Assembler::LE;
 135   case TemplateTable::greater_equal: return Assembler::LT;
 136   }
 137   ShouldNotReachHere();
 138   return Assembler::EQ;
 139 }
 140 
 141 
 142 // Miscelaneous helper routines
 143 // Store an oop (or NULL) at the Address described by obj.
 144 // If val == noreg this means store a NULL
 145 static void do_oop_store(InterpreterMacroAssembler* _masm,
 146                          Address dst,
 147                          Register val,
 148                          DecoratorSet decorators) {
 149   assert(val == noreg || val == r0, "parameter is just for looks");
 150   __ store_heap_oop(dst, val, r10, r1, decorators);
 151 }
 152 
 153 static void do_oop_load(InterpreterMacroAssembler* _masm,
 154                         Address src,
 155                         Register dst,
 156                         DecoratorSet decorators) {
 157   __ load_heap_oop(dst, src, r10, r1, decorators);
 158 }
 159 
 160 Address TemplateTable::at_bcp(int offset) {
 161   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 162   return Address(rbcp, offset);
 163 }
 164 
 165 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 166                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 167                                    int byte_no)
 168 {
 169   if (!RewriteBytecodes)  return;
 170   Label L_patch_done;
 171 
 172   switch (bc) {
 173   case Bytecodes::_fast_aputfield:
 174   case Bytecodes::_fast_bputfield:
 175   case Bytecodes::_fast_zputfield:
 176   case Bytecodes::_fast_cputfield:
 177   case Bytecodes::_fast_dputfield:
 178   case Bytecodes::_fast_fputfield:
 179   case Bytecodes::_fast_iputfield:
 180   case Bytecodes::_fast_lputfield:
 181   case Bytecodes::_fast_sputfield:
 182     {
 183       // We skip bytecode quickening for putfield instructions when
 184       // the put_code written to the constant pool cache is zero.
 185       // This is required so that every execution of this instruction
 186       // calls out to InterpreterRuntime::resolve_get_put to do
 187       // additional, required work.
 188       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 189       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 190       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 191       __ movw(bc_reg, bc);
 192       __ cbzw(temp_reg, L_patch_done);  // don't patch
 193     }
 194     break;
 195   default:
 196     assert(byte_no == -1, "sanity");
 197     // the pair bytecodes have already done the load.
 198     if (load_bc_into_bc_reg) {
 199       __ movw(bc_reg, bc);
 200     }
 201   }
 202 
 203   if (JvmtiExport::can_post_breakpoint()) {
 204     Label L_fast_patch;
 205     // if a breakpoint is present we can't rewrite the stream directly
 206     __ load_unsigned_byte(temp_reg, at_bcp(0));
 207     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 208     __ br(Assembler::NE, L_fast_patch);
 209     // Let breakpoint table handling rewrite to quicker bytecode
 210     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 211     __ b(L_patch_done);
 212     __ bind(L_fast_patch);
 213   }
 214 
 215 #ifdef ASSERT
 216   Label L_okay;
 217   __ load_unsigned_byte(temp_reg, at_bcp(0));
 218   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 219   __ br(Assembler::EQ, L_okay);
 220   __ cmpw(temp_reg, bc_reg);
 221   __ br(Assembler::EQ, L_okay);
 222   __ stop("patching the wrong bytecode");
 223   __ bind(L_okay);
 224 #endif
 225 
 226   // patch bytecode
 227   __ strb(bc_reg, at_bcp(0));
 228   __ bind(L_patch_done);
 229 }
 230 
 231 
 232 // Individual instructions
 233 
 234 void TemplateTable::nop() {
 235   transition(vtos, vtos);
 236   // nothing to do
 237 }
 238 
 239 void TemplateTable::shouldnotreachhere() {
 240   transition(vtos, vtos);
 241   __ stop("shouldnotreachhere bytecode");
 242 }
 243 
 244 void TemplateTable::aconst_null()
 245 {
 246   transition(vtos, atos);
 247   __ mov(r0, 0);
 248 }
 249 
 250 void TemplateTable::iconst(int value)
 251 {
 252   transition(vtos, itos);
 253   __ mov(r0, value);
 254 }
 255 
 256 void TemplateTable::lconst(int value)
 257 {
 258   __ mov(r0, value);
 259 }
 260 
 261 void TemplateTable::fconst(int value)
 262 {
 263   transition(vtos, ftos);
 264   switch (value) {
 265   case 0:
 266     __ fmovs(v0, zr);
 267     break;
 268   case 1:
 269     __ fmovs(v0, 1.0);
 270     break;
 271   case 2:
 272     __ fmovs(v0, 2.0);
 273     break;
 274   default:
 275     ShouldNotReachHere();
 276     break;
 277   }
 278 }
 279 
 280 void TemplateTable::dconst(int value)
 281 {
 282   transition(vtos, dtos);
 283   switch (value) {
 284   case 0:
 285     __ fmovd(v0, zr);
 286     break;
 287   case 1:
 288     __ fmovd(v0, 1.0);
 289     break;
 290   case 2:
 291     __ fmovd(v0, 2.0);
 292     break;
 293   default:
 294     ShouldNotReachHere();
 295     break;
 296   }
 297 }
 298 
 299 void TemplateTable::bipush()
 300 {
 301   transition(vtos, itos);
 302   __ load_signed_byte32(r0, at_bcp(1));
 303 }
 304 
 305 void TemplateTable::sipush()
 306 {
 307   transition(vtos, itos);
 308   __ load_unsigned_short(r0, at_bcp(1));
 309   __ revw(r0, r0);
 310   __ asrw(r0, r0, 16);
 311 }
 312 
 313 void TemplateTable::ldc(bool wide)
 314 {
 315   transition(vtos, vtos);
 316   Label call_ldc, notFloat, notClass, notInt, Done;
 317 
 318   if (wide) {
 319     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 320   } else {
 321     __ load_unsigned_byte(r1, at_bcp(1));
 322   }
 323   __ get_cpool_and_tags(r2, r0);
 324 
 325   const int base_offset = ConstantPool::header_size() * wordSize;
 326   const int tags_offset = Array<u1>::base_offset_in_bytes();
 327 
 328   // get type
 329   __ add(r3, r1, tags_offset);
 330   __ lea(r3, Address(r0, r3));
 331   __ ldarb(r3, r3);
 332 
 333   // unresolved class - get the resolved class
 334   __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
 335   __ br(Assembler::EQ, call_ldc);
 336 
 337   // unresolved class in error state - call into runtime to throw the error
 338   // from the first resolution attempt
 339   __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
 340   __ br(Assembler::EQ, call_ldc);
 341 
 342   // resolved class - need to call vm to get java mirror of the class
 343   __ cmp(r3, JVM_CONSTANT_Class);
 344   __ br(Assembler::NE, notClass);
 345 
 346   __ bind(call_ldc);
 347   __ mov(c_rarg1, wide);
 348   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 349   __ push_ptr(r0);
 350   __ verify_oop(r0);
 351   __ b(Done);
 352 
 353   __ bind(notClass);
 354   __ cmp(r3, JVM_CONSTANT_Float);
 355   __ br(Assembler::NE, notFloat);
 356   // ftos
 357   __ adds(r1, r2, r1, Assembler::LSL, 3);
 358   __ ldrs(v0, Address(r1, base_offset));
 359   __ push_f();
 360   __ b(Done);
 361 
 362   __ bind(notFloat);
 363 
 364   __ cmp(r3, JVM_CONSTANT_Integer);
 365   __ br(Assembler::NE, notInt);
 366 
 367   // itos
 368   __ adds(r1, r2, r1, Assembler::LSL, 3);
 369   __ ldrw(r0, Address(r1, base_offset));
 370   __ push_i(r0);
 371   __ b(Done);
 372 
 373   __ bind(notInt);
 374   condy_helper(Done);
 375 
 376   __ bind(Done);
 377 }
 378 
 379 // Fast path for caching oop constants.
 380 void TemplateTable::fast_aldc(bool wide)
 381 {
 382   transition(vtos, atos);
 383 
 384   Register result = r0;
 385   Register tmp = r1;
 386   Register rarg = r2;
 387 
 388   int index_size = wide ? sizeof(u2) : sizeof(u1);
 389 
 390   Label resolved;
 391 
 392   // We are resolved if the resolved reference cache entry contains a
 393   // non-null object (String, MethodType, etc.)
 394   assert_different_registers(result, tmp);
 395   __ get_cache_index_at_bcp(tmp, 1, index_size);
 396   __ load_resolved_reference_at_index(result, tmp);
 397   __ cbnz(result, resolved);
 398 
 399   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 400 
 401   // first time invocation - must resolve first
 402   __ mov(rarg, (int)bytecode());
 403   __ call_VM(result, entry, rarg);
 404 
 405   __ bind(resolved);
 406 
 407   { // Check for the null sentinel.
 408     // If we just called the VM, it already did the mapping for us,
 409     // but it's harmless to retry.
