1 /*
   2  * Copyright (c) 2013, Red Hat Inc.
   3  * Copyright (c) 2003, 2011, Oracle and/or its affiliates.
   4  * All rights reserved.
   5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6  *
   7  * This code is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License version 2 only, as
   9  * published by the Free Software Foundation.
  10  *
  11  * This code is distributed in the hope that it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14  * version 2 for more details (a copy is included in the LICENSE file that
  15  * accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License version
  18  * 2 along with this work; if not, write to the Free Software Foundation,
  19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  20  *
  21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  22  * or visit www.oracle.com if you need additional information or have any
  23  * questions.
  24  *
  25  */
  26 
  27 #include "precompiled.hpp"
  28 #include "asm/macroAssembler.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateTable.hpp"
  32 #include "memory/universe.inline.hpp"
  33 #include "oops/methodData.hpp"
  34 #include "oops/method.hpp"
  35 #include "oops/objArrayKlass.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/methodHandles.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "runtime/synchronizer.hpp"
  41 
  42 #ifndef CC_INTERP
  43 
  44 #define __ _masm->
  45 
  46 // Platform-dependent initialization
  47 
  48 void TemplateTable::pd_initialize() {
  49   // No amd64 specific initialization
  50 }
  51 
  52 // Address computation: local variables
  53 
  54 static inline Address iaddress(int n) {
  55   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  56 }
  57 
  58 static inline Address laddress(int n) {
  59   return iaddress(n + 1);
  60 }
  61 
  62 static inline Address faddress(int n) {
  63   return iaddress(n);
  64 }
  65 
  66 static inline Address daddress(int n) {
  67   return laddress(n);
  68 }
  69 
  70 static inline Address aaddress(int n) {
  71   return iaddress(n);
  72 }
  73 
  74 static inline Address iaddress(Register r) {
  75   return Address(rlocals, r, Address::lsl(3));
  76 }
  77 
  78 static inline Address laddress(Register r, Register scratch,
  79                                InterpreterMacroAssembler* _masm) {
  80   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  81   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  82 }
  83 
  84 static inline Address faddress(Register r) {
  85   return iaddress(r);
  86 }
  87 
  88 static inline Address daddress(Register r, Register scratch,
  89                                InterpreterMacroAssembler* _masm) {
  90   return laddress(r, scratch, _masm);
  91 }
  92 
  93 static inline Address aaddress(Register r) {
  94   return iaddress(r);
  95 }
  96 
  97 static inline Address at_rsp() {
  98   return Address(esp, 0);
  99 }
 100 
 101 // At top of Java expression stack which may be different than esp().  It
 102 // isn't for category 1 objects.
 103 static inline Address at_tos   () {
 104   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 105 }
 106 
 107 static inline Address at_tos_p1() {
 108   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 109 }
 110 
 111 static inline Address at_tos_p2() {
 112   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 113 }
 114 
 115 static inline Address at_tos_p3() {
 116   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 117 }
 118 
 119 static inline Address at_tos_p4() {
 120   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 121 }
 122 
 123 static inline Address at_tos_p5() {
 124   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 125 }
 126 
 127 // Condition conversion
 128 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 129   switch (cc) {
 130   case TemplateTable::equal        : return Assembler::NE;
 131   case TemplateTable::not_equal    : return Assembler::EQ;
 132   case TemplateTable::less         : return Assembler::GE;
 133   case TemplateTable::less_equal   : return Assembler::GT;
 134   case TemplateTable::greater      : return Assembler::LE;
 135   case TemplateTable::greater_equal: return Assembler::LT;
 136   }
 137   ShouldNotReachHere();
 138   return Assembler::EQ;
 139 }
 140 
 141 
 142 // Miscelaneous helper routines
 143 // Store an oop (or NULL) at the Address described by obj.
 144 // If val == noreg this means store a NULL
 145 static void do_oop_store(InterpreterMacroAssembler* _masm,
 146                          Address obj,
 147                          Register val,
 148                          BarrierSet::Name barrier,
 149                          bool precise) {
 150   assert(val == noreg || val == r0, "parameter is just for looks");
 151   switch (barrier) {
 152 #if INCLUDE_ALL_GCS
 153     case BarrierSet::G1SATBCT:
 154     case BarrierSet::G1SATBCTLogging:
 155     case BarrierSet::ShenandoahBarrierSet:
 156       {
 157         // flatten object address if needed
 158         if (obj.index() == noreg && obj.offset() == 0) {
 159           if (obj.base() != r3) {
 160             __ mov(r3, obj.base());
 161           }
 162         } else {
 163           __ lea(r3, obj);
 164         }
 165         __ g1_write_barrier_pre(r3 /* obj */,
 166                                 r1 /* pre_val */,
 167                                 rthread /* thread */,
 168                                 r10  /* tmp */,
 169                                 val != noreg /* tosca_live */,
 170                                 false /* expand_call */);
 171         if (val == noreg) {
 172           __ store_heap_oop_null(Address(r3, 0));
 173         } else {
 174           // G1 barrier needs uncompressed oop for region cross check.
 175           Register new_val = val;
 176           if (UseCompressedOops) {
 177             new_val = rscratch1;
 178             __ mov(new_val, val);
 179           }
 180           // For Shenandoah, make sure we only store refs into to-space.
 181           oopDesc::bs()->interpreter_read_barrier(_masm, val);
 182 
 183           __ store_heap_oop(Address(r3, 0), val);
 184 
 185           __ g1_write_barrier_post(r3 /* store_adr */,
 186                                    new_val /* new_val */,
 187                                    rthread /* thread */,
 188                                    r10 /* tmp */,
 189                                    r1 /* tmp2 */);
 190         }
 191 
 192       }
 193       break;
 194 #endif // INCLUDE_ALL_GCS
 195     case BarrierSet::CardTableModRef:
 196     case BarrierSet::CardTableExtension:
 197       {
 198         if (val == noreg) {
 199           __ store_heap_oop_null(obj);
 200         } else {
 201           __ store_heap_oop(obj, val);
 202           // flatten object address if needed
 203           if (!precise || (obj.index() == noreg && obj.offset() == 0)) {
 204             __ store_check(obj.base());
 205           } else {
 206             __ lea(r3, obj);
 207             __ store_check(r3);
 208           }
 209         }
 210       }
 211       break;
 212     case BarrierSet::ModRef:
 213     case BarrierSet::Other:
 214       if (val == noreg) {
 215         __ store_heap_oop_null(obj);
 216       } else {
 217         __ store_heap_oop(obj, val);
 218       }
 219       break;
 220     default      :
 221       ShouldNotReachHere();
 222 
 223   }
 224 }
 225 
 226 Address TemplateTable::at_bcp(int offset) {
 227   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 228   return Address(rbcp, offset);
 229 }
 230 
 231 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 232                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 233                                    int byte_no)
 234 {
 235   if (!RewriteBytecodes)  return;
 236   Label L_patch_done;
 237 
 238   switch (bc) {
 239   case Bytecodes::_fast_aputfield:
 240   case Bytecodes::_fast_bputfield:
 241   case Bytecodes::_fast_zputfield:
 242   case Bytecodes::_fast_cputfield:
 243   case Bytecodes::_fast_dputfield:
 244   case Bytecodes::_fast_fputfield:
 245   case Bytecodes::_fast_iputfield:
 246   case Bytecodes::_fast_lputfield:
 247   case Bytecodes::_fast_sputfield:
 248     {
 249       // We skip bytecode quickening for putfield instructions when
 250       // the put_code written to the constant pool cache is zero.
 251       // This is required so that every execution of this instruction
 252       // calls out to InterpreterRuntime::resolve_get_put to do
 253       // additional, required work.
 254       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 255       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 256       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 257       __ movw(bc_reg, bc);
 258       __ cmpw(temp_reg, (unsigned) 0);
 259       __ br(Assembler::EQ, L_patch_done);  // don't patch
 260     }
 261     break;
 262   default:
 263     assert(byte_no == -1, "sanity");
 264     // the pair bytecodes have already done the load.
 265     if (load_bc_into_bc_reg) {
 266       __ movw(bc_reg, bc);
 267     }
 268   }
 269 
 270   if (JvmtiExport::can_post_breakpoint()) {
 271     Label L_fast_patch;
 272     // if a breakpoint is present we can't rewrite the stream directly
 273     __ load_unsigned_byte(temp_reg, at_bcp(0));
 274     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 275     __ br(Assembler::NE, L_fast_patch);
 276     // Let breakpoint table handling rewrite to quicker bytecode
 277     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 278     __ b(L_patch_done);
 279     __ bind(L_fast_patch);
 280   }
 281 
 282 #ifdef ASSERT
 283   Label L_okay;
 284   __ load_unsigned_byte(temp_reg, at_bcp(0));
 285   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 286   __ br(Assembler::EQ, L_okay);
 287   __ cmpw(temp_reg, bc_reg);
 288   __ br(Assembler::EQ, L_okay);
 289   __ stop("patching the wrong bytecode");
 290   __ bind(L_okay);
 291 #endif
 292 
 293   // patch bytecode
 294   __ strb(bc_reg, at_bcp(0));
 295   __ bind(L_patch_done);
 296 }
 297 
 298 
 299 // Individual instructions
 300 
 301 void TemplateTable::nop() {
 302   transition(vtos, vtos);
 303   // nothing to do
 304 }
 305 
 306 void TemplateTable::shouldnotreachhere() {
 307   transition(vtos, vtos);
 308   __ stop("shouldnotreachhere bytecode");
 309 }
 310 
 311 void TemplateTable::aconst_null()
 312 {
 313   transition(vtos, atos);
 314   __ mov(r0, 0);
 315 }
 316 
 317 void TemplateTable::iconst(int value)
 318 {
 319   transition(vtos, itos);
 320   __ mov(r0, value);
 321 }
 322 
 323 void TemplateTable::lconst(int value)
 324 {
 325   __ mov(r0, value);
 326 }
 327 
 328 void TemplateTable::fconst(int value)
 329 {
 330   transition(vtos, ftos);
 331   switch (value) {
 332   case 0:
 333     __ fmovs(v0, zr);
 334     break;
 335   case 1:
 336     __ fmovs(v0, 1.0);
 337     break;
 338   case 2:
 339     __ fmovs(v0, 2.0);
 340     break;
 341   default:
 342     ShouldNotReachHere();
 343     break;
 344   }
 345 }
 346 
 347 void TemplateTable::dconst(int value)
 348 {
 349   transition(vtos, dtos);
 350   switch (value) {
 351   case 0:
 352     __ fmovd(v0, zr);
 353     break;
 354   case 1:
 355     __ fmovd(v0, 1.0);
 356     break;
 357   case 2:
 358     __ fmovd(v0, 2.0);
 359     break;
 360   default:
 361     ShouldNotReachHere();
 362     break;
 363   }
 364 }
 365 
 366 void TemplateTable::bipush()
 367 {
 368   transition(vtos, itos);
 369   __ load_signed_byte32(r0, at_bcp(1));
 370 }
 371 
 372 void TemplateTable::sipush()
 373 {
 374   transition(vtos, itos);
 375   __ load_unsigned_short(r0, at_bcp(1));
 376   __ revw(r0, r0);
 377   __ asrw(r0, r0, 16);
 378 }
 379 
 380 void TemplateTable::ldc(bool wide)
 381 {
 382   transition(vtos, vtos);
 383   Label call_ldc, notFloat, notClass, Done;
 384 
 385   if (wide) {
 386     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 387   } else {
 388     __ load_unsigned_byte(r1, at_bcp(1));
 389   }
 390   __ get_cpool_and_tags(r2, r0);
 391 
 392   const int base_offset = ConstantPool::header_size() * wordSize;
 393   const int tags_offset = Array<u1>::base_offset_in_bytes();
 394 
 395   // get type
 396   __ add(r3, r1, tags_offset);
 397   __ lea(r3, Address(r0, r3));
 398   __ ldarb(r3, r3);
 399 
 400   // unresolved class - get the resolved class
 401   __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
 402   __ br(Assembler::EQ, call_ldc);
 403 
 404   // unresolved class in error state - call into runtime to throw the error
 405   // from the first resolution attempt
 406   __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
 407   __ br(Assembler::EQ, call_ldc);
 408 
 409   // resolved class - need to call vm to get java mirror of the class
 410   __ cmp(r3, JVM_CONSTANT_Class);
 411   __ br(Assembler::NE, notClass);
 412 
 413   __ bind(call_ldc);
 414   __ mov(c_rarg1, wide);
 415   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 416   __ push_ptr(r0);
 417   __ verify_oop(r0);
 418   __ b(Done);
 419 
 420   __ bind(notClass);
 421   __ cmp(r3, JVM_CONSTANT_Float);
 422   __ br(Assembler::NE, notFloat);
 423   // ftos
 424   __ adds(r1, r2, r1, Assembler::LSL, 3);
 425   __ ldrs(v0, Address(r1, base_offset));
 426   __ push_f();
 427   __ b(Done);
 428 
 429   __ bind(notFloat);
 430 #ifdef ASSERT
 431   {
 432     Label L;
 433     __ cmp(r3, JVM_CONSTANT_Integer);
 434     __ br(Assembler::EQ, L);
 435     // String and Object are rewritten to fast_aldc
 436     __ stop("unexpected tag type in ldc");
 437     __ bind(L);
 438   }
 439 #endif
 440   // itos JVM_CONSTANT_Integer only
 441   __ adds(r1, r2, r1, Assembler::LSL, 3);
 442   __ ldrw(r0, Address(r1, base_offset));
 443   __ push_i(r0);
 444   __ bind(Done);
 445 }
 446 
 447 // Fast path for caching oop constants.
