8248238: Adding Windows support to OpenJDK on AArch64

Summary: LP64 vs LLP64 changes to add Windows support

Contributed-by: Monica Beckwith <monica.beckwith@microsoft.com>, Ludovic Henry <luhenry@microsoft.com>
Reviewed-by:

   1 /*
   2  * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "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 #include "utilities/powerOfTwo.hpp"
  44 
  45 #define __ _masm->
  46 
  47 // Platform-dependent initialization
  48 
  49 void TemplateTable::pd_initialize() {
  50   // No aarch64 specific initialization
  51 }
  52 
  53 // Address computation: local variables
  54 
  55 static inline Address iaddress(int n) {
  56   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  57 }
  58 
  59 static inline Address laddress(int n) {
  60   return iaddress(n + 1);
  61 }
  62 
  63 static inline Address faddress(int n) {
  64   return iaddress(n);
  65 }
  66 
  67 static inline Address daddress(int n) {
  68   return laddress(n);
  69 }
  70 
  71 static inline Address aaddress(int n) {
  72   return iaddress(n);
  73 }
  74 
  75 static inline Address iaddress(Register r) {
  76   return Address(rlocals, r, Address::lsl(3));
  77 }
  78 
  79 static inline Address laddress(Register r, Register scratch,
  80                                InterpreterMacroAssembler* _masm) {
  81   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  82   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  83 }
  84 
  85 static inline Address faddress(Register r) {
  86   return iaddress(r);
  87 }
  88 
  89 static inline Address daddress(Register r, Register scratch,
  90                                InterpreterMacroAssembler* _masm) {
  91   return laddress(r, scratch, _masm);
  92 }
  93 
  94 static inline Address aaddress(Register r) {
  95   return iaddress(r);
  96 }
  97 
  98 static inline Address at_rsp() {
  99   return Address(esp, 0);
 100 }
 101 
 102 // At top of Java expression stack which may be different than esp().  It
 103 // isn't for category 1 objects.
 104 static inline Address at_tos   () {
 105   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 106 }
 107 
 108 static inline Address at_tos_p1() {
 109   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 110 }
 111 
 112 static inline Address at_tos_p2() {
 113   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 114 }
 115 
 116 static inline Address at_tos_p3() {
 117   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 118 }
 119 
 120 static inline Address at_tos_p4() {
 121   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 122 }
 123 
 124 static inline Address at_tos_p5() {
 125   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 126 }
 127 
 128 // Condition conversion
 129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 130   switch (cc) {
 131   case TemplateTable::equal        : return Assembler::NE;
 132   case TemplateTable::not_equal    : return Assembler::EQ;
 133   case TemplateTable::less         : return Assembler::GE;
 134   case TemplateTable::less_equal   : return Assembler::GT;
 135   case TemplateTable::greater      : return Assembler::LE;
 136   case TemplateTable::greater_equal: return Assembler::LT;
 137   }
 138   ShouldNotReachHere();
 139   return Assembler::EQ;
 140 }
 141 
 142 
 143 // Miscelaneous helper routines
 144 // Store an oop (or NULL) at the Address described by obj.
 145 // If val == noreg this means store a NULL
 146 static void do_oop_store(InterpreterMacroAssembler* _masm,
 147                          Address dst,
 148                          Register val,
 149                          DecoratorSet decorators) {
 150   assert(val == noreg || val == r0, "parameter is just for looks");
 151   __ store_heap_oop(dst, val, r10, r1, decorators);
 152 }
 153 
 154 static void do_oop_load(InterpreterMacroAssembler* _masm,
 155                         Address src,
 156                         Register dst,
 157                         DecoratorSet decorators) {
 158   __ load_heap_oop(dst, src, r10, r1, decorators);
 159 }
 160 
 161 Address TemplateTable::at_bcp(int offset) {
 162   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 163   return Address(rbcp, offset);
 164 }
 165 
 166 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 167                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 168                                    int byte_no)
 169 {
 170   if (!RewriteBytecodes)  return;
 171   Label L_patch_done;
 172 
 173   switch (bc) {
 174   case Bytecodes::_fast_aputfield:
 175   case Bytecodes::_fast_bputfield:
 176   case Bytecodes::_fast_zputfield:
 177   case Bytecodes::_fast_cputfield:
 178   case Bytecodes::_fast_dputfield:
 179   case Bytecodes::_fast_fputfield:
 180   case Bytecodes::_fast_iputfield:
 181   case Bytecodes::_fast_lputfield:
 182   case Bytecodes::_fast_sputfield:
 183     {
 184       // We skip bytecode quickening for putfield instructions when
 185       // the put_code written to the constant pool cache is zero.
 186       // This is required so that every execution of this instruction
 187       // calls out to InterpreterRuntime::resolve_get_put to do
 188       // additional, required work.
 189       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 190       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 191       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 192       __ movw(bc_reg, bc);
 193       __ cbzw(temp_reg, L_patch_done);  // don't patch
 194     }
 195     break;
 196   default:
 197     assert(byte_no == -1, "sanity");
 198     // the pair bytecodes have already done the load.
 199     if (load_bc_into_bc_reg) {
 200       __ movw(bc_reg, bc);
 201     }
 202   }
 203 
 204   if (JvmtiExport::can_post_breakpoint()) {
 205     Label L_fast_patch;
 206     // if a breakpoint is present we can't rewrite the stream directly
 207     __ load_unsigned_byte(temp_reg, at_bcp(0));
 208     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 209     __ br(Assembler::NE, L_fast_patch);
 210     // Let breakpoint table handling rewrite to quicker bytecode
 211     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 212     __ b(L_patch_done);
 213     __ bind(L_fast_patch);
 214   }
 215 
 216 #ifdef ASSERT
 217   Label L_okay;
 218   __ load_unsigned_byte(temp_reg, at_bcp(0));
 219   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 220   __ br(Assembler::EQ, L_okay);
 221   __ cmpw(temp_reg, bc_reg);
 222   __ br(Assembler::EQ, L_okay);
 223   __ stop("patching the wrong bytecode");
 224   __ bind(L_okay);
 225 #endif
 226 
 227   // patch bytecode
 228   __ strb(bc_reg, at_bcp(0));
 229   __ bind(L_patch_done);
 230 }
 231 
 232 
 233 // Individual instructions
 234 
 235 void TemplateTable::nop() {
 236   transition(vtos, vtos);
 237   // nothing to do
 238 }
 239 
 240 void TemplateTable::shouldnotreachhere() {
 241   transition(vtos, vtos);
 242   __ stop("shouldnotreachhere bytecode");
 243 }
 244 
 245 void TemplateTable::aconst_null()
 246 {
 247   transition(vtos, atos);
 248   __ mov(r0, 0);
 249 }
 250 
 251 void TemplateTable::iconst(int value)
 252 {
 253   transition(vtos, itos);
 254   __ mov(r0, value);
 255 }
 256 
 257 void TemplateTable::lconst(int value)
 258 {
 259   __ mov(r0, value);
 260 }
 261 
 262 void TemplateTable::fconst(int value)
 263 {
 264   transition(vtos, ftos);
 265   switch (value) {
 266   case 0:
 267     __ fmovs(v0, zr);
 268     break;
 269   case 1:
 270     __ fmovs(v0, 1.0);
 271     break;
 272   case 2:
 273     __ fmovs(v0, 2.0);
 274     break;
 275   default:
 276     ShouldNotReachHere();
 277     break;
 278   }
 279 }
 280 
 281 void TemplateTable::dconst(int value)
 282 {
 283   transition(vtos, dtos);
 284   switch (value) {
 285   case 0:
 286     __ fmovd(v0, zr);
 287     break;
 288   case 1:
 289     __ fmovd(v0, 1.0);
 290     break;
 291   case 2:
 292     __ fmovd(v0, 2.0);
 293     break;
 294   default:
 295     ShouldNotReachHere();
 296     break;
 297   }
 298 }
 299 
 300 void TemplateTable::bipush()
 301 {
 302   transition(vtos, itos);
 303   __ load_signed_byte32(r0, at_bcp(1));
 304 }
 305 
 306 void TemplateTable::sipush()
 307 {
 308   transition(vtos, itos);
 309   __ load_unsigned_short(r0, at_bcp(1));
 310   __ revw(r0, r0);
 311   __ asrw(r0, r0, 16);
 312 }
 313 
 314 void TemplateTable::ldc(bool wide)
 315 {
 316   transition(vtos, vtos);
 317   Label call_ldc, notFloat, notClass, notInt, Done;
 318 
 319   if (wide) {
 320     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 321   } else {
 322     __ load_unsigned_byte(r1, at_bcp(1));
 323   }
 324   __ get_cpool_and_tags(r2, r0);
 325 
 326   const int base_offset = ConstantPool::header_size() * wordSize;
 327   const int tags_offset = Array<u1>::base_offset_in_bytes();
 328 
 329   // get type
 330   __ add(r3, r1, tags_offset);
 331   __ lea(r3, Address(r0, r3));
 332   __ ldarb(r3, r3);
 333 
 334   // unresolved class - get the resolved class
 335   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClass);
 336   __ br(Assembler::EQ, call_ldc);
 337 
 338   // unresolved class in error state - call into runtime to throw the error
 339   // from the first resolution attempt
 340   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClassInError);
 341   __ br(Assembler::EQ, call_ldc);
 342 
 343   // resolved class - need to call vm to get java mirror of the class
 344   __ cmp(r3, (u1)JVM_CONSTANT_Class);
 345   __ br(Assembler::NE, notClass);
 346 
 347   __ bind(call_ldc);
 348   __ mov(c_rarg1, wide);
 349   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 350   __ push_ptr(r0);
 351   __ verify_oop(r0);
 352   __ b(Done);
 353 
 354   __ bind(notClass);
 355   __ cmp(r3, (u1)JVM_CONSTANT_Float);
 356   __ br(Assembler::NE, notFloat);
 357   // ftos
 358   __ adds(r1, r2, r1, Assembler::LSL, 3);
 359   __ ldrs(v0, Address(r1, base_offset));
 360   __ push_f();
 361   __ b(Done);
 362 
 363   __ bind(notFloat);
 364 
 365   __ cmp(r3, (u1)JVM_CONSTANT_Integer);
 366   __ br(Assembler::NE, notInt);
 367 
 368   // itos
 369   __ adds(r1, r2, r1, Assembler::LSL, 3);
 370   __ ldrw(r0, Address(r1, base_offset));
 371   __ push_i(r0);
 372   __ b(Done);
 373 
 374   __ bind(notInt);
 375   condy_helper(Done);
 376 
 377   __ bind(Done);
 378 }
 379 
 380 // Fast path for caching oop constants.
 381 void TemplateTable::fast_aldc(bool wide)
 382 {
 383   transition(vtos, atos);
 384 
 385   Register result = r0;
 386   Register tmp = r1;
 387   Register rarg = r2;
 388 
 389   int index_size = wide ? sizeof(u2) : sizeof(u1);
 390 
 391   Label resolved;
 392 
 393   // We are resolved if the resolved reference cache entry contains a
 394   // non-null object (String, MethodType, etc.)
 395   assert_different_registers(result, tmp);
 396   __ get_cache_index_at_bcp(tmp, 1, index_size);
 397   __ load_resolved_reference_at_index(result, tmp);
 398   __ cbnz(result, resolved);
 399 
 400   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 401 
 402   // first time invocation - must resolve first
 403   __ mov(rarg, (int)bytecode());
 404   __ call_VM(result, entry, rarg);
 405 
 406   __ bind(resolved);
 407 
 408   { // Check for the null sentinel.
 409     // If we just called the VM, it already did the mapping for us,
 410     // but it's harmless to retry.
 411     Label notNull;
 412 
 413     // Stash null_sentinel address to get its value later
 414     __ movptr(rarg, (uintptr_t)Universe::the_null_sentinel_addr());
 415     __ ldr(tmp, Address(rarg));
 416     __ cmpoop(result, tmp);
 417     __ br(Assembler::NE, notNull);
 418     __ mov(result, 0);  // NULL object reference
 419     __ bind(notNull);
 420   }
 421 
 422   if (VerifyOops) {
 423     // Safe to call with 0 result
 424     __ verify_oop(result);
 425   }
 426 }
 427 
 428 void TemplateTable::ldc2_w()
 429 {
 430   transition(vtos, vtos);
 431   Label notDouble, notLong, Done;
 432   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 433 
 434   __ get_cpool_and_tags(r1, r2);
 435   const int base_offset = ConstantPool::header_size() * wordSize;
 436   const int tags_offset = Array<u1>::base_offset_in_bytes();
 437 
 438   // get type
 439   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 440   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 441   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 442   __ br(Assembler::NE, notDouble);
 443 
 444   // dtos
 445   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 446   __ ldrd(v0, Address(r2, base_offset));
 447   __ push_d();
 448   __ b(Done);
 449 
 450   __ bind(notDouble);
 451   __ cmpw(r2, (int)JVM_CONSTANT_Long);
 452   __ br(Assembler::NE, notLong);
 453 
 454   // ltos
 455   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 456   __ ldr(r0, Address(r0, base_offset));
 457   __ push_l();
 458   __ b(Done);
 459 
 460   __ bind(notLong);
 461   condy_helper(Done);
 462 
 463   __ bind(Done);
 464 }
 465 
 466 void TemplateTable::condy_helper(Label& Done)
 467 {
 468   Register obj = r0;
 469   Register rarg = r1;
 470   Register flags = r2;
 471   Register off = r3;
 472 
 473   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 474 
 475   __ mov(rarg, (int) bytecode());
 476   __ call_VM(obj, entry, rarg);
 477 
 478   __ get_vm_result_2(flags, rthread);
 479 
 480   // VMr = obj = base address to find primitive value to push
 481   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 482   __ mov(off, flags);
 483   __ andw(off, off, ConstantPoolCacheEntry::field_index_mask);
 484 
 485   const Address field(obj, off);
 486 
 487   // What sort of thing are we loading?
 488   // x86 uses a shift and mask or wings it with a shift plus assert
 489   // the mask is not needed. aarch64 just uses bitfield extract
 490   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,
 491            ConstantPoolCacheEntry::tos_state_bits);
 492 
 493   switch (bytecode()) {
 494     case Bytecodes::_ldc:
 495     case Bytecodes::_ldc_w:
 496       {
 497         // tos in (itos, ftos, stos, btos, ctos, ztos)
 498         Label notInt, notFloat, notShort, notByte, notChar, notBool;
 499         __ cmpw(flags, itos);
 500         __ br(Assembler::NE, notInt);
 501         // itos
 502         __ ldrw(r0, field);
 503         __ push(itos);
 504         __ b(Done);
 505 
 506         __ bind(notInt);
 507         __ cmpw(flags, ftos);
 508         __ br(Assembler::NE, notFloat);
 509         // ftos
 510         __ load_float(field);
 511         __ push(ftos);
 512         __ b(Done);
 513 
 514         __ bind(notFloat);
 515         __ cmpw(flags, stos);
 516         __ br(Assembler::NE, notShort);
 517         // stos
 518         __ load_signed_short(r0, field);
 519         __ push(stos);
 520         __ b(Done);
 521 
 522         __ bind(notShort);
 523         __ cmpw(flags, btos);
 524         __ br(Assembler::NE, notByte);
 525         // btos
 526         __ load_signed_byte(r0, field);
 527         __ push(btos);
 528         __ b(Done);
 529 
 530         __ bind(notByte);
 531         __ cmpw(flags, ctos);
 532         __ br(Assembler::NE, notChar);
 533         // ctos
 534         __ load_unsigned_short(r0, field);
 535         __ push(ctos);
 536         __ b(Done);
 537 
 538         __ bind(notChar);
 539         __ cmpw(flags, ztos);
 540         __ br(Assembler::NE, notBool);
 541         // ztos
 542         __ load_signed_byte(r0, field);
 543         __ push(ztos);
 544         __ b(Done);
 545 
 546         __ bind(notBool);
 547         break;
 548       }
 549 
 550     case Bytecodes::_ldc2_w:
 551       {
 552         Label notLong, notDouble;
 553         __ cmpw(flags, ltos);
 554         __ br(Assembler::NE, notLong);
 555         // ltos
 556         __ ldr(r0, field);
 557         __ push(ltos);
 558         __ b(Done);
 559 
 560         __ bind(notLong);
 561         __ cmpw(flags, dtos);
 562         __ br(Assembler::NE, notDouble);
 563         // dtos
 564         __ load_double(field);
 565         __ push(dtos);
 566         __ b(Done);
 567 
 568        __ bind(notDouble);
 569         break;
 570       }
 571 
 572     default:
 573       ShouldNotReachHere();
 574     }
 575 
 576     __ stop("bad ldc/condy");
 577 }
 578 
 579 void TemplateTable::locals_index(Register reg, int offset)
 580 {
 581   __ ldrb(reg, at_bcp(offset));
 582   __ neg(reg, reg);
 583 }
 584 
 585 void TemplateTable::iload() {
 586   iload_internal();
 587 }
 588 
 589 void TemplateTable::nofast_iload() {
 590   iload_internal(may_not_rewrite);
 591 }
 592 
 593 void TemplateTable::iload_internal(RewriteControl rc) {
 594   transition(vtos, itos);
 595   if (RewriteFrequentPairs && rc == may_rewrite) {
 596     Label rewrite, done;
 597     Register bc = r4;
 598 
 599     // get next bytecode
 600     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 601 
 602     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 603     // last two iloads in a pair.  Comparing against fast_iload means that
 604     // the next bytecode is neither an iload or a caload, and therefore
 605     // an iload pair.