 410     Label notNull;
 411 
 412     // Stash null_sentinel address to get its value later
 413     __ movptr(rarg, (uintptr_t)Universe::the_null_sentinel_addr());
 414     __ ldr(tmp, Address(rarg));
 415     __ cmp(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);
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);
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);
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);
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);
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);
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_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1482                          0, 2, MacroAssembler::ret_type_float);
1483     break;
1484   default:
1485     ShouldNotReachHere();
1486     break;
1487   }
1488 }
1489 
1490 void TemplateTable::dop2(Operation op)
1491 {
1492   transition(dtos, dtos);
1493   switch (op) {
1494   case add:
1495     // n.b. use ldrd because this is a 64 bit slot
1496     __ pop_d(v1);
1497     __ faddd(v0, v1, v0);
1498     break;
1499   case sub:
1500     __ pop_d(v1);
1501     __ fsubd(v0, v1, v0);
1502     break;
1503   case mul:
1504     __ pop_d(v1);
1505     __ fmuld(v0, v1, v0);
1506     break;
1507   case div:
1508     __ pop_d(v1);
1509     __ fdivd(v0, v1, v0);
1510     break;
1511   case rem:
1512     __ fmovd(v1, v0);
1513     __ pop_d(v0);
1514     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1515                          0, 2, MacroAssembler::ret_type_double);
1516     break;
1517   default:
1518     ShouldNotReachHere();
1519     break;
1520   }
1521 }
1522 
1523 void TemplateTable::ineg()
1524 {
1525   transition(itos, itos);
1526   __ negw(r0, r0);
1527 
1528 }
1529 
1530 void TemplateTable::lneg()
1531 {
1532   transition(ltos, ltos);
1533   __ neg(r0, r0);
1534 }
1535 
1536 void TemplateTable::fneg()
1537 {
1538   transition(ftos, ftos);
1539   __ fnegs(v0, v0);
1540 }
1541 
1542 void TemplateTable::dneg()
1543 {
1544   transition(dtos, dtos);
1545   __ fnegd(v0, v0);
1546 }
1547 
1548 void TemplateTable::iinc()
1549 {
1550   transition(vtos, vtos);
1551   __ load_signed_byte(r1, at_bcp(2)); // get constant
1552   locals_index(r2);
1553   __ ldr(r0, iaddress(r2));
1554   __ addw(r0, r0, r1);
1555   __ str(r0, iaddress(r2));
1556 }
1557 
1558 void TemplateTable::wide_iinc()
1559 {
1560   transition(vtos, vtos);
1561   // __ mov(r1, zr);
1562   __ ldrw(r1, at_bcp(2)); // get constant and index
1563   __ rev16(r1, r1);
1564   __ ubfx(r2, r1, 0, 16);
1565   __ neg(r2, r2);
1566   __ sbfx(r1, r1, 16, 16);
1567   __ ldr(r0, iaddress(r2));
1568   __ addw(r0, r0, r1);
1569   __ str(r0, iaddress(r2));
1570 }
1571 
1572 void TemplateTable::convert()
1573 {
1574   // Checking
1575 #ifdef ASSERT
1576   {
1577     TosState tos_in  = ilgl;
1578     TosState tos_out = ilgl;
1579     switch (bytecode()) {
1580     case Bytecodes::_i2l: // fall through
1581     case Bytecodes::_i2f: // fall through
1582     case Bytecodes::_i2d: // fall through
1583     case Bytecodes::_i2b: // fall through
1584     case Bytecodes::_i2c: // fall through
1585     case Bytecodes::_i2s: tos_in = itos; break;
1586     case Bytecodes::_l2i: // fall through
1587     case Bytecodes::_l2f: // fall through
1588     case Bytecodes::_l2d: tos_in = ltos; break;
1589     case Bytecodes::_f2i: // fall through
1590     case Bytecodes::_f2l: // fall through
1591     case Bytecodes::_f2d: tos_in = ftos; break;
1592     case Bytecodes::_d2i: // fall through
1593     case Bytecodes::_d2l: // fall through
1594     case Bytecodes::_d2f: tos_in = dtos; break;
1595     default             : ShouldNotReachHere();
1596     }
1597     switch (bytecode()) {
1598     case Bytecodes::_l2i: // fall through
1599     case Bytecodes::_f2i: // fall through
1600     case Bytecodes::_d2i: // fall through
1601     case Bytecodes::_i2b: // fall through
1602     case Bytecodes::_i2c: // fall through
1603     case Bytecodes::_i2s: tos_out = itos; break;
1604     case Bytecodes::_i2l: // fall through
1605     case Bytecodes::_f2l: // fall through
1606     case Bytecodes::_d2l: tos_out = ltos; break;
1607     case Bytecodes::_i2f: // fall through
1608     case Bytecodes::_l2f: // fall through
1609     case Bytecodes::_d2f: tos_out = ftos; break;
1610     case Bytecodes::_i2d: // fall through
1611     case Bytecodes::_l2d: // fall through
1612     case Bytecodes::_f2d: tos_out = dtos; break;
1613     default             : ShouldNotReachHere();
1614     }
1615     transition(tos_in, tos_out);
1616   }
1617 #endif // ASSERT
1618   // static const int64_t is_nan = 0x8000000000000000L;
1619 
1620   // Conversion
1621   switch (bytecode()) {
1622   case Bytecodes::_i2l:
1623     __ sxtw(r0, r0);
1624     break;
1625   case Bytecodes::_i2f:
1626     __ scvtfws(v0, r0);
1627     break;
1628   case Bytecodes::_i2d:
1629     __ scvtfwd(v0, r0);
1630     break;
1631   case Bytecodes::_i2b:
1632     __ sxtbw(r0, r0);
1633     break;
1634   case Bytecodes::_i2c:
1635     __ uxthw(r0, r0);
1636     break;
1637   case Bytecodes::_i2s:
1638     __ sxthw(r0, r0);
1639     break;
1640   case Bytecodes::_l2i:
1641     __ uxtw(r0, r0);
1642     break;
1643   case Bytecodes::_l2f:
1644     __ scvtfs(v0, r0);
1645     break;
1646   case Bytecodes::_l2d:
1647     __ scvtfd(v0, r0);
1648     break;
1649   case Bytecodes::_f2i:
1650   {
1651     Label L_Okay;
1652     __ clear_fpsr();
1653     __ fcvtzsw(r0, v0);
1654     __ get_fpsr(r1);
1655     __ cbzw(r1, L_Okay);
1656     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1657                          0, 1, MacroAssembler::ret_type_integral);
1658     __ bind(L_Okay);
1659   }
1660     break;
1661   case Bytecodes::_f2l:
1662   {
1663     Label L_Okay;
1664     __ clear_fpsr();
1665     __ fcvtzs(r0, v0);
1666     __ get_fpsr(r1);
1667     __ cbzw(r1, L_Okay);
1668     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1669                          0, 1, MacroAssembler::ret_type_integral);
1670     __ bind(L_Okay);
1671   }
1672     break;
1673   case Bytecodes::_f2d:
1674     __ fcvts(v0, v0);
1675     break;
1676   case Bytecodes::_d2i:
1677   {
1678     Label L_Okay;
1679     __ clear_fpsr();
1680     __ fcvtzdw(r0, v0);
1681     __ get_fpsr(r1);
1682     __ cbzw(r1, L_Okay);
1683     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1684                          0, 1, MacroAssembler::ret_type_integral);
1685     __ bind(L_Okay);
1686   }
1687     break;
1688   case Bytecodes::_d2l:
1689   {
1690     Label L_Okay;
1691     __ clear_fpsr();
1692     __ fcvtzd(r0, v0);
1693     __ get_fpsr(r1);
1694     __ cbzw(r1, L_Okay);
1695     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1696                          0, 1, MacroAssembler::ret_type_integral);
1697     __ bind(L_Okay);
1698   }
1699     break;
1700   case Bytecodes::_d2f:
1701     __ fcvtd(v0, v0);
1702     break;
1703   default:
1704     ShouldNotReachHere();
1705   }
1706 }
1707 
1708 void TemplateTable::lcmp()
1709 {
1710   transition(ltos, itos);
1711   Label done;
1712   __ pop_l(r1);
1713   __ cmp(r1, r0);
1714   __ mov(r0, (u_int64_t)-1L);
1715   __ br(Assembler::LT, done);
1716   // __ mov(r0, 1UL);
1717   // __ csel(r0, r0, zr, Assembler::NE);
1718   // and here is a faster way
1719   __ csinc(r0, zr, zr, Assembler::EQ);
1720   __ bind(done);
1721 }
1722 
1723 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1724 {
1725   Label done;
1726   if (is_float) {
1727     // XXX get rid of pop here, use ... reg, mem32
1728     __ pop_f(v1);
1729     __ fcmps(v1, v0);
1730   } else {
1731     // XXX get rid of pop here, use ... reg, mem64
1732     __ pop_d(v1);
1733     __ fcmpd(v1, v0);
1734   }
1735   if (unordered_result < 0) {
1736     // we want -1 for unordered or less than, 0 for equal and 1 for
1737     // greater than.
1738     __ mov(r0, (u_int64_t)-1L);
1739     // for FP LT tests less than or unordered
1740     __ br(Assembler::LT, done);
1741     // install 0 for EQ otherwise 1
1742     __ csinc(r0, zr, zr, Assembler::EQ);
1743   } else {
1744     // we want -1 for less than, 0 for equal and 1 for unordered or
1745     // greater than.
1746     __ mov(r0, 1L);
1747     // for FP HI tests greater than or unordered
1748     __ br(Assembler::HI, done);
1749     // install 0 for EQ otherwise ~0
1750     __ csinv(r0, zr, zr, Assembler::EQ);
1751 
1752   }
1753   __ bind(done);
1754 }
1755 
1756 void TemplateTable::branch(bool is_jsr, bool is_wide)
1757 {
1758   // We might be moving to a safepoint.  The thread which calls
1759   // Interpreter::notice_safepoints() will effectively flush its cache
1760   // when it makes a system call, but we need to do something to
1761   // ensure that we see the changed dispatch table.
1762   __ membar(MacroAssembler::LoadLoad);
1763 
1764   __ profile_taken_branch(r0, r1);
1765   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1766                              InvocationCounter::counter_offset();
1767   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1768                               InvocationCounter::counter_offset();
1769 
1770   // load branch displacement
1771   if (!is_wide) {
1772     __ ldrh(r2, at_bcp(1));
1773     __ rev16(r2, r2);
1774     // sign extend the 16 bit value in r2
1775     __ sbfm(r2, r2, 0, 15);
1776   } else {
1777     __ ldrw(r2, at_bcp(1));
1778     __ revw(r2, r2);
1779     // sign extend the 32 bit value in r2
1780     __ sbfm(r2, r2, 0, 31);
1781   }
1782 
1783   // Handle all the JSR stuff here, then exit.
1784   // It's much shorter and cleaner than intermingling with the non-JSR
1785   // normal-branch stuff occurring below.
1786 
1787   if (is_jsr) {
1788     // Pre-load the next target bytecode into rscratch1
1789     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1790     // compute return address as bci
1791     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1792     __ add(rscratch2, rscratch2,
1793            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1794     __ sub(r1, rbcp, rscratch2);
1795     __ push_i(r1);
1796     // Adjust the bcp by the 16-bit displacement in r2
1797     __ add(rbcp, rbcp, r2);
1798     __ dispatch_only(vtos, /*generate_poll*/true);
1799     return;
1800   }
1801 
1802   // Normal (non-jsr) branch handling
1803 
1804   // Adjust the bcp by the displacement in r2
1805   __ add(rbcp, rbcp, r2);
1806 
1807   assert(UseLoopCounter || !UseOnStackReplacement,
1808          "on-stack-replacement requires loop counters");
1809   Label backedge_counter_overflow;
1810   Label profile_method;
1811   Label dispatch;
1812   if (UseLoopCounter) {
1813     // increment backedge counter for backward branches
1814     // r0: MDO
1815     // w1: MDO bumped taken-count
1816     // r2: target offset
1817     __ cmp(r2, zr);
1818     __ br(Assembler::GT, dispatch); // count only if backward branch
1819 
1820     // ECN: FIXME: This code smells
1821     // check if MethodCounters exists
1822     Label has_counters;
1823     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1824     __ cbnz(rscratch1, has_counters);
1825     __ push(r0);
1826     __ push(r1);
1827     __ push(r2);
1828     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1829             InterpreterRuntime::build_method_counters), rmethod);
1830     __ pop(r2);
1831     __ pop(r1);
1832     __ pop(r0);
1833     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1834     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1835     __ bind(has_counters);
1836 
1837     if (TieredCompilation) {
1838       Label no_mdo;
1839       int increment = InvocationCounter::count_increment;
1840       if (ProfileInterpreter) {
1841         // Are we profiling?
1842         __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1843         __ cbz(r1, no_mdo);
1844         // Increment the MDO backedge counter
1845         const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1846                                            in_bytes(InvocationCounter::counter_offset()));
1847         const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1848         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1849                                    r0, rscratch1, false, Assembler::EQ,
1850                                    UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1851         __ b(dispatch);
1852       }
1853       __ bind(no_mdo);
1854       // Increment backedge counter in MethodCounters*
1855       __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1856       const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1857       __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1858                                  r0, rscratch2, false, Assembler::EQ,
1859                                  UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1860     } else { // not TieredCompilation
1861       // increment counter
1862       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1863       __ ldrw(r0, Address(rscratch2, be_offset));        // load backedge counter
1864       __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1865       __ strw(rscratch1, Address(rscratch2, be_offset));        // store counter
1866 
1867       __ ldrw(r0, Address(rscratch2, inv_offset));    // load invocation counter
1868       __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1869       __ addw(r0, r0, rscratch1);        // add both counters
1870 
1871       if (ProfileInterpreter) {
1872         // Test to see if we should create a method data oop
1873         __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1874         __ cmpw(r0, rscratch1);
1875         __ br(Assembler::LT, dispatch);
1876 
1877         // if no method data exists, go to profile method
1878         __ test_method_data_pointer(r0, profile_method);
1879 
1880         if (UseOnStackReplacement) {
1881           // check for overflow against w1 which is the MDO taken count
1882           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1883           __ cmpw(r1, rscratch1);
1884           __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1885 
1886           // When ProfileInterpreter is on, the backedge_count comes
1887           // from the MethodData*, which value does not get reset on
1888           // the call to frequency_counter_overflow().  To avoid
1889           // excessive calls to the overflow routine while the method is
1890           // being compiled, add a second test to make sure the overflow
1891           // function is called only once every overflow_frequency.