 448 void TemplateTable::fast_aldc(bool wide)
 449 {
 450   transition(vtos, atos);
 451 
 452   Register result = r0;
 453   Register tmp = r1;
 454   int index_size = wide ? sizeof(u2) : sizeof(u1);
 455 
 456   Label resolved;
 457 
 458   // We are resolved if the resolved reference cache entry contains a
 459   // non-null object (String, MethodType, etc.)
 460   assert_different_registers(result, tmp);
 461   __ get_cache_index_at_bcp(tmp, 1, index_size);
 462   __ load_resolved_reference_at_index(result, tmp);
 463   __ cbnz(result, resolved);
 464 
 465   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 466 
 467   // first time invocation - must resolve first
 468   __ mov(tmp, (int)bytecode());
 469   __ call_VM(result, entry, tmp);
 470 
 471   __ bind(resolved);
 472 
 473   if (VerifyOops) {
 474     __ verify_oop(result);
 475   }
 476 }
 477 
 478 void TemplateTable::ldc2_w()
 479 {
 480   transition(vtos, vtos);
 481   Label Long, Done;
 482   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 483 
 484   __ get_cpool_and_tags(r1, r2);
 485   const int base_offset = ConstantPool::header_size() * wordSize;
 486   const int tags_offset = Array<u1>::base_offset_in_bytes();
 487 
 488   // get type
 489   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 490   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 491   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 492   __ br(Assembler::NE, Long);
 493   // dtos
 494   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 495   __ ldrd(v0, Address(r2, base_offset));
 496   __ push_d();
 497   __ b(Done);
 498 
 499   __ bind(Long);
 500   // ltos
 501   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 502   __ ldr(r0, Address(r0, base_offset));
 503   __ push_l();
 504 
 505   __ bind(Done);
 506 }
 507 
 508 void TemplateTable::locals_index(Register reg, int offset)
 509 {
 510   __ ldrb(reg, at_bcp(offset));
 511   __ neg(reg, reg);
 512 }
 513 
 514 void TemplateTable::iload()
 515 {
 516   transition(vtos, itos);
 517   if (RewriteFrequentPairs) {
 518     Label rewrite, done;
 519     Register bc = r4;
 520 
 521     // get next bytecode
 522     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 523 
 524     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 525     // last two iloads in a pair.  Comparing against fast_iload means that
 526     // the next bytecode is neither an iload or a caload, and therefore
 527     // an iload pair.
 528     __ cmpw(r1, Bytecodes::_iload);
 529     __ br(Assembler::EQ, done);
 530 
 531     // if _fast_iload rewrite to _fast_iload2
 532     __ cmpw(r1, Bytecodes::_fast_iload);
 533     __ movw(bc, Bytecodes::_fast_iload2);
 534     __ br(Assembler::EQ, rewrite);
 535 
 536     // if _caload rewrite to _fast_icaload
 537     __ cmpw(r1, Bytecodes::_caload);
 538     __ movw(bc, Bytecodes::_fast_icaload);
 539     __ br(Assembler::EQ, rewrite);
 540 
 541     // else rewrite to _fast_iload
 542     __ movw(bc, Bytecodes::_fast_iload);
 543 
 544     // rewrite
 545     // bc: new bytecode
 546     __ bind(rewrite);
 547     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 548     __ bind(done);
 549 
 550   }
 551 
 552   // do iload, get the local value into tos
 553   locals_index(r1);
 554   __ ldr(r0, iaddress(r1));
 555 
 556 }
 557 
 558 void TemplateTable::fast_iload2()
 559 {
 560   transition(vtos, itos);
 561   locals_index(r1);
 562   __ ldr(r0, iaddress(r1));
 563   __ push(itos);
 564   locals_index(r1, 3);
 565   __ ldr(r0, iaddress(r1));
 566 }
 567 
 568 void TemplateTable::fast_iload()
 569 {
 570   transition(vtos, itos);
 571   locals_index(r1);
 572   __ ldr(r0, iaddress(r1));
 573 }
 574 
 575 void TemplateTable::lload()
 576 {
 577   transition(vtos, ltos);
 578   __ ldrb(r1, at_bcp(1));
 579   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 580   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 581 }
 582 
 583 void TemplateTable::fload()
 584 {
 585   transition(vtos, ftos);
 586   locals_index(r1);
 587   // n.b. we use ldrd here because this is a 64 bit slot
 588   // this is comparable to the iload case
 589   __ ldrd(v0, faddress(r1));
 590 }
 591 
 592 void TemplateTable::dload()
 593 {
 594   transition(vtos, dtos);
 595   __ ldrb(r1, at_bcp(1));
 596   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 597   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 598 }
 599 
 600 void TemplateTable::aload()
 601 {
 602   transition(vtos, atos);
 603   locals_index(r1);
 604   __ ldr(r0, iaddress(r1));
 605 }
 606 
 607 void TemplateTable::locals_index_wide(Register reg) {
 608   __ ldrh(reg, at_bcp(2));
 609   __ rev16w(reg, reg);
 610   __ neg(reg, reg);
 611 }
 612 
 613 void TemplateTable::wide_iload() {
 614   transition(vtos, itos);
 615   locals_index_wide(r1);
 616   __ ldr(r0, iaddress(r1));
 617 }
 618 
 619 void TemplateTable::wide_lload()
 620 {
 621   transition(vtos, ltos);
 622   __ ldrh(r1, at_bcp(2));
 623   __ rev16w(r1, r1);
 624   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 625   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 626 }
 627 
 628 void TemplateTable::wide_fload()
 629 {
 630   transition(vtos, ftos);
 631   locals_index_wide(r1);
 632   // n.b. we use ldrd here because this is a 64 bit slot
 633   // this is comparable to the iload case
 634   __ ldrd(v0, faddress(r1));
 635 }
 636 
 637 void TemplateTable::wide_dload()
 638 {
 639   transition(vtos, dtos);
 640   __ ldrh(r1, at_bcp(2));
 641   __ rev16w(r1, r1);
 642   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 643   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 644 }
 645 
 646 void TemplateTable::wide_aload()
 647 {
 648   transition(vtos, atos);
 649   locals_index_wide(r1);
 650   __ ldr(r0, aaddress(r1));
 651 }
 652 
 653 void TemplateTable::index_check(Register array, Register index)
 654 {
 655   // destroys r1, rscratch1
 656   // check array
 657   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 658   // sign extend index for use by indexed load
 659   // __ movl2ptr(index, index);
 660   // check index
 661   Register length = rscratch1;
 662   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 663   __ cmpw(index, length);
 664   if (index != r1) {
 665     // ??? convention: move aberrant index into r1 for exception message
 666     assert(r1 != array, "different registers");
 667     __ mov(r1, index);
 668   }
 669   Label ok;
 670   __ br(Assembler::LO, ok);
 671   __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 672   __ br(rscratch1);
 673   __ bind(ok);
 674 }
 675 
 676 void TemplateTable::iaload()
 677 {
 678   transition(itos, itos);
 679   __ mov(r1, r0);
 680   __ pop_ptr(r0);
 681   // r0: array
 682   // r1: index
 683   index_check(r0, r1); // leaves index in r1, kills rscratch1
 684   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 685   __ lea(r1, Address(r0, r1, Address::uxtw(2)));
 686   __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
 687 }
 688 
 689 void TemplateTable::laload()
 690 {
 691   transition(itos, ltos);
 692   __ mov(r1, r0);
 693   __ pop_ptr(r0);
 694   // r0: array
 695   // r1: index
 696   index_check(r0, r1); // leaves index in r1, kills rscratch1
 697   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 698   __ lea(r1, Address(r0, r1, Address::uxtw(3)));
 699   __ ldr(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_LONG)));
 700 }
 701 
 702 void TemplateTable::faload()
 703 {
 704   transition(itos, ftos);
 705   __ mov(r1, r0);
 706   __ pop_ptr(r0);
 707   // r0: array
 708   // r1: index
 709   index_check(r0, r1); // leaves index in r1, kills rscratch1
 710   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 711   __ lea(r1,  Address(r0, r1, Address::uxtw(2)));
 712   __ ldrs(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 713 }
 714 
 715 void TemplateTable::daload()
 716 {
 717   transition(itos, dtos);
 718   __ mov(r1, r0);
 719   __ pop_ptr(r0);
 720   // r0: array
 721   // r1: index
 722   index_check(r0, r1); // leaves index in r1, kills rscratch1
 723   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 724   __ lea(r1,  Address(r0, r1, Address::uxtw(3)));
 725   __ ldrd(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 726 }
 727 
 728 void TemplateTable::aaload()
 729 {
 730   transition(itos, atos);
 731   __ mov(r1, r0);
 732   __ pop_ptr(r0);
 733   // r0: array
 734   // r1: index
 735   index_check(r0, r1); // leaves index in r1, kills rscratch1
 736   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 737   int s = (UseCompressedOops ? 2 : 3);
 738   __ lea(r1, Address(r0, r1, Address::uxtw(s)));
 739   __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 740 }
 741 
 742 void TemplateTable::baload()
 743 {
 744   transition(itos, itos);
 745   __ mov(r1, r0);
 746   __ pop_ptr(r0);
 747   // r0: array
 748   // r1: index
 749   index_check(r0, r1); // leaves index in r1, kills rscratch1
 750   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 751   __ lea(r1,  Address(r0, r1, Address::uxtw(0)));
 752   __ load_signed_byte(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 753 }
 754 
 755 void TemplateTable::caload()
 756 {
 757   transition(itos, itos);
 758   __ mov(r1, r0);
 759   __ pop_ptr(r0);
 760   // r0: array
 761   // r1: index
 762   index_check(r0, r1); // leaves index in r1, kills rscratch1
 763   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 764   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 765   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 766 }
 767 
 768 // iload followed by caload frequent pair
 769 void TemplateTable::fast_icaload()
 770 {
 771   transition(vtos, itos);
 772   // load index out of locals
 773   locals_index(r2);
 774   __ ldr(r1, iaddress(r2));
 775 
 776   __ pop_ptr(r0);
 777 
 778   // r0: array
 779   // r1: index
 780   index_check(r0, r1); // leaves index in r1, kills rscratch1
 781   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 782   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 783   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 784 }
 785 
 786 void TemplateTable::saload()
 787 {
 788   transition(itos, itos);
 789   __ mov(r1, r0);
 790   __ pop_ptr(r0);
 791   // r0: array
 792   // r1: index
 793   index_check(r0, r1); // leaves index in r1, kills rscratch1
 794   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
 795   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 796   __ load_signed_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 797 }
 798 
 799 void TemplateTable::iload(int n)
 800 {
 801   transition(vtos, itos);
 802   __ ldr(r0, iaddress(n));
 803 }
 804 
 805 void TemplateTable::lload(int n)
 806 {
 807   transition(vtos, ltos);
 808   __ ldr(r0, laddress(n));
 809 }
 810 
 811 void TemplateTable::fload(int n)
 812 {
 813   transition(vtos, ftos);
 814   __ ldrs(v0, faddress(n));
 815 }
 816 
 817 void TemplateTable::dload(int n)
 818 {
 819   transition(vtos, dtos);
 820   __ ldrd(v0, daddress(n));
 821 }
 822 
 823 void TemplateTable::aload(int n)
 824 {
 825   transition(vtos, atos);
 826   __ ldr(r0, iaddress(n));
 827 }
 828 
 829 void TemplateTable::aload_0()
 830 {
 831   // According to bytecode histograms, the pairs:
 832   //
 833   // _aload_0, _fast_igetfield
 834   // _aload_0, _fast_agetfield
 835   // _aload_0, _fast_fgetfield
 836   //
 837   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 838   // _aload_0 bytecode checks if the next bytecode is either
 839   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 840   // rewrites the current bytecode into a pair bytecode; otherwise it
 841   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 842   // the pair check anymore.
 843   //
 844   // Note: If the next bytecode is _getfield, the rewrite must be
 845   //       delayed, otherwise we may miss an opportunity for a pair.