 606     __ cmpw(r1, Bytecodes::_iload);
 607     __ br(Assembler::EQ, done);
 608 
 609     // if _fast_iload rewrite to _fast_iload2
 610     __ cmpw(r1, Bytecodes::_fast_iload);
 611     __ movw(bc, Bytecodes::_fast_iload2);
 612     __ br(Assembler::EQ, rewrite);
 613 
 614     // if _caload rewrite to _fast_icaload
 615     __ cmpw(r1, Bytecodes::_caload);
 616     __ movw(bc, Bytecodes::_fast_icaload);
 617     __ br(Assembler::EQ, rewrite);
 618 
 619     // else rewrite to _fast_iload
 620     __ movw(bc, Bytecodes::_fast_iload);
 621 
 622     // rewrite
 623     // bc: new bytecode
 624     __ bind(rewrite);
 625     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 626     __ bind(done);
 627 
 628   }
 629 
 630   // do iload, get the local value into tos
 631   locals_index(r1);
 632   __ ldr(r0, iaddress(r1));
 633 
 634 }
 635 
 636 void TemplateTable::fast_iload2()
 637 {
 638   transition(vtos, itos);
 639   locals_index(r1);
 640   __ ldr(r0, iaddress(r1));
 641   __ push(itos);
 642   locals_index(r1, 3);
 643   __ ldr(r0, iaddress(r1));
 644 }
 645 
 646 void TemplateTable::fast_iload()
 647 {
 648   transition(vtos, itos);
 649   locals_index(r1);
 650   __ ldr(r0, iaddress(r1));
 651 }
 652 
 653 void TemplateTable::lload()
 654 {
 655   transition(vtos, ltos);
 656   __ ldrb(r1, at_bcp(1));
 657   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 658   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 659 }
 660 
 661 void TemplateTable::fload()
 662 {
 663   transition(vtos, ftos);
 664   locals_index(r1);
 665   // n.b. we use ldrd here because this is a 64 bit slot
 666   // this is comparable to the iload case
 667   __ ldrd(v0, faddress(r1));
 668 }
 669 
 670 void TemplateTable::dload()
 671 {
 672   transition(vtos, dtos);
 673   __ ldrb(r1, at_bcp(1));
 674   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 675   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 676 }
 677 
 678 void TemplateTable::aload()
 679 {
 680   transition(vtos, atos);
 681   locals_index(r1);
 682   __ ldr(r0, iaddress(r1));
 683 }
 684 
 685 void TemplateTable::locals_index_wide(Register reg) {
 686   __ ldrh(reg, at_bcp(2));
 687   __ rev16w(reg, reg);
 688   __ neg(reg, reg);
 689 }
 690 
 691 void TemplateTable::wide_iload() {
 692   transition(vtos, itos);
 693   locals_index_wide(r1);
 694   __ ldr(r0, iaddress(r1));
 695 }
 696 
 697 void TemplateTable::wide_lload()
 698 {
 699   transition(vtos, ltos);
 700   __ ldrh(r1, at_bcp(2));
 701   __ rev16w(r1, r1);
 702   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 703   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 704 }
 705 
 706 void TemplateTable::wide_fload()
 707 {
 708   transition(vtos, ftos);
 709   locals_index_wide(r1);
 710   // n.b. we use ldrd here because this is a 64 bit slot
 711   // this is comparable to the iload case
 712   __ ldrd(v0, faddress(r1));
 713 }
 714 
 715 void TemplateTable::wide_dload()
 716 {
 717   transition(vtos, dtos);
 718   __ ldrh(r1, at_bcp(2));
 719   __ rev16w(r1, r1);
 720   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 721   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 722 }
 723 
 724 void TemplateTable::wide_aload()
 725 {
 726   transition(vtos, atos);
 727   locals_index_wide(r1);
 728   __ ldr(r0, aaddress(r1));
 729 }
 730 
 731 void TemplateTable::index_check(Register array, Register index)
 732 {
 733   // destroys r1, rscratch1
 734   // check array
 735   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 736   // sign extend index for use by indexed load
 737   // __ movl2ptr(index, index);
 738   // check index
 739   Register length = rscratch1;
 740   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 741   __ cmpw(index, length);
 742   if (index != r1) {
 743     // ??? convention: move aberrant index into r1 for exception message
 744     assert(r1 != array, "different registers");
 745     __ mov(r1, index);
 746   }
 747   Label ok;
 748   __ br(Assembler::LO, ok);
 749     // ??? convention: move array into r3 for exception message
 750   __ mov(r3, array);
 751   __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 752   __ br(rscratch1);
 753   __ bind(ok);
 754 }
 755 
 756 void TemplateTable::iaload()
 757 {
 758   transition(itos, itos);
 759   __ mov(r1, r0);
 760   __ pop_ptr(r0);
 761   // r0: array
 762   // r1: index
 763   index_check(r0, r1); // leaves index in r1, kills rscratch1
 764   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
 765   __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 766 }
 767 
 768 void TemplateTable::laload()
 769 {
 770   transition(itos, ltos);
 771   __ mov(r1, r0);
 772   __ pop_ptr(r0);
 773   // r0: array
 774   // r1: index
 775   index_check(r0, r1); // leaves index in r1, kills rscratch1
 776   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
 777   __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 778 }
 779 
 780 void TemplateTable::faload()
 781 {
 782   transition(itos, ftos);
 783   __ mov(r1, r0);
 784   __ pop_ptr(r0);
 785   // r0: array
 786   // r1: index
 787   index_check(r0, r1); // leaves index in r1, kills rscratch1
 788   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
 789   __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 790 }
 791 
 792 void TemplateTable::daload()
 793 {
 794   transition(itos, dtos);
 795   __ mov(r1, r0);
 796   __ pop_ptr(r0);
 797   // r0: array
 798   // r1: index
 799   index_check(r0, r1); // leaves index in r1, kills rscratch1
 800   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
 801   __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 802 }
 803 
 804 void TemplateTable::aaload()
 805 {
 806   transition(itos, atos);
 807   __ mov(r1, r0);
 808   __ pop_ptr(r0);
 809   // r0: array
 810   // r1: index
 811   index_check(r0, r1); // leaves index in r1, kills rscratch1
 812   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 813   do_oop_load(_masm,
 814               Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)),
 815               r0,
 816               IS_ARRAY);
 817 }
 818 
 819 void TemplateTable::baload()
 820 {
 821   transition(itos, itos);
 822   __ mov(r1, r0);
 823   __ pop_ptr(r0);
 824   // r0: array
 825   // r1: index
 826   index_check(r0, r1); // leaves index in r1, kills rscratch1
 827   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
 828   __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(0)), noreg, noreg);
 829 }
 830 
 831 void TemplateTable::caload()
 832 {
 833   transition(itos, itos);
 834   __ mov(r1, r0);
 835   __ pop_ptr(r0);
 836   // r0: array
 837   // r1: index
 838   index_check(r0, r1); // leaves index in r1, kills rscratch1
 839   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 840   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 841 }
 842 
 843 // iload followed by caload frequent pair
 844 void TemplateTable::fast_icaload()
 845 {
 846   transition(vtos, itos);
 847   // load index out of locals
 848   locals_index(r2);
 849   __ ldr(r1, iaddress(r2));
 850 
 851   __ pop_ptr(r0);
 852 
 853   // r0: array
 854   // r1: index
 855   index_check(r0, r1); // leaves index in r1, kills rscratch1
 856   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 857   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 858 }
 859 
 860 void TemplateTable::saload()
 861 {
 862   transition(itos, itos);
 863   __ mov(r1, r0);
 864   __ pop_ptr(r0);
 865   // r0: array
 866   // r1: index
 867   index_check(r0, r1); // leaves index in r1, kills rscratch1
 868   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_SHORT) >> 1);
 869   __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 870 }
 871 
 872 void TemplateTable::iload(int n)
 873 {
 874   transition(vtos, itos);
 875   __ ldr(r0, iaddress(n));
 876 }
 877 
 878 void TemplateTable::lload(int n)
 879 {
 880   transition(vtos, ltos);
 881   __ ldr(r0, laddress(n));
 882 }
 883 
 884 void TemplateTable::fload(int n)
 885 {
 886   transition(vtos, ftos);
 887   __ ldrs(v0, faddress(n));
 888 }
 889 
 890 void TemplateTable::dload(int n)
 891 {
 892   transition(vtos, dtos);
 893   __ ldrd(v0, daddress(n));
 894 }
 895 
 896 void TemplateTable::aload(int n)
 897 {
 898   transition(vtos, atos);
 899   __ ldr(r0, iaddress(n));
 900 }
 901 
 902 void TemplateTable::aload_0() {
 903   aload_0_internal();
 904 }
 905 
 906 void TemplateTable::nofast_aload_0() {
 907   aload_0_internal(may_not_rewrite);
 908 }
 909 
 910 void TemplateTable::aload_0_internal(RewriteControl rc) {
 911   // According to bytecode histograms, the pairs:
 912   //
 913   // _aload_0, _fast_igetfield
 914   // _aload_0, _fast_agetfield
 915   // _aload_0, _fast_fgetfield
 916   //
 917   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 918   // _aload_0 bytecode checks if the next bytecode is either
 919   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 920   // rewrites the current bytecode into a pair bytecode; otherwise it
 921   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 922   // the pair check anymore.
 923   //
 924   // Note: If the next bytecode is _getfield, the rewrite must be
 925   //       delayed, otherwise we may miss an opportunity for a pair.
 926   //
 927   // Also rewrite frequent pairs
 928   //   aload_0, aload_1
 929   //   aload_0, iload_1
 930   // These bytecodes with a small amount of code are most profitable
 931   // to rewrite
 932   if (RewriteFrequentPairs && rc == may_rewrite) {
 933     Label rewrite, done;
 934     const Register bc = r4;
 935 
 936     // get next bytecode
 937     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 938 
 939     // if _getfield then wait with rewrite
 940     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 941     __ br(Assembler::EQ, done);
 942 
 943     // if _igetfield then rewrite to _fast_iaccess_0
 944     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 945     __ cmpw(r1, Bytecodes::_fast_igetfield);
 946     __ movw(bc, Bytecodes::_fast_iaccess_0);
 947     __ br(Assembler::EQ, rewrite);
 948 
 949     // if _agetfield then rewrite to _fast_aaccess_0
 950     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 951     __ cmpw(r1, Bytecodes::_fast_agetfield);
 952     __ movw(bc, Bytecodes::_fast_aaccess_0);
 953     __ br(Assembler::EQ, rewrite);
 954 
 955     // if _fgetfield then rewrite to _fast_faccess_0
 956     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 957     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 958     __ movw(bc, Bytecodes::_fast_faccess_0);
 959     __ br(Assembler::EQ, rewrite);
 960 
 961     // else rewrite to _fast_aload0
 962     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 963     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 964 
 965     // rewrite
 966     // bc: new bytecode
 967     __ bind(rewrite);
 968     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 969 
 970     __ bind(done);
 971   }
 972 
 973   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 974   aload(0);
 975 }
 976 
 977 void TemplateTable::istore()
 978 {
 979   transition(itos, vtos);
 980   locals_index(r1);
 981   // FIXME: We're being very pernickerty here storing a jint in a
 982   // local with strw, which costs an extra instruction over what we'd
 983   // be able to do with a simple str.  We should just store the whole
 984   // word.
 985   __ lea(rscratch1, iaddress(r1));
 986   __ strw(r0, Address(rscratch1));
 987 }
 988 
 989 void TemplateTable::lstore()
 990 {
 991   transition(ltos, vtos);
 992   locals_index(r1);
 993   __ str(r0, laddress(r1, rscratch1, _masm));
 994 }
 995 
 996 void TemplateTable::fstore() {
 997   transition(ftos, vtos);
 998   locals_index(r1);
 999   __ lea(rscratch1, iaddress(r1));
1000   __ strs(v0, Address(rscratch1));
1001 }
1002 
1003 void TemplateTable::dstore() {
1004   transition(dtos, vtos);
1005   locals_index(r1);
1006   __ strd(v0, daddress(r1, rscratch1, _masm));
1007 }
1008 
1009 void TemplateTable::astore()
1010 {
1011   transition(vtos, vtos);
1012   __ pop_ptr(r0);
1013   locals_index(r1);
1014   __ str(r0, aaddress(r1));
1015 }
1016 
1017 void TemplateTable::wide_istore() {
1018   transition(vtos, vtos);
1019   __ pop_i();
1020   locals_index_wide(r1);
1021   __ lea(rscratch1, iaddress(r1));
1022   __ strw(r0, Address(rscratch1));
1023 }
1024 
1025 void TemplateTable::wide_lstore() {
1026   transition(vtos, vtos);
1027   __ pop_l();
1028   locals_index_wide(r1);
1029   __ str(r0, laddress(r1, rscratch1, _masm));
1030 }
1031 
1032 void TemplateTable::wide_fstore() {
1033   transition(vtos, vtos);
1034   __ pop_f();
1035   locals_index_wide(r1);
1036   __ lea(rscratch1, faddress(r1));
1037   __ strs(v0, rscratch1);
1038 }
1039 
1040 void TemplateTable::wide_dstore() {
1041   transition(vtos, vtos);
1042   __ pop_d();
1043   locals_index_wide(r1);
1044   __ strd(v0, daddress(r1, rscratch1, _masm));
1045 }
1046 
1047 void TemplateTable::wide_astore() {
1048   transition(vtos, vtos);
1049   __ pop_ptr(r0);
1050   locals_index_wide(r1);
1051   __ str(r0, aaddress(r1));
1052 }
1053 
1054 void TemplateTable::iastore() {
1055   transition(itos, vtos);
1056   __ pop_i(r1);
1057   __ pop_ptr(r3);
1058   // r0: value
1059   // r1: index
1060   // r3: array
1061   index_check(r3, r1); // prefer index in r1
1062   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
1063   __ access_store_at(T_INT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), r0, noreg, noreg);
1064 }
1065 
1066 void TemplateTable::lastore() {
1067   transition(ltos, vtos);
1068   __ pop_i(r1);
1069   __ pop_ptr(r3);
1070   // r0: value
1071   // r1: index
1072   // r3: array
1073   index_check(r3, r1); // prefer index in r1
1074   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
1075   __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), r0, noreg, noreg);
1076 }
1077 
1078 void TemplateTable::fastore() {
1079   transition(ftos, vtos);
1080   __ pop_i(r1);
1081   __ pop_ptr(r3);
1082   // v0: value
1083   // r1:  index
1084   // r3:  array
1085   index_check(r3, r1); // prefer index in r1
1086   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
1087   __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), noreg /* ftos */, noreg, noreg);
1088 }
1089 
1090 void TemplateTable::dastore() {
1091   transition(dtos, vtos);
1092   __ pop_i(r1);
1093   __ pop_ptr(r3);
1094   // v0: value
1095   // r1:  index
1096   // r3:  array
1097   index_check(r3, r1); // prefer index in r1
1098   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
1099   __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), noreg /* dtos */, noreg, noreg);
1100 }
1101 
1102 void TemplateTable::aastore() {
1103   Label is_null, ok_is_subtype, done;
1104   transition(vtos, vtos);
1105   // stack: ..., array, index, value
1106   __ ldr(r0, at_tos());    // value
1107   __ ldr(r2, at_tos_p1()); // index
1108   __ ldr(r3, at_tos_p2()); // array
1109 
1110   Address element_address(r3, r4, Address::uxtw(LogBytesPerHeapOop));
1111 
1112   index_check(r3, r2);     // kills r1
1113   __ add(r4, r2, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
1114 
1115   // do array store check - check for NULL value first
1116   __ cbz(r0, is_null);
1117 
1118   // Move subklass into r1
1119   __ load_klass(r1, r0);
1120   // Move superklass into r0
1121   __ load_klass(r0, r3);
1122   __ ldr(r0, Address(r0,
1123                      ObjArrayKlass::element_klass_offset()));
1124   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1125 
1126   // Generate subtype check.  Blows r2, r5
1127   // Superklass in r0.  Subklass in r1.