1892           const int overflow_frequency = 1024;
1893           __ andsw(r1, r1, overflow_frequency - 1);
1894           __ br(Assembler::EQ, backedge_counter_overflow);
1895 
1896         }
1897       } else {
1898         if (UseOnStackReplacement) {
1899           // check for overflow against w0, which is the sum of the
1900           // counters
1901           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1902           __ cmpw(r0, rscratch1);
1903           __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1904         }
1905       }
1906     }
1907     __ bind(dispatch);
1908   }
1909 
1910   // Pre-load the next target bytecode into rscratch1
1911   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1912 
1913   // continue with the bytecode @ target
1914   // rscratch1: target bytecode
1915   // rbcp: target bcp
1916   __ dispatch_only(vtos, /*generate_poll*/true);
1917 
1918   if (UseLoopCounter) {
1919     if (ProfileInterpreter) {
1920       // Out-of-line code to allocate method data oop.
1921       __ bind(profile_method);
1922       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1923       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1924       __ set_method_data_pointer_for_bcp();
1925       __ b(dispatch);
1926     }
1927 
1928     if (UseOnStackReplacement) {
1929       // invocation counter overflow
1930       __ bind(backedge_counter_overflow);
1931       __ neg(r2, r2);
1932       __ add(r2, r2, rbcp);     // branch bcp
1933       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1934       __ call_VM(noreg,
1935                  CAST_FROM_FN_PTR(address,
1936                                   InterpreterRuntime::frequency_counter_overflow),
1937                  r2);
1938       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1939 
1940       // r0: osr nmethod (osr ok) or NULL (osr not possible)
1941       // w1: target bytecode
1942       // r2: scratch
1943       __ cbz(r0, dispatch);     // test result -- no osr if null
1944       // nmethod may have been invalidated (VM may block upon call_VM return)
1945       __ ldrb(r2, Address(r0, nmethod::state_offset()));
1946       if (nmethod::in_use != 0)
1947         __ sub(r2, r2, nmethod::in_use);
1948       __ cbnz(r2, dispatch);
1949 
1950       // We have the address of an on stack replacement routine in r0
1951       // We need to prepare to execute the OSR method. First we must
1952       // migrate the locals and monitors off of the stack.
1953 
1954       __ mov(r19, r0);                             // save the nmethod
1955 
1956       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1957 
1958       // r0 is OSR buffer, move it to expected parameter location
1959       __ mov(j_rarg0, r0);
1960 
1961       // remove activation
1962       // get sender esp
1963       __ ldr(esp,
1964           Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1965       // remove frame anchor
1966       __ leave();
1967       // Ensure compiled code always sees stack at proper alignment
1968       __ andr(sp, esp, -16);
1969 
1970       // and begin the OSR nmethod
1971       __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1972       __ br(rscratch1);
1973     }
1974   }
1975 }
1976 
1977 
1978 void TemplateTable::if_0cmp(Condition cc)
1979 {
1980   transition(itos, vtos);
1981   // assume branch is more often taken than not (loops use backward branches)
1982   Label not_taken;
1983   if (cc == equal)
1984     __ cbnzw(r0, not_taken);
1985   else if (cc == not_equal)
1986     __ cbzw(r0, not_taken);
1987   else {
1988     __ andsw(zr, r0, r0);
1989     __ br(j_not(cc), not_taken);
1990   }
1991 
1992   branch(false, false);
1993   __ bind(not_taken);
1994   __ profile_not_taken_branch(r0);
1995 }
1996 
1997 void TemplateTable::if_icmp(Condition cc)
1998 {
1999   transition(itos, vtos);
2000   // assume branch is more often taken than not (loops use backward branches)
2001   Label not_taken;
2002   __ pop_i(r1);
2003   __ cmpw(r1, r0, Assembler::LSL);
2004   __ br(j_not(cc), not_taken);
2005   branch(false, false);
2006   __ bind(not_taken);
2007   __ profile_not_taken_branch(r0);
2008 }
2009 
2010 void TemplateTable::if_nullcmp(Condition cc)
2011 {
2012   transition(atos, vtos);
2013   // assume branch is more often taken than not (loops use backward branches)
2014   Label not_taken;
2015   if (cc == equal)
2016     __ cbnz(r0, not_taken);
2017   else
2018     __ cbz(r0, not_taken);
2019   branch(false, false);
2020   __ bind(not_taken);
2021   __ profile_not_taken_branch(r0);
2022 }
2023 
2024 void TemplateTable::if_acmp(Condition cc)
2025 {
2026   transition(atos, vtos);
2027   // assume branch is more often taken than not (loops use backward branches)
2028   Label not_taken;
2029   __ pop_ptr(r1);
2030   __ cmpoop(r1, r0);
2031   __ br(j_not(cc), not_taken);
2032   branch(false, false);
2033   __ bind(not_taken);
2034   __ profile_not_taken_branch(r0);
2035 }
2036 
2037 void TemplateTable::ret() {
2038   transition(vtos, vtos);
2039   // We might be moving to a safepoint.  The thread which calls
2040   // Interpreter::notice_safepoints() will effectively flush its cache
2041   // when it makes a system call, but we need to do something to
2042   // ensure that we see the changed dispatch table.
2043   __ membar(MacroAssembler::LoadLoad);
2044 
2045   locals_index(r1);
2046   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2047   __ profile_ret(r1, r2);
2048   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2049   __ lea(rbcp, Address(rbcp, r1));
2050   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2051   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2052 }
2053 
2054 void TemplateTable::wide_ret() {
2055   transition(vtos, vtos);
2056   locals_index_wide(r1);
2057   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2058   __ profile_ret(r1, r2);
2059   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2060   __ lea(rbcp, Address(rbcp, r1));
2061   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2062   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2063 }
2064 
2065 
2066 void TemplateTable::tableswitch() {
2067   Label default_case, continue_execution;
2068   transition(itos, vtos);
2069   // align rbcp
2070   __ lea(r1, at_bcp(BytesPerInt));
2071   __ andr(r1, r1, -BytesPerInt);
2072   // load lo & hi
2073   __ ldrw(r2, Address(r1, BytesPerInt));
2074   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2075   __ rev32(r2, r2);
2076   __ rev32(r3, r3);
2077   // check against lo & hi
2078   __ cmpw(r0, r2);
2079   __ br(Assembler::LT, default_case);
2080   __ cmpw(r0, r3);
2081   __ br(Assembler::GT, default_case);
2082   // lookup dispatch offset
2083   __ subw(r0, r0, r2);
2084   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2085   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2086   __ profile_switch_case(r0, r1, r2);
2087   // continue execution
2088   __ bind(continue_execution);
2089   __ rev32(r3, r3);
2090   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2091   __ add(rbcp, rbcp, r3, ext::sxtw);
2092   __ dispatch_only(vtos, /*generate_poll*/true);
2093   // handle default
2094   __ bind(default_case);
2095   __ profile_switch_default(r0);
2096   __ ldrw(r3, Address(r1, 0));
2097   __ b(continue_execution);
2098 }
2099 
2100 void TemplateTable::lookupswitch() {
2101   transition(itos, itos);
2102   __ stop("lookupswitch bytecode should have been rewritten");
2103 }
2104 
2105 void TemplateTable::fast_linearswitch() {
2106   transition(itos, vtos);
2107   Label loop_entry, loop, found, continue_execution;
2108   // bswap r0 so we can avoid bswapping the table entries
2109   __ rev32(r0, r0);
2110   // align rbcp
2111   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2112                                     // this instruction (change offsets
2113                                     // below)
2114   __ andr(r19, r19, -BytesPerInt);
2115   // set counter
2116   __ ldrw(r1, Address(r19, BytesPerInt));
2117   __ rev32(r1, r1);
2118   __ b(loop_entry);
2119   // table search
2120   __ bind(loop);
2121   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2122   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2123   __ cmpw(r0, rscratch1);
2124   __ br(Assembler::EQ, found);
2125   __ bind(loop_entry);
2126   __ subs(r1, r1, 1);
2127   __ br(Assembler::PL, loop);
2128   // default case
2129   __ profile_switch_default(r0);
2130   __ ldrw(r3, Address(r19, 0));
2131   __ b(continue_execution);
2132   // entry found -> get offset
2133   __ bind(found);
2134   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2135   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2136   __ profile_switch_case(r1, r0, r19);
2137   // continue execution
2138   __ bind(continue_execution);
2139   __ rev32(r3, r3);
2140   __ add(rbcp, rbcp, r3, ext::sxtw);
2141   __ ldrb(rscratch1, Address(rbcp, 0));
2142   __ dispatch_only(vtos, /*generate_poll*/true);
2143 }
2144 
2145 void TemplateTable::fast_binaryswitch() {
2146   transition(itos, vtos);
2147   // Implementation using the following core algorithm:
2148   //
2149   // int binary_search(int key, LookupswitchPair* array, int n) {
2150   //   // Binary search according to "Methodik des Programmierens" by
2151   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2152   //   int i = 0;
2153   //   int j = n;
2154   //   while (i+1 < j) {
2155   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2156   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2157   //     // where a stands for the array and assuming that the (inexisting)
2158   //     // element a[n] is infinitely big.
2159   //     int h = (i + j) >> 1;
2160   //     // i < h < j
2161   //     if (key < array[h].fast_match()) {
2162   //       j = h;
2163   //     } else {
2164   //       i = h;
2165   //     }
2166   //   }
2167   //   // R: a[i] <= key < a[i+1] or Q
2168   //   // (i.e., if key is within array, i is the correct index)
2169   //   return i;
2170   // }
2171 
2172   // Register allocation
2173   const Register key   = r0; // already set (tosca)
2174   const Register array = r1;
2175   const Register i     = r2;
2176   const Register j     = r3;
2177   const Register h     = rscratch1;
2178   const Register temp  = rscratch2;
2179 
2180   // Find array start
2181   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2182                                           // get rid of this
2183                                           // instruction (change
2184                                           // offsets below)
2185   __ andr(array, array, -BytesPerInt);
2186 
2187   // Initialize i & j
2188   __ mov(i, 0);                            // i = 0;
2189   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2190 
2191   // Convert j into native byteordering
2192   __ rev32(j, j);
2193 
2194   // And start
2195   Label entry;
2196   __ b(entry);
2197 
2198   // binary search loop
2199   {
2200     Label loop;
2201     __ bind(loop);
2202     // int h = (i + j) >> 1;
2203     __ addw(h, i, j);                           // h = i + j;
2204     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2205     // if (key < array[h].fast_match()) {
2206     //   j = h;
2207     // } else {
2208     //   i = h;
2209     // }
2210     // Convert array[h].match to native byte-ordering before compare
2211     __ ldr(temp, Address(array, h, Address::lsl(3)));
2212     __ rev32(temp, temp);
2213     __ cmpw(key, temp);
2214     // j = h if (key <  array[h].fast_match())
2215     __ csel(j, h, j, Assembler::LT);
2216     // i = h if (key >= array[h].fast_match())
2217     __ csel(i, h, i, Assembler::GE);
2218     // while (i+1 < j)
2219     __ bind(entry);
2220     __ addw(h, i, 1);          // i+1
2221     __ cmpw(h, j);             // i+1 < j
2222     __ br(Assembler::LT, loop);
2223   }
2224 
2225   // end of binary search, result index is i (must check again!)