 846   //
 847   // Also rewrite frequent pairs
 848   //   aload_0, aload_1
 849   //   aload_0, iload_1
 850   // These bytecodes with a small amount of code are most profitable
 851   // to rewrite
 852   if (RewriteFrequentPairs) {
 853     Label rewrite, done;
 854     const Register bc = r4;
 855 
 856     // get next bytecode
 857     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 858 
 859     // do actual aload_0
 860     aload(0);
 861 
 862     // if _getfield then wait with rewrite
 863     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 864     __ br(Assembler::EQ, done);
 865 
 866     // if _igetfield then reqrite to _fast_iaccess_0
 867     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 868     __ cmpw(r1, Bytecodes::_fast_igetfield);
 869     __ movw(bc, Bytecodes::_fast_iaccess_0);
 870     __ br(Assembler::EQ, rewrite);
 871 
 872     // if _agetfield then reqrite to _fast_aaccess_0
 873     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 874     __ cmpw(r1, Bytecodes::_fast_agetfield);
 875     __ movw(bc, Bytecodes::_fast_aaccess_0);
 876     __ br(Assembler::EQ, rewrite);
 877 
 878     // if _fgetfield then reqrite to _fast_faccess_0
 879     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 880     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 881     __ movw(bc, Bytecodes::_fast_faccess_0);
 882     __ br(Assembler::EQ, rewrite);
 883 
 884     // else rewrite to _fast_aload0
 885     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 886     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 887 
 888     // rewrite
 889     // bc: new bytecode
 890     __ bind(rewrite);
 891     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 892 
 893     __ bind(done);
 894   } else {
 895     aload(0);
 896   }
 897 }
 898 
 899 void TemplateTable::istore()
 900 {
 901   transition(itos, vtos);
 902   locals_index(r1);
 903   // FIXME: We're being very pernickerty here storing a jint in a
 904   // local with strw, which costs an extra instruction over what we'd
 905   // be able to do with a simple str.  We should just store the whole
 906   // word.
 907   __ lea(rscratch1, iaddress(r1));
 908   __ strw(r0, Address(rscratch1));
 909 }
 910 
 911 void TemplateTable::lstore()
 912 {
 913   transition(ltos, vtos);
 914   locals_index(r1);
 915   __ str(r0, laddress(r1, rscratch1, _masm));
 916 }
 917 
 918 void TemplateTable::fstore() {
 919   transition(ftos, vtos);
 920   locals_index(r1);
 921   __ lea(rscratch1, iaddress(r1));
 922   __ strs(v0, Address(rscratch1));
 923 }
 924 
 925 void TemplateTable::dstore() {
 926   transition(dtos, vtos);
 927   locals_index(r1);
 928   __ strd(v0, daddress(r1, rscratch1, _masm));
 929 }
 930 
 931 void TemplateTable::astore()
 932 {
 933   transition(vtos, vtos);
 934   __ pop_ptr(r0);
 935   locals_index(r1);
 936   __ str(r0, aaddress(r1));
 937 }
 938 
 939 void TemplateTable::wide_istore() {
 940   transition(vtos, vtos);
 941   __ pop_i();
 942   locals_index_wide(r1);
 943   __ lea(rscratch1, iaddress(r1));
 944   __ strw(r0, Address(rscratch1));
 945 }
 946 
 947 void TemplateTable::wide_lstore() {
 948   transition(vtos, vtos);
 949   __ pop_l();
 950   locals_index_wide(r1);
 951   __ str(r0, laddress(r1, rscratch1, _masm));
 952 }
 953 
 954 void TemplateTable::wide_fstore() {
 955   transition(vtos, vtos);
 956   __ pop_f();
 957   locals_index_wide(r1);
 958   __ lea(rscratch1, faddress(r1));
 959   __ strs(v0, rscratch1);
 960 }
 961 
 962 void TemplateTable::wide_dstore() {
 963   transition(vtos, vtos);
 964   __ pop_d();
 965   locals_index_wide(r1);
 966   __ strd(v0, daddress(r1, rscratch1, _masm));
 967 }
 968 
 969 void TemplateTable::wide_astore() {
 970   transition(vtos, vtos);
 971   __ pop_ptr(r0);
 972   locals_index_wide(r1);
 973   __ str(r0, aaddress(r1));
 974 }
 975 
 976 void TemplateTable::iastore() {
 977   transition(itos, vtos);
 978   __ pop_i(r1);
 979   __ pop_ptr(r3);
 980   // r0: value
 981   // r1: index
 982   // r3: array
 983   index_check(r3, r1); // prefer index in r1
 984   oopDesc::bs()->interpreter_write_barrier(_masm, r3);
 985   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
 986   __ strw(r0, Address(rscratch1,
 987                       arrayOopDesc::base_offset_in_bytes(T_INT)));
 988 }
 989 
 990 void TemplateTable::lastore() {
 991   transition(ltos, vtos);
 992   __ pop_i(r1);
 993   __ pop_ptr(r3);
 994   // r0: value
 995   // r1: index
 996   // r3: array
 997   index_check(r3, r1); // prefer index in r1
 998   oopDesc::bs()->interpreter_write_barrier(_masm, r3);
 999   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1000   __ str(r0, Address(rscratch1,
1001                       arrayOopDesc::base_offset_in_bytes(T_LONG)));
1002 }
1003 
1004 void TemplateTable::fastore() {
1005   transition(ftos, vtos);
1006   __ pop_i(r1);
1007   __ pop_ptr(r3);
1008   // v0: value
1009   // r1:  index
1010   // r3:  array
1011   index_check(r3, r1); // prefer index in r1
1012   oopDesc::bs()->interpreter_write_barrier(_masm, r3);
1013   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1014   __ strs(v0, Address(rscratch1,
1015                       arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1016 }
1017 
1018 void TemplateTable::dastore() {
1019   transition(dtos, vtos);
1020   __ pop_i(r1);
1021   __ pop_ptr(r3);
1022   // v0: value
1023   // r1:  index
1024   // r3:  array
1025   index_check(r3, r1); // prefer index in r1
1026   oopDesc::bs()->interpreter_write_barrier(_masm, r3);
1027   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1028   __ strd(v0, Address(rscratch1,
1029                       arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1030 }
1031 
1032 void TemplateTable::aastore() {
1033   Label is_null, ok_is_subtype, done;
1034   transition(vtos, vtos);
1035   // stack: ..., array, index, value
1036   __ ldr(r0, at_tos());    // value
1037   __ ldr(r2, at_tos_p1()); // index
1038   __ ldr(r3, at_tos_p2()); // array
1039 
1040   Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1041 
1042   index_check(r3, r2);     // kills r1
1043   oopDesc::bs()->interpreter_write_barrier(_masm, r3);
1044   __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1045 
1046   // do array store check - check for NULL value first
1047   __ cbz(r0, is_null);
1048 
1049   // Move subklass into r1
1050   __ load_klass(r1, r0);
1051   // Move superklass into r0
1052   __ load_klass(r0, r3);
1053   __ ldr(r0, Address(r0,
1054                      ObjArrayKlass::element_klass_offset()));
1055   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1056 
1057   // Generate subtype check.  Blows r2, r5
1058   // Superklass in r0.  Subklass in r1.
1059   __ gen_subtype_check(r1, ok_is_subtype);
1060 
1061   // Come here on failure
1062   // object is at TOS
1063   __ b(Interpreter::_throw_ArrayStoreException_entry);
1064 
1065   // Come here on success
1066   __ bind(ok_is_subtype);
1067 
1068   // Get the value we will store
1069   __ ldr(r0, at_tos());
1070   // Now store using the appropriate barrier
1071   do_oop_store(_masm, element_address, r0, _bs->kind(), true);
1072   __ b(done);
1073 
1074   // Have a NULL in r0, r3=array, r2=index.  Store NULL at ary[idx]
1075   __ bind(is_null);
1076   __ profile_null_seen(r2);
1077 
1078   // Store a NULL
1079   do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1080 
1081   // Pop stack arguments
1082   __ bind(done);
1083   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1084 }
1085 
1086 void TemplateTable::bastore()
1087 {
1088   transition(itos, vtos);
1089   __ pop_i(r1);
1090   __ pop_ptr(r3);
1091   // r0: value
1092   // r1: index
1093   // r3: array
1094   oopDesc::bs()->interpreter_write_barrier(_masm, r3);
1095   index_check(r3, r1); // prefer index in r1
1096 
1097   // Need to check whether array is boolean or byte
1098   // since both types share the bastore bytecode.
1099   __ load_klass(r2, r3);
1100   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1101   int diffbit = Klass::layout_helper_boolean_diffbit();
1102   __ andw(rscratch1, r2, diffbit);
1103   Label L_skip;
1104   __ cbzw(rscratch1, L_skip);
1105   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1106   __ bind(L_skip);
1107 
1108   __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1109   __ strb(r0, Address(rscratch1,
1110                       arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1111 }
1112 
1113 void TemplateTable::castore()
1114 {
1115   transition(itos, vtos);
1116   __ pop_i(r1);
1117   __ pop_ptr(r3);
1118   // r0: value
1119   // r1: index
1120   // r3: array
1121   index_check(r3, r1); // prefer index in r1
1122   oopDesc::bs()->interpreter_write_barrier(_masm, r3);
1123   __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1124   __ strh(r0, Address(rscratch1,
1125                       arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1126 }
1127 
1128 void TemplateTable::sastore()
1129 {
1130   castore();
1131 }
1132 
1133 void TemplateTable::istore(int n)
1134 {
1135   transition(itos, vtos);
1136   __ str(r0, iaddress(n));
1137 }
1138 
1139 void TemplateTable::lstore(int n)
1140 {
1141   transition(ltos, vtos);
1142   __ str(r0, laddress(n));
1143 }
1144 
1145 void TemplateTable::fstore(int n)
1146 {
1147   transition(ftos, vtos);
1148   __ strs(v0, faddress(n));
1149 }
1150 
1151 void TemplateTable::dstore(int n)
1152 {
1153   transition(dtos, vtos);
1154   __ strd(v0, daddress(n));
1155 }
1156 
1157 void TemplateTable::astore(int n)
1158 {
1159   transition(vtos, vtos);
1160   __ pop_ptr(r0);
1161   __ str(r0, iaddress(n));
1162 }
1163 
1164 void TemplateTable::pop()
1165 {
1166   transition(vtos, vtos);
1167   __ add(esp, esp, Interpreter::stackElementSize);
1168 }
1169 
1170 void TemplateTable::pop2()
1171 {
1172   transition(vtos, vtos);
1173   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1174 }
1175 
1176 void TemplateTable::dup()
1177 {
1178   transition(vtos, vtos);
1179   __ ldr(r0, Address(esp, 0));
1180   __ push(r0);
1181   // stack: ..., a, a
1182 }
1183 
1184 void TemplateTable::dup_x1()
1185 {
1186   transition(vtos, vtos);
1187   // stack: ..., a, b
1188   __ ldr(r0, at_tos());  // load b
1189   __ ldr(r2, at_tos_p1());  // load a
1190   __ str(r0, at_tos_p1());  // store b
1191   __ str(r2, at_tos());  // store a
1192   __ push(r0);                  // push b
1193   // stack: ..., b, a, b
1194 }
1195 
1196 void TemplateTable::dup_x2()
1197 {
1198   transition(vtos, vtos);
1199   // stack: ..., a, b, c
1200   __ ldr(r0, at_tos());  // load c
1201   __ ldr(r2, at_tos_p2());  // load a
1202   __ str(r0, at_tos_p2());  // store c in a
1203   __ push(r0);      // push c
1204   // stack: ..., c, b, c, c
1205   __ ldr(r0, at_tos_p2());  // load b
1206   __ str(r2, at_tos_p2());  // store a in b
1207   // stack: ..., c, a, c, c
1208   __ str(r0, at_tos_p1());  // store b in c
1209   // stack: ..., c, a, b, c
1210 }
1211 
1212 void TemplateTable::dup2()
1213 {
1214   transition(vtos, vtos);
1215   // stack: ..., a, b
1216   __ ldr(r0, at_tos_p1());  // load a
1217   __ push(r0);                  // push a
1218   __ ldr(r0, at_tos_p1());  // load b
1219   __ push(r0);                  // push b
1220   // stack: ..., a, b, a, b
1221 }
1222 
1223 void TemplateTable::dup2_x1()
1224 {
1225   transition(vtos, vtos);
1226   // stack: ..., a, b, c
1227   __ ldr(r2, at_tos());  // load c
1228   __ ldr(r0, at_tos_p1());  // load b
1229   __ push(r0);                  // push b
1230   __ push(r2);                  // push c
1231   // stack: ..., a, b, c, b, c
1232   __ str(r2, at_tos_p3());  // store c in b
1233   // stack: ..., a, c, c, b, c
1234   __ ldr(r2, at_tos_p4());  // load a
1235   __ str(r2, at_tos_p2());  // store a in 2nd c
1236   // stack: ..., a, c, a, b, c
1237   __ str(r0, at_tos_p4());  // store b in a
1238   // stack: ..., b, c, a, b, c
1239 }
1240 
1241 void TemplateTable::dup2_x2()
1242 {
1243   transition(vtos, vtos);
1244   // stack: ..., a, b, c, d