1128   __ gen_subtype_check(r1, ok_is_subtype);
1129 
1130   // Come here on failure
1131   // object is at TOS
1132   __ b(Interpreter::_throw_ArrayStoreException_entry);
1133 
1134   // Come here on success
1135   __ bind(ok_is_subtype);
1136 
1137   // Get the value we will store
1138   __ ldr(r0, at_tos());
1139   // Now store using the appropriate barrier
1140   do_oop_store(_masm, element_address, r0, IS_ARRAY);
1141   __ b(done);
1142 
1143   // Have a NULL in r0, r3=array, r2=index.  Store NULL at ary[idx]
1144   __ bind(is_null);
1145   __ profile_null_seen(r2);
1146 
1147   // Store a NULL
1148   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1149 
1150   // Pop stack arguments
1151   __ bind(done);
1152   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1153 }
1154 
1155 void TemplateTable::bastore()
1156 {
1157   transition(itos, vtos);
1158   __ pop_i(r1);
1159   __ pop_ptr(r3);
1160   // r0: value
1161   // r1: index
1162   // r3: array
1163   index_check(r3, r1); // prefer index in r1
1164 
1165   // Need to check whether array is boolean or byte
1166   // since both types share the bastore bytecode.
1167   __ load_klass(r2, r3);
1168   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1169   int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1170   Label L_skip;
1171   __ tbz(r2, diffbit_index, L_skip);
1172   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1173   __ bind(L_skip);
1174 
1175   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
1176   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(0)), r0, noreg, noreg);
1177 }
1178 
1179 void TemplateTable::castore()
1180 {
1181   transition(itos, vtos);
1182   __ pop_i(r1);
1183   __ pop_ptr(r3);
1184   // r0: value
1185   // r1: index
1186   // r3: array
1187   index_check(r3, r1); // prefer index in r1
1188   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
1189   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(1)), r0, noreg, noreg);
1190 }
1191 
1192 void TemplateTable::sastore()
1193 {
1194   castore();
1195 }
1196 
1197 void TemplateTable::istore(int n)
1198 {
1199   transition(itos, vtos);
1200   __ str(r0, iaddress(n));
1201 }
1202 
1203 void TemplateTable::lstore(int n)
1204 {
1205   transition(ltos, vtos);
1206   __ str(r0, laddress(n));
1207 }
1208 
1209 void TemplateTable::fstore(int n)
1210 {
1211   transition(ftos, vtos);
1212   __ strs(v0, faddress(n));
1213 }
1214 
1215 void TemplateTable::dstore(int n)
1216 {
1217   transition(dtos, vtos);
1218   __ strd(v0, daddress(n));
1219 }
1220 
1221 void TemplateTable::astore(int n)
1222 {
1223   transition(vtos, vtos);
1224   __ pop_ptr(r0);
1225   __ str(r0, iaddress(n));
1226 }
1227 
1228 void TemplateTable::pop()
1229 {
1230   transition(vtos, vtos);
1231   __ add(esp, esp, Interpreter::stackElementSize);
1232 }
1233 
1234 void TemplateTable::pop2()
1235 {
1236   transition(vtos, vtos);
1237   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1238 }
1239 
1240 void TemplateTable::dup()
1241 {
1242   transition(vtos, vtos);
1243   __ ldr(r0, Address(esp, 0));
1244   __ push(r0);
1245   // stack: ..., a, a
1246 }
1247 
1248 void TemplateTable::dup_x1()
1249 {
1250   transition(vtos, vtos);
1251   // stack: ..., a, b
1252   __ ldr(r0, at_tos());  // load b
1253   __ ldr(r2, at_tos_p1());  // load a
1254   __ str(r0, at_tos_p1());  // store b
1255   __ str(r2, at_tos());  // store a
1256   __ push(r0);                  // push b
1257   // stack: ..., b, a, b
1258 }
1259 
1260 void TemplateTable::dup_x2()
1261 {
1262   transition(vtos, vtos);
1263   // stack: ..., a, b, c
1264   __ ldr(r0, at_tos());  // load c
1265   __ ldr(r2, at_tos_p2());  // load a
1266   __ str(r0, at_tos_p2());  // store c in a
1267   __ push(r0);      // push c
1268   // stack: ..., c, b, c, c
1269   __ ldr(r0, at_tos_p2());  // load b
1270   __ str(r2, at_tos_p2());  // store a in b
1271   // stack: ..., c, a, c, c
1272   __ str(r0, at_tos_p1());  // store b in c
1273   // stack: ..., c, a, b, c
1274 }
1275 
1276 void TemplateTable::dup2()
1277 {
1278   transition(vtos, vtos);
1279   // stack: ..., a, b
1280   __ ldr(r0, at_tos_p1());  // load a
1281   __ push(r0);                  // push a
1282   __ ldr(r0, at_tos_p1());  // load b
1283   __ push(r0);                  // push b
1284   // stack: ..., a, b, a, b
1285 }
1286 
1287 void TemplateTable::dup2_x1()
1288 {
1289   transition(vtos, vtos);
1290   // stack: ..., a, b, c
1291   __ ldr(r2, at_tos());  // load c
1292   __ ldr(r0, at_tos_p1());  // load b
1293   __ push(r0);                  // push b
1294   __ push(r2);                  // push c
1295   // stack: ..., a, b, c, b, c
1296   __ str(r2, at_tos_p3());  // store c in b
1297   // stack: ..., a, c, c, b, c
1298   __ ldr(r2, at_tos_p4());  // load a
1299   __ str(r2, at_tos_p2());  // store a in 2nd c
1300   // stack: ..., a, c, a, b, c
1301   __ str(r0, at_tos_p4());  // store b in a
1302   // stack: ..., b, c, a, b, c
1303 }
1304 
1305 void TemplateTable::dup2_x2()
1306 {
1307   transition(vtos, vtos);
1308   // stack: ..., a, b, c, d
1309   __ ldr(r2, at_tos());  // load d
1310   __ ldr(r0, at_tos_p1());  // load c
1311   __ push(r0)            ;      // push c
1312   __ push(r2);                  // push d
1313   // stack: ..., a, b, c, d, c, d
1314   __ ldr(r0, at_tos_p4());  // load b
1315   __ str(r0, at_tos_p2());  // store b in d
1316   __ str(r2, at_tos_p4());  // store d in b
1317   // stack: ..., a, d, c, b, c, d
1318   __ ldr(r2, at_tos_p5());  // load a
1319   __ ldr(r0, at_tos_p3());  // load c
1320   __ str(r2, at_tos_p3());  // store a in c
1321   __ str(r0, at_tos_p5());  // store c in a
1322   // stack: ..., c, d, a, b, c, d
1323 }
1324 
1325 void TemplateTable::swap()
1326 {
1327   transition(vtos, vtos);
1328   // stack: ..., a, b
1329   __ ldr(r2, at_tos_p1());  // load a
1330   __ ldr(r0, at_tos());  // load b
1331   __ str(r2, at_tos());  // store a in b
1332   __ str(r0, at_tos_p1());  // store b in a
1333   // stack: ..., b, a
1334 }
1335 
1336 void TemplateTable::iop2(Operation op)
1337 {
1338   transition(itos, itos);
1339   // r0 <== r1 op r0
1340   __ pop_i(r1);
1341   switch (op) {
1342   case add  : __ addw(r0, r1, r0); break;
1343   case sub  : __ subw(r0, r1, r0); break;
1344   case mul  : __ mulw(r0, r1, r0); break;
1345   case _and : __ andw(r0, r1, r0); break;
1346   case _or  : __ orrw(r0, r1, r0); break;
1347   case _xor : __ eorw(r0, r1, r0); break;
1348   case shl  : __ lslvw(r0, r1, r0); break;
1349   case shr  : __ asrvw(r0, r1, r0); break;
1350   case ushr : __ lsrvw(r0, r1, r0);break;
1351   default   : ShouldNotReachHere();
1352   }
1353 }
1354 
1355 void TemplateTable::lop2(Operation op)
1356 {
1357   transition(ltos, ltos);
1358   // r0 <== r1 op r0
1359   __ pop_l(r1);
1360   switch (op) {
1361   case add  : __ add(r0, r1, r0); break;
1362   case sub  : __ sub(r0, r1, r0); break;
1363   case mul  : __ mul(r0, r1, r0); break;
1364   case _and : __ andr(r0, r1, r0); break;
1365   case _or  : __ orr(r0, r1, r0); break;
1366   case _xor : __ eor(r0, r1, r0); break;
1367   default   : ShouldNotReachHere();
1368   }
1369 }
1370 
1371 void TemplateTable::idiv()
1372 {
1373   transition(itos, itos);
1374   // explicitly check for div0
1375   Label no_div0;
1376   __ cbnzw(r0, no_div0);
1377   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1378   __ br(rscratch1);
1379   __ bind(no_div0);
1380   __ pop_i(r1);
1381   // r0 <== r1 idiv r0
1382   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1383 }
1384 
1385 void TemplateTable::irem()
1386 {
1387   transition(itos, itos);
1388   // explicitly check for div0
1389   Label no_div0;
1390   __ cbnzw(r0, no_div0);
1391   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1392   __ br(rscratch1);
1393   __ bind(no_div0);
1394   __ pop_i(r1);
1395   // r0 <== r1 irem r0
1396   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1397 }
1398 
1399 void TemplateTable::lmul()
1400 {
1401   transition(ltos, ltos);
1402   __ pop_l(r1);
1403   __ mul(r0, r0, r1);
1404 }
1405 
1406 void TemplateTable::ldiv()
1407 {
1408   transition(ltos, ltos);
1409   // explicitly check for div0
1410   Label no_div0;
1411   __ cbnz(r0, no_div0);
1412   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1413   __ br(rscratch1);
1414   __ bind(no_div0);
1415   __ pop_l(r1);
1416   // r0 <== r1 ldiv r0
1417   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1418 }
1419 
1420 void TemplateTable::lrem()
1421 {
1422   transition(ltos, ltos);
1423   // explicitly check for div0
1424   Label no_div0;
1425   __ cbnz(r0, no_div0);
1426   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1427   __ br(rscratch1);
1428   __ bind(no_div0);
1429   __ pop_l(r1);
1430   // r0 <== r1 lrem r0
1431   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1432 }
1433 
1434 void TemplateTable::lshl()
1435 {
1436   transition(itos, ltos);
1437   // shift count is in r0
1438   __ pop_l(r1);
1439   __ lslv(r0, r1, r0);
1440 }
1441 
1442 void TemplateTable::lshr()
1443 {
1444   transition(itos, ltos);
1445   // shift count is in r0
1446   __ pop_l(r1);
1447   __ asrv(r0, r1, r0);
1448 }
1449 
1450 void TemplateTable::lushr()
1451 {
1452   transition(itos, ltos);
1453   // shift count is in r0
1454   __ pop_l(r1);
1455   __ lsrv(r0, r1, r0);
1456 }
1457 
1458 void TemplateTable::fop2(Operation op)
1459 {
1460   transition(ftos, ftos);
1461   switch (op) {
1462   case add:
1463     // n.b. use ldrd because this is a 64 bit slot
1464     __ pop_f(v1);
1465     __ fadds(v0, v1, v0);
1466     break;
1467   case sub:
1468     __ pop_f(v1);
1469     __ fsubs(v0, v1, v0);
1470     break;
1471   case mul:
1472     __ pop_f(v1);
1473     __ fmuls(v0, v1, v0);
1474     break;
1475   case div:
1476     __ pop_f(v1);
1477     __ fdivs(v0, v1, v0);
1478     break;
1479   case rem:
1480     __ fmovs(v1, v0);
1481     __ pop_f(v0);
1482     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
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(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1515     break;
1516   default:
1517     ShouldNotReachHere();
1518     break;
1519   }
1520 }
1521 
1522 void TemplateTable::ineg()
1523 {
1524   transition(itos, itos);
1525   __ negw(r0, r0);
1526 
1527 }
1528 
1529 void TemplateTable::lneg()
1530 {
1531   transition(ltos, ltos);
1532   __ neg(r0, r0);
1533 }
1534 
1535 void TemplateTable::fneg()
1536 {
1537   transition(ftos, ftos);
1538   __ fnegs(v0, v0);
1539 }
1540 
1541 void TemplateTable::dneg()
1542 {
1543   transition(dtos, dtos);
1544   __ fnegd(v0, v0);
1545 }
1546 
1547 void TemplateTable::iinc()
1548 {
1549   transition(vtos, vtos);
1550   __ load_signed_byte(r1, at_bcp(2)); // get constant
1551   locals_index(r2);
1552   __ ldr(r0, iaddress(r2));
1553   __ addw(r0, r0, r1);
1554   __ str(r0, iaddress(r2));
1555 }
1556 
1557 void TemplateTable::wide_iinc()
1558 {
1559   transition(vtos, vtos);
1560   // __ mov(r1, zr);
1561   __ ldrw(r1, at_bcp(2)); // get constant and index
1562   __ rev16(r1, r1);
1563   __ ubfx(r2, r1, 0, 16);
1564   __ neg(r2, r2);
1565   __ sbfx(r1, r1, 16, 16);
1566   __ ldr(r0, iaddress(r2));
1567   __ addw(r0, r0, r1);
1568   __ str(r0, iaddress(r2));
1569 }
1570 
1571 void TemplateTable::convert()
1572 {
1573   // Checking
1574 #ifdef ASSERT
1575   {
1576     TosState tos_in  = ilgl;
1577     TosState tos_out = ilgl;
1578     switch (bytecode()) {
1579     case Bytecodes::_i2l: // fall through
1580     case Bytecodes::_i2f: // fall through
1581     case Bytecodes::_i2d: // fall through
1582     case Bytecodes::_i2b: // fall through
1583     case Bytecodes::_i2c: // fall through
1584     case Bytecodes::_i2s: tos_in = itos; break;
1585     case Bytecodes::_l2i: // fall through
1586     case Bytecodes::_l2f: // fall through
1587     case Bytecodes::_l2d: tos_in = ltos; break;
1588     case Bytecodes::_f2i: // fall through
1589     case Bytecodes::_f2l: // fall through
1590     case Bytecodes::_f2d: tos_in = ftos; break;
1591     case Bytecodes::_d2i: // fall through
1592     case Bytecodes::_d2l: // fall through
1593     case Bytecodes::_d2f: tos_in = dtos; break;
1594     default             : ShouldNotReachHere();
1595     }
1596     switch (bytecode()) {
1597     case Bytecodes::_l2i: // fall through
1598     case Bytecodes::_f2i: // fall through
1599     case Bytecodes::_d2i: // fall through
1600     case Bytecodes::_i2b: // fall through
1601     case Bytecodes::_i2c: // fall through
1602     case Bytecodes::_i2s: tos_out = itos; break;
1603     case Bytecodes::_i2l: // fall through
1604     case Bytecodes::_f2l: // fall through
1605     case Bytecodes::_d2l: tos_out = ltos; break;
1606     case Bytecodes::_i2f: // fall through
1607     case Bytecodes::_l2f: // fall through
1608     case Bytecodes::_d2f: tos_out = ftos; break;
1609     case Bytecodes::_i2d: // fall through
1610     case Bytecodes::_l2d: // fall through
1611     case Bytecodes::_f2d: tos_out = dtos; break;
1612     default             : ShouldNotReachHere();
1613     }
1614     transition(tos_in, tos_out);
1615   }
1616 #endif // ASSERT
1617   // static const int64_t is_nan = 0x8000000000000000L;
1618 
1619   // Conversion
1620   switch (bytecode()) {
1621   case Bytecodes::_i2l:
1622     __ sxtw(r0, r0);
1623     break;
1624   case Bytecodes::_i2f:
1625     __ scvtfws(v0, r0);
1626     break;
1627   case Bytecodes::_i2d:
1628     __ scvtfwd(v0, r0);
1629     break;
1630   case Bytecodes::_i2b:
1631     __ sxtbw(r0, r0);
1632     break;
1633   case Bytecodes::_i2c:
1634     __ uxthw(r0, r0);
1635     break;
1636   case Bytecodes::_i2s:
1637     __ sxthw(r0, r0);
1638     break;
1639   case Bytecodes::_l2i:
1640     __ uxtw(r0, r0);
1641     break;
1642   case Bytecodes::_l2f:
1643     __ scvtfs(v0, r0);
1644     break;
1645   case Bytecodes::_l2d:
1646     __ scvtfd(v0, r0);
1647     break;
1648   case Bytecodes::_f2i:
1649   {
1650     Label L_Okay;
1651     __ clear_fpsr();
1652     __ fcvtzsw(r0, v0);
1653     __ get_fpsr(r1);
1654     __ cbzw(r1, L_Okay);