2226   Label default_case;
2227   // Convert array[i].match to native byte-ordering before compare
2228   __ ldr(temp, Address(array, i, Address::lsl(3)));
2229   __ rev32(temp, temp);
2230   __ cmpw(key, temp);
2231   __ br(Assembler::NE, default_case);
2232 
2233   // entry found -> j = offset
2234   __ add(j, array, i, ext::uxtx, 3);
2235   __ ldrw(j, Address(j, BytesPerInt));
2236   __ profile_switch_case(i, key, array);
2237   __ rev32(j, j);
2238   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2239   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2240   __ dispatch_only(vtos, /*generate_poll*/true);
2241 
2242   // default case -> j = default offset
2243   __ bind(default_case);
2244   __ profile_switch_default(i);
2245   __ ldrw(j, Address(array, -2 * BytesPerInt));
2246   __ rev32(j, j);
2247   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2248   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2249   __ dispatch_only(vtos, /*generate_poll*/true);
2250 }
2251 
2252 
2253 void TemplateTable::_return(TosState state)
2254 {
2255   transition(state, state);
2256   assert(_desc->calls_vm(),
2257          "inconsistent calls_vm information"); // call in remove_activation
2258 
2259   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2260     assert(state == vtos, "only valid state");
2261 
2262     __ ldr(c_rarg1, aaddress(0));
2263     __ load_klass(r3, c_rarg1);
2264     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2265     Label skip_register_finalizer;
2266     __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2267 
2268     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2269 
2270     __ bind(skip_register_finalizer);
2271   }
2272 
2273   // Issue a StoreStore barrier after all stores but before return
2274   // from any constructor for any class with a final field.  We don't
2275   // know if this is a finalizer, so we always do so.
2276   if (_desc->bytecode() == Bytecodes::_return)
2277     __ membar(MacroAssembler::StoreStore);
2278 
2279   // Narrow result if state is itos but result type is smaller.
2280   // Need to narrow in the return bytecode rather than in generate_return_entry
2281   // since compiled code callers expect the result to already be narrowed.
2282   if (state == itos) {
2283     __ narrow(r0);
2284   }
2285 
2286   __ remove_activation(state);
2287   __ ret(lr);
2288 }
2289 
2290 // ----------------------------------------------------------------------------
2291 // Volatile variables demand their effects be made known to all CPU's
2292 // in order.  Store buffers on most chips allow reads & writes to
2293 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2294 // without some kind of memory barrier (i.e., it's not sufficient that
2295 // the interpreter does not reorder volatile references, the hardware
2296 // also must not reorder them).
2297 //
2298 // According to the new Java Memory Model (JMM):
2299 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2300 //     writes act as aquire & release, so:
2301 // (2) A read cannot let unrelated NON-volatile memory refs that
2302 //     happen after the read float up to before the read.  It's OK for
2303 //     non-volatile memory refs that happen before the volatile read to
2304 //     float down below it.
2305 // (3) Similar a volatile write cannot let unrelated NON-volatile
2306 //     memory refs that happen BEFORE the write float down to after the
2307 //     write.  It's OK for non-volatile memory refs that happen after the
2308 //     volatile write to float up before it.
2309 //
2310 // We only put in barriers around volatile refs (they are expensive),
2311 // not _between_ memory refs (that would require us to track the
2312 // flavor of the previous memory refs).  Requirements (2) and (3)
2313 // require some barriers before volatile stores and after volatile
2314 // loads.  These nearly cover requirement (1) but miss the
2315 // volatile-store-volatile-load case.  This final case is placed after
2316 // volatile-stores although it could just as well go before
2317 // volatile-loads.
2318 
2319 void TemplateTable::resolve_cache_and_index(int byte_no,
2320                                             Register Rcache,
2321                                             Register index,
2322                                             size_t index_size) {
2323   const Register temp = r19;
2324   assert_different_registers(Rcache, index, temp);
2325 
2326   Label resolved;
2327 
2328   Bytecodes::Code code = bytecode();
2329   switch (code) {
2330   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2331   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2332   }
2333 
2334   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2335   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2336   __ cmp(temp, (int) code);  // have we resolved this bytecode?
2337   __ br(Assembler::EQ, resolved);
2338 
2339   // resolve first time through
2340   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2341   __ mov(temp, (int) code);
2342   __ call_VM(noreg, entry, temp);
2343 
2344   // Update registers with resolved info
2345   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2346   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2347   // so all clients ofthis method must be modified accordingly
2348   __ bind(resolved);
2349 }
2350 
2351 // The Rcache and index registers must be set before call
2352 // n.b unlike x86 cache already includes the index offset
2353 void TemplateTable::load_field_cp_cache_entry(Register obj,
2354                                               Register cache,
2355                                               Register index,
2356                                               Register off,
2357                                               Register flags,
2358                                               bool is_static = false) {
2359   assert_different_registers(cache, index, flags, off);
2360 
2361   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2362   // Field offset
2363   __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2364                                           ConstantPoolCacheEntry::f2_offset())));
2365   // Flags
2366   __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2367                                            ConstantPoolCacheEntry::flags_offset())));
2368 
2369   // klass overwrite register
2370   if (is_static) {
2371     __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2372                                         ConstantPoolCacheEntry::f1_offset())));
2373     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2374     __ ldr(obj, Address(obj, mirror_offset));
2375     __ resolve_oop_handle(obj);
2376   }
2377 }
2378 
2379 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2380                                                Register method,
2381                                                Register itable_index,
2382                                                Register flags,
2383                                                bool is_invokevirtual,
2384                                                bool is_invokevfinal, /*unused*/
2385                                                bool is_invokedynamic) {
2386   // setup registers
2387   const Register cache = rscratch2;
2388   const Register index = r4;
2389   assert_different_registers(method, flags);
2390   assert_different_registers(method, cache, index);
2391   assert_different_registers(itable_index, flags);
2392   assert_different_registers(itable_index, cache, index);
2393   // determine constant pool cache field offsets
2394   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2395   const int method_offset = in_bytes(
2396     ConstantPoolCache::base_offset() +
2397       (is_invokevirtual
2398        ? ConstantPoolCacheEntry::f2_offset()
2399        : ConstantPoolCacheEntry::f1_offset()));
2400   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2401                                     ConstantPoolCacheEntry::flags_offset());
2402   // access constant pool cache fields
2403   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2404                                     ConstantPoolCacheEntry::f2_offset());
2405 
2406   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2407   resolve_cache_and_index(byte_no, cache, index, index_size);
2408   __ ldr(method, Address(cache, method_offset));
2409 
2410   if (itable_index != noreg) {
2411     __ ldr(itable_index, Address(cache, index_offset));
2412   }
2413   __ ldrw(flags, Address(cache, flags_offset));
2414 }
2415 
2416 
2417 // The registers cache and index expected to be set before call.
2418 // Correct values of the cache and index registers are preserved.
2419 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2420                                             bool is_static, bool has_tos) {
2421   // do the JVMTI work here to avoid disturbing the register state below
2422   // We use c_rarg registers here because we want to use the register used in
2423   // the call to the VM
2424   if (JvmtiExport::can_post_field_access()) {
2425     // Check to see if a field access watch has been set before we
2426     // take the time to call into the VM.
2427     Label L1;
2428     assert_different_registers(cache, index, r0);
2429     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2430     __ ldrw(r0, Address(rscratch1));
2431     __ cbzw(r0, L1);
2432 
2433     __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2434     __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2435 
2436     if (is_static) {
2437       __ mov(c_rarg1, zr); // NULL object reference
2438     } else {
2439       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2440       __ verify_oop(c_rarg1);
2441     }
2442     // c_rarg1: object pointer or NULL
2443     // c_rarg2: cache entry pointer
2444     // c_rarg3: jvalue object on the stack
2445     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2446                                        InterpreterRuntime::post_field_access),
2447                c_rarg1, c_rarg2, c_rarg3);
2448     __ get_cache_and_index_at_bcp(cache, index, 1);
2449     __ bind(L1);
2450   }
2451 }
2452 
2453 void TemplateTable::pop_and_check_object(Register r)
2454 {
2455   __ pop_ptr(r);
2456   __ null_check(r);  // for field access must check obj.
2457   __ verify_oop(r);
2458 }
2459 
2460 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2461 {
2462   const Register cache = r2;
2463   const Register index = r3;
2464   const Register obj   = r4;
2465   const Register off   = r19;
2466   const Register flags = r0;
2467   const Register raw_flags = r6;
2468   const Register bc    = r4; // uses same reg as obj, so don't mix them
2469 
2470   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2471   jvmti_post_field_access(cache, index, is_static, false);
2472   load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2473 
2474   if (!is_static) {
2475     // obj is on the stack
2476     pop_and_check_object(obj);
2477   }
2478 
2479   // 8179954: We need to make sure that the code generated for
2480   // volatile accesses forms a sequentially-consistent set of
2481   // operations when combined with STLR and LDAR.  Without a leading
2482   // membar it's possible for a simple Dekker test to fail if loads
2483   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2484   // the stores in one method and we interpret the loads in another.
2485   if (! UseBarriersForVolatile) {
2486     Label notVolatile;
2487     __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2488     __ membar(MacroAssembler::AnyAny);
2489     __ bind(notVolatile);
2490   }
2491 
2492   const Address field(obj, off);
2493 
2494   Label Done, notByte, notBool, notInt, notShort, notChar,
2495               notLong, notFloat, notObj, notDouble;
2496 
2497   // x86 uses a shift and mask or wings it with a shift plus assert
2498   // the mask is not needed. aarch64 just uses bitfield extract
2499   __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2500            ConstantPoolCacheEntry::tos_state_bits);
2501 
2502   assert(btos == 0, "change code, btos != 0");
2503   __ cbnz(flags, notByte);
2504 
2505   // Don't rewrite getstatic, only getfield
2506   if (is_static) rc = may_not_rewrite;
2507 
2508   // btos
2509   __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
2510   __ push(btos);
2511   // Rewrite bytecode to be faster
2512   if (rc == may_rewrite) {
2513     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2514   }
2515   __ b(Done);
2516 
2517   __ bind(notByte);
2518   __ cmp(flags, ztos);
2519   __ br(Assembler::NE, notBool);
2520 
2521   // ztos (same code as btos)
2522   __ access_load_at(T_BOOLEAN, IN_HEAP, r0, field, noreg, noreg);
2523   __ push(ztos);
2524   // Rewrite bytecode to be faster
2525   if (rc == may_rewrite) {
2526     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2527     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2528   }
2529   __ b(Done);
2530 
2531   __ bind(notBool);
2532   __ cmp(flags, atos);
2533   __ br(Assembler::NE, notObj);
2534   // atos
2535   do_oop_load(_masm, field, r0, IN_HEAP);
2536   __ push(atos);
2537   if (rc == may_rewrite) {
2538     patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2539   }
2540   __ b(Done);
2541 
2542   __ bind(notObj);
2543   __ cmp(flags, itos);
2544   __ br(Assembler::NE, notInt);
2545   // itos
2546   __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
2547   __ push(itos);
2548   // Rewrite bytecode to be faster
2549   if (rc == may_rewrite) {
2550     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2551   }
2552   __ b(Done);
2553 
2554   __ bind(notInt);
2555   __ cmp(flags, ctos);
2556   __ br(Assembler::NE, notChar);
2557   // ctos
2558   __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
2559   __ push(ctos);
2560   // Rewrite bytecode to be faster
2561   if (rc == may_rewrite) {
2562     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2563   }
2564   __ b(Done);
2565 
2566   __ bind(notChar);
2567   __ cmp(flags, stos);
2568   __ br(Assembler::NE, notShort);
2569   // stos
2570   __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
2571   __ push(stos);
2572   // Rewrite bytecode to be faster
2573   if (rc == may_rewrite) {
2574     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2575   }
2576   __ b(Done);
2577 
2578   __ bind(notShort);
2579   __ cmp(flags, ltos);
2580   __ br(Assembler::NE, notLong);
2581   // ltos
2582   __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
2583   __ push(ltos);
2584   // Rewrite bytecode to be faster
2585   if (rc == may_rewrite) {
2586     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2587   }
2588   __ b(Done);
2589 
2590   __ bind(notLong);
2591   __ cmp(flags, ftos);
2592   __ br(Assembler::NE, notFloat);
2593   // ftos
2594   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2595   __ push(ftos);
2596   // Rewrite bytecode to be faster
2597   if (rc == may_rewrite) {
2598     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2599   }
2600   __ b(Done);
2601 
2602   __ bind(notFloat);
2603 #ifdef ASSERT
2604   __ cmp(flags, dtos);
2605   __ br(Assembler::NE, notDouble);
2606 #endif
2607   // dtos
2608   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2609   __ push(dtos);
2610   // Rewrite bytecode to be faster
2611   if (rc == may_rewrite) {
2612     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2613   }
2614 #ifdef ASSERT
2615   __ b(Done);
2616 
2617   __ bind(notDouble);
2618   __ stop("Bad state");
2619 #endif
2620 
2621   __ bind(Done);
2622 
2623   Label notVolatile;
2624   __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2625   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2626   __ bind(notVolatile);
2627 }
2628 
2629 
2630 void TemplateTable::getfield(int byte_no)
2631 {
2632   getfield_or_static(byte_no, false);
2633 }
2634 
2635 void TemplateTable::nofast_getfield(int byte_no) {
2636   getfield_or_static(byte_no, false, may_not_rewrite);
2637 }
2638 
2639 void TemplateTable::getstatic(int byte_no)
2640 {
2641   getfield_or_static(byte_no, true);
2642 }
2643 
2644 // The registers cache and index expected to be set before call.