1245   __ ldr(r2, at_tos());  // load d
1246   __ ldr(r0, at_tos_p1());  // load c
1247   __ push(r0)            ;      // push c
1248   __ push(r2);                  // push d
1249   // stack: ..., a, b, c, d, c, d
1250   __ ldr(r0, at_tos_p4());  // load b
1251   __ str(r0, at_tos_p2());  // store b in d
1252   __ str(r2, at_tos_p4());  // store d in b
1253   // stack: ..., a, d, c, b, c, d
1254   __ ldr(r2, at_tos_p5());  // load a
1255   __ ldr(r0, at_tos_p3());  // load c
1256   __ str(r2, at_tos_p3());  // store a in c
1257   __ str(r0, at_tos_p5());  // store c in a
1258   // stack: ..., c, d, a, b, c, d
1259 }
1260 
1261 void TemplateTable::swap()
1262 {
1263   transition(vtos, vtos);
1264   // stack: ..., a, b
1265   __ ldr(r2, at_tos_p1());  // load a
1266   __ ldr(r0, at_tos());  // load b
1267   __ str(r2, at_tos());  // store a in b
1268   __ str(r0, at_tos_p1());  // store b in a
1269   // stack: ..., b, a
1270 }
1271 
1272 void TemplateTable::iop2(Operation op)
1273 {
1274   transition(itos, itos);
1275   // r0 <== r1 op r0
1276   __ pop_i(r1);
1277   switch (op) {
1278   case add  : __ addw(r0, r1, r0); break;
1279   case sub  : __ subw(r0, r1, r0); break;
1280   case mul  : __ mulw(r0, r1, r0); break;
1281   case _and : __ andw(r0, r1, r0); break;
1282   case _or  : __ orrw(r0, r1, r0); break;
1283   case _xor : __ eorw(r0, r1, r0); break;
1284   case shl  : __ lslvw(r0, r1, r0); break;
1285   case shr  : __ asrvw(r0, r1, r0); break;
1286   case ushr : __ lsrvw(r0, r1, r0);break;
1287   default   : ShouldNotReachHere();
1288   }
1289 }
1290 
1291 void TemplateTable::lop2(Operation op)
1292 {
1293   transition(ltos, ltos);
1294   // r0 <== r1 op r0
1295   __ pop_l(r1);
1296   switch (op) {
1297   case add  : __ add(r0, r1, r0); break;
1298   case sub  : __ sub(r0, r1, r0); break;
1299   case mul  : __ mul(r0, r1, r0); break;
1300   case _and : __ andr(r0, r1, r0); break;
1301   case _or  : __ orr(r0, r1, r0); break;
1302   case _xor : __ eor(r0, r1, r0); break;
1303   default   : ShouldNotReachHere();
1304   }
1305 }
1306 
1307 void TemplateTable::idiv()
1308 {
1309   transition(itos, itos);
1310   // explicitly check for div0
1311   Label no_div0;
1312   __ cbnzw(r0, no_div0);
1313   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1314   __ br(rscratch1);
1315   __ bind(no_div0);
1316   __ pop_i(r1);
1317   // r0 <== r1 idiv r0
1318   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1319 }
1320 
1321 void TemplateTable::irem()
1322 {
1323   transition(itos, itos);
1324   // explicitly check for div0
1325   Label no_div0;
1326   __ cbnzw(r0, no_div0);
1327   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1328   __ br(rscratch1);
1329   __ bind(no_div0);
1330   __ pop_i(r1);
1331   // r0 <== r1 irem r0
1332   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1333 }
1334 
1335 void TemplateTable::lmul()
1336 {
1337   transition(ltos, ltos);
1338   __ pop_l(r1);
1339   __ mul(r0, r0, r1);
1340 }
1341 
1342 void TemplateTable::ldiv()
1343 {
1344   transition(ltos, ltos);
1345   // explicitly check for div0
1346   Label no_div0;
1347   __ cbnz(r0, no_div0);
1348   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1349   __ br(rscratch1);
1350   __ bind(no_div0);
1351   __ pop_l(r1);
1352   // r0 <== r1 ldiv r0
1353   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1354 }
1355 
1356 void TemplateTable::lrem()
1357 {
1358   transition(ltos, ltos);
1359   // explicitly check for div0
1360   Label no_div0;
1361   __ cbnz(r0, no_div0);
1362   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1363   __ br(rscratch1);
1364   __ bind(no_div0);
1365   __ pop_l(r1);
1366   // r0 <== r1 lrem r0
1367   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1368 }
1369 
1370 void TemplateTable::lshl()
1371 {
1372   transition(itos, ltos);
1373   // shift count is in r0
1374   __ pop_l(r1);
1375   __ lslv(r0, r1, r0);
1376 }
1377 
1378 void TemplateTable::lshr()
1379 {
1380   transition(itos, ltos);
1381   // shift count is in r0
1382   __ pop_l(r1);
1383   __ asrv(r0, r1, r0);
1384 }
1385 
1386 void TemplateTable::lushr()
1387 {
1388   transition(itos, ltos);
1389   // shift count is in r0
1390   __ pop_l(r1);
1391   __ lsrv(r0, r1, r0);
1392 }
1393 
1394 void TemplateTable::fop2(Operation op)
1395 {
1396   transition(ftos, ftos);
1397   switch (op) {
1398   case add:
1399     // n.b. use ldrd because this is a 64 bit slot
1400     __ pop_f(v1);
1401     __ fadds(v0, v1, v0);
1402     break;
1403   case sub:
1404     __ pop_f(v1);
1405     __ fsubs(v0, v1, v0);
1406     break;
1407   case mul:
1408     __ pop_f(v1);
1409     __ fmuls(v0, v1, v0);
1410     break;
1411   case div:
1412     __ pop_f(v1);
1413     __ fdivs(v0, v1, v0);
1414     break;
1415   case rem:
1416     __ fmovs(v1, v0);
1417     __ pop_f(v0);
1418     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1419                          0, 2, MacroAssembler::ret_type_float);
1420     break;
1421   default:
1422     ShouldNotReachHere();
1423     break;
1424   }
1425 }
1426 
1427 void TemplateTable::dop2(Operation op)
1428 {
1429   transition(dtos, dtos);
1430   switch (op) {
1431   case add:
1432     // n.b. use ldrd because this is a 64 bit slot
1433     __ pop_d(v1);
1434     __ faddd(v0, v1, v0);
1435     break;
1436   case sub:
1437     __ pop_d(v1);
1438     __ fsubd(v0, v1, v0);
1439     break;
1440   case mul:
1441     __ pop_d(v1);
1442     __ fmuld(v0, v1, v0);
1443     break;
1444   case div:
1445     __ pop_d(v1);
1446     __ fdivd(v0, v1, v0);
1447     break;
1448   case rem:
1449     __ fmovd(v1, v0);
1450     __ pop_d(v0);
1451     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1452                          0, 2, MacroAssembler::ret_type_double);
1453     break;
1454   default:
1455     ShouldNotReachHere();
1456     break;
1457   }
1458 }
1459 
1460 void TemplateTable::ineg()
1461 {
1462   transition(itos, itos);
1463   __ negw(r0, r0);
1464 
1465 }
1466 
1467 void TemplateTable::lneg()
1468 {
1469   transition(ltos, ltos);
1470   __ neg(r0, r0);
1471 }
1472 
1473 void TemplateTable::fneg()
1474 {
1475   transition(ftos, ftos);
1476   __ fnegs(v0, v0);
1477 }
1478 
1479 void TemplateTable::dneg()
1480 {
1481   transition(dtos, dtos);
1482   __ fnegd(v0, v0);
1483 }
1484 
1485 void TemplateTable::iinc()
1486 {
1487   transition(vtos, vtos);
1488   __ load_signed_byte(r1, at_bcp(2)); // get constant
1489   locals_index(r2);
1490   __ ldr(r0, iaddress(r2));
1491   __ addw(r0, r0, r1);
1492   __ str(r0, iaddress(r2));
1493 }
1494 
1495 void TemplateTable::wide_iinc()
1496 {
1497   transition(vtos, vtos);
1498   // __ mov(r1, zr);
1499   __ ldrw(r1, at_bcp(2)); // get constant and index
1500   __ rev16(r1, r1);
1501   __ ubfx(r2, r1, 0, 16);
1502   __ neg(r2, r2);
1503   __ sbfx(r1, r1, 16, 16);
1504   __ ldr(r0, iaddress(r2));
1505   __ addw(r0, r0, r1);
1506   __ str(r0, iaddress(r2));
1507 }
1508 
1509 void TemplateTable::convert()
1510 {
1511   // Checking
1512 #ifdef ASSERT
1513   {
1514     TosState tos_in  = ilgl;
1515     TosState tos_out = ilgl;
1516     switch (bytecode()) {
1517     case Bytecodes::_i2l: // fall through
1518     case Bytecodes::_i2f: // fall through
1519     case Bytecodes::_i2d: // fall through
1520     case Bytecodes::_i2b: // fall through
1521     case Bytecodes::_i2c: // fall through
1522     case Bytecodes::_i2s: tos_in = itos; break;
1523     case Bytecodes::_l2i: // fall through
1524     case Bytecodes::_l2f: // fall through
1525     case Bytecodes::_l2d: tos_in = ltos; break;
1526     case Bytecodes::_f2i: // fall through
1527     case Bytecodes::_f2l: // fall through
1528     case Bytecodes::_f2d: tos_in = ftos; break;
1529     case Bytecodes::_d2i: // fall through
1530     case Bytecodes::_d2l: // fall through
1531     case Bytecodes::_d2f: tos_in = dtos; break;
1532     default             : ShouldNotReachHere();
1533     }
1534     switch (bytecode()) {
1535     case Bytecodes::_l2i: // fall through
1536     case Bytecodes::_f2i: // fall through
1537     case Bytecodes::_d2i: // fall through
1538     case Bytecodes::_i2b: // fall through
1539     case Bytecodes::_i2c: // fall through
1540     case Bytecodes::_i2s: tos_out = itos; break;
1541     case Bytecodes::_i2l: // fall through
1542     case Bytecodes::_f2l: // fall through
1543     case Bytecodes::_d2l: tos_out = ltos; break;
1544     case Bytecodes::_i2f: // fall through
1545     case Bytecodes::_l2f: // fall through
1546     case Bytecodes::_d2f: tos_out = ftos; break;
1547     case Bytecodes::_i2d: // fall through
1548     case Bytecodes::_l2d: // fall through
1549     case Bytecodes::_f2d: tos_out = dtos; break;
1550     default             : ShouldNotReachHere();
1551     }
1552     transition(tos_in, tos_out);
1553   }
1554 #endif // ASSERT
1555   // static const int64_t is_nan = 0x8000000000000000L;
1556 
1557   // Conversion
1558   switch (bytecode()) {
1559   case Bytecodes::_i2l:
1560     __ sxtw(r0, r0);
1561     break;
1562   case Bytecodes::_i2f:
1563     __ scvtfws(v0, r0);
1564     break;
1565   case Bytecodes::_i2d:
1566     __ scvtfwd(v0, r0);
1567     break;
1568   case Bytecodes::_i2b:
1569     __ sxtbw(r0, r0);
1570     break;
1571   case Bytecodes::_i2c:
1572     __ uxthw(r0, r0);
1573     break;
1574   case Bytecodes::_i2s:
1575     __ sxthw(r0, r0);
1576     break;
1577   case Bytecodes::_l2i:
1578     __ uxtw(r0, r0);
1579     break;
1580   case Bytecodes::_l2f:
1581     __ scvtfs(v0, r0);
1582     break;
1583   case Bytecodes::_l2d:
1584     __ scvtfd(v0, r0);
1585     break;
1586   case Bytecodes::_f2i:
1587   {
1588     Label L_Okay;
1589     __ clear_fpsr();
1590     __ fcvtzsw(r0, v0);
1591     __ get_fpsr(r1);
1592     __ cbzw(r1, L_Okay);
1593     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1594                          0, 1, MacroAssembler::ret_type_integral);
1595     __ bind(L_Okay);
1596   }
1597     break;
1598   case Bytecodes::_f2l:
1599   {
1600     Label L_Okay;
1601     __ clear_fpsr();
1602     __ fcvtzs(r0, v0);
1603     __ get_fpsr(r1);
1604     __ cbzw(r1, L_Okay);
1605     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1606                          0, 1, MacroAssembler::ret_type_integral);
1607     __ bind(L_Okay);
1608   }
1609     break;
1610   case Bytecodes::_f2d:
1611     __ fcvts(v0, v0);
1612     break;
1613   case Bytecodes::_d2i:
1614   {
1615     Label L_Okay;
1616     __ clear_fpsr();
1617     __ fcvtzdw(r0, v0);
1618     __ get_fpsr(r1);
1619     __ cbzw(r1, L_Okay);
1620     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1621                          0, 1, MacroAssembler::ret_type_integral);
1622     __ bind(L_Okay);
1623   }
1624     break;
1625   case Bytecodes::_d2l:
1626   {
1627     Label L_Okay;
1628     __ clear_fpsr();
1629     __ fcvtzd(r0, v0);
1630     __ get_fpsr(r1);
1631     __ cbzw(r1, L_Okay);
1632     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1633                          0, 1, MacroAssembler::ret_type_integral);
1634     __ bind(L_Okay);
1635   }
1636     break;
1637   case Bytecodes::_d2f:
1638     __ fcvtd(v0, v0);
1639     break;
1640   default:
1641     ShouldNotReachHere();
1642   }
1643 }
1644 
1645 void TemplateTable::lcmp()
1646 {
1647   transition(ltos, itos);
1648   Label done;
1649   __ pop_l(r1);
1650   __ cmp(r1, r0);
1651   __ mov(r0, (u_int64_t)-1L);
1652   __ br(Assembler::LT, done);
1653   // __ mov(r0, 1UL);
1654   // __ csel(r0, r0, zr, Assembler::NE);
1655   // and here is a faster way
1656   __ csinc(r0, zr, zr, Assembler::EQ);
1657   __ bind(done);
1658 }
1659 
1660 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1661 {
1662   Label done;
1663   if (is_float) {
1664     // XXX get rid of pop here, use ... reg, mem32
1665     __ pop_f(v1);
1666     __ fcmps(v1, v0);
1667   } else {
1668     // XXX get rid of pop here, use ... reg, mem64
1669     __ pop_d(v1);
1670     __ fcmpd(v1, v0);
1671   }
1672   if (unordered_result < 0) {
1673     // we want -1 for unordered or less than, 0 for equal and 1 for
1674     // greater than.