1655     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i));
1656     __ bind(L_Okay);
1657   }
1658     break;
1659   case Bytecodes::_f2l:
1660   {
1661     Label L_Okay;
1662     __ clear_fpsr();
1663     __ fcvtzs(r0, v0);
1664     __ get_fpsr(r1);
1665     __ cbzw(r1, L_Okay);
1666     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1667     __ bind(L_Okay);
1668   }
1669     break;
1670   case Bytecodes::_f2d:
1671     __ fcvts(v0, v0);
1672     break;
1673   case Bytecodes::_d2i:
1674   {
1675     Label L_Okay;
1676     __ clear_fpsr();
1677     __ fcvtzdw(r0, v0);
1678     __ get_fpsr(r1);
1679     __ cbzw(r1, L_Okay);
1680     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1681     __ bind(L_Okay);
1682   }
1683     break;
1684   case Bytecodes::_d2l:
1685   {
1686     Label L_Okay;
1687     __ clear_fpsr();
1688     __ fcvtzd(r0, v0);
1689     __ get_fpsr(r1);
1690     __ cbzw(r1, L_Okay);
1691     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1692     __ bind(L_Okay);
1693   }
1694     break;
1695   case Bytecodes::_d2f:
1696     __ fcvtd(v0, v0);
1697     break;
1698   default:
1699     ShouldNotReachHere();
1700   }
1701 }
1702 
1703 void TemplateTable::lcmp()
1704 {
1705   transition(ltos, itos);
1706   Label done;
1707   __ pop_l(r1);
1708   __ cmp(r1, r0);
1709   __ mov(r0, (u_int64_t)-1L);
1710   __ br(Assembler::LT, done);
1711   // __ mov(r0, 1UL);
1712   // __ csel(r0, r0, zr, Assembler::NE);
1713   // and here is a faster way
1714   __ csinc(r0, zr, zr, Assembler::EQ);
1715   __ bind(done);
1716 }
1717 
1718 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1719 {
1720   Label done;
1721   if (is_float) {
1722     // XXX get rid of pop here, use ... reg, mem32
1723     __ pop_f(v1);
1724     __ fcmps(v1, v0);
1725   } else {
1726     // XXX get rid of pop here, use ... reg, mem64
1727     __ pop_d(v1);
1728     __ fcmpd(v1, v0);
1729   }
1730   if (unordered_result < 0) {
1731     // we want -1 for unordered or less than, 0 for equal and 1 for
1732     // greater than.
1733     __ mov(r0, (u_int64_t)-1L);
1734     // for FP LT tests less than or unordered
1735     __ br(Assembler::LT, done);
1736     // install 0 for EQ otherwise 1
1737     __ csinc(r0, zr, zr, Assembler::EQ);
1738   } else {
1739     // we want -1 for less than, 0 for equal and 1 for unordered or
1740     // greater than.
1741     __ mov(r0, 1L);
1742     // for FP HI tests greater than or unordered
1743     __ br(Assembler::HI, done);
1744     // install 0 for EQ otherwise ~0
1745     __ csinv(r0, zr, zr, Assembler::EQ);
1746 
1747   }
1748   __ bind(done);
1749 }
1750 
1751 void TemplateTable::branch(bool is_jsr, bool is_wide)
1752 {
1753   // We might be moving to a safepoint.  The thread which calls
1754   // Interpreter::notice_safepoints() will effectively flush its cache
1755   // when it makes a system call, but we need to do something to
1756   // ensure that we see the changed dispatch table.
1757   __ membar(MacroAssembler::LoadLoad);
1758 
1759   __ profile_taken_branch(r0, r1);
1760   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1761                              InvocationCounter::counter_offset();
1762   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1763                               InvocationCounter::counter_offset();
1764 
1765   // load branch displacement
1766   if (!is_wide) {
1767     __ ldrh(r2, at_bcp(1));
1768     __ rev16(r2, r2);
1769     // sign extend the 16 bit value in r2
1770     __ sbfm(r2, r2, 0, 15);
1771   } else {
1772     __ ldrw(r2, at_bcp(1));
1773     __ revw(r2, r2);
1774     // sign extend the 32 bit value in r2
1775     __ sbfm(r2, r2, 0, 31);
1776   }
1777 
1778   // Handle all the JSR stuff here, then exit.
1779   // It's much shorter and cleaner than intermingling with the non-JSR
1780   // normal-branch stuff occurring below.
1781 
1782   if (is_jsr) {
1783     // Pre-load the next target bytecode into rscratch1
1784     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1785     // compute return address as bci
1786     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1787     __ add(rscratch2, rscratch2,
1788            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1789     __ sub(r1, rbcp, rscratch2);
1790     __ push_i(r1);
1791     // Adjust the bcp by the 16-bit displacement in r2
1792     __ add(rbcp, rbcp, r2);
1793     __ dispatch_only(vtos, /*generate_poll*/true);
1794     return;
1795   }
1796 
1797   // Normal (non-jsr) branch handling
1798 
1799   // Adjust the bcp by the displacement in r2
1800   __ add(rbcp, rbcp, r2);
1801 
1802   assert(UseLoopCounter || !UseOnStackReplacement,
1803          "on-stack-replacement requires loop counters");
1804   Label backedge_counter_overflow;
1805   Label profile_method;
1806   Label dispatch;
1807   if (UseLoopCounter) {
1808     // increment backedge counter for backward branches
1809     // r0: MDO
1810     // w1: MDO bumped taken-count
1811     // r2: target offset
1812     __ cmp(r2, zr);
1813     __ br(Assembler::GT, dispatch); // count only if backward branch
1814 
1815     // ECN: FIXME: This code smells
1816     // check if MethodCounters exists
1817     Label has_counters;
1818     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1819     __ cbnz(rscratch1, has_counters);
1820     __ push(r0);
1821     __ push(r1);
1822     __ push(r2);
1823     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1824             InterpreterRuntime::build_method_counters), rmethod);
1825     __ pop(r2);
1826     __ pop(r1);
1827     __ pop(r0);
1828     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1829     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1830     __ bind(has_counters);
1831 
1832     if (TieredCompilation) {
1833       Label no_mdo;
1834       int increment = InvocationCounter::count_increment;
1835       if (ProfileInterpreter) {
1836         // Are we profiling?
1837         __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1838         __ cbz(r1, no_mdo);
1839         // Increment the MDO backedge counter
1840         const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1841                                            in_bytes(InvocationCounter::counter_offset()));
1842         const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1843         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1844                                    r0, rscratch1, false, Assembler::EQ,
1845                                    UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1846         __ b(dispatch);
1847       }
1848       __ bind(no_mdo);
1849       // Increment backedge counter in MethodCounters*
1850       __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1851       const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1852       __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1853                                  r0, rscratch2, false, Assembler::EQ,
1854                                  UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1855     } else { // not TieredCompilation
1856       // increment counter
1857       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1858       __ ldrw(r0, Address(rscratch2, be_offset));        // load backedge counter
1859       __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1860       __ strw(rscratch1, Address(rscratch2, be_offset));        // store counter
1861 
1862       __ ldrw(r0, Address(rscratch2, inv_offset));    // load invocation counter
1863       __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1864       __ addw(r0, r0, rscratch1);        // add both counters
1865 
1866       if (ProfileInterpreter) {
1867         // Test to see if we should create a method data oop
1868         __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1869         __ cmpw(r0, rscratch1);
1870         __ br(Assembler::LT, dispatch);
1871 
1872         // if no method data exists, go to profile method
1873         __ test_method_data_pointer(r0, profile_method);
1874 
1875         if (UseOnStackReplacement) {
1876           // check for overflow against w1 which is the MDO taken count
1877           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1878           __ cmpw(r1, rscratch1);
1879           __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1880 
1881           // When ProfileInterpreter is on, the backedge_count comes
1882           // from the MethodData*, which value does not get reset on
1883           // the call to frequency_counter_overflow().  To avoid
1884           // excessive calls to the overflow routine while the method is
1885           // being compiled, add a second test to make sure the overflow
1886           // function is called only once every overflow_frequency.
1887           const int overflow_frequency = 1024;
1888           __ andsw(r1, r1, overflow_frequency - 1);
1889           __ br(Assembler::EQ, backedge_counter_overflow);
1890 
1891         }
1892       } else {
1893         if (UseOnStackReplacement) {
1894           // check for overflow against w0, which is the sum of the
1895           // counters
1896           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1897           __ cmpw(r0, rscratch1);
1898           __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1899         }
1900       }
1901     }
1902     __ bind(dispatch);
1903   }
1904 
1905   // Pre-load the next target bytecode into rscratch1
1906   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1907 
1908   // continue with the bytecode @ target
1909   // rscratch1: target bytecode
1910   // rbcp: target bcp
1911   __ dispatch_only(vtos, /*generate_poll*/true);
1912 
1913   if (UseLoopCounter) {
1914     if (ProfileInterpreter) {
1915       // Out-of-line code to allocate method data oop.
1916       __ bind(profile_method);
1917       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1918       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1919       __ set_method_data_pointer_for_bcp();
1920       __ b(dispatch);
1921     }
1922 
1923     if (UseOnStackReplacement) {
1924       // invocation counter overflow
1925       __ bind(backedge_counter_overflow);
1926       __ neg(r2, r2);
1927       __ add(r2, r2, rbcp);     // branch bcp
1928       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1929       __ call_VM(noreg,
1930                  CAST_FROM_FN_PTR(address,
1931                                   InterpreterRuntime::frequency_counter_overflow),
1932                  r2);
1933       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1934 
1935       // r0: osr nmethod (osr ok) or NULL (osr not possible)
1936       // w1: target bytecode
1937       // r2: scratch
1938       __ cbz(r0, dispatch);     // test result -- no osr if null
1939       // nmethod may have been invalidated (VM may block upon call_VM return)
1940       __ ldrb(r2, Address(r0, nmethod::state_offset()));
1941       if (nmethod::in_use != 0)
1942         __ sub(r2, r2, nmethod::in_use);
1943       __ cbnz(r2, dispatch);
1944 
1945       // We have the address of an on stack replacement routine in r0
1946       // We need to prepare to execute the OSR method. First we must
1947       // migrate the locals and monitors off of the stack.
1948 
1949       __ mov(r19, r0);                             // save the nmethod
1950 
1951       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1952 
1953       // r0 is OSR buffer, move it to expected parameter location
1954       __ mov(j_rarg0, r0);
1955 
1956       // remove activation
1957       // get sender esp
1958       __ ldr(esp,
1959           Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1960       // remove frame anchor
1961       __ leave();
1962       // Ensure compiled code always sees stack at proper alignment
1963       __ andr(sp, esp, -16);
1964 
1965       // and begin the OSR nmethod
1966       __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1967       __ br(rscratch1);
1968     }
1969   }
1970 }
1971 
1972 
1973 void TemplateTable::if_0cmp(Condition cc)
1974 {
1975   transition(itos, vtos);
1976   // assume branch is more often taken than not (loops use backward branches)
1977   Label not_taken;
1978   if (cc == equal)
1979     __ cbnzw(r0, not_taken);
1980   else if (cc == not_equal)
1981     __ cbzw(r0, not_taken);
1982   else {
1983     __ andsw(zr, r0, r0);
1984     __ br(j_not(cc), not_taken);
1985   }
1986 
1987   branch(false, false);
1988   __ bind(not_taken);
1989   __ profile_not_taken_branch(r0);
1990 }
1991 
1992 void TemplateTable::if_icmp(Condition cc)
1993 {
1994   transition(itos, vtos);
1995   // assume branch is more often taken than not (loops use backward branches)
1996   Label not_taken;
1997   __ pop_i(r1);
1998   __ cmpw(r1, r0, Assembler::LSL);
1999   __ br(j_not(cc), not_taken);
2000   branch(false, false);
2001   __ bind(not_taken);
2002   __ profile_not_taken_branch(r0);
2003 }
2004 
2005 void TemplateTable::if_nullcmp(Condition cc)
2006 {
2007   transition(atos, vtos);
2008   // assume branch is more often taken than not (loops use backward branches)
2009   Label not_taken;
2010   if (cc == equal)
2011     __ cbnz(r0, not_taken);
2012   else
2013     __ cbz(r0, not_taken);
2014   branch(false, false);
2015   __ bind(not_taken);
2016   __ profile_not_taken_branch(r0);
2017 }
2018 
2019 void TemplateTable::if_acmp(Condition cc)
2020 {
2021   transition(atos, vtos);
2022   // assume branch is more often taken than not (loops use backward branches)
2023   Label not_taken;
2024   __ pop_ptr(r1);
2025   __ cmpoop(r1, r0);
2026   __ br(j_not(cc), not_taken);
2027   branch(false, false);
2028   __ bind(not_taken);
2029   __ profile_not_taken_branch(r0);
2030 }
2031 
2032 void TemplateTable::ret() {
2033   transition(vtos, vtos);
2034   // We might be moving to a safepoint.  The thread which calls
2035   // Interpreter::notice_safepoints() will effectively flush its cache
2036   // when it makes a system call, but we need to do something to
2037   // ensure that we see the changed dispatch table.