2645 // The function may destroy various registers, just not the cache and index registers.
2646 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2647   transition(vtos, vtos);
2648 
2649   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2650 
2651   if (JvmtiExport::can_post_field_modification()) {
2652     // Check to see if a field modification watch has been set before
2653     // we take the time to call into the VM.
2654     Label L1;
2655     assert_different_registers(cache, index, r0);
2656     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2657     __ ldrw(r0, Address(rscratch1));
2658     __ cbz(r0, L1);
2659 
2660     __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2661 
2662     if (is_static) {
2663       // Life is simple.  Null out the object pointer.
2664       __ mov(c_rarg1, zr);
2665     } else {
2666       // Life is harder. The stack holds the value on top, followed by
2667       // the object.  We don't know the size of the value, though; it
2668       // could be one or two words depending on its type. As a result,
2669       // we must find the type to determine where the object is.
2670       __ ldrw(c_rarg3, Address(c_rarg2,
2671                                in_bytes(cp_base_offset +
2672                                         ConstantPoolCacheEntry::flags_offset())));
2673       __ lsr(c_rarg3, c_rarg3,
2674              ConstantPoolCacheEntry::tos_state_shift);
2675       ConstantPoolCacheEntry::verify_tos_state_shift();
2676       Label nope2, done, ok;
2677       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2678       __ cmpw(c_rarg3, ltos);
2679       __ br(Assembler::EQ, ok);
2680       __ cmpw(c_rarg3, dtos);
2681       __ br(Assembler::NE, nope2);
2682       __ bind(ok);
2683       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2684       __ bind(nope2);
2685     }
2686     // cache entry pointer
2687     __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2688     // object (tos)
2689     __ mov(c_rarg3, esp);
2690     // c_rarg1: object pointer set up above (NULL if static)
2691     // c_rarg2: cache entry pointer
2692     // c_rarg3: jvalue object on the stack
2693     __ call_VM(noreg,
2694                CAST_FROM_FN_PTR(address,
2695                                 InterpreterRuntime::post_field_modification),
2696                c_rarg1, c_rarg2, c_rarg3);
2697     __ get_cache_and_index_at_bcp(cache, index, 1);
2698     __ bind(L1);
2699   }
2700 }
2701 
2702 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2703   transition(vtos, vtos);
2704 
2705   const Register cache = r2;
2706   const Register index = r3;
2707   const Register obj   = r2;
2708   const Register off   = r19;
2709   const Register flags = r0;
2710   const Register bc    = r4;
2711 
2712   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2713   jvmti_post_field_mod(cache, index, is_static);
2714   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2715 
2716   Label Done;
2717   __ mov(r5, flags);
2718 
2719   {
2720     Label notVolatile;
2721     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2722     __ membar(MacroAssembler::StoreStore);
2723     __ bind(notVolatile);
2724   }
2725 
2726   // field address
2727   const Address field(obj, off);
2728 
2729   Label notByte, notBool, notInt, notShort, notChar,
2730         notLong, notFloat, notObj, notDouble;
2731 
2732   // x86 uses a shift and mask or wings it with a shift plus assert
2733   // the mask is not needed. aarch64 just uses bitfield extract
2734   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2735 
2736   assert(btos == 0, "change code, btos != 0");
2737   __ cbnz(flags, notByte);
2738 
2739   // Don't rewrite putstatic, only putfield
2740   if (is_static) rc = may_not_rewrite;
2741 
2742   // btos
2743   {
2744     __ pop(btos);
2745     if (!is_static) pop_and_check_object(obj);
2746     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg);
2747     if (rc == may_rewrite) {
2748       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2749     }
2750     __ b(Done);
2751   }
2752 
2753   __ bind(notByte);
2754   __ cmp(flags, ztos);
2755   __ br(Assembler::NE, notBool);
2756 
2757   // ztos
2758   {
2759     __ pop(ztos);
2760     if (!is_static) pop_and_check_object(obj);
2761     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg);
2762     if (rc == may_rewrite) {
2763       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2764     }
2765     __ b(Done);
2766   }
2767 
2768   __ bind(notBool);
2769   __ cmp(flags, atos);
2770   __ br(Assembler::NE, notObj);
2771 
2772   // atos
2773   {
2774     __ pop(atos);
2775     if (!is_static) pop_and_check_object(obj);
2776     // Store into the field
2777     do_oop_store(_masm, field, r0, IN_HEAP);
2778     if (rc == may_rewrite) {
2779       patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2780     }
2781     __ b(Done);
2782   }
2783 
2784   __ bind(notObj);
2785   __ cmp(flags, itos);
2786   __ br(Assembler::NE, notInt);
2787 
2788   // itos
2789   {
2790     __ pop(itos);
2791     if (!is_static) pop_and_check_object(obj);
2792     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg);
2793     if (rc == may_rewrite) {
2794       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2795     }
2796     __ b(Done);
2797   }
2798 
2799   __ bind(notInt);
2800   __ cmp(flags, ctos);
2801   __ br(Assembler::NE, notChar);
2802 
2803   // ctos
2804   {
2805     __ pop(ctos);
2806     if (!is_static) pop_and_check_object(obj);
2807     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg);
2808     if (rc == may_rewrite) {
2809       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2810     }
2811     __ b(Done);
2812   }
2813 
2814   __ bind(notChar);
2815   __ cmp(flags, stos);
2816   __ br(Assembler::NE, notShort);
2817 
2818   // stos
2819   {
2820     __ pop(stos);
2821     if (!is_static) pop_and_check_object(obj);
2822     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg);
2823     if (rc == may_rewrite) {
2824       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2825     }
2826     __ b(Done);
2827   }
2828 
2829   __ bind(notShort);
2830   __ cmp(flags, ltos);
2831   __ br(Assembler::NE, notLong);
2832 
2833   // ltos
2834   {
2835     __ pop(ltos);
2836     if (!is_static) pop_and_check_object(obj);
2837     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg);
2838     if (rc == may_rewrite) {
2839       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2840     }
2841     __ b(Done);
2842   }
2843 
2844   __ bind(notLong);
2845   __ cmp(flags, ftos);
2846   __ br(Assembler::NE, notFloat);
2847 
2848   // ftos
2849   {
2850     __ pop(ftos);
2851     if (!is_static) pop_and_check_object(obj);
2852     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
2853     if (rc == may_rewrite) {
2854       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2855     }
2856     __ b(Done);
2857   }
2858 
2859   __ bind(notFloat);
2860 #ifdef ASSERT
2861   __ cmp(flags, dtos);
2862   __ br(Assembler::NE, notDouble);
2863 #endif
2864 
2865   // dtos
2866   {
2867     __ pop(dtos);
2868     if (!is_static) pop_and_check_object(obj);
2869     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
2870     if (rc == may_rewrite) {
2871       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2872     }
2873   }
2874 
2875 #ifdef ASSERT
2876   __ b(Done);
2877 
2878   __ bind(notDouble);
2879   __ stop("Bad state");
2880 #endif
2881 
2882   __ bind(Done);
2883 
2884   {
2885     Label notVolatile;
2886     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2887     __ membar(MacroAssembler::StoreLoad);
2888     __ bind(notVolatile);
2889   }
2890 }
2891 
2892 void TemplateTable::putfield(int byte_no)
2893 {
2894   putfield_or_static(byte_no, false);
2895 }
2896 
2897 void TemplateTable::nofast_putfield(int byte_no) {
2898   putfield_or_static(byte_no, false, may_not_rewrite);
2899 }
2900 
2901 void TemplateTable::putstatic(int byte_no) {
2902   putfield_or_static(byte_no, true);
2903 }
2904 
2905 void TemplateTable::jvmti_post_fast_field_mod()
2906 {
2907   if (JvmtiExport::can_post_field_modification()) {
2908     // Check to see if a field modification watch has been set before
2909     // we take the time to call into the VM.
2910     Label L2;
2911     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2912     __ ldrw(c_rarg3, Address(rscratch1));
2913     __ cbzw(c_rarg3, L2);
2914     __ pop_ptr(r19);                  // copy the object pointer from tos
2915     __ verify_oop(r19);
2916     __ push_ptr(r19);                 // put the object pointer back on tos
2917     // Save tos values before call_VM() clobbers them. Since we have
2918     // to do it for every data type, we use the saved values as the
2919     // jvalue object.