1675     __ mov(r0, (u_int64_t)-1L);
1676     // for FP LT tests less than or unordered
1677     __ br(Assembler::LT, done);
1678     // install 0 for EQ otherwise 1
1679     __ csinc(r0, zr, zr, Assembler::EQ);
1680   } else {
1681     // we want -1 for less than, 0 for equal and 1 for unordered or
1682     // greater than.
1683     __ mov(r0, 1L);
1684     // for FP HI tests greater than or unordered
1685     __ br(Assembler::HI, done);
1686     // install 0 for EQ otherwise ~0
1687     __ csinv(r0, zr, zr, Assembler::EQ);
1688 
1689   }
1690   __ bind(done);
1691 }
1692 
1693 void TemplateTable::branch(bool is_jsr, bool is_wide)
1694 {
1695   // We might be moving to a safepoint.  The thread which calls
1696   // Interpreter::notice_safepoints() will effectively flush its cache
1697   // when it makes a system call, but we need to do something to
1698   // ensure that we see the changed dispatch table.
1699   __ membar(MacroAssembler::LoadLoad);
1700 
1701   __ profile_taken_branch(r0, r1);
1702   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1703                              InvocationCounter::counter_offset();
1704   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1705                               InvocationCounter::counter_offset();
1706 
1707   // load branch displacement
1708   if (!is_wide) {
1709     __ ldrh(r2, at_bcp(1));
1710     __ rev16(r2, r2);
1711     // sign extend the 16 bit value in r2
1712     __ sbfm(r2, r2, 0, 15);
1713   } else {
1714     __ ldrw(r2, at_bcp(1));
1715     __ revw(r2, r2);
1716     // sign extend the 32 bit value in r2
1717     __ sbfm(r2, r2, 0, 31);
1718   }
1719 
1720   // Handle all the JSR stuff here, then exit.
1721   // It's much shorter and cleaner than intermingling with the non-JSR
1722   // normal-branch stuff occurring below.
1723 
1724   if (is_jsr) {
1725     // Pre-load the next target bytecode into rscratch1
1726     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1727     // compute return address as bci
1728     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1729     __ add(rscratch2, rscratch2,
1730            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1731     __ sub(r1, rbcp, rscratch2);
1732     __ push_i(r1);
1733     // Adjust the bcp by the 16-bit displacement in r2
1734     __ add(rbcp, rbcp, r2);
1735     __ dispatch_only(vtos);
1736     return;
1737   }
1738 
1739   // Normal (non-jsr) branch handling
1740 
1741   // Adjust the bcp by the displacement in r2
1742   __ add(rbcp, rbcp, r2);
1743 
1744   assert(UseLoopCounter || !UseOnStackReplacement,
1745          "on-stack-replacement requires loop counters");
1746   Label backedge_counter_overflow;
1747   Label profile_method;
1748   Label dispatch;
1749   if (UseLoopCounter) {
1750     // increment backedge counter for backward branches
1751     // r0: MDO
1752     // w1: MDO bumped taken-count
1753     // r2: target offset
1754     __ cmp(r2, zr);
1755     __ br(Assembler::GT, dispatch); // count only if backward branch
1756 
1757     // ECN: FIXME: This code smells
1758     // check if MethodCounters exists
1759     Label has_counters;
1760     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1761     __ cbnz(rscratch1, has_counters);
1762     __ push(r0);
1763     __ push(r1);
1764     __ push(r2);
1765     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1766             InterpreterRuntime::build_method_counters), rmethod);
1767     __ pop(r2);
1768     __ pop(r1);
1769     __ pop(r0);
1770     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1771     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1772     __ bind(has_counters);
1773 
1774     if (TieredCompilation) {
1775       Label no_mdo;
1776       int increment = InvocationCounter::count_increment;
1777       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1778       if (ProfileInterpreter) {
1779         // Are we profiling?
1780         __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1781         __ cbz(r1, no_mdo);
1782         // Increment the MDO backedge counter
1783         const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1784                                            in_bytes(InvocationCounter::counter_offset()));
1785         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1786                                    r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1787         __ b(dispatch);
1788       }
1789       __ bind(no_mdo);
1790       // Increment backedge counter in MethodCounters*
1791       __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1792       __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1793                                  r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1794     } else {
1795       // increment counter
1796       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1797       __ ldrw(r0, Address(rscratch2, be_offset));        // load backedge counter
1798       __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1799       __ strw(rscratch1, Address(rscratch2, be_offset));        // store counter
1800 
1801       __ ldrw(r0, Address(rscratch2, inv_offset));    // load invocation counter
1802       __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1803       __ addw(r0, r0, rscratch1);        // add both counters
1804 
1805       if (ProfileInterpreter) {
1806         // Test to see if we should create a method data oop
1807         __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1808         __ ldrw(rscratch1, rscratch1);
1809         __ cmpw(r0, rscratch1);
1810         __ br(Assembler::LT, dispatch);
1811 
1812         // if no method data exists, go to profile method
1813         __ test_method_data_pointer(r0, profile_method);
1814 
1815         if (UseOnStackReplacement) {
1816           // check for overflow against w1 which is the MDO taken count
1817           __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1818           __ ldrw(rscratch1, rscratch1);
1819           __ cmpw(r1, rscratch1);
1820           __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1821 
1822           // When ProfileInterpreter is on, the backedge_count comes
1823           // from the MethodData*, which value does not get reset on
1824           // the call to frequency_counter_overflow().  To avoid
1825           // excessive calls to the overflow routine while the method is
1826           // being compiled, add a second test to make sure the overflow
1827           // function is called only once every overflow_frequency.
1828           const int overflow_frequency = 1024;
1829           __ andsw(r1, r1, overflow_frequency - 1);
1830           __ br(Assembler::EQ, backedge_counter_overflow);
1831 
1832         }
1833       } else {
1834         if (UseOnStackReplacement) {
1835           // check for overflow against w0, which is the sum of the
1836           // counters
1837           __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1838           __ ldrw(rscratch1, rscratch1);
1839           __ cmpw(r0, rscratch1);
1840           __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1841         }
1842       }
1843     }
1844   }
1845   __ bind(dispatch);
1846 
1847   // Pre-load the next target bytecode into rscratch1
1848   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1849 
1850   // continue with the bytecode @ target
1851   // rscratch1: target bytecode
1852   // rbcp: target bcp
1853   __ dispatch_only(vtos);
1854 
1855   if (UseLoopCounter) {
1856     if (ProfileInterpreter) {
1857       // Out-of-line code to allocate method data oop.
1858       __ bind(profile_method);
1859       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1860       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1861       __ set_method_data_pointer_for_bcp();
1862       __ b(dispatch);
1863     }
1864 
1865     if (TieredCompilation || UseOnStackReplacement) {
1866       // invocation counter overflow
1867       __ bind(backedge_counter_overflow);
1868       __ neg(r2, r2);
1869       __ add(r2, r2, rbcp);     // branch bcp
1870       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1871       __ call_VM(noreg,
1872                  CAST_FROM_FN_PTR(address,
1873                                   InterpreterRuntime::frequency_counter_overflow),
1874                  r2);
1875       if (!UseOnStackReplacement)
1876         __ b(dispatch);
1877     }
1878 
1879     if (UseOnStackReplacement) {
1880       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1881 
1882       // r0: osr nmethod (osr ok) or NULL (osr not possible)
1883       // w1: target bytecode
1884       // r2: scratch
1885       __ cbz(r0, dispatch);     // test result -- no osr if null
1886       // nmethod may have been invalidated (VM may block upon call_VM return)
1887       __ ldrw(r2, Address(r0, nmethod::entry_bci_offset()));
1888       // InvalidOSREntryBci == -2 which overflows cmpw as unsigned
1889       // use cmnw against -InvalidOSREntryBci which does the same thing
1890       __ cmn(r2, -InvalidOSREntryBci);
1891       __ br(Assembler::EQ, dispatch);
1892 
1893       // We have the address of an on stack replacement routine in r0
1894       // We need to prepare to execute the OSR method. First we must
1895       // migrate the locals and monitors off of the stack.
1896 
1897       __ mov(r19, r0);                             // save the nmethod
1898 
1899       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1900 
1901       // r0 is OSR buffer, move it to expected parameter location
1902       __ mov(j_rarg0, r0);
1903 
1904       // remove activation
1905       // get sender esp
1906       __ ldr(esp,
1907           Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1908       // remove frame anchor
1909       __ leave();
1910       // Ensure compiled code always sees stack at proper alignment
1911       __ andr(sp, esp, -16);
1912 
1913       // and begin the OSR nmethod
1914       __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1915       __ br(rscratch1);
1916     }
1917   }
1918 }
1919 
1920 
1921 void TemplateTable::if_0cmp(Condition cc)
1922 {
1923   transition(itos, vtos);
1924   // assume branch is more often taken than not (loops use backward branches)
1925   Label not_taken;
1926   if (cc == equal)
1927     __ cbnzw(r0, not_taken);
1928   else if (cc == not_equal)
1929     __ cbzw(r0, not_taken);
1930   else {
1931     __ andsw(zr, r0, r0);
1932     __ br(j_not(cc), not_taken);
1933   }
1934 
1935   branch(false, false);
1936   __ bind(not_taken);
1937   __ profile_not_taken_branch(r0);
1938 }
1939 
1940 void TemplateTable::if_icmp(Condition cc)
1941 {
1942   transition(itos, vtos);
1943   // assume branch is more often taken than not (loops use backward branches)
1944   Label not_taken;
1945   __ pop_i(r1);
1946   __ cmpw(r1, r0, Assembler::LSL);
1947   __ br(j_not(cc), not_taken);
1948   branch(false, false);
1949   __ bind(not_taken);
1950   __ profile_not_taken_branch(r0);
1951 }
1952 
1953 void TemplateTable::if_nullcmp(Condition cc)
1954 {
1955   transition(atos, vtos);
1956   // assume branch is more often taken than not (loops use backward branches)
1957   Label not_taken;
1958   if (cc == equal)
1959     __ cbnz(r0, not_taken);
1960   else
1961     __ cbz(r0, not_taken);
1962   branch(false, false);
1963   __ bind(not_taken);
1964   __ profile_not_taken_branch(r0);
1965 }
1966 
1967 void TemplateTable::if_acmp(Condition cc)
1968 {
1969   transition(atos, vtos);
1970   // assume branch is more often taken than not (loops use backward branches)
1971   Label not_taken;
1972   __ pop_ptr(r1);
1973   __ cmpoops(r1, r0);
1974   __ br(j_not(cc), not_taken);
1975   branch(false, false);
1976   __ bind(not_taken);
1977   __ profile_not_taken_branch(r0);
1978 }
1979 
1980 void TemplateTable::ret() {
1981   transition(vtos, vtos);
1982   // We might be moving to a safepoint.  The thread which calls
1983   // Interpreter::notice_safepoints() will effectively flush its cache
1984   // when it makes a system call, but we need to do something to
1985   // ensure that we see the changed dispatch table.