2038   __ membar(MacroAssembler::LoadLoad);
2039 
2040   locals_index(r1);
2041   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2042   __ profile_ret(r1, r2);
2043   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2044   __ lea(rbcp, Address(rbcp, r1));
2045   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2046   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2047 }
2048 
2049 void TemplateTable::wide_ret() {
2050   transition(vtos, vtos);
2051   locals_index_wide(r1);
2052   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2053   __ profile_ret(r1, r2);
2054   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2055   __ lea(rbcp, Address(rbcp, r1));
2056   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2057   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2058 }
2059 
2060 
2061 void TemplateTable::tableswitch() {
2062   Label default_case, continue_execution;
2063   transition(itos, vtos);
2064   // align rbcp
2065   __ lea(r1, at_bcp(BytesPerInt));
2066   __ andr(r1, r1, -BytesPerInt);
2067   // load lo & hi
2068   __ ldrw(r2, Address(r1, BytesPerInt));
2069   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2070   __ rev32(r2, r2);
2071   __ rev32(r3, r3);
2072   // check against lo & hi
2073   __ cmpw(r0, r2);
2074   __ br(Assembler::LT, default_case);
2075   __ cmpw(r0, r3);
2076   __ br(Assembler::GT, default_case);
2077   // lookup dispatch offset
2078   __ subw(r0, r0, r2);
2079   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2080   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2081   __ profile_switch_case(r0, r1, r2);
2082   // continue execution
2083   __ bind(continue_execution);
2084   __ rev32(r3, r3);
2085   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2086   __ add(rbcp, rbcp, r3, ext::sxtw);
2087   __ dispatch_only(vtos, /*generate_poll*/true);
2088   // handle default
2089   __ bind(default_case);
2090   __ profile_switch_default(r0);
2091   __ ldrw(r3, Address(r1, 0));
2092   __ b(continue_execution);
2093 }
2094 
2095 void TemplateTable::lookupswitch() {
2096   transition(itos, itos);
2097   __ stop("lookupswitch bytecode should have been rewritten");
2098 }
2099 
2100 void TemplateTable::fast_linearswitch() {
2101   transition(itos, vtos);
2102   Label loop_entry, loop, found, continue_execution;
2103   // bswap r0 so we can avoid bswapping the table entries
2104   __ rev32(r0, r0);
2105   // align rbcp
2106   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2107                                     // this instruction (change offsets
2108                                     // below)
2109   __ andr(r19, r19, -BytesPerInt);
2110   // set counter
2111   __ ldrw(r1, Address(r19, BytesPerInt));
2112   __ rev32(r1, r1);
2113   __ b(loop_entry);
2114   // table search
2115   __ bind(loop);
2116   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2117   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2118   __ cmpw(r0, rscratch1);
2119   __ br(Assembler::EQ, found);
2120   __ bind(loop_entry);
2121   __ subs(r1, r1, 1);
2122   __ br(Assembler::PL, loop);
2123   // default case
2124   __ profile_switch_default(r0);
2125   __ ldrw(r3, Address(r19, 0));
2126   __ b(continue_execution);
2127   // entry found -> get offset
2128   __ bind(found);
2129   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2130   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2131   __ profile_switch_case(r1, r0, r19);
2132   // continue execution
2133   __ bind(continue_execution);
2134   __ rev32(r3, r3);
2135   __ add(rbcp, rbcp, r3, ext::sxtw);
2136   __ ldrb(rscratch1, Address(rbcp, 0));
2137   __ dispatch_only(vtos, /*generate_poll*/true);
2138 }
2139 
2140 void TemplateTable::fast_binaryswitch() {
2141   transition(itos, vtos);
2142   // Implementation using the following core algorithm:
2143   //
2144   // int binary_search(int key, LookupswitchPair* array, int n) {
2145   //   // Binary search according to "Methodik des Programmierens" by
2146   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2147   //   int i = 0;
2148   //   int j = n;
2149   //   while (i+1 < j) {
2150   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2151   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2152   //     // where a stands for the array and assuming that the (inexisting)
2153   //     // element a[n] is infinitely big.
2154   //     int h = (i + j) >> 1;
2155   //     // i < h < j
2156   //     if (key < array[h].fast_match()) {
2157   //       j = h;
2158   //     } else {
2159   //       i = h;
2160   //     }
2161   //   }
2162   //   // R: a[i] <= key < a[i+1] or Q
2163   //   // (i.e., if key is within array, i is the correct index)
2164   //   return i;
2165   // }
2166 
2167   // Register allocation
2168   const Register key   = r0; // already set (tosca)
2169   const Register array = r1;
2170   const Register i     = r2;
2171   const Register j     = r3;
2172   const Register h     = rscratch1;
2173   const Register temp  = rscratch2;
2174 
2175   // Find array start
2176   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2177                                           // get rid of this
2178                                           // instruction (change
2179                                           // offsets below)
2180   __ andr(array, array, -BytesPerInt);
2181 
2182   // Initialize i & j
2183   __ mov(i, 0);                            // i = 0;
2184   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2185 
2186   // Convert j into native byteordering
2187   __ rev32(j, j);
2188 
2189   // And start
2190   Label entry;
2191   __ b(entry);
2192 
2193   // binary search loop
2194   {
2195     Label loop;
2196     __ bind(loop);
2197     // int h = (i + j) >> 1;
2198     __ addw(h, i, j);                           // h = i + j;
2199     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2200     // if (key < array[h].fast_match()) {
2201     //   j = h;
2202     // } else {
2203     //   i = h;
2204     // }
2205     // Convert array[h].match to native byte-ordering before compare
2206     __ ldr(temp, Address(array, h, Address::lsl(3)));
2207     __ rev32(temp, temp);
2208     __ cmpw(key, temp);
2209     // j = h if (key <  array[h].fast_match())
2210     __ csel(j, h, j, Assembler::LT);
2211     // i = h if (key >= array[h].fast_match())
2212     __ csel(i, h, i, Assembler::GE);
2213     // while (i+1 < j)
2214     __ bind(entry);
2215     __ addw(h, i, 1);          // i+1
2216     __ cmpw(h, j);             // i+1 < j
2217     __ br(Assembler::LT, loop);
2218   }
2219 
2220   // end of binary search, result index is i (must check again!)
2221   Label default_case;
2222   // Convert array[i].match to native byte-ordering before compare
2223   __ ldr(temp, Address(array, i, Address::lsl(3)));
2224   __ rev32(temp, temp);
2225   __ cmpw(key, temp);
2226   __ br(Assembler::NE, default_case);
2227 
2228   // entry found -> j = offset
2229   __ add(j, array, i, ext::uxtx, 3);
2230   __ ldrw(j, Address(j, BytesPerInt));
2231   __ profile_switch_case(i, key, array);
2232   __ rev32(j, j);
2233   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2234   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2235   __ dispatch_only(vtos, /*generate_poll*/true);
2236 
2237   // default case -> j = default offset
2238   __ bind(default_case);
2239   __ profile_switch_default(i);
2240   __ ldrw(j, Address(array, -2 * BytesPerInt));
2241   __ rev32(j, j);
2242   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2243   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2244   __ dispatch_only(vtos, /*generate_poll*/true);
2245 }
2246 
2247 
2248 void TemplateTable::_return(TosState state)
2249 {
2250   transition(state, state);
2251   assert(_desc->calls_vm(),
2252          "inconsistent calls_vm information"); // call in remove_activation
2253 
2254   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2255     assert(state == vtos, "only valid state");
2256 
2257     __ ldr(c_rarg1, aaddress(0));
2258     __ load_klass(r3, c_rarg1);
2259     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2260     Label skip_register_finalizer;
2261     __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2262 
2263     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2264 
2265     __ bind(skip_register_finalizer);
2266   }
2267 
2268   // Issue a StoreStore barrier after all stores but before return
2269   // from any constructor for any class with a final field.  We don't
2270   // know if this is a finalizer, so we always do so.
2271   if (_desc->bytecode() == Bytecodes::_return)
2272     __ membar(MacroAssembler::StoreStore);
2273 
2274   // Narrow result if state is itos but result type is smaller.
2275   // Need to narrow in the return bytecode rather than in generate_return_entry
2276   // since compiled code callers expect the result to already be narrowed.
2277   if (state == itos) {
2278     __ narrow(r0);
2279   }
2280 
2281   __ remove_activation(state);
2282   __ ret(lr);
2283 }
2284 
2285 // ----------------------------------------------------------------------------
2286 // Volatile variables demand their effects be made known to all CPU's
2287 // in order.  Store buffers on most chips allow reads & writes to
2288 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2289 // without some kind of memory barrier (i.e., it's not sufficient that
2290 // the interpreter does not reorder volatile references, the hardware
2291 // also must not reorder them).
2292 //
2293 // According to the new Java Memory Model (JMM):
2294 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2295 //     writes act as aquire & release, so:
2296 // (2) A read cannot let unrelated NON-volatile memory refs that
2297 //     happen after the read float up to before the read.  It's OK for
2298 //     non-volatile memory refs that happen before the volatile read to
2299 //     float down below it.
2300 // (3) Similar a volatile write cannot let unrelated NON-volatile
2301 //     memory refs that happen BEFORE the write float down to after the
2302 //     write.  It's OK for non-volatile memory refs that happen after the
2303 //     volatile write to float up before it.
2304 //
2305 // We only put in barriers around volatile refs (they are expensive),
2306 // not _between_ memory refs (that would require us to track the
2307 // flavor of the previous memory refs).  Requirements (2) and (3)
2308 // require some barriers before volatile stores and after volatile
2309 // loads.  These nearly cover requirement (1) but miss the
2310 // volatile-store-volatile-load case.  This final case is placed after
2311 // volatile-stores although it could just as well go before
2312 // volatile-loads.
2313 
2314 void TemplateTable::resolve_cache_and_index(int byte_no,
2315                                             Register Rcache,
2316                                             Register index,
2317                                             size_t index_size) {
2318   const Register temp = r19;
2319   assert_different_registers(Rcache, index, temp);
2320 
2321   Label resolved, clinit_barrier_slow;
2322 
2323   Bytecodes::Code code = bytecode();
2324   switch (code) {
2325   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2326   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2327   default: break;
2328   }
2329 
2330   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2331   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2332   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2333   __ br(Assembler::EQ, resolved);
2334 
2335   // resolve first time through
2336   // Class initialization barrier slow path lands here as well.
2337   __ bind(clinit_barrier_slow);
2338   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2339   __ mov(temp, (int) code);
2340   __ call_VM(noreg, entry, temp);
2341 
2342   // Update registers with resolved info
2343   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2344   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2345   // so all clients ofthis method must be modified accordingly
2346   __ bind(resolved);
2347 
2348   // Class initialization barrier for static methods
2349   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2350     __ load_resolved_method_at_index(byte_no, temp, Rcache);
2351     __ load_method_holder(temp, temp);
2352     __ clinit_barrier(temp, rscratch1, NULL, &clinit_barrier_slow);
2353   }
2354 }
2355 
2356 // The Rcache and index registers must be set before call
2357 // n.b unlike x86 cache already includes the index offset
2358 void TemplateTable::load_field_cp_cache_entry(Register obj,
2359                                               Register cache,
2360                                               Register index,
2361                                               Register off,
2362                                               Register flags,
2363                                               bool is_static = false) {
2364   assert_different_registers(cache, index, flags, off);
2365 
2366   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2367   // Field offset
2368   __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2369                                           ConstantPoolCacheEntry::f2_offset())));
2370   // Flags
2371   __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2372                                            ConstantPoolCacheEntry::flags_offset())));
2373 
2374   // klass overwrite register
2375   if (is_static) {
2376     __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2377                                         ConstantPoolCacheEntry::f1_offset())));
2378     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2379     __ ldr(obj, Address(obj, mirror_offset));
2380     __ resolve_oop_handle(obj);
2381   }
2382 }
2383 
2384 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2385                                                Register method,
2386                                                Register itable_index,
2387                                                Register flags,
2388                                                bool is_invokevirtual,
2389                                                bool is_invokevfinal, /*unused*/
2390                                                bool is_invokedynamic) {
2391   // setup registers
2392   const Register cache = rscratch2;
2393   const Register index = r4;
2394   assert_different_registers(method, flags);
2395   assert_different_registers(method, cache, index);
2396   assert_different_registers(itable_index, flags);
2397   assert_different_registers(itable_index, cache, index);
2398   // determine constant pool cache field offsets
2399   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2400   const int method_offset = in_bytes(
2401     ConstantPoolCache::base_offset() +
2402       (is_invokevirtual
2403        ? ConstantPoolCacheEntry::f2_offset()
2404        : ConstantPoolCacheEntry::f1_offset()));
2405   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2406                                     ConstantPoolCacheEntry::flags_offset());
2407   // access constant pool cache fields
2408   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2409                                     ConstantPoolCacheEntry::f2_offset());
2410 
2411   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2412   resolve_cache_and_index(byte_no, cache, index, index_size);
2413   __ ldr(method, Address(cache, method_offset));
2414 
2415   if (itable_index != noreg) {
2416     __ ldr(itable_index, Address(cache, index_offset));
2417   }
2418   __ ldrw(flags, Address(cache, flags_offset));
2419 }
2420 
2421 
2422 // The registers cache and index expected to be set before call.
2423 // Correct values of the cache and index registers are preserved.
2424 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2425                                             bool is_static, bool has_tos) {
2426   // do the JVMTI work here to avoid disturbing the register state below
2427   // We use c_rarg registers here because we want to use the register used in
2428   // the call to the VM
2429   if (JvmtiExport::can_post_field_access()) {
2430     // Check to see if a field access watch has been set before we
2431     // take the time to call into the VM.
2432     Label L1;
2433     assert_different_registers(cache, index, r0);
2434     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2435     __ ldrw(r0, Address(rscratch1));
2436     __ cbzw(r0, L1);
2437 
2438     __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2439     __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2440 
2441     if (is_static) {
2442       __ mov(c_rarg1, zr); // NULL object reference
2443     } else {
2444       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2445       __ verify_oop(c_rarg1);
2446     }
2447     // c_rarg1: object pointer or NULL
2448     // c_rarg2: cache entry pointer
2449     // c_rarg3: jvalue object on the stack
2450     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2451                                        InterpreterRuntime::post_field_access),
2452                c_rarg1, c_rarg2, c_rarg3);
2453     __ get_cache_and_index_at_bcp(cache, index, 1);
2454     __ bind(L1);
2455   }
2456 }
2457 
2458 void TemplateTable::pop_and_check_object(Register r)
2459 {
2460   __ pop_ptr(r);
2461   __ null_check(r);  // for field access must check obj.
2462   __ verify_oop(r);
2463 }
2464 
2465 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2466 {
2467   const Register cache = r2;
2468   const Register index = r3;
2469   const Register obj   = r4;
2470   const Register off   = r19;
2471   const Register flags = r0;
2472   const Register raw_flags = r6;
2473   const Register bc    = r4; // uses same reg as obj, so don't mix them
2474 
2475   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2476   jvmti_post_field_access(cache, index, is_static, false);
2477   load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2478 
2479   if (!is_static) {
2480     // obj is on the stack
2481     pop_and_check_object(obj);
2482   }
2483 
2484   // 8179954: We need to make sure that the code generated for
2485   // volatile accesses forms a sequentially-consistent set of
2486   // operations when combined with STLR and LDAR.  Without a leading
2487   // membar it's possible for a simple Dekker test to fail if loads
2488   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2489   // the stores in one method and we interpret the loads in another.