2920     switch (bytecode()) {          // load values into the jvalue object
2921     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2922     case Bytecodes::_fast_bputfield: // fall through
2923     case Bytecodes::_fast_zputfield: // fall through
2924     case Bytecodes::_fast_sputfield: // fall through
2925     case Bytecodes::_fast_cputfield: // fall through
2926     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2927     case Bytecodes::_fast_dputfield: __ push_d(); break;
2928     case Bytecodes::_fast_fputfield: __ push_f(); break;
2929     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2930 
2931     default:
2932       ShouldNotReachHere();
2933     }
2934     __ mov(c_rarg3, esp);             // points to jvalue on the stack
2935     // access constant pool cache entry
2936     __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2937     __ verify_oop(r19);
2938     // r19: object pointer copied above
2939     // c_rarg2: cache entry pointer
2940     // c_rarg3: jvalue object on the stack
2941     __ call_VM(noreg,
2942                CAST_FROM_FN_PTR(address,
2943                                 InterpreterRuntime::post_field_modification),
2944                r19, c_rarg2, c_rarg3);
2945 
2946     switch (bytecode()) {             // restore tos values
2947     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2948     case Bytecodes::_fast_bputfield: // fall through
2949     case Bytecodes::_fast_zputfield: // fall through
2950     case Bytecodes::_fast_sputfield: // fall through
2951     case Bytecodes::_fast_cputfield: // fall through
2952     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2953     case Bytecodes::_fast_dputfield: __ pop_d(); break;
2954     case Bytecodes::_fast_fputfield: __ pop_f(); break;
2955     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2956     }
2957     __ bind(L2);
2958   }
2959 }
2960 
2961 void TemplateTable::fast_storefield(TosState state)
2962 {
2963   transition(state, vtos);
2964 
2965   ByteSize base = ConstantPoolCache::base_offset();
2966 
2967   jvmti_post_fast_field_mod();
2968 
2969   // access constant pool cache
2970   __ get_cache_and_index_at_bcp(r2, r1, 1);
2971 
2972   // test for volatile with r3
2973   __ ldrw(r3, Address(r2, in_bytes(base +
2974                                    ConstantPoolCacheEntry::flags_offset())));
2975 
2976   // replace index with field offset from cache entry
2977   __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2978 
2979   {
2980     Label notVolatile;
2981     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2982     __ membar(MacroAssembler::StoreStore);
2983     __ bind(notVolatile);
2984   }
2985 
2986   Label notVolatile;
2987 
2988   // Get object from stack
2989   pop_and_check_object(r2);
2990 
2991   // field address
2992   const Address field(r2, r1);
2993 
2994   // access field
2995   switch (bytecode()) {
2996   case Bytecodes::_fast_aputfield:
2997     do_oop_store(_masm, field, r0, IN_HEAP);
2998     break;
2999   case Bytecodes::_fast_lputfield:
3000     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg);
3001     break;
3002   case Bytecodes::_fast_iputfield:
3003     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg);
3004     break;
3005   case Bytecodes::_fast_zputfield:
3006     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg);
3007     break;
3008   case Bytecodes::_fast_bputfield:
3009     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg);
3010     break;
3011   case Bytecodes::_fast_sputfield:
3012     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg);
3013     break;
3014   case Bytecodes::_fast_cputfield:
3015     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg);
3016     break;
3017   case Bytecodes::_fast_fputfield:
3018     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3019     break;
3020   case Bytecodes::_fast_dputfield:
3021     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
3022     break;
3023   default:
3024     ShouldNotReachHere();
3025   }
3026 
3027   {
3028     Label notVolatile;
3029     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3030     __ membar(MacroAssembler::StoreLoad);
3031     __ bind(notVolatile);
3032   }
3033 }
3034 
3035 
3036 void TemplateTable::fast_accessfield(TosState state)
3037 {
3038   transition(atos, state);
3039   // Do the JVMTI work here to avoid disturbing the register state below
3040   if (JvmtiExport::can_post_field_access()) {
3041     // Check to see if a field access watch has been set before we
3042     // take the time to call into the VM.
3043     Label L1;
3044     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3045     __ ldrw(r2, Address(rscratch1));
3046     __ cbzw(r2, L1);
3047     // access constant pool cache entry
3048     __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
3049     __ verify_oop(r0);
3050     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
3051     __ mov(c_rarg1, r0);
3052     // c_rarg1: object pointer copied above
3053     // c_rarg2: cache entry pointer
3054     __ call_VM(noreg,
3055                CAST_FROM_FN_PTR(address,
3056                                 InterpreterRuntime::post_field_access),
3057                c_rarg1, c_rarg2);
3058     __ pop_ptr(r0); // restore object pointer
3059     __ bind(L1);
3060   }
3061 
3062   // access constant pool cache
3063   __ get_cache_and_index_at_bcp(r2, r1, 1);
3064   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3065                                   ConstantPoolCacheEntry::f2_offset())));
3066   __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3067                                    ConstantPoolCacheEntry::flags_offset())));
3068 
3069   // r0: object
3070   __ verify_oop(r0);
3071   __ null_check(r0);
3072   const Address field(r0, r1);
3073 
3074   // 8179954: We need to make sure that the code generated for
3075   // volatile accesses forms a sequentially-consistent set of
3076   // operations when combined with STLR and LDAR.  Without a leading
3077   // membar it's possible for a simple Dekker test to fail if loads
3078   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3079   // the stores in one method and we interpret the loads in another.
3080   if (! UseBarriersForVolatile) {
3081     Label notVolatile;
3082     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3083     __ membar(MacroAssembler::AnyAny);
3084     __ bind(notVolatile);
3085   }
3086 
3087   // access field
3088   switch (bytecode()) {
3089   case Bytecodes::_fast_agetfield:
3090     do_oop_load(_masm, field, r0, IN_HEAP);
3091     __ verify_oop(r0);
3092     break;
3093   case Bytecodes::_fast_lgetfield:
3094     __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
3095     break;
3096   case Bytecodes::_fast_igetfield:
3097     __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
3098     break;
3099   case Bytecodes::_fast_bgetfield:
3100     __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
3101     break;
3102   case Bytecodes::_fast_sgetfield:
3103     __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
3104     break;
3105   case Bytecodes::_fast_cgetfield:
3106     __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
3107     break;
3108   case Bytecodes::_fast_fgetfield:
3109     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3110     break;
3111   case Bytecodes::_fast_dgetfield:
3112     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3113     break;
3114   default:
3115     ShouldNotReachHere();
3116   }
3117   {
3118     Label notVolatile;
3119     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3120     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3121     __ bind(notVolatile);
3122   }
3123 }
3124 
3125 void TemplateTable::fast_xaccess(TosState state)
3126 {
3127   transition(vtos, state);
3128 
3129   // get receiver
3130   __ ldr(r0, aaddress(0));
3131   // access constant pool cache
3132   __ get_cache_and_index_at_bcp(r2, r3, 2);
3133   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3134                                   ConstantPoolCacheEntry::f2_offset())));
3135 
3136   // 8179954: We need to make sure that the code generated for
3137   // volatile accesses forms a sequentially-consistent set of
3138   // operations when combined with STLR and LDAR.  Without a leading
3139   // membar it's possible for a simple Dekker test to fail if loads
3140   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3141   // the stores in one method and we interpret the loads in another.
3142   if (! UseBarriersForVolatile) {
3143     Label notVolatile;
3144     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3145                                      ConstantPoolCacheEntry::flags_offset())));
3146     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3147     __ membar(MacroAssembler::AnyAny);
3148     __ bind(notVolatile);
3149   }
3150 
3151   // make sure exception is reported in correct bcp range (getfield is
3152   // next instruction)
3153   __ increment(rbcp);
3154   __ null_check(r0);
3155   switch (state) {
3156   case itos:
3157     __ access_load_at(T_INT, IN_HEAP, r0, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3158     break;
3159   case atos:
3160     do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3161     __ verify_oop(r0);
3162     break;
3163   case ftos:
3164     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3165     break;
3166   default:
3167     ShouldNotReachHere();
3168   }
3169 
3170   {
3171     Label notVolatile;
3172     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3173                                      ConstantPoolCacheEntry::flags_offset())));
3174     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3175     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3176     __ bind(notVolatile);
3177   }
3178 
3179   __ decrement(rbcp);
3180 }
3181 
3182 
3183 
3184 //-----------------------------------------------------------------------------
3185 // Calls
3186 
3187 void TemplateTable::count_calls(Register method, Register temp)
3188 {
3189   __ call_Unimplemented();
3190 }
3191 
3192 void TemplateTable::prepare_invoke(int byte_no,
3193                                    Register method, // linked method (or i-klass)
3194                                    Register index,  // itable index, MethodType, etc.
3195                                    Register recv,   // if caller wants to see it
3196                                    Register flags   // if caller wants to test it
3197                                    ) {
3198   // determine flags
3199   Bytecodes::Code code = bytecode();
3200   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3201   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3202   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3203   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3204   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3205   const bool load_receiver       = (recv  != noreg);
3206   const bool save_flags          = (flags != noreg);
3207   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3208   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3209   assert(flags == noreg || flags == r3, "");
3210   assert(recv  == noreg || recv  == r2, "");
3211 
3212   // setup registers & access constant pool cache
3213   if (recv  == noreg)  recv  = r2;
3214   if (flags == noreg)  flags = r3;
3215   assert_different_registers(method, index, recv, flags);
3216 
3217   // save 'interpreter return address'
3218   __ save_bcp();
3219 
3220   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3221 
3222   // maybe push appendix to arguments (just before return address)
3223   if (is_invokedynamic || is_invokehandle) {
3224     Label L_no_push;
3225     __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3226     // Push the appendix as a trailing parameter.
3227     // This must be done before we get the receiver,
3228     // since the parameter_size includes it.
3229     __ push(r19);
3230     __ mov(r19, index);
3231     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3232     __ load_resolved_reference_at_index(index, r19);
3233     __ pop(r19);
3234     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3235     __ bind(L_no_push);
3236   }
3237 
3238   // load receiver if needed (note: no return address pushed yet)
3239   if (load_receiver) {
3240     __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3241     // FIXME -- is this actually correct? looks like it should be 2
3242     // const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3243     // const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3244     // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3245     // __ movptr(recv, recv_addr);
3246     __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3247     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3248     __ verify_oop(recv);
3249   }
3250 
3251   // compute return type
3252   // x86 uses a shift and mask or wings it with a shift plus assert
3253   // the mask is not needed. aarch64 just uses bitfield extract
3254   __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
3255   // load return address
3256   {
3257     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3258     __ mov(rscratch1, table_addr);
3259     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3260   }
3261 }
3262 
3263 
3264 void TemplateTable::invokevirtual_helper(Register index,
3265                                          Register recv,
3266                                          Register flags)
3267 {
3268   // Uses temporary registers r0, r3
3269   assert_different_registers(index, recv, r0, r3);
3270   // Test for an invoke of a final method
3271   Label notFinal;
3272   __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3273 
3274   const Register method = index;  // method must be rmethod
3275   assert(method == rmethod,
3276          "methodOop must be rmethod for interpreter calling convention");
3277 
3278   // do the call - the index is actually the method to call
3279   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3280 
3281   // It's final, need a null check here!
3282   __ null_check(recv);
3283 
3284   // profile this call
3285   __ profile_final_call(r0);
3286   __ profile_arguments_type(r0, method, r4, true);
3287 
3288   __ jump_from_interpreted(method, r0);
3289 
3290   __ bind(notFinal);
3291 
3292   // get receiver klass
3293   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3294   __ load_klass(r0, recv);
3295 
3296   // profile this call
3297   __ profile_virtual_call(r0, rlocals, r3);
3298 
3299   // get target methodOop & entry point
3300   __ lookup_virtual_method(r0, index, method);
3301   __ profile_arguments_type(r3, method, r4, true);
3302   // FIXME -- this looks completely redundant. is it?