1986   __ membar(MacroAssembler::LoadLoad);
1987 
1988   locals_index(r1);
1989   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1990   __ profile_ret(r1, r2);
1991   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1992   __ lea(rbcp, Address(rbcp, r1));
1993   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1994   __ dispatch_next(vtos);
1995 }
1996 
1997 void TemplateTable::wide_ret() {
1998   transition(vtos, vtos);
1999   locals_index_wide(r1);
2000   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2001   __ profile_ret(r1, r2);
2002   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2003   __ lea(rbcp, Address(rbcp, r1));
2004   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2005   __ dispatch_next(vtos);
2006 }
2007 
2008 
2009 void TemplateTable::tableswitch() {
2010   Label default_case, continue_execution;
2011   transition(itos, vtos);
2012   // align rbcp
2013   __ lea(r1, at_bcp(BytesPerInt));
2014   __ andr(r1, r1, -BytesPerInt);
2015   // load lo & hi
2016   __ ldrw(r2, Address(r1, BytesPerInt));
2017   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2018   __ rev32(r2, r2);
2019   __ rev32(r3, r3);
2020   // check against lo & hi
2021   __ cmpw(r0, r2);
2022   __ br(Assembler::LT, default_case);
2023   __ cmpw(r0, r3);
2024   __ br(Assembler::GT, default_case);
2025   // lookup dispatch offset
2026   __ subw(r0, r0, r2);
2027   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2028   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2029   __ profile_switch_case(r0, r1, r2);
2030   // continue execution
2031   __ bind(continue_execution);
2032   __ rev32(r3, r3);
2033   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2034   __ add(rbcp, rbcp, r3, ext::sxtw);
2035   __ dispatch_only(vtos);
2036   // handle default
2037   __ bind(default_case);
2038   __ profile_switch_default(r0);
2039   __ ldrw(r3, Address(r1, 0));
2040   __ b(continue_execution);
2041 }
2042 
2043 void TemplateTable::lookupswitch() {
2044   transition(itos, itos);
2045   __ stop("lookupswitch bytecode should have been rewritten");
2046 }
2047 
2048 void TemplateTable::fast_linearswitch() {
2049   transition(itos, vtos);
2050   Label loop_entry, loop, found, continue_execution;
2051   // bswap r0 so we can avoid bswapping the table entries
2052   __ rev32(r0, r0);
2053   // align rbcp
2054   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2055                                     // this instruction (change offsets
2056                                     // below)
2057   __ andr(r19, r19, -BytesPerInt);
2058   // set counter
2059   __ ldrw(r1, Address(r19, BytesPerInt));
2060   __ rev32(r1, r1);
2061   __ b(loop_entry);
2062   // table search
2063   __ bind(loop);
2064   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2065   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2066   __ cmpw(r0, rscratch1);
2067   __ br(Assembler::EQ, found);
2068   __ bind(loop_entry);
2069   __ subs(r1, r1, 1);
2070   __ br(Assembler::PL, loop);
2071   // default case
2072   __ profile_switch_default(r0);
2073   __ ldrw(r3, Address(r19, 0));
2074   __ b(continue_execution);
2075   // entry found -> get offset
2076   __ bind(found);
2077   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2078   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2079   __ profile_switch_case(r1, r0, r19);
2080   // continue execution
2081   __ bind(continue_execution);
2082   __ rev32(r3, r3);
2083   __ add(rbcp, rbcp, r3, ext::sxtw);
2084   __ ldrb(rscratch1, Address(rbcp, 0));
2085   __ dispatch_only(vtos);
2086 }
2087 
2088 void TemplateTable::fast_binaryswitch() {
2089   transition(itos, vtos);
2090   // Implementation using the following core algorithm:
2091   //
2092   // int binary_search(int key, LookupswitchPair* array, int n) {
2093   //   // Binary search according to "Methodik des Programmierens" by
2094   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2095   //   int i = 0;
2096   //   int j = n;
2097   //   while (i+1 < j) {
2098   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2099   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2100   //     // where a stands for the array and assuming that the (inexisting)
2101   //     // element a[n] is infinitely big.
2102   //     int h = (i + j) >> 1;
2103   //     // i < h < j
2104   //     if (key < array[h].fast_match()) {
2105   //       j = h;
2106   //     } else {
2107   //       i = h;
2108   //     }
2109   //   }
2110   //   // R: a[i] <= key < a[i+1] or Q
2111   //   // (i.e., if key is within array, i is the correct index)
2112   //   return i;
2113   // }
2114 
2115   // Register allocation
2116   const Register key   = r0; // already set (tosca)
2117   const Register array = r1;
2118   const Register i     = r2;
2119   const Register j     = r3;
2120   const Register h     = rscratch1;
2121   const Register temp  = rscratch2;
2122 
2123   // Find array start
2124   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2125                                           // get rid of this
2126                                           // instruction (change
2127                                           // offsets below)
2128   __ andr(array, array, -BytesPerInt);
2129 
2130   // Initialize i & j
2131   __ mov(i, 0);                            // i = 0;
2132   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2133 
2134   // Convert j into native byteordering
2135   __ rev32(j, j);
2136 
2137   // And start
2138   Label entry;
2139   __ b(entry);
2140 
2141   // binary search loop
2142   {
2143     Label loop;
2144     __ bind(loop);
2145     // int h = (i + j) >> 1;
2146     __ addw(h, i, j);                           // h = i + j;
2147     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2148     // if (key < array[h].fast_match()) {
2149     //   j = h;
2150     // } else {
2151     //   i = h;
2152     // }
2153     // Convert array[h].match to native byte-ordering before compare
2154     __ ldr(temp, Address(array, h, Address::lsl(3)));
2155     __ rev32(temp, temp);
2156     __ cmpw(key, temp);
2157     // j = h if (key <  array[h].fast_match())
2158     __ csel(j, h, j, Assembler::LT);
2159     // i = h if (key >= array[h].fast_match())
2160     __ csel(i, h, i, Assembler::GE);
2161     // while (i+1 < j)
2162     __ bind(entry);
2163     __ addw(h, i, 1);          // i+1
2164     __ cmpw(h, j);             // i+1 < j
2165     __ br(Assembler::LT, loop);
2166   }
2167 
2168   // end of binary search, result index is i (must check again!)
2169   Label default_case;
2170   // Convert array[i].match to native byte-ordering before compare
2171   __ ldr(temp, Address(array, i, Address::lsl(3)));
2172   __ rev32(temp, temp);
2173   __ cmpw(key, temp);
2174   __ br(Assembler::NE, default_case);
2175 
2176   // entry found -> j = offset
2177   __ add(j, array, i, ext::uxtx, 3);
2178   __ ldrw(j, Address(j, BytesPerInt));
2179   __ profile_switch_case(i, key, array);
2180   __ rev32(j, j);
2181   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2182   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2183   __ dispatch_only(vtos);
2184 
2185   // default case -> j = default offset
2186   __ bind(default_case);
2187   __ profile_switch_default(i);
2188   __ ldrw(j, Address(array, -2 * BytesPerInt));
2189   __ rev32(j, j);
2190   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2191   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2192   __ dispatch_only(vtos);
2193 }
2194 
2195 
2196 void TemplateTable::_return(TosState state)
2197 {
2198   transition(state, state);
2199   assert(_desc->calls_vm(),
2200          "inconsistent calls_vm information"); // call in remove_activation
2201 
2202   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2203     assert(state == vtos, "only valid state");
2204 
2205     __ ldr(c_rarg1, aaddress(0));
2206     __ load_klass(r3, c_rarg1);
2207     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2208     __ tst(r3, JVM_ACC_HAS_FINALIZER);
2209     Label skip_register_finalizer;
2210     __ br(Assembler::EQ, skip_register_finalizer);
2211 
2212     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2213 
2214     __ bind(skip_register_finalizer);
2215   }
2216 
2217   // Issue a StoreStore barrier after all stores but before return
2218   // from any constructor for any class with a final field.  We don't
2219   // know if this is a finalizer, so we always do so.
2220   if (_desc->bytecode() == Bytecodes::_return)
2221     __ membar(MacroAssembler::StoreStore);
2222 
2223   // Narrow result if state is itos but result type is smaller.
2224   // Need to narrow in the return bytecode rather than in generate_return_entry
2225   // since compiled code callers expect the result to already be narrowed.
2226   if (state == itos) {
2227     __ narrow(r0);
2228   }
2229 
2230   __ remove_activation(state);
2231   __ ret(lr);
2232 }
2233 
2234 // ----------------------------------------------------------------------------
2235 // Volatile variables demand their effects be made known to all CPU's
2236 // in order.  Store buffers on most chips allow reads & writes to
2237 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2238 // without some kind of memory barrier (i.e., it's not sufficient that
2239 // the interpreter does not reorder volatile references, the hardware
2240 // also must not reorder them).
2241 //
2242 // According to the new Java Memory Model (JMM):
2243 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2244 //     writes act as aquire & release, so:
2245 // (2) A read cannot let unrelated NON-volatile memory refs that
2246 //     happen after the read float up to before the read.  It's OK for
2247 //     non-volatile memory refs that happen before the volatile read to
2248 //     float down below it.
2249 // (3) Similar a volatile write cannot let unrelated NON-volatile
2250 //     memory refs that happen BEFORE the write float down to after the
2251 //     write.  It's OK for non-volatile memory refs that happen after the
2252 //     volatile write to float up before it.
2253 //
2254 // We only put in barriers around volatile refs (they are expensive),
2255 // not _between_ memory refs (that would require us to track the
2256 // flavor of the previous memory refs).  Requirements (2) and (3)
2257 // require some barriers before volatile stores and after volatile
2258 // loads.  These nearly cover requirement (1) but miss the
2259 // volatile-store-volatile-load case.  This final case is placed after
2260 // volatile-stores although it could just as well go before
2261 // volatile-loads.
2262 
2263 void TemplateTable::resolve_cache_and_index(int byte_no,
2264                                             Register Rcache,
2265                                             Register index,
2266                                             size_t index_size) {
2267   const Register temp = r19;
2268   assert_different_registers(Rcache, index, temp);
2269 
2270   Label resolved;
2271   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2272   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2273   __ cmp(temp, (int) bytecode());  // have we resolved this bytecode?
2274   __ br(Assembler::EQ, resolved);
2275 
2276   // resolve first time through
2277   address entry;
2278   switch (bytecode()) {
2279   case Bytecodes::_getstatic:
2280   case Bytecodes::_putstatic:
2281   case Bytecodes::_getfield:
2282   case Bytecodes::_putfield:
2283     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2284     break;
2285   case Bytecodes::_invokevirtual:
2286   case Bytecodes::_invokespecial:
2287   case Bytecodes::_invokestatic:
2288   case Bytecodes::_invokeinterface:
2289     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2290     break;
2291   case Bytecodes::_invokehandle:
2292     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2293     break;
2294   case Bytecodes::_invokedynamic:
2295     entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2296     break;
2297   default:
2298     fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2299     break;
2300   }
2301   __ mov(temp, (int) bytecode());
2302   __ call_VM(noreg, entry, temp);
2303 
2304   // Update registers with resolved info
2305   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2306   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2307   // so all clients ofthis method must be modified accordingly
2308   __ bind(resolved);
2309 }
2310 
2311 // The Rcache and index registers must be set before call
2312 // n.b unlike x86 cache already includes the index offset
2313 void TemplateTable::load_field_cp_cache_entry(Register obj,
2314                                               Register cache,
2315                                               Register index,
2316                                               Register off,
2317                                               Register flags,
2318                                               bool is_static = false) {
2319   assert_different_registers(cache, index, flags, off);
2320 
2321   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2322   // Field offset
2323   __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2324                                           ConstantPoolCacheEntry::f2_offset())));
2325   // Flags
2326   __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2327                                            ConstantPoolCacheEntry::flags_offset())));
2328 
2329   // klass overwrite register
2330   if (is_static) {
2331     __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2332                                         ConstantPoolCacheEntry::f1_offset())));
2333     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2334     __ ldr(obj, Address(obj, mirror_offset));
2335   }
2336 }
2337 
2338 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2339                                                Register method,
2340                                                Register itable_index,
2341                                                Register flags,
2342                                                bool is_invokevirtual,
2343                                                bool is_invokevfinal, /*unused*/
2344                                                bool is_invokedynamic) {
2345   // setup registers
2346   const Register cache = rscratch2;
2347   const Register index = r4;
2348   assert_different_registers(method, flags);
2349   assert_different_registers(method, cache, index);
2350   assert_different_registers(itable_index, flags);
2351   assert_different_registers(itable_index, cache, index);
2352   // determine constant pool cache field offsets
2353   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2354   const int method_offset = in_bytes(
2355     ConstantPoolCache::base_offset() +
2356       (is_invokevirtual
2357        ? ConstantPoolCacheEntry::f2_offset()
2358        : ConstantPoolCacheEntry::f1_offset()));
2359   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2360                                     ConstantPoolCacheEntry::flags_offset());
2361   // access constant pool cache fields
2362   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2363                                     ConstantPoolCacheEntry::f2_offset());
2364 
2365   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2366   resolve_cache_and_index(byte_no, cache, index, index_size);
2367   __ ldr(method, Address(cache, method_offset));
2368 
2369   if (itable_index != noreg) {
2370     __ ldr(itable_index, Address(cache, index_offset));
2371   }
2372   __ ldrw(flags, Address(cache, flags_offset));
2373 }
2374 
2375 
2376 // The registers cache and index expected to be set before call.