2490   if (!is_c1_or_interpreter_only()){
2491     Label notVolatile;
2492     __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2493     __ membar(MacroAssembler::AnyAny);
2494     __ bind(notVolatile);
2495   }
2496 
2497   const Address field(obj, off);
2498 
2499   Label Done, notByte, notBool, notInt, notShort, notChar,
2500               notLong, notFloat, notObj, notDouble;
2501 
2502   // x86 uses a shift and mask or wings it with a shift plus assert
2503   // the mask is not needed. aarch64 just uses bitfield extract
2504   __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2505            ConstantPoolCacheEntry::tos_state_bits);
2506 
2507   assert(btos == 0, "change code, btos != 0");
2508   __ cbnz(flags, notByte);
2509 
2510   // Don't rewrite getstatic, only getfield
2511   if (is_static) rc = may_not_rewrite;
2512 
2513   // btos
2514   __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
2515   __ push(btos);
2516   // Rewrite bytecode to be faster
2517   if (rc == may_rewrite) {
2518     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2519   }
2520   __ b(Done);
2521 
2522   __ bind(notByte);
2523   __ cmp(flags, (u1)ztos);
2524   __ br(Assembler::NE, notBool);
2525 
2526   // ztos (same code as btos)
2527   __ access_load_at(T_BOOLEAN, IN_HEAP, r0, field, noreg, noreg);
2528   __ push(ztos);
2529   // Rewrite bytecode to be faster
2530   if (rc == may_rewrite) {
2531     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2532     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2533   }
2534   __ b(Done);
2535 
2536   __ bind(notBool);
2537   __ cmp(flags, (u1)atos);
2538   __ br(Assembler::NE, notObj);
2539   // atos
2540   do_oop_load(_masm, field, r0, IN_HEAP);
2541   __ push(atos);
2542   if (rc == may_rewrite) {
2543     patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2544   }
2545   __ b(Done);
2546 
2547   __ bind(notObj);
2548   __ cmp(flags, (u1)itos);
2549   __ br(Assembler::NE, notInt);
2550   // itos
2551   __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
2552   __ push(itos);
2553   // Rewrite bytecode to be faster
2554   if (rc == may_rewrite) {
2555     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2556   }
2557   __ b(Done);
2558 
2559   __ bind(notInt);
2560   __ cmp(flags, (u1)ctos);
2561   __ br(Assembler::NE, notChar);
2562   // ctos
2563   __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
2564   __ push(ctos);
2565   // Rewrite bytecode to be faster
2566   if (rc == may_rewrite) {
2567     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2568   }
2569   __ b(Done);
2570 
2571   __ bind(notChar);
2572   __ cmp(flags, (u1)stos);
2573   __ br(Assembler::NE, notShort);
2574   // stos
2575   __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
2576   __ push(stos);
2577   // Rewrite bytecode to be faster
2578   if (rc == may_rewrite) {
2579     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2580   }
2581   __ b(Done);
2582 
2583   __ bind(notShort);
2584   __ cmp(flags, (u1)ltos);
2585   __ br(Assembler::NE, notLong);
2586   // ltos
2587   __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
2588   __ push(ltos);
2589   // Rewrite bytecode to be faster
2590   if (rc == may_rewrite) {
2591     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2592   }
2593   __ b(Done);
2594 
2595   __ bind(notLong);
2596   __ cmp(flags, (u1)ftos);
2597   __ br(Assembler::NE, notFloat);
2598   // ftos
2599   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2600   __ push(ftos);
2601   // Rewrite bytecode to be faster
2602   if (rc == may_rewrite) {
2603     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2604   }
2605   __ b(Done);
2606 
2607   __ bind(notFloat);
2608 #ifdef ASSERT
2609   __ cmp(flags, (u1)dtos);
2610   __ br(Assembler::NE, notDouble);
2611 #endif
2612   // dtos
2613   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2614   __ push(dtos);
2615   // Rewrite bytecode to be faster
2616   if (rc == may_rewrite) {
2617     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2618   }
2619 #ifdef ASSERT
2620   __ b(Done);
2621 
2622   __ bind(notDouble);
2623   __ stop("Bad state");
2624 #endif
2625 
2626   __ bind(Done);
2627 
2628   Label notVolatile;
2629   __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2630   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2631   __ bind(notVolatile);
2632 }
2633 
2634 
2635 void TemplateTable::getfield(int byte_no)
2636 {
2637   getfield_or_static(byte_no, false);
2638 }
2639 
2640 void TemplateTable::nofast_getfield(int byte_no) {
2641   getfield_or_static(byte_no, false, may_not_rewrite);
2642 }
2643 
2644 void TemplateTable::getstatic(int byte_no)
2645 {
2646   getfield_or_static(byte_no, true);
2647 }
2648 
2649 // The registers cache and index expected to be set before call.
2650 // The function may destroy various registers, just not the cache and index registers.
2651 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2652   transition(vtos, vtos);
2653 
2654   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2655 
2656   if (JvmtiExport::can_post_field_modification()) {
2657     // Check to see if a field modification watch has been set before
2658     // we take the time to call into the VM.
2659     Label L1;
2660     assert_different_registers(cache, index, r0);
2661     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2662     __ ldrw(r0, Address(rscratch1));
2663     __ cbz(r0, L1);
2664 
2665     __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2666 
2667     if (is_static) {
2668       // Life is simple.  Null out the object pointer.
2669       __ mov(c_rarg1, zr);
2670     } else {
2671       // Life is harder. The stack holds the value on top, followed by
2672       // the object.  We don't know the size of the value, though; it
2673       // could be one or two words depending on its type. As a result,
2674       // we must find the type to determine where the object is.
2675       __ ldrw(c_rarg3, Address(c_rarg2,
2676                                in_bytes(cp_base_offset +
2677                                         ConstantPoolCacheEntry::flags_offset())));
2678       __ lsr(c_rarg3, c_rarg3,
2679              ConstantPoolCacheEntry::tos_state_shift);
2680       ConstantPoolCacheEntry::verify_tos_state_shift();
2681       Label nope2, done, ok;
2682       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2683       __ cmpw(c_rarg3, ltos);
2684       __ br(Assembler::EQ, ok);
2685       __ cmpw(c_rarg3, dtos);
2686       __ br(Assembler::NE, nope2);
2687       __ bind(ok);
2688       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2689       __ bind(nope2);
2690     }
2691     // cache entry pointer
2692     __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2693     // object (tos)
2694     __ mov(c_rarg3, esp);
2695     // c_rarg1: object pointer set up above (NULL if static)
2696     // c_rarg2: cache entry pointer
2697     // c_rarg3: jvalue object on the stack
2698     __ call_VM(noreg,
2699                CAST_FROM_FN_PTR(address,
2700                                 InterpreterRuntime::post_field_modification),
2701                c_rarg1, c_rarg2, c_rarg3);
2702     __ get_cache_and_index_at_bcp(cache, index, 1);
2703     __ bind(L1);
2704   }
2705 }
2706 
2707 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2708   transition(vtos, vtos);
2709 
2710   const Register cache = r2;
2711   const Register index = r3;
2712   const Register obj   = r2;
2713   const Register off   = r19;
2714   const Register flags = r0;
2715   const Register bc    = r4;
2716 
2717   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2718   jvmti_post_field_mod(cache, index, is_static);
2719   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2720 
2721   Label Done;
2722   __ mov(r5, flags);
2723 
2724   {
2725     Label notVolatile;
2726     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2727     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2728     __ bind(notVolatile);
2729   }
2730 
2731   // field address
2732   const Address field(obj, off);
2733 
2734   Label notByte, notBool, notInt, notShort, notChar,
2735         notLong, notFloat, notObj, notDouble;
2736 
2737   // x86 uses a shift and mask or wings it with a shift plus assert
2738   // the mask is not needed. aarch64 just uses bitfield extract
2739   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2740 
2741   assert(btos == 0, "change code, btos != 0");
2742   __ cbnz(flags, notByte);
2743 
2744   // Don't rewrite putstatic, only putfield
2745   if (is_static) rc = may_not_rewrite;
2746 
2747   // btos
2748   {
2749     __ pop(btos);
2750     if (!is_static) pop_and_check_object(obj);
2751     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg);
2752     if (rc == may_rewrite) {
2753       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2754     }
2755     __ b(Done);
2756   }
2757 
2758   __ bind(notByte);
2759   __ cmp(flags, (u1)ztos);
2760   __ br(Assembler::NE, notBool);
2761 
2762   // ztos
2763   {
2764     __ pop(ztos);
2765     if (!is_static) pop_and_check_object(obj);
2766     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg);
2767     if (rc == may_rewrite) {
2768       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2769     }
2770     __ b(Done);
2771   }
2772 
2773   __ bind(notBool);
2774   __ cmp(flags, (u1)atos);
2775   __ br(Assembler::NE, notObj);
2776 
2777   // atos
2778   {
2779     __ pop(atos);
2780     if (!is_static) pop_and_check_object(obj);
2781     // Store into the field
2782     do_oop_store(_masm, field, r0, IN_HEAP);
2783     if (rc == may_rewrite) {
2784       patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2785     }
2786     __ b(Done);
2787   }
2788 
2789   __ bind(notObj);
2790   __ cmp(flags, (u1)itos);
2791   __ br(Assembler::NE, notInt);
2792 
2793   // itos
2794   {
2795     __ pop(itos);
2796     if (!is_static) pop_and_check_object(obj);
2797     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg);
2798     if (rc == may_rewrite) {
2799       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2800     }
2801     __ b(Done);
2802   }
2803 
2804   __ bind(notInt);
2805   __ cmp(flags, (u1)ctos);
2806   __ br(Assembler::NE, notChar);
2807 
2808   // ctos
2809   {
2810     __ pop(ctos);
2811     if (!is_static) pop_and_check_object(obj);
2812     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg);
2813     if (rc == may_rewrite) {
2814       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2815     }
2816     __ b(Done);
2817   }
2818 
2819   __ bind(notChar);
2820   __ cmp(flags, (u1)stos);
2821   __ br(Assembler::NE, notShort);
2822 
2823   // stos
2824   {
2825     __ pop(stos);
2826     if (!is_static) pop_and_check_object(obj);
2827     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg);
2828     if (rc == may_rewrite) {
2829       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2830     }
2831     __ b(Done);
2832   }
2833 
2834   __ bind(notShort);
2835   __ cmp(flags, (u1)ltos);
2836   __ br(Assembler::NE, notLong);
2837 
2838   // ltos
2839   {
2840     __ pop(ltos);
2841     if (!is_static) pop_and_check_object(obj);
2842     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg);
2843     if (rc == may_rewrite) {
2844       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2845     }
2846     __ b(Done);
2847   }
2848 
2849   __ bind(notLong);
2850   __ cmp(flags, (u1)ftos);
2851   __ br(Assembler::NE, notFloat);
2852 
2853   // ftos
2854   {
2855     __ pop(ftos);
2856     if (!is_static) pop_and_check_object(obj);
2857     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
2858     if (rc == may_rewrite) {
2859       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2860     }
2861     __ b(Done);
2862   }
2863 
2864   __ bind(notFloat);
2865 #ifdef ASSERT
2866   __ cmp(flags, (u1)dtos);
2867   __ br(Assembler::NE, notDouble);
2868 #endif
2869 
2870   // dtos
2871   {
2872     __ pop(dtos);
2873     if (!is_static) pop_and_check_object(obj);
2874     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
2875     if (rc == may_rewrite) {
2876       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2877     }
2878   }
2879 
2880 #ifdef ASSERT
2881   __ b(Done);
2882 
2883   __ bind(notDouble);
2884   __ stop("Bad state");
2885 #endif
2886 
2887   __ bind(Done);
2888 
2889   {
2890     Label notVolatile;
2891     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2892     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2893     __ bind(notVolatile);
2894   }
2895 }
2896 
2897 void TemplateTable::putfield(int byte_no)
2898 {
2899   putfield_or_static(byte_no, false);
2900 }
2901 
2902 void TemplateTable::nofast_putfield(int byte_no) {
2903   putfield_or_static(byte_no, false, may_not_rewrite);
2904 }
2905 
2906 void TemplateTable::putstatic(int byte_no) {
2907   putfield_or_static(byte_no, true);
2908 }
2909 
2910 void TemplateTable::jvmti_post_fast_field_mod()
2911 {
2912   if (JvmtiExport::can_post_field_modification()) {
2913     // Check to see if a field modification watch has been set before
2914     // we take the time to call into the VM.
2915     Label L2;
2916     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2917     __ ldrw(c_rarg3, Address(rscratch1));
2918     __ cbzw(c_rarg3, L2);
2919     __ pop_ptr(r19);                  // copy the object pointer from tos
2920     __ verify_oop(r19);
2921     __ push_ptr(r19);                 // put the object pointer back on tos
2922     // Save tos values before call_VM() clobbers them. Since we have
2923     // to do it for every data type, we use the saved values as the
2924     // jvalue object.
2925     switch (bytecode()) {          // load values into the jvalue object
2926     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2927     case Bytecodes::_fast_bputfield: // fall through
2928     case Bytecodes::_fast_zputfield: // fall through
2929     case Bytecodes::_fast_sputfield: // fall through
2930     case Bytecodes::_fast_cputfield: // fall through
2931     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2932     case Bytecodes::_fast_dputfield: __ push_d(); break;
2933     case Bytecodes::_fast_fputfield: __ push_f(); break;
2934     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2935 
2936     default:
2937       ShouldNotReachHere();
2938     }
2939     __ mov(c_rarg3, esp);             // points to jvalue on the stack
2940     // access constant pool cache entry
2941     __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2942     __ verify_oop(r19);
2943     // r19: object pointer copied above
2944     // c_rarg2: cache entry pointer
2945     // c_rarg3: jvalue object on the stack
2946     __ call_VM(noreg,
2947                CAST_FROM_FN_PTR(address,
2948                                 InterpreterRuntime::post_field_modification),
2949                r19, c_rarg2, c_rarg3);
2950 
2951     switch (bytecode()) {             // restore tos values
2952     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2953     case Bytecodes::_fast_bputfield: // fall through
2954     case Bytecodes::_fast_zputfield: // fall through
2955     case Bytecodes::_fast_sputfield: // fall through
2956     case Bytecodes::_fast_cputfield: // fall through
2957     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2958     case Bytecodes::_fast_dputfield: __ pop_d(); break;
2959     case Bytecodes::_fast_fputfield: __ pop_f(); break;
2960     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2961     default: break;
2962     }
2963     __ bind(L2);
2964   }
2965 }
2966 
2967 void TemplateTable::fast_storefield(TosState state)
2968 {
2969   transition(state, vtos);
2970 
2971   ByteSize base = ConstantPoolCache::base_offset();
2972 
2973   jvmti_post_fast_field_mod();
2974 
2975   // access constant pool cache
2976   __ get_cache_and_index_at_bcp(r2, r1, 1);
2977 
2978   // test for volatile with r3
2979   __ ldrw(r3, Address(r2, in_bytes(base +
2980                                    ConstantPoolCacheEntry::flags_offset())));
2981 
2982   // replace index with field offset from cache entry
2983   __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2984 
2985   {
2986     Label notVolatile;
2987     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2988     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2989     __ bind(notVolatile);
2990   }
2991 
2992   Label notVolatile;
2993 
2994   // Get object from stack
2995   pop_and_check_object(r2);
2996 
2997   // field address
2998   const Address field(r2, r1);
2999 
3000   // access field
3001   switch (bytecode()) {
3002   case Bytecodes::_fast_aputfield:
3003     do_oop_store(_masm, field, r0, IN_HEAP);
3004     break;
3005   case Bytecodes::_fast_lputfield:
3006     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg);
3007     break;
3008   case Bytecodes::_fast_iputfield:
3009     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg);
3010     break;
3011   case Bytecodes::_fast_zputfield:
3012     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg);
3013     break;
3014   case Bytecodes::_fast_bputfield:
3015     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg);
3016     break;
3017   case Bytecodes::_fast_sputfield:
3018     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg);
3019     break;
3020   case Bytecodes::_fast_cputfield:
3021     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg);
3022     break;
3023   case Bytecodes::_fast_fputfield:
3024     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3025     break;
3026   case Bytecodes::_fast_dputfield:
3027     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
3028     break;
3029   default:
3030     ShouldNotReachHere();
3031   }
3032 
3033   {
3034     Label notVolatile;
3035     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3036     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3037     __ bind(notVolatile);
3038   }
3039 }
3040 
3041 
3042 void TemplateTable::fast_accessfield(TosState state)
3043 {
3044   transition(atos, state);
3045   // Do the JVMTI work here to avoid disturbing the register state below
3046   if (JvmtiExport::can_post_field_access()) {
3047     // Check to see if a field access watch has been set before we
3048     // take the time to call into the VM.