3303   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3304   __ jump_from_interpreted(method, r3);
3305 }
3306 
3307 void TemplateTable::invokevirtual(int byte_no)
3308 {
3309   transition(vtos, vtos);
3310   assert(byte_no == f2_byte, "use this argument");
3311 
3312   prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3313 
3314   // rmethod: index (actually a Method*)
3315   // r2: receiver
3316   // r3: flags
3317 
3318   invokevirtual_helper(rmethod, r2, r3);
3319 }
3320 
3321 void TemplateTable::invokespecial(int byte_no)
3322 {
3323   transition(vtos, vtos);
3324   assert(byte_no == f1_byte, "use this argument");
3325 
3326   prepare_invoke(byte_no, rmethod, noreg,  // get f1 Method*
3327                  r2);  // get receiver also for null check
3328   __ verify_oop(r2);
3329   __ null_check(r2);
3330   // do the call
3331   __ profile_call(r0);
3332   __ profile_arguments_type(r0, rmethod, rbcp, false);
3333   __ jump_from_interpreted(rmethod, r0);
3334 }
3335 
3336 void TemplateTable::invokestatic(int byte_no)
3337 {
3338   transition(vtos, vtos);
3339   assert(byte_no == f1_byte, "use this argument");
3340 
3341   prepare_invoke(byte_no, rmethod);  // get f1 Method*
3342   // do the call
3343   __ profile_call(r0);
3344   __ profile_arguments_type(r0, rmethod, r4, false);
3345   __ jump_from_interpreted(rmethod, r0);
3346 }
3347 
3348 void TemplateTable::fast_invokevfinal(int byte_no)
3349 {
3350   __ call_Unimplemented();
3351 }
3352 
3353 void TemplateTable::invokeinterface(int byte_no) {
3354   transition(vtos, vtos);
3355   assert(byte_no == f1_byte, "use this argument");
3356 
3357   prepare_invoke(byte_no, r0, rmethod,  // get f1 Klass*, f2 Method*
3358                  r2, r3); // recv, flags
3359 
3360   // r0: interface klass (from f1)
3361   // rmethod: method (from f2)
3362   // r2: receiver
3363   // r3: flags
3364 
3365   // First check for Object case, then private interface method,
3366   // then regular interface method.
3367 
3368   // Special case of invokeinterface called for virtual method of
3369   // java.lang.Object.  See cpCache.cpp for details.
3370   Label notObjectMethod;
3371   __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notObjectMethod);
3372 
3373   invokevirtual_helper(rmethod, r2, r3);
3374   __ bind(notObjectMethod);
3375 
3376   Label no_such_interface;
3377 
3378   // Check for private method invocation - indicated by vfinal
3379   Label notVFinal;
3380   __ tbz(r3, ConstantPoolCacheEntry::is_vfinal_shift, notVFinal);
3381 
3382   // Get receiver klass into r3 - also a null check
3383   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3384   __ load_klass(r3, r2);
3385 
3386   Label subtype;
3387   __ check_klass_subtype(r3, r0, r4, subtype);
3388   // If we get here the typecheck failed
3389   __ b(no_such_interface);
3390   __ bind(subtype);
3391 
3392   __ profile_final_call(r0);
3393   __ profile_arguments_type(r0, rmethod, r4, true);
3394   __ jump_from_interpreted(rmethod, r0);
3395 
3396   __ bind(notVFinal);
3397 
3398   // Get receiver klass into r3 - also a null check
3399   __ restore_locals();
3400   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3401   __ load_klass(r3, r2);
3402 
3403   Label no_such_method;
3404 
3405   // Preserve method for throw_AbstractMethodErrorVerbose.
3406   __ mov(r16, rmethod);
3407   // Receiver subtype check against REFC.
3408   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3409   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3410                              r3, r0, noreg,
3411                              // outputs: scan temp. reg, scan temp. reg
3412                              rscratch2, r13,
3413                              no_such_interface,
3414                              /*return_method=*/false);
3415 
3416   // profile this call
3417   __ profile_virtual_call(r3, r13, r19);
3418 
3419   // Get declaring interface class from method, and itable index
3420   __ ldr(r0, Address(rmethod, Method::const_offset()));
3421   __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
3422   __ ldr(r0, Address(r0, ConstantPool::pool_holder_offset_in_bytes()));
3423   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3424   __ subw(rmethod, rmethod, Method::itable_index_max);
3425   __ negw(rmethod, rmethod);
3426 
3427   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3428   __ mov(rlocals, r3);
3429   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3430                              rlocals, r0, rmethod,
3431                              // outputs: method, scan temp. reg
3432                              rmethod, r13,
3433                              no_such_interface);
3434 
3435   // rmethod,: methodOop to call
3436   // r2: receiver
3437   // Check for abstract method error
3438   // Note: This should be done more efficiently via a throw_abstract_method_error
3439   //       interpreter entry point and a conditional jump to it in case of a null
3440   //       method.
3441   __ cbz(rmethod, no_such_method);
3442 
3443   __ profile_arguments_type(r3, rmethod, r13, true);
3444 
3445   // do the call
3446   // r2: receiver
3447   // rmethod,: methodOop
3448   __ jump_from_interpreted(rmethod, r3);
3449   __ should_not_reach_here();
3450 
3451   // exception handling code follows...
3452   // note: must restore interpreter registers to canonical
3453   //       state for exception handling to work correctly!
3454 
3455   __ bind(no_such_method);
3456   // throw exception
3457   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3458   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3459   // Pass arguments for generating a verbose error message.
3460   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3461   // the call_VM checks for exception, so we should never return here.
3462   __ should_not_reach_here();
3463 
3464   __ bind(no_such_interface);
3465   // throw exception
3466   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3467   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3468   // Pass arguments for generating a verbose error message.
3469   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3470                    InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3471   // the call_VM checks for exception, so we should never return here.
3472   __ should_not_reach_here();
3473   return;
3474 }
3475 
3476 void TemplateTable::invokehandle(int byte_no) {
3477   transition(vtos, vtos);
3478   assert(byte_no == f1_byte, "use this argument");
3479 
3480   prepare_invoke(byte_no, rmethod, r0, r2);
3481   __ verify_method_ptr(r2);
3482   __ verify_oop(r2);
3483   __ null_check(r2);
3484 
3485   // FIXME: profile the LambdaForm also
3486 
3487   // r13 is safe to use here as a scratch reg because it is about to
3488   // be clobbered by jump_from_interpreted().
3489   __ profile_final_call(r13);
3490   __ profile_arguments_type(r13, rmethod, r4, true);
3491 
3492   __ jump_from_interpreted(rmethod, r0);
3493 }
3494 
3495 void TemplateTable::invokedynamic(int byte_no) {
3496   transition(vtos, vtos);
3497   assert(byte_no == f1_byte, "use this argument");
3498 
3499   prepare_invoke(byte_no, rmethod, r0);
3500 
3501   // r0: CallSite object (from cpool->resolved_references[])
3502   // rmethod: MH.linkToCallSite method (from f2)
3503 
3504   // Note:  r0_callsite is already pushed by prepare_invoke
3505 
3506   // %%% should make a type profile for any invokedynamic that takes a ref argument
3507   // profile this call
3508   __ profile_call(rbcp);
3509   __ profile_arguments_type(r3, rmethod, r13, false);
3510 
3511   __ verify_oop(r0);
3512 
3513   __ jump_from_interpreted(rmethod, r0);
3514 }
3515 
3516 
3517 //-----------------------------------------------------------------------------
3518 // Allocation
3519 
3520 void TemplateTable::_new() {
3521   transition(vtos, atos);
3522 
3523   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3524   Label slow_case;
3525   Label done;
3526   Label initialize_header;
3527   Label initialize_object; // including clearing the fields
3528 
3529   __ get_cpool_and_tags(r4, r0);
3530   // Make sure the class we're about to instantiate has been resolved.
3531   // This is done before loading InstanceKlass to be consistent with the order
3532   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3533   const int tags_offset = Array<u1>::base_offset_in_bytes();
3534   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3535   __ lea(rscratch1, Address(rscratch1, tags_offset));
3536   __ ldarb(rscratch1, rscratch1);
3537   __ cmp(rscratch1, JVM_CONSTANT_Class);
3538   __ br(Assembler::NE, slow_case);
3539 
3540   // get InstanceKlass
3541   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3542 
3543   // make sure klass is initialized & doesn't have finalizer
3544   // make sure klass is fully initialized
3545   __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3546   __ cmp(rscratch1, InstanceKlass::fully_initialized);
3547   __ br(Assembler::NE, slow_case);
3548 
3549   // get instance_size in InstanceKlass (scaled to a count of bytes)
3550   __ ldrw(r3,
3551           Address(r4,
3552                   Klass::layout_helper_offset()));
3553   // test to see if it has a finalizer or is malformed in some way
3554   __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3555 
3556   // Allocate the instance:
3557   //  If TLAB is enabled:
3558   //    Try to allocate in the TLAB.
3559   //    If fails, go to the slow path.
3560   //  Else If inline contiguous allocations are enabled:
3561   //    Try to allocate in eden.
3562   //    If fails due to heap end, go to slow path.
3563   //
3564   //  If TLAB is enabled OR inline contiguous is enabled:
3565   //    Initialize the allocation.
3566   //    Exit.
3567   //
3568   //  Go to slow path.
3569   const bool allow_shared_alloc =
3570     Universe::heap()->supports_inline_contig_alloc();
3571 
3572   if (UseTLAB) {
3573     __ tlab_allocate(r0, r3, 0, noreg, r1, slow_case);
3574 
3575     if (ZeroTLAB) {
3576       // the fields have been already cleared
3577       __ b(initialize_header);
3578     } else {
3579       // initialize both the header and fields
3580       __ b(initialize_object);
3581     }
3582   } else {
3583     // Allocation in the shared Eden, if allowed.
3584     //
3585     // r3: instance size in bytes
3586     if (allow_shared_alloc) {
3587       __ eden_allocate(r0, r3, 0, r10, slow_case);
3588     }
3589   }
3590 
3591   // If UseTLAB or allow_shared_alloc are true, the object is created above and
3592   // there is an initialize need. Otherwise, skip and go to the slow path.
3593   if (UseTLAB || allow_shared_alloc) {
3594     // The object is initialized before the header.  If the object size is
3595     // zero, go directly to the header initialization.
3596     __ bind(initialize_object);
3597     __ sub(r3, r3, sizeof(oopDesc));
3598     __ cbz(r3, initialize_header);
3599 
3600     // Initialize object fields
3601     {
3602       __ add(r2, r0, sizeof(oopDesc));
3603       Label loop;
3604       __ bind(loop);
3605       __ str(zr, Address(__ post(r2, BytesPerLong)));
3606       __ sub(r3, r3, BytesPerLong);
3607       __ cbnz(r3, loop);
3608     }
3609 
3610     // initialize object header only.
3611     __ bind(initialize_header);
3612     if (UseBiasedLocking) {
3613       __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3614     } else {
3615       __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3616     }
3617     __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3618     __ store_klass_gap(r0, zr);  // zero klass gap for compressed oops
3619     __ store_klass(r0, r4);      // store klass last
3620 
3621     {
3622       SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3623       // Trigger dtrace event for fastpath
3624       __ push(atos); // save the return value
3625       __ call_VM_leaf(
3626            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3627       __ pop(atos); // restore the return value
3628 
3629     }
3630     __ b(done);
3631   }
3632 
3633   // slow case
3634   __ bind(slow_case);
3635   __ get_constant_pool(c_rarg1);
3636   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3637   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3638   __ verify_oop(r0);
3639 
3640   // continue
3641   __ bind(done);
3642   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3643   __ membar(Assembler::StoreStore);
3644 }
3645 
3646 void TemplateTable::newarray() {
3647   transition(itos, atos);
3648   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3649   __ mov(c_rarg2, r0);
3650   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3651           c_rarg1, c_rarg2);
3652   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3653   __ membar(Assembler::StoreStore);
3654 }
3655 
3656 void TemplateTable::anewarray() {
3657   transition(itos, atos);
3658   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3659   __ get_constant_pool(c_rarg1);
3660   __ mov(c_rarg3, r0);
3661   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3662           c_rarg1, c_rarg2, c_rarg3);
3663   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3664   __ membar(Assembler::StoreStore);
3665 }
3666 
3667 void TemplateTable::arraylength() {
3668   transition(atos, itos);
3669   __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3670   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3671 }
3672 
3673 void TemplateTable::checkcast()
3674 {
3675   transition(atos, atos);
3676   Label done, is_null, ok_is_subtype, quicked, resolved;
3677   __ cbz(r0, is_null);
3678 
3679   // Get cpool & tags index
3680   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3681   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3682   // See if bytecode has already been quicked
3683   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3684   __ lea(r1, Address(rscratch1, r19));
3685   __ ldarb(r1, r1);
3686   __ cmp(r1, JVM_CONSTANT_Class);
3687   __ br(Assembler::EQ, quicked);
3688 
3689   __ push(atos); // save receiver for result, and for GC
3690   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3691   // vm_result_2 has metadata result
3692   __ get_vm_result_2(r0, rthread);
3693   __ pop(r3); // restore receiver
3694   __ b(resolved);
3695 
3696   // Get superklass in r0 and subklass in r3
3697   __ bind(quicked);
3698   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3699   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3700 
3701   __ bind(resolved);
3702   __ load_klass(r19, r3);
3703 
3704   // Generate subtype check.  Blows r2, r5.  Object in r3.