2377 // Correct values of the cache and index registers are preserved.
2378 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2379                                             bool is_static, bool has_tos) {
2380   // do the JVMTI work here to avoid disturbing the register state below
2381   // We use c_rarg registers here because we want to use the register used in
2382   // the call to the VM
2383   if (JvmtiExport::can_post_field_access()) {
2384     // Check to see if a field access watch has been set before we
2385     // take the time to call into the VM.
2386     Label L1;
2387     assert_different_registers(cache, index, r0);
2388     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2389     __ ldrw(r0, Address(rscratch1));
2390     __ cbzw(r0, L1);
2391 
2392     __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2393     __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2394 
2395     if (is_static) {
2396       __ mov(c_rarg1, zr); // NULL object reference
2397     } else {
2398       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2399       __ verify_oop(c_rarg1);
2400     }
2401     // c_rarg1: object pointer or NULL
2402     // c_rarg2: cache entry pointer
2403     // c_rarg3: jvalue object on the stack
2404     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2405                                        InterpreterRuntime::post_field_access),
2406                c_rarg1, c_rarg2, c_rarg3);
2407     __ get_cache_and_index_at_bcp(cache, index, 1);
2408     __ bind(L1);
2409   }
2410 }
2411 
2412 void TemplateTable::pop_and_check_object(Register r)
2413 {
2414   __ pop_ptr(r);
2415   __ null_check(r);  // for field access must check obj.
2416   __ verify_oop(r);
2417 }
2418 
2419 void TemplateTable::getfield_or_static(int byte_no, bool is_static)
2420 {
2421   const Register cache = r2;
2422   const Register index = r3;
2423   const Register obj   = r4;
2424   const Register off   = r19;
2425   const Register flags = r0;
2426   const Register bc    = r4; // uses same reg as obj, so don't mix them
2427 
2428   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2429   jvmti_post_field_access(cache, index, is_static, false);
2430   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2431 
2432   if (!is_static) {
2433     // obj is on the stack
2434     pop_and_check_object(obj);
2435   }
2436   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, obj);
2437 
2438   const Address field(obj, off);
2439 
2440   Label Done, notByte, notBool, notInt, notShort, notChar,
2441               notLong, notFloat, notObj, notDouble;
2442 
2443   // x86 uses a shift and mask or wings it with a shift plus assert
2444   // the mask is not needed. aarch64 just uses bitfield extract
2445   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2446 
2447   assert(btos == 0, "change code, btos != 0");
2448   __ cbnz(flags, notByte);
2449 
2450   // btos
2451   __ load_signed_byte(r0, field);
2452   __ push(btos);
2453   // Rewrite bytecode to be faster
2454   if (!is_static) {
2455     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2456   }
2457   __ b(Done);
2458 
2459   __ bind(notByte);
2460   __ cmp(flags, ztos);
2461   __ br(Assembler::NE, notBool);
2462 
2463   // ztos (same code as btos)
2464   __ ldrsb(r0, field);
2465   __ push(ztos);
2466   // Rewrite bytecode to be faster
2467   if (!is_static) {
2468     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2469     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2470   }
2471   __ b(Done);
2472 
2473   __ bind(notBool);
2474   __ cmp(flags, atos);
2475   __ br(Assembler::NE, notObj);
2476   // atos
2477   __ load_heap_oop(r0, field);
2478   __ push(atos);
2479   if (!is_static) {
2480     patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2481   }
2482   __ b(Done);
2483 
2484   __ bind(notObj);
2485   __ cmp(flags, itos);
2486   __ br(Assembler::NE, notInt);
2487   // itos
2488   __ ldrw(r0, field);
2489   __ push(itos);
2490   // Rewrite bytecode to be faster
2491   if (!is_static) {
2492     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2493   }
2494   __ b(Done);
2495 
2496   __ bind(notInt);
2497   __ cmp(flags, ctos);
2498   __ br(Assembler::NE, notChar);
2499   // ctos
2500   __ load_unsigned_short(r0, field);
2501   __ push(ctos);
2502   // Rewrite bytecode to be faster
2503   if (!is_static) {
2504     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2505   }
2506   __ b(Done);
2507 
2508   __ bind(notChar);
2509   __ cmp(flags, stos);
2510   __ br(Assembler::NE, notShort);
2511   // stos
2512   __ load_signed_short(r0, field);
2513   __ push(stos);
2514   // Rewrite bytecode to be faster
2515   if (!is_static) {
2516     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2517   }
2518   __ b(Done);
2519 
2520   __ bind(notShort);
2521   __ cmp(flags, ltos);
2522   __ br(Assembler::NE, notLong);
2523   // ltos
2524   __ ldr(r0, field);
2525   __ push(ltos);
2526   // Rewrite bytecode to be faster
2527   if (!is_static) {
2528     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2529   }
2530   __ b(Done);
2531 
2532   __ bind(notLong);
2533   __ cmp(flags, ftos);
2534   __ br(Assembler::NE, notFloat);
2535   // ftos
2536   __ ldrs(v0, field);
2537   __ push(ftos);
2538   // Rewrite bytecode to be faster
2539   if (!is_static) {
2540     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2541   }
2542   __ b(Done);
2543 
2544   __ bind(notFloat);
2545 #ifdef ASSERT
2546   __ cmp(flags, dtos);
2547   __ br(Assembler::NE, notDouble);
2548 #endif
2549   // dtos
2550   __ ldrd(v0, field);
2551   __ push(dtos);
2552   // Rewrite bytecode to be faster
2553   if (!is_static) {
2554     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2555   }
2556 #ifdef ASSERT
2557   __ b(Done);
2558 
2559   __ bind(notDouble);
2560   __ stop("Bad state");
2561 #endif
2562 
2563   __ bind(Done);
2564   // It's really not worth bothering to check whether this field
2565   // really is volatile in the slow case.
2566   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2567 }
2568 
2569 
2570 void TemplateTable::getfield(int byte_no)
2571 {
2572   getfield_or_static(byte_no, false);
2573 }
2574 
2575 void TemplateTable::getstatic(int byte_no)
2576 {
2577   getfield_or_static(byte_no, true);
2578 }
2579 
2580 // The registers cache and index expected to be set before call.
2581 // The function may destroy various registers, just not the cache and index registers.
2582 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2583   transition(vtos, vtos);
2584 
2585   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2586 
2587   if (JvmtiExport::can_post_field_modification()) {
2588     // Check to see if a field modification watch has been set before
2589     // we take the time to call into the VM.
2590     Label L1;
2591     assert_different_registers(cache, index, r0);
2592     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2593     __ ldrw(r0, Address(rscratch1));
2594     __ cbz(r0, L1);
2595 
2596     __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2597 
2598     if (is_static) {
2599       // Life is simple.  Null out the object pointer.
2600       __ mov(c_rarg1, zr);
2601     } else {
2602       // Life is harder. The stack holds the value on top, followed by
2603       // the object.  We don't know the size of the value, though; it
2604       // could be one or two words depending on its type. As a result,
2605       // we must find the type to determine where the object is.
2606       __ ldrw(c_rarg3, Address(c_rarg2,
2607                                in_bytes(cp_base_offset +
2608                                         ConstantPoolCacheEntry::flags_offset())));
2609       __ lsr(c_rarg3, c_rarg3,
2610              ConstantPoolCacheEntry::tos_state_shift);
2611       ConstantPoolCacheEntry::verify_tos_state_shift();
2612       Label nope2, done, ok;
2613       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2614       __ cmpw(c_rarg3, ltos);
2615       __ br(Assembler::EQ, ok);
2616       __ cmpw(c_rarg3, dtos);
2617       __ br(Assembler::NE, nope2);
2618       __ bind(ok);
2619       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2620       __ bind(nope2);
2621     }
2622     // cache entry pointer
2623     __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2624     // object (tos)
2625     __ mov(c_rarg3, esp);
2626     // c_rarg1: object pointer set up above (NULL if static)
2627     // c_rarg2: cache entry pointer
2628     // c_rarg3: jvalue object on the stack
2629     __ call_VM(noreg,
2630                CAST_FROM_FN_PTR(address,
2631                                 InterpreterRuntime::post_field_modification),
2632                c_rarg1, c_rarg2, c_rarg3);
2633     __ get_cache_and_index_at_bcp(cache, index, 1);
2634     __ bind(L1);
2635   }
2636 }
2637 
2638 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2639   transition(vtos, vtos);
2640 
2641   const Register cache = r2;
2642   const Register index = r3;
2643   const Register obj   = r2;
2644   const Register off   = r19;
2645   const Register flags = r0;
2646   const Register bc    = r4;
2647 
2648   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2649   jvmti_post_field_mod(cache, index, is_static);
2650   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2651 
2652   Label Done;
2653   __ mov(r5, flags);
2654 
2655   {
2656     Label notVolatile;
2657     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2658     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2659     __ bind(notVolatile);
2660   }
2661 
2662   // field address
2663   const Address field(obj, off);
2664 
2665   Label notByte, notBool, notInt, notShort, notChar,
2666         notLong, notFloat, notObj, notDouble;
2667 
2668   // x86 uses a shift and mask or wings it with a shift plus assert
2669   // the mask is not needed. aarch64 just uses bitfield extract
2670   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2671 
2672   assert(btos == 0, "change code, btos != 0");
2673   __ cbnz(flags, notByte);
2674 
2675   // btos
2676   {
2677     __ pop(btos);
2678     if (!is_static) pop_and_check_object(obj);
2679     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2680     __ strb(r0, field);
2681     if (!is_static) {
2682       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2683     }
2684     __ b(Done);
2685   }
2686 
2687   __ bind(notByte);
2688   __ cmp(flags, ztos);
2689   __ br(Assembler::NE, notBool);
2690 
2691   // ztos
2692   {
2693     __ pop(ztos);
2694     if (!is_static) pop_and_check_object(obj);
2695     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2696     __ andw(r0, r0, 0x1);
2697     __ strb(r0, field);
2698     if (!is_static) {
2699       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2700     }
2701     __ b(Done);
2702   }
2703 
2704   __ bind(notBool);
2705   __ cmp(flags, atos);
2706   __ br(Assembler::NE, notObj);
2707 
2708   // atos
2709   {
2710     __ pop(atos);
2711     if (!is_static) pop_and_check_object(obj);
2712     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2713     // Store into the field
2714     do_oop_store(_masm, field, r0, _bs->kind(), false);
2715     if (!is_static) {
2716       patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2717     }
2718     __ b(Done);
2719   }
2720 
2721   __ bind(notObj);
2722   __ cmp(flags, itos);
2723   __ br(Assembler::NE, notInt);
2724 
2725   // itos
2726   {
2727     __ pop(itos);
2728     if (!is_static) pop_and_check_object(obj);
2729     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2730     __ strw(r0, field);
2731     if (!is_static) {
2732       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2733     }
2734     __ b(Done);
2735   }
2736 
2737   __ bind(notInt);
2738   __ cmp(flags, ctos);
2739   __ br(Assembler::NE, notChar);
2740 
2741   // ctos
2742   {
2743     __ pop(ctos);
2744     if (!is_static) pop_and_check_object(obj);
2745     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2746     __ strh(r0, field);
2747     if (!is_static) {
2748       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2749     }
2750     __ b(Done);
2751   }
2752 
2753   __ bind(notChar);
2754   __ cmp(flags, stos);
2755   __ br(Assembler::NE, notShort);
2756 
2757   // stos
2758   {
2759     __ pop(stos);
2760     if (!is_static) pop_and_check_object(obj);
2761     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2762     __ strh(r0, field);
2763     if (!is_static) {
2764       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2765     }
2766     __ b(Done);
2767   }
2768 
2769   __ bind(notShort);
2770   __ cmp(flags, ltos);
2771   __ br(Assembler::NE, notLong);
2772 
2773   // ltos
2774   {
2775     __ pop(ltos);
2776     if (!is_static) pop_and_check_object(obj);
2777     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2778     __ str(r0, field);
2779     if (!is_static) {
2780       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2781     }
2782     __ b(Done);
2783   }
2784 
2785   __ bind(notLong);
2786   __ cmp(flags, ftos);
2787   __ br(Assembler::NE, notFloat);
2788 
2789   // ftos
2790   {
2791     __ pop(ftos);
2792     if (!is_static) pop_and_check_object(obj);
2793     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2794     __ strs(v0, field);
2795     if (!is_static) {
2796       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2797     }
2798     __ b(Done);
2799   }
2800 
2801   __ bind(notFloat);
2802 #ifdef ASSERT
2803   __ cmp(flags, dtos);
2804   __ br(Assembler::NE, notDouble);
2805 #endif
2806 
2807   // dtos
2808   {
2809     __ pop(dtos);
2810     if (!is_static) pop_and_check_object(obj);
2811     oopDesc::bs()->interpreter_write_barrier(_masm, obj);
2812     __ strd(v0, field);
2813     if (!is_static) {
2814       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2815     }
2816   }
2817 
2818 #ifdef ASSERT
2819   __ b(Done);
2820 
2821   __ bind(notDouble);
2822   __ stop("Bad state");
2823 #endif
2824 
2825   __ bind(Done);
2826 
2827   {
2828     Label notVolatile;
2829     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2830     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2831     __ bind(notVolatile);
2832   }
2833 }
2834 
2835 void TemplateTable::putfield(int byte_no)
2836 {
2837   putfield_or_static(byte_no, false);
2838 }
2839 
2840 void TemplateTable::putstatic(int byte_no) {
2841   putfield_or_static(byte_no, true);
2842 }
2843 
2844 void TemplateTable::jvmti_post_fast_field_mod()
2845 {
2846   if (JvmtiExport::can_post_field_modification()) {
2847     // Check to see if a field modification watch has been set before
2848     // we take the time to call into the VM.