3049     Label L1;
3050     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3051     __ ldrw(r2, Address(rscratch1));
3052     __ cbzw(r2, L1);
3053     // access constant pool cache entry
3054     __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
3055     __ verify_oop(r0);
3056     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
3057     __ mov(c_rarg1, r0);
3058     // c_rarg1: object pointer copied above
3059     // c_rarg2: cache entry pointer
3060     __ call_VM(noreg,
3061                CAST_FROM_FN_PTR(address,
3062                                 InterpreterRuntime::post_field_access),
3063                c_rarg1, c_rarg2);
3064     __ pop_ptr(r0); // restore object pointer
3065     __ bind(L1);
3066   }
3067 
3068   // access constant pool cache
3069   __ get_cache_and_index_at_bcp(r2, r1, 1);
3070   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3071                                   ConstantPoolCacheEntry::f2_offset())));
3072   __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3073                                    ConstantPoolCacheEntry::flags_offset())));
3074 
3075   // r0: object
3076   __ verify_oop(r0);
3077   __ null_check(r0);
3078   const Address field(r0, r1);
3079 
3080   // 8179954: We need to make sure that the code generated for
3081   // volatile accesses forms a sequentially-consistent set of
3082   // operations when combined with STLR and LDAR.  Without a leading
3083   // membar it's possible for a simple Dekker test to fail if loads
3084   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3085   // the stores in one method and we interpret the loads in another.
3086   if (!is_c1_or_interpreter_only()) {
3087     Label notVolatile;
3088     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3089     __ membar(MacroAssembler::AnyAny);
3090     __ bind(notVolatile);
3091   }
3092 
3093   // access field
3094   switch (bytecode()) {
3095   case Bytecodes::_fast_agetfield:
3096     do_oop_load(_masm, field, r0, IN_HEAP);
3097     __ verify_oop(r0);
3098     break;
3099   case Bytecodes::_fast_lgetfield:
3100     __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
3101     break;
3102   case Bytecodes::_fast_igetfield:
3103     __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
3104     break;
3105   case Bytecodes::_fast_bgetfield:
3106     __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
3107     break;
3108   case Bytecodes::_fast_sgetfield:
3109     __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
3110     break;
3111   case Bytecodes::_fast_cgetfield:
3112     __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
3113     break;
3114   case Bytecodes::_fast_fgetfield:
3115     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3116     break;
3117   case Bytecodes::_fast_dgetfield:
3118     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3119     break;
3120   default:
3121     ShouldNotReachHere();
3122   }
3123   {
3124     Label notVolatile;
3125     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3126     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3127     __ bind(notVolatile);
3128   }
3129 }
3130 
3131 void TemplateTable::fast_xaccess(TosState state)
3132 {
3133   transition(vtos, state);
3134 
3135   // get receiver
3136   __ ldr(r0, aaddress(0));
3137   // access constant pool cache
3138   __ get_cache_and_index_at_bcp(r2, r3, 2);
3139   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3140                                   ConstantPoolCacheEntry::f2_offset())));
3141 
3142   // 8179954: We need to make sure that the code generated for
3143   // volatile accesses forms a sequentially-consistent set of
3144   // operations when combined with STLR and LDAR.  Without a leading
3145   // membar it's possible for a simple Dekker test to fail if loads
3146   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3147   // the stores in one method and we interpret the loads in another.
3148   if (!is_c1_or_interpreter_only()) {
3149     Label notVolatile;
3150     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3151                                      ConstantPoolCacheEntry::flags_offset())));
3152     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3153     __ membar(MacroAssembler::AnyAny);
3154     __ bind(notVolatile);
3155   }
3156 
3157   // make sure exception is reported in correct bcp range (getfield is
3158   // next instruction)
3159   __ increment(rbcp);
3160   __ null_check(r0);
3161   switch (state) {
3162   case itos:
3163     __ access_load_at(T_INT, IN_HEAP, r0, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3164     break;
3165   case atos:
3166     do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3167     __ verify_oop(r0);
3168     break;
3169   case ftos:
3170     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3171     break;
3172   default:
3173     ShouldNotReachHere();
3174   }
3175 
3176   {
3177     Label notVolatile;
3178     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3179                                      ConstantPoolCacheEntry::flags_offset())));
3180     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3181     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3182     __ bind(notVolatile);
3183   }
3184 
3185   __ decrement(rbcp);
3186 }
3187 
3188 
3189 
3190 //-----------------------------------------------------------------------------
3191 // Calls
3192 
3193 void TemplateTable::count_calls(Register method, Register temp)
3194 {
3195   __ call_Unimplemented();
3196 }
3197 
3198 void TemplateTable::prepare_invoke(int byte_no,
3199                                    Register method, // linked method (or i-klass)
3200                                    Register index,  // itable index, MethodType, etc.
3201                                    Register recv,   // if caller wants to see it
3202                                    Register flags   // if caller wants to test it
3203                                    ) {
3204   // determine flags
3205   Bytecodes::Code code = bytecode();
3206   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3207   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3208   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3209   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3210   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3211   const bool load_receiver       = (recv  != noreg);
3212   const bool save_flags          = (flags != noreg);
3213   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3214   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3215   assert(flags == noreg || flags == r3, "");
3216   assert(recv  == noreg || recv  == r2, "");
3217 
3218   // setup registers & access constant pool cache
3219   if (recv  == noreg)  recv  = r2;
3220   if (flags == noreg)  flags = r3;
3221   assert_different_registers(method, index, recv, flags);
3222 
3223   // save 'interpreter return address'
3224   __ save_bcp();
3225 
3226   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3227 
3228   // maybe push appendix to arguments (just before return address)
3229   if (is_invokedynamic || is_invokehandle) {
3230     Label L_no_push;
3231     __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3232     // Push the appendix as a trailing parameter.
3233     // This must be done before we get the receiver,
3234     // since the parameter_size includes it.
3235     __ push(r19);
3236     __ mov(r19, index);
3237     __ load_resolved_reference_at_index(index, r19);
3238     __ pop(r19);
3239     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3240     __ bind(L_no_push);
3241   }
3242 
3243   // load receiver if needed (note: no return address pushed yet)
3244   if (load_receiver) {
3245     __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3246     // FIXME -- is this actually correct? looks like it should be 2
3247     // const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3248     // const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3249     // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3250     // __ movptr(recv, recv_addr);
3251     __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3252     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3253     __ verify_oop(recv);
3254   }
3255 
3256   // compute return type
3257   // x86 uses a shift and mask or wings it with a shift plus assert
3258   // the mask is not needed. aarch64 just uses bitfield extract
3259   __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
3260   // load return address
3261   {
3262     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3263     __ mov(rscratch1, table_addr);
3264     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3265   }
3266 }
3267 
3268 
3269 void TemplateTable::invokevirtual_helper(Register index,
3270                                          Register recv,
3271                                          Register flags)
3272 {
3273   // Uses temporary registers r0, r3
3274   assert_different_registers(index, recv, r0, r3);
3275   // Test for an invoke of a final method
3276   Label notFinal;
3277   __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3278 
3279   const Register method = index;  // method must be rmethod
3280   assert(method == rmethod,
3281          "methodOop must be rmethod for interpreter calling convention");
3282 
3283   // do the call - the index is actually the method to call
3284   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3285 
3286   // It's final, need a null check here!
3287   __ null_check(recv);
3288 
3289   // profile this call
3290   __ profile_final_call(r0);
3291   __ profile_arguments_type(r0, method, r4, true);
3292 
3293   __ jump_from_interpreted(method, r0);
3294 
3295   __ bind(notFinal);
3296 
3297   // get receiver klass
3298   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3299   __ load_klass(r0, recv);
3300 
3301   // profile this call
3302   __ profile_virtual_call(r0, rlocals, r3);
3303 
3304   // get target methodOop & entry point
3305   __ lookup_virtual_method(r0, index, method);
3306   __ profile_arguments_type(r3, method, r4, true);
3307   // FIXME -- this looks completely redundant. is it?
3308   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3309   __ jump_from_interpreted(method, r3);
3310 }
3311 
3312 void TemplateTable::invokevirtual(int byte_no)
3313 {
3314   transition(vtos, vtos);
3315   assert(byte_no == f2_byte, "use this argument");
3316 
3317   prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3318 
3319   // rmethod: index (actually a Method*)
3320   // r2: receiver
3321   // r3: flags
3322 
3323   invokevirtual_helper(rmethod, r2, r3);
3324 }
3325 
3326 void TemplateTable::invokespecial(int byte_no)
3327 {
3328   transition(vtos, vtos);
3329   assert(byte_no == f1_byte, "use this argument");
3330 
3331   prepare_invoke(byte_no, rmethod, noreg,  // get f1 Method*
3332                  r2);  // get receiver also for null check
3333   __ verify_oop(r2);
3334   __ null_check(r2);
3335   // do the call
3336   __ profile_call(r0);
3337   __ profile_arguments_type(r0, rmethod, rbcp, false);
3338   __ jump_from_interpreted(rmethod, r0);
3339 }
3340 
3341 void TemplateTable::invokestatic(int byte_no)
3342 {
3343   transition(vtos, vtos);
3344   assert(byte_no == f1_byte, "use this argument");
3345 
3346   prepare_invoke(byte_no, rmethod);  // get f1 Method*
3347   // do the call
3348   __ profile_call(r0);
3349   __ profile_arguments_type(r0, rmethod, r4, false);
3350   __ jump_from_interpreted(rmethod, r0);
3351 }
3352 
3353 void TemplateTable::fast_invokevfinal(int byte_no)
3354 {
3355   __ call_Unimplemented();
3356 }
3357 
3358 void TemplateTable::invokeinterface(int byte_no) {
3359   transition(vtos, vtos);
3360   assert(byte_no == f1_byte, "use this argument");
3361 
3362   prepare_invoke(byte_no, r0, rmethod,  // get f1 Klass*, f2 Method*
3363                  r2, r3); // recv, flags
3364 
3365   // r0: interface klass (from f1)
3366   // rmethod: method (from f2)
3367   // r2: receiver
3368   // r3: flags
3369 
3370   // First check for Object case, then private interface method,
3371   // then regular interface method.
3372 
3373   // Special case of invokeinterface called for virtual method of
3374   // java.lang.Object.  See cpCache.cpp for details.
3375   Label notObjectMethod;
3376   __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notObjectMethod);
3377 
3378   invokevirtual_helper(rmethod, r2, r3);
3379   __ bind(notObjectMethod);
3380 
3381   Label no_such_interface;
3382 
3383   // Check for private method invocation - indicated by vfinal
3384   Label notVFinal;
3385   __ tbz(r3, ConstantPoolCacheEntry::is_vfinal_shift, notVFinal);
3386 
3387   // Get receiver klass into r3 - also a null check
3388   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3389   __ load_klass(r3, r2);
3390 
3391   Label subtype;
3392   __ check_klass_subtype(r3, r0, r4, subtype);
3393   // If we get here the typecheck failed
3394   __ b(no_such_interface);
3395   __ bind(subtype);
3396 
3397   __ profile_final_call(r0);
3398   __ profile_arguments_type(r0, rmethod, r4, true);
3399   __ jump_from_interpreted(rmethod, r0);
3400 
3401   __ bind(notVFinal);
3402 
3403   // Get receiver klass into r3 - also a null check
3404   __ restore_locals();
3405   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3406   __ load_klass(r3, r2);
3407 
3408   Label no_such_method;
3409 
3410   // Preserve method for throw_AbstractMethodErrorVerbose.
3411   __ mov(r16, rmethod);
3412   // Receiver subtype check against REFC.
3413   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3414   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3415                              r3, r0, noreg,
3416                              // outputs: scan temp. reg, scan temp. reg
3417                              rscratch2, r13,
3418                              no_such_interface,
3419                              /*return_method=*/false);
3420 
3421   // profile this call
3422   __ profile_virtual_call(r3, r13, r19);
3423 
3424   // Get declaring interface class from method, and itable index
3425 
3426   __ load_method_holder(r0, rmethod);
3427   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3428   __ subw(rmethod, rmethod, Method::itable_index_max);
3429   __ negw(rmethod, rmethod);
3430 
3431   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3432   __ mov(rlocals, r3);
3433   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3434                              rlocals, r0, rmethod,
3435                              // outputs: method, scan temp. reg
3436                              rmethod, r13,
3437                              no_such_interface);
3438 
3439   // rmethod,: methodOop to call
3440   // r2: receiver
3441   // Check for abstract method error
3442   // Note: This should be done more efficiently via a throw_abstract_method_error
3443   //       interpreter entry point and a conditional jump to it in case of a null
3444   //       method.
3445   __ cbz(rmethod, no_such_method);
3446 
3447   __ profile_arguments_type(r3, rmethod, r13, true);
3448 
3449   // do the call
3450   // r2: receiver
3451   // rmethod,: methodOop
3452   __ jump_from_interpreted(rmethod, r3);
3453   __ should_not_reach_here();
3454 
3455   // exception handling code follows...
3456   // note: must restore interpreter registers to canonical
3457   //       state for exception handling to work correctly!
3458 
3459   __ bind(no_such_method);
3460   // throw exception
3461   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3462   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3463   // Pass arguments for generating a verbose error message.
3464   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3465   // the call_VM checks for exception, so we should never return here.
3466   __ should_not_reach_here();
3467 
3468   __ bind(no_such_interface);
3469   // throw exception
3470   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3471   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3472   // Pass arguments for generating a verbose error message.
3473   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3474                    InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3475   // the call_VM checks for exception, so we should never return here.
3476   __ should_not_reach_here();
3477   return;
3478 }
3479 
3480 void TemplateTable::invokehandle(int byte_no) {
3481   transition(vtos, vtos);
3482   assert(byte_no == f1_byte, "use this argument");
3483 
3484   prepare_invoke(byte_no, rmethod, r0, r2);
3485   __ verify_method_ptr(r2);
3486   __ verify_oop(r2);
3487   __ null_check(r2);
3488 
3489   // FIXME: profile the LambdaForm also
3490 
3491   // r13 is safe to use here as a scratch reg because it is about to
3492   // be clobbered by jump_from_interpreted().
3493   __ profile_final_call(r13);
3494   __ profile_arguments_type(r13, rmethod, r4, true);
3495 
3496   __ jump_from_interpreted(rmethod, r0);
3497 }
3498 
3499 void TemplateTable::invokedynamic(int byte_no) {
3500   transition(vtos, vtos);
3501   assert(byte_no == f1_byte, "use this argument");
3502 
3503   prepare_invoke(byte_no, rmethod, r0);
3504 
3505   // r0: CallSite object (from cpool->resolved_references[])
3506   // rmethod: MH.linkToCallSite method (from f2)
3507 
3508   // Note:  r0_callsite is already pushed by prepare_invoke
3509 
3510   // %%% should make a type profile for any invokedynamic that takes a ref argument
3511   // profile this call
3512   __ profile_call(rbcp);
3513   __ profile_arguments_type(r3, rmethod, r13, false);
3514 
3515   __ verify_oop(r0);
3516 
3517   __ jump_from_interpreted(rmethod, r0);
3518 }
3519 
3520 
3521 //-----------------------------------------------------------------------------
3522 // Allocation
3523 
3524 void TemplateTable::_new() {
3525   transition(vtos, atos);
3526 
3527   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3528   Label slow_case;
3529   Label done;
3530   Label initialize_header;
3531   Label initialize_object; // including clearing the fields
3532 
3533   __ get_cpool_and_tags(r4, r0);
3534   // Make sure the class we're about to instantiate has been resolved.