3705   // Superklass in r0.  Subklass in r19.
3706   __ gen_subtype_check(r19, ok_is_subtype);
3707 
3708   // Come here on failure
3709   __ push(r3);
3710   // object is at TOS
3711   __ b(Interpreter::_throw_ClassCastException_entry);
3712 
3713   // Come here on success
3714   __ bind(ok_is_subtype);
3715   __ mov(r0, r3); // Restore object in r3
3716 
3717   // Collect counts on whether this test sees NULLs a lot or not.
3718   if (ProfileInterpreter) {
3719     __ b(done);
3720     __ bind(is_null);
3721     __ profile_null_seen(r2);
3722   } else {
3723     __ bind(is_null);   // same as 'done'
3724   }
3725   __ bind(done);
3726 }
3727 
3728 void TemplateTable::instanceof() {
3729   transition(atos, itos);
3730   Label done, is_null, ok_is_subtype, quicked, resolved;
3731   __ cbz(r0, is_null);
3732 
3733   // Get cpool & tags index
3734   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3735   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3736   // See if bytecode has already been quicked
3737   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3738   __ lea(r1, Address(rscratch1, r19));
3739   __ ldarb(r1, r1);
3740   __ cmp(r1, JVM_CONSTANT_Class);
3741   __ br(Assembler::EQ, quicked);
3742 
3743   __ push(atos); // save receiver for result, and for GC
3744   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3745   // vm_result_2 has metadata result
3746   __ get_vm_result_2(r0, rthread);
3747   __ pop(r3); // restore receiver
3748   __ verify_oop(r3);
3749   __ load_klass(r3, r3);
3750   __ b(resolved);
3751 
3752   // Get superklass in r0 and subklass in r3
3753   __ bind(quicked);
3754   __ load_klass(r3, r0);
3755   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3756 
3757   __ bind(resolved);
3758 
3759   // Generate subtype check.  Blows r2, r5
3760   // Superklass in r0.  Subklass in r3.
3761   __ gen_subtype_check(r3, ok_is_subtype);
3762 
3763   // Come here on failure
3764   __ mov(r0, 0);
3765   __ b(done);
3766   // Come here on success
3767   __ bind(ok_is_subtype);
3768   __ mov(r0, 1);
3769 
3770   // Collect counts on whether this test sees NULLs a lot or not.
3771   if (ProfileInterpreter) {
3772     __ b(done);
3773     __ bind(is_null);
3774     __ profile_null_seen(r2);
3775   } else {
3776     __ bind(is_null);   // same as 'done'
3777   }
3778   __ bind(done);
3779   // r0 = 0: obj == NULL or  obj is not an instanceof the specified klass
3780   // r0 = 1: obj != NULL and obj is     an instanceof the specified klass
3781 }
3782 
3783 //-----------------------------------------------------------------------------
3784 // Breakpoints
3785 void TemplateTable::_breakpoint() {
3786   // Note: We get here even if we are single stepping..
3787   // jbug inists on setting breakpoints at every bytecode
3788   // even if we are in single step mode.
3789 
3790   transition(vtos, vtos);
3791 
3792   // get the unpatched byte code
3793   __ get_method(c_rarg1);
3794   __ call_VM(noreg,
3795              CAST_FROM_FN_PTR(address,
3796                               InterpreterRuntime::get_original_bytecode_at),
3797              c_rarg1, rbcp);
3798   __ mov(r19, r0);
3799 
3800   // post the breakpoint event
3801   __ call_VM(noreg,
3802              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3803              rmethod, rbcp);
3804 
3805   // complete the execution of original bytecode
3806   __ mov(rscratch1, r19);
3807   __ dispatch_only_normal(vtos);
3808 }
3809 
3810 //-----------------------------------------------------------------------------
3811 // Exceptions
3812 
3813 void TemplateTable::athrow() {
3814   transition(atos, vtos);
3815   __ null_check(r0);
3816   __ b(Interpreter::throw_exception_entry());
3817 }
3818 
3819 //-----------------------------------------------------------------------------
3820 // Synchronization
3821 //
3822 // Note: monitorenter & exit are symmetric routines; which is reflected
3823 //       in the assembly code structure as well
3824 //
3825 // Stack layout:
3826 //
3827 // [expressions  ] <--- esp               = expression stack top
3828 // ..
3829 // [expressions  ]
3830 // [monitor entry] <--- monitor block top = expression stack bot
3831 // ..
3832 // [monitor entry]
3833 // [frame data   ] <--- monitor block bot
3834 // ...
3835 // [saved rbp    ] <--- rbp
3836 void TemplateTable::monitorenter()
3837 {
3838   transition(atos, vtos);
3839 
3840   // check for NULL object
3841   __ null_check(r0);
3842 
3843   // We need to preemptively evacuate the object, because we later compare
3844   // it to objects in the BasicObjectLock list, and we might get false negatives
3845   // if another thread evacuates the object in the meantime. See acmp.
3846   __ resolve_for_write(0, r0);
3847 
3848   const Address monitor_block_top(
3849         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3850   const Address monitor_block_bot(
3851         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3852   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3853 
3854   Label allocated;
3855 
3856   // initialize entry pointer
3857   __ mov(c_rarg1, zr); // points to free slot or NULL
3858 
3859   // find a free slot in the monitor block (result in c_rarg1)
3860   {
3861     Label entry, loop, exit;
3862     __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3863                                         // starting with top-most entry
3864     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3865 
3866     __ b(entry);
3867 
3868     __ bind(loop);
3869     // check if current entry is used
3870     // if not used then remember entry in c_rarg1
3871     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3872     __ cmp(zr, rscratch1);
3873     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3874     // check if current entry is for same object
3875     __ cmp(r0, rscratch1);
3876     // if same object then stop searching
3877     __ br(Assembler::EQ, exit);
3878     // otherwise advance to next entry
3879     __ add(c_rarg3, c_rarg3, entry_size);
3880     __ bind(entry);
3881     // check if bottom reached
3882     __ cmp(c_rarg3, c_rarg2);
3883     // if not at bottom then check this entry
3884     __ br(Assembler::NE, loop);
3885     __ bind(exit);
3886   }
3887 
3888   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3889                             // if found, continue with that on
3890 
3891   // allocate one if there's no free slot
3892   {
3893     Label entry, loop;
3894     // 1. compute new pointers            // rsp: old expression stack top
3895     __ ldr(c_rarg1, monitor_block_bot);   // c_rarg1: old expression stack bottom
3896     __ sub(esp, esp, entry_size);         // move expression stack top
3897     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3898     __ mov(c_rarg3, esp);                 // set start value for copy loop
3899     __ str(c_rarg1, monitor_block_bot);   // set new monitor block bottom
3900 
3901     __ sub(sp, sp, entry_size);           // make room for the monitor
3902 
3903     __ b(entry);
3904     // 2. move expression stack contents
3905     __ bind(loop);
3906     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3907                                                    // word from old location
3908     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
3909     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
3910     __ bind(entry);
3911     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
3912     __ br(Assembler::NE, loop);      // if not at bottom then
3913                                      // copy next word
3914   }
3915 
3916   // call run-time routine
3917   // c_rarg1: points to monitor entry
3918   __ bind(allocated);
3919 
3920   // Increment bcp to point to the next bytecode, so exception
3921   // handling for async. exceptions work correctly.
3922   // The object has already been poped from the stack, so the
3923   // expression stack looks correct.
3924   __ increment(rbcp);
3925 
3926   // store object
3927   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3928   __ lock_object(c_rarg1);
3929 
3930   // check to make sure this monitor doesn't cause stack overflow after locking
3931   __ save_bcp();  // in case of exception
3932   __ generate_stack_overflow_check(0);
3933 
3934   // The bcp has already been incremented. Just need to dispatch to
3935   // next instruction.
3936   __ dispatch_next(vtos);
3937 }
3938 
3939 
3940 void TemplateTable::monitorexit()
3941 {
3942   transition(atos, vtos);
3943 
3944   // check for NULL object
3945   __ null_check(r0);
3946 
3947   // We need to preemptively evacuate the object, because we later compare
3948   // it to objects in the BasicObjectLock list, and we might get false negatives
3949   // if another thread evacuates the object in the meantime. See acmp.
3950   __ resolve_for_write(0, r0);
3951 
3952   const Address monitor_block_top(
3953         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3954   const Address monitor_block_bot(
3955         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3956   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3957 
3958   Label found;
3959 
3960   // find matching slot
3961   {
3962     Label entry, loop;
3963     __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3964                                         // starting with top-most entry
3965     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3966                                         // of monitor block
3967     __ b(entry);
3968 
3969     __ bind(loop);
3970     // check if current entry is for same object
3971     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3972     __ cmp(r0, rscratch1);
3973     // if same object then stop searching
3974     __ br(Assembler::EQ, found);
3975     // otherwise advance to next entry
3976     __ add(c_rarg1, c_rarg1, entry_size);
3977     __ bind(entry);
3978     // check if bottom reached
3979     __ cmp(c_rarg1, c_rarg2);
3980     // if not at bottom then check this entry
3981     __ br(Assembler::NE, loop);
3982   }
3983 
3984   // error handling. Unlocking was not block-structured
3985   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3986                    InterpreterRuntime::throw_illegal_monitor_state_exception));
3987   __ should_not_reach_here();
3988 
3989   // call run-time routine
3990   __ bind(found);
3991   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3992   __ unlock_object(c_rarg1);
3993   __ pop_ptr(r0); // discard object
3994 }
3995 
3996 
3997 // Wide instructions
3998 void TemplateTable::wide()
3999 {
4000   __ load_unsigned_byte(r19, at_bcp(1));
4001   __ mov(rscratch1, (address)Interpreter::_wentry_point);
4002   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
4003   __ br(rscratch1);
4004 }
4005 
4006 
4007 // Multi arrays
4008 void TemplateTable::multianewarray() {
4009   transition(vtos, atos);
4010   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
4011   // last dim is on top of stack; we want address of first one:
4012   // first_addr = last_addr + (ndims - 1) * wordSize
4013   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
4014   __ sub(c_rarg1, c_rarg1, wordSize);
4015   call_VM(r0,
4016           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
4017           c_rarg1);
4018   __ load_unsigned_byte(r1, at_bcp(3));
4019   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
4020 }