3535   // This is done before loading InstanceKlass to be consistent with the order
3536   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3537   const int tags_offset = Array<u1>::base_offset_in_bytes();
3538   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3539   __ lea(rscratch1, Address(rscratch1, tags_offset));
3540   __ ldarb(rscratch1, rscratch1);
3541   __ cmp(rscratch1, (u1)JVM_CONSTANT_Class);
3542   __ br(Assembler::NE, slow_case);
3543 
3544   // get InstanceKlass
3545   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3546 
3547   // make sure klass is initialized & doesn't have finalizer
3548   // make sure klass is fully initialized
3549   __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3550   __ cmp(rscratch1, (u1)InstanceKlass::fully_initialized);
3551   __ br(Assembler::NE, slow_case);
3552 
3553   // get instance_size in InstanceKlass (scaled to a count of bytes)
3554   __ ldrw(r3,
3555           Address(r4,
3556                   Klass::layout_helper_offset()));
3557   // test to see if it has a finalizer or is malformed in some way
3558   __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3559 
3560   // Allocate the instance:
3561   //  If TLAB is enabled:
3562   //    Try to allocate in the TLAB.
3563   //    If fails, go to the slow path.
3564   //  Else If inline contiguous allocations are enabled:
3565   //    Try to allocate in eden.
3566   //    If fails due to heap end, go to slow path.
3567   //
3568   //  If TLAB is enabled OR inline contiguous is enabled:
3569   //    Initialize the allocation.
3570   //    Exit.
3571   //
3572   //  Go to slow path.
3573   const bool allow_shared_alloc =
3574     Universe::heap()->supports_inline_contig_alloc();
3575 
3576   if (UseTLAB) {
3577     __ tlab_allocate(r0, r3, 0, noreg, r1, slow_case);
3578 
3579     if (ZeroTLAB) {
3580       // the fields have been already cleared
3581       __ b(initialize_header);
3582     } else {
3583       // initialize both the header and fields
3584       __ b(initialize_object);
3585     }
3586   } else {
3587     // Allocation in the shared Eden, if allowed.
3588     //
3589     // r3: instance size in bytes
3590     if (allow_shared_alloc) {
3591       __ eden_allocate(r0, r3, 0, r10, slow_case);
3592     }
3593   }
3594 
3595   // If UseTLAB or allow_shared_alloc are true, the object is created above and
3596   // there is an initialize need. Otherwise, skip and go to the slow path.
3597   if (UseTLAB || allow_shared_alloc) {
3598     // The object is initialized before the header.  If the object size is
3599     // zero, go directly to the header initialization.
3600     __ bind(initialize_object);
3601     __ sub(r3, r3, sizeof(oopDesc));
3602     __ cbz(r3, initialize_header);
3603 
3604     // Initialize object fields
3605     {
3606       __ add(r2, r0, sizeof(oopDesc));
3607       Label loop;
3608       __ bind(loop);
3609       __ str(zr, Address(__ post(r2, BytesPerLong)));
3610       __ sub(r3, r3, BytesPerLong);
3611       __ cbnz(r3, loop);
3612     }
3613 
3614     // initialize object header only.
3615     __ bind(initialize_header);
3616     if (UseBiasedLocking) {
3617       __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3618     } else {
3619       __ mov(rscratch1, (intptr_t)markWord::prototype().value());
3620     }
3621     __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3622     __ store_klass_gap(r0, zr);  // zero klass gap for compressed oops
3623     __ store_klass(r0, r4);      // store klass last
3624 
3625     {
3626       SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3627       // Trigger dtrace event for fastpath
3628       __ push(atos); // save the return value
3629       __ call_VM_leaf(
3630            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3631       __ pop(atos); // restore the return value
3632 
3633     }
3634     __ b(done);
3635   }
3636 
3637   // slow case
3638   __ bind(slow_case);
3639   __ get_constant_pool(c_rarg1);
3640   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3641   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3642   __ verify_oop(r0);
3643 
3644   // continue
3645   __ bind(done);
3646   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3647   __ membar(Assembler::StoreStore);
3648 }
3649 
3650 void TemplateTable::newarray() {
3651   transition(itos, atos);
3652   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3653   __ mov(c_rarg2, r0);
3654   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3655           c_rarg1, c_rarg2);
3656   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3657   __ membar(Assembler::StoreStore);
3658 }
3659 
3660 void TemplateTable::anewarray() {
3661   transition(itos, atos);
3662   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3663   __ get_constant_pool(c_rarg1);
3664   __ mov(c_rarg3, r0);
3665   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3666           c_rarg1, c_rarg2, c_rarg3);
3667   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3668   __ membar(Assembler::StoreStore);
3669 }
3670 
3671 void TemplateTable::arraylength() {
3672   transition(atos, itos);
3673   __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3674   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3675 }
3676 
3677 void TemplateTable::checkcast()
3678 {
3679   transition(atos, atos);
3680   Label done, is_null, ok_is_subtype, quicked, resolved;
3681   __ cbz(r0, is_null);
3682 
3683   // Get cpool & tags index
3684   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3685   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3686   // See if bytecode has already been quicked
3687   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3688   __ lea(r1, Address(rscratch1, r19));
3689   __ ldarb(r1, r1);
3690   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3691   __ br(Assembler::EQ, quicked);
3692 
3693   __ push(atos); // save receiver for result, and for GC
3694   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3695   // vm_result_2 has metadata result
3696   __ get_vm_result_2(r0, rthread);
3697   __ pop(r3); // restore receiver
3698   __ b(resolved);
3699 
3700   // Get superklass in r0 and subklass in r3
3701   __ bind(quicked);
3702   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3703   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3704 
3705   __ bind(resolved);
3706   __ load_klass(r19, r3);
3707 
3708   // Generate subtype check.  Blows r2, r5.  Object in r3.
3709   // Superklass in r0.  Subklass in r19.
3710   __ gen_subtype_check(r19, ok_is_subtype);
3711 
3712   // Come here on failure
3713   __ push(r3);
3714   // object is at TOS
3715   __ b(Interpreter::_throw_ClassCastException_entry);
3716 
3717   // Come here on success
3718   __ bind(ok_is_subtype);
3719   __ mov(r0, r3); // Restore object in r3
3720 
3721   // Collect counts on whether this test sees NULLs a lot or not.
3722   if (ProfileInterpreter) {
3723     __ b(done);
3724     __ bind(is_null);
3725     __ profile_null_seen(r2);
3726   } else {
3727     __ bind(is_null);   // same as 'done'
3728   }
3729   __ bind(done);
3730 }
3731 
3732 void TemplateTable::instanceof() {
3733   transition(atos, itos);
3734   Label done, is_null, ok_is_subtype, quicked, resolved;
3735   __ cbz(r0, is_null);
3736 
3737   // Get cpool & tags index
3738   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3739   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3740   // See if bytecode has already been quicked
3741   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3742   __ lea(r1, Address(rscratch1, r19));
3743   __ ldarb(r1, r1);
3744   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3745   __ br(Assembler::EQ, quicked);
3746 
3747   __ push(atos); // save receiver for result, and for GC
3748   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3749   // vm_result_2 has metadata result
3750   __ get_vm_result_2(r0, rthread);
3751   __ pop(r3); // restore receiver
3752   __ verify_oop(r3);
3753   __ load_klass(r3, r3);
3754   __ b(resolved);
3755 
3756   // Get superklass in r0 and subklass in r3
3757   __ bind(quicked);
3758   __ load_klass(r3, r0);
3759   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3760 
3761   __ bind(resolved);
3762 
3763   // Generate subtype check.  Blows r2, r5
3764   // Superklass in r0.  Subklass in r3.
3765   __ gen_subtype_check(r3, ok_is_subtype);
3766 
3767   // Come here on failure
3768   __ mov(r0, 0);
3769   __ b(done);
3770   // Come here on success
3771   __ bind(ok_is_subtype);
3772   __ mov(r0, 1);
3773 
3774   // Collect counts on whether this test sees NULLs a lot or not.
3775   if (ProfileInterpreter) {
3776     __ b(done);
3777     __ bind(is_null);
3778     __ profile_null_seen(r2);
3779   } else {
3780     __ bind(is_null);   // same as 'done'
3781   }
3782   __ bind(done);
3783   // r0 = 0: obj == NULL or  obj is not an instanceof the specified klass
3784   // r0 = 1: obj != NULL and obj is     an instanceof the specified klass
3785 }
3786 
3787 //-----------------------------------------------------------------------------
3788 // Breakpoints
3789 void TemplateTable::_breakpoint() {
3790   // Note: We get here even if we are single stepping..
3791   // jbug inists on setting breakpoints at every bytecode
3792   // even if we are in single step mode.
3793 
3794   transition(vtos, vtos);
3795 
3796   // get the unpatched byte code
3797   __ get_method(c_rarg1);
3798   __ call_VM(noreg,
3799              CAST_FROM_FN_PTR(address,
3800                               InterpreterRuntime::get_original_bytecode_at),
3801              c_rarg1, rbcp);
3802   __ mov(r19, r0);
3803 
3804   // post the breakpoint event
3805   __ call_VM(noreg,
3806              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3807              rmethod, rbcp);
3808 
3809   // complete the execution of original bytecode
3810   __ mov(rscratch1, r19);
3811   __ dispatch_only_normal(vtos);
3812 }
3813 
3814 //-----------------------------------------------------------------------------
3815 // Exceptions
3816 
3817 void TemplateTable::athrow() {
3818   transition(atos, vtos);
3819   __ null_check(r0);
3820   __ b(Interpreter::throw_exception_entry());
3821 }
3822 
3823 //-----------------------------------------------------------------------------
3824 // Synchronization
3825 //
3826 // Note: monitorenter & exit are symmetric routines; which is reflected
3827 //       in the assembly code structure as well
3828 //
3829 // Stack layout:
3830 //
3831 // [expressions  ] <--- esp               = expression stack top
3832 // ..
3833 // [expressions  ]
3834 // [monitor entry] <--- monitor block top = expression stack bot
3835 // ..
3836 // [monitor entry]
3837 // [frame data   ] <--- monitor block bot
3838 // ...
3839 // [saved rbp    ] <--- rbp
3840 void TemplateTable::monitorenter()
3841 {
3842   transition(atos, vtos);
3843 
3844   // check for NULL object
3845   __ null_check(r0);
3846 
3847   __ resolve(IS_NOT_NULL, r0);
3848 
3849   const Address monitor_block_top(
3850         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3851   const Address monitor_block_bot(
3852         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3853   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3854 
3855   Label allocated;
3856 
3857   // initialize entry pointer
3858   __ mov(c_rarg1, zr); // points to free slot or NULL
3859 
3860   // find a free slot in the monitor block (result in c_rarg1)
3861   {
3862     Label entry, loop, exit;
3863     __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3864                                         // starting with top-most entry
3865     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3866 
3867     __ b(entry);
3868 
3869     __ bind(loop);
3870     // check if current entry is used
3871     // if not used then remember entry in c_rarg1
3872     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3873     __ cmp(zr, rscratch1);
3874     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3875     // check if current entry is for same object
3876     __ cmp(r0, rscratch1);
3877     // if same object then stop searching
3878     __ br(Assembler::EQ, exit);
3879     // otherwise advance to next entry
3880     __ add(c_rarg3, c_rarg3, entry_size);
3881     __ bind(entry);
3882     // check if bottom reached
3883     __ cmp(c_rarg3, c_rarg2);
3884     // if not at bottom then check this entry
3885     __ br(Assembler::NE, loop);
3886     __ bind(exit);
3887   }
3888 
3889   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3890                             // if found, continue with that on
3891 
3892   // allocate one if there's no free slot
3893   {
3894     Label entry, loop;
3895     // 1. compute new pointers            // rsp: old expression stack top
3896     __ ldr(c_rarg1, monitor_block_bot);   // c_rarg1: old expression stack bottom
3897     __ sub(esp, esp, entry_size);         // move expression stack top
3898     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3899     __ mov(c_rarg3, esp);                 // set start value for copy loop
3900     __ str(c_rarg1, monitor_block_bot);   // set new monitor block bottom
3901 
3902     __ sub(sp, sp, entry_size);           // make room for the monitor
3903 
3904     __ b(entry);
3905     // 2. move expression stack contents
3906     __ bind(loop);
3907     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3908                                                    // word from old location
3909     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
3910     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
3911     __ bind(entry);
3912     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
3913     __ br(Assembler::NE, loop);      // if not at bottom then
3914                                      // copy next word
3915   }
3916 
3917   // call run-time routine
3918   // c_rarg1: points to monitor entry
3919   __ bind(allocated);
3920 
3921   // Increment bcp to point to the next bytecode, so exception
3922   // handling for async. exceptions work correctly.
3923   // The object has already been poped from the stack, so the
3924   // expression stack looks correct.
3925   __ increment(rbcp);
3926 
3927   // store object
3928   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3929   __ lock_object(c_rarg1);
3930 
3931   // check to make sure this monitor doesn't cause stack overflow after locking
3932   __ save_bcp();  // in case of exception
3933   __ generate_stack_overflow_check(0);
3934 
3935   // The bcp has already been incremented. Just need to dispatch to
3936   // next instruction.
3937   __ dispatch_next(vtos);
3938 }
3939 
3940 
3941 void TemplateTable::monitorexit()
3942 {
3943   transition(atos, vtos);
3944 
3945   // check for NULL object
3946   __ null_check(r0);
3947 
3948   __ resolve(IS_NOT_NULL, r0);
3949 
3950   const Address monitor_block_top(
3951         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3952   const Address monitor_block_bot(
3953         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3954   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3955 
3956   Label found;
3957 
3958   // find matching slot
3959   {
3960     Label entry, loop;
3961     __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3962                                         // starting with top-most entry
3963     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3964                                         // of monitor block
3965     __ b(entry);
3966 
3967     __ bind(loop);
3968     // check if current entry is for same object
3969     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3970     __ cmp(r0, rscratch1);
3971     // if same object then stop searching
3972     __ br(Assembler::EQ, found);
3973     // otherwise advance to next entry
3974     __ add(c_rarg1, c_rarg1, entry_size);
3975     __ bind(entry);
3976     // check if bottom reached
3977     __ cmp(c_rarg1, c_rarg2);
3978     // if not at bottom then check this entry
3979     __ br(Assembler::NE, loop);
3980   }
3981 
3982   // error handling. Unlocking was not block-structured
3983   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3984                    InterpreterRuntime::throw_illegal_monitor_state_exception));
3985   __ should_not_reach_here();
3986 
3987   // call run-time routine
3988   __ bind(found);
3989   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3990   __ unlock_object(c_rarg1);
3991   __ pop_ptr(r0); // discard object
3992 }
3993 
3994 
3995 // Wide instructions
3996 void TemplateTable::wide()
3997 {
3998   __ load_unsigned_byte(r19, at_bcp(1));
3999   __ mov(rscratch1, (address)Interpreter::_wentry_point);
4000   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
4001   __ br(rscratch1);
4002 }
4003 
4004 
4005 // Multi arrays
4006 void TemplateTable::multianewarray() {
4007   transition(vtos, atos);
4008   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
4009   // last dim is on top of stack; we want address of first one:
4010   // first_addr = last_addr + (ndims - 1) * wordSize
4011   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
4012   __ sub(c_rarg1, c_rarg1, wordSize);
4013   call_VM(r0,
4014           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
4015           c_rarg1);
4016   __ load_unsigned_byte(r1, at_bcp(3));
4017   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
4018 }
--- EOF ---