2849     Label L2;
2850     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2851     __ ldrw(c_rarg3, Address(rscratch1));
2852     __ cbzw(c_rarg3, L2);
2853     __ pop_ptr(r19);                  // copy the object pointer from tos
2854     __ verify_oop(r19);
2855     __ push_ptr(r19);                 // put the object pointer back on tos
2856     // Save tos values before call_VM() clobbers them. Since we have
2857     // to do it for every data type, we use the saved values as the
2858     // jvalue object.
2859     switch (bytecode()) {          // load values into the jvalue object
2860     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2861     case Bytecodes::_fast_bputfield: // fall through
2862     case Bytecodes::_fast_zputfield: // fall through
2863     case Bytecodes::_fast_sputfield: // fall through
2864     case Bytecodes::_fast_cputfield: // fall through
2865     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2866     case Bytecodes::_fast_dputfield: __ push_d(); break;
2867     case Bytecodes::_fast_fputfield: __ push_f(); break;
2868     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2869 
2870     default:
2871       ShouldNotReachHere();
2872     }
2873     __ mov(c_rarg3, esp);             // points to jvalue on the stack
2874     // access constant pool cache entry
2875     __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2876     __ verify_oop(r19);
2877     // r19: object pointer copied above
2878     // c_rarg2: cache entry pointer
2879     // c_rarg3: jvalue object on the stack
2880     __ call_VM(noreg,
2881                CAST_FROM_FN_PTR(address,
2882                                 InterpreterRuntime::post_field_modification),
2883                r19, c_rarg2, c_rarg3);
2884 
2885     switch (bytecode()) {             // restore tos values
2886     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2887     case Bytecodes::_fast_bputfield: // fall through
2888     case Bytecodes::_fast_zputfield: // fall through
2889     case Bytecodes::_fast_sputfield: // fall through
2890     case Bytecodes::_fast_cputfield: // fall through
2891     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2892     case Bytecodes::_fast_dputfield: __ pop_d(); break;
2893     case Bytecodes::_fast_fputfield: __ pop_f(); break;
2894     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2895     }
2896     __ bind(L2);
2897   }
2898 }
2899 
2900 void TemplateTable::fast_storefield(TosState state)
2901 {
2902   transition(state, vtos);
2903 
2904   ByteSize base = ConstantPoolCache::base_offset();
2905 
2906   jvmti_post_fast_field_mod();
2907 
2908   // access constant pool cache
2909   __ get_cache_and_index_at_bcp(r2, r1, 1);
2910 
2911   // test for volatile with r3
2912   __ ldrw(r3, Address(r2, in_bytes(base +
2913                                    ConstantPoolCacheEntry::flags_offset())));
2914 
2915   // replace index with field offset from cache entry
2916   __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2917 
2918   {
2919     Label notVolatile;
2920     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2921     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2922     __ bind(notVolatile);
2923   }
2924 
2925   Label notVolatile;
2926 
2927   // Get object from stack
2928   pop_and_check_object(r2);
2929   oopDesc::bs()->interpreter_write_barrier(_masm, r2);
2930 
2931   // field address
2932   const Address field(r2, r1);
2933 
2934   // access field
2935   switch (bytecode()) {
2936   case Bytecodes::_fast_aputfield:
2937     do_oop_store(_masm, field, r0, _bs->kind(), false);
2938     break;
2939   case Bytecodes::_fast_lputfield:
2940     __ str(r0, field);
2941     break;
2942   case Bytecodes::_fast_iputfield:
2943     __ strw(r0, field);
2944     break;
2945   case Bytecodes::_fast_zputfield:
2946     __ andw(r0, r0, 0x1);  // boolean is true if LSB is 1
2947     // fall through to bputfield
2948   case Bytecodes::_fast_bputfield:
2949     __ strb(r0, field);
2950     break;
2951   case Bytecodes::_fast_sputfield:
2952     // fall through
2953   case Bytecodes::_fast_cputfield:
2954     __ strh(r0, field);
2955     break;
2956   case Bytecodes::_fast_fputfield:
2957     __ strs(v0, field);
2958     break;
2959   case Bytecodes::_fast_dputfield:
2960     __ strd(v0, field);
2961     break;
2962   default:
2963     ShouldNotReachHere();
2964   }
2965 
2966   {
2967     Label notVolatile;
2968     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2969     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2970     __ bind(notVolatile);
2971   }
2972 }
2973 
2974 
2975 void TemplateTable::fast_accessfield(TosState state)
2976 {
2977   transition(atos, state);
2978   // Do the JVMTI work here to avoid disturbing the register state below
2979   if (JvmtiExport::can_post_field_access()) {
2980     // Check to see if a field access watch has been set before we
2981     // take the time to call into the VM.
2982     Label L1;
2983     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2984     __ ldrw(r2, Address(rscratch1));
2985     __ cbzw(r2, L1);
2986     // access constant pool cache entry
2987     __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2988     __ verify_oop(r0);
2989     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
2990     __ mov(c_rarg1, r0);
2991     // c_rarg1: object pointer copied above
2992     // c_rarg2: cache entry pointer
2993     __ call_VM(noreg,
2994                CAST_FROM_FN_PTR(address,
2995                                 InterpreterRuntime::post_field_access),
2996                c_rarg1, c_rarg2);
2997     __ pop_ptr(r0); // restore object pointer
2998     __ bind(L1);
2999   }
3000 
3001   // access constant pool cache
3002   __ get_cache_and_index_at_bcp(r2, r1, 1);
3003   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3004                                   ConstantPoolCacheEntry::f2_offset())));
3005   __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3006                                    ConstantPoolCacheEntry::flags_offset())));
3007 
3008   // r0: object
3009   __ verify_oop(r0);
3010   __ null_check(r0);
3011   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
3012   const Address field(r0, r1);
3013 
3014   // access field
3015   switch (bytecode()) {
3016   case Bytecodes::_fast_agetfield:
3017     __ load_heap_oop(r0, field);
3018     __ verify_oop(r0);
3019     break;
3020   case Bytecodes::_fast_lgetfield:
3021     __ ldr(r0, field);
3022     break;
3023   case Bytecodes::_fast_igetfield:
3024     __ ldrw(r0, field);
3025     break;
3026   case Bytecodes::_fast_bgetfield:
3027     __ load_signed_byte(r0, field);
3028     break;
3029   case Bytecodes::_fast_sgetfield:
3030     __ load_signed_short(r0, field);
3031     break;
3032   case Bytecodes::_fast_cgetfield:
3033     __ load_unsigned_short(r0, field);
3034     break;
3035   case Bytecodes::_fast_fgetfield:
3036     __ ldrs(v0, field);
3037     break;
3038   case Bytecodes::_fast_dgetfield:
3039     __ ldrd(v0, field);
3040     break;
3041   default:
3042     ShouldNotReachHere();
3043   }
3044   {
3045     Label notVolatile;
3046     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3047     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3048     __ bind(notVolatile);
3049   }
3050 }
3051 
3052 void TemplateTable::fast_xaccess(TosState state)
3053 {
3054   transition(vtos, state);
3055 
3056   // get receiver
3057   __ ldr(r0, aaddress(0));
3058   // access constant pool cache
3059   __ get_cache_and_index_at_bcp(r2, r3, 2);
3060   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3061                                   ConstantPoolCacheEntry::f2_offset())));
3062   // make sure exception is reported in correct bcp range (getfield is
3063   // next instruction)
3064   __ increment(rbcp);
3065   __ null_check(r0);
3066   oopDesc::bs()->interpreter_read_barrier_not_null(_masm, r0);
3067   switch (state) {
3068   case itos:
3069     __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3070     break;
3071   case atos:
3072     __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0)));
3073     __ verify_oop(r0);
3074     break;
3075   case ftos:
3076     __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3077     break;
3078   default:
3079     ShouldNotReachHere();
3080   }
3081 
3082   {
3083     Label notVolatile;
3084     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3085                                      ConstantPoolCacheEntry::flags_offset())));
3086     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3087     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3088     __ bind(notVolatile);
3089   }
3090 
3091   __ decrement(rbcp);
3092 }
3093 
3094 
3095 
3096 //-----------------------------------------------------------------------------
3097 // Calls
3098 
3099 void TemplateTable::count_calls(Register method, Register temp)
3100 {
3101   __ call_Unimplemented();
3102 }
3103 
3104 void TemplateTable::prepare_invoke(int byte_no,
3105                                    Register method, // linked method (or i-klass)
3106                                    Register index,  // itable index, MethodType, etc.
3107                                    Register recv,   // if caller wants to see it
3108                                    Register flags   // if caller wants to test it
3109                                    ) {
3110   // determine flags
3111   Bytecodes::Code code = bytecode();
3112   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3113   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3114   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3115   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3116   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3117   const bool load_receiver       = (recv  != noreg);
3118   const bool save_flags          = (flags != noreg);
3119   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3120   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3121   assert(flags == noreg || flags == r3, "");
3122   assert(recv  == noreg || recv  == r2, "");
3123 
3124   // setup registers & access constant pool cache
3125   if (recv  == noreg)  recv  = r2;
3126   if (flags == noreg)  flags = r3;
3127   assert_different_registers(method, index, recv, flags);
3128 
3129   // save 'interpreter return address'
3130   __ save_bcp();
3131 
3132   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3133 
3134   // maybe push appendix to arguments (just before return address)
3135   if (is_invokedynamic || is_invokehandle) {
3136     Label L_no_push;
3137     __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3138     // Push the appendix as a trailing parameter.
3139     // This must be done before we get the receiver,
3140     // since the parameter_size includes it.
3141     __ push(r19);
3142     __ mov(r19, index);
3143     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3144     __ load_resolved_reference_at_index(index, r19);
3145     __ pop(r19);
3146     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3147     __ bind(L_no_push);
3148   }
3149 
3150   // load receiver if needed (note: no return address pushed yet)
3151   if (load_receiver) {
3152     __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3153     // FIXME -- is this actually correct? looks like it should be 2
3154     // const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3155     // const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3156     // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3157     // __ movptr(recv, recv_addr);
3158     __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3159     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3160     __ verify_oop(recv);
3161   }
3162 
3163   // compute return type
3164   // x86 uses a shift and mask or wings it with a shift plus assert
3165   // the mask is not needed. aarch64 just uses bitfield extract
3166   __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
3167   // load return address
3168   {
3169     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3170     __ mov(rscratch1, table_addr);
3171     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3172   }
3173 }
3174 
3175 
3176 void TemplateTable::invokevirtual_helper(Register index,
3177                                          Register recv,
3178                                          Register flags)
3179 {
3180   // Uses temporary registers r0, r3
3181   assert_different_registers(index, recv, r0, r3);
3182   // Test for an invoke of a final method
3183   Label notFinal;
3184   __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3185 
3186   const Register method = index;  // method must be rmethod
3187   assert(method == rmethod,
3188          "methodOop must be rmethod for interpreter calling convention");
3189 
3190   // do the call - the index is actually the method to call
3191   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3192 
3193   // It's final, need a null check here!
3194   __ null_check(recv);
3195 
3196   // profile this call
3197   __ profile_final_call(r0);
3198   __ profile_arguments_type(r0, method, r4, true);
3199 
3200   __ jump_from_interpreted(method, r0);
3201 
3202   __ bind(notFinal);
3203 
3204   // get receiver klass
3205   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3206   __ load_klass(r0, recv);
3207 
3208   // profile this call
3209   __ profile_virtual_call(r0, rlocals, r3);
3210 
3211   // get target methodOop & entry point
3212   __ lookup_virtual_method(r0, index, method);
3213   __ profile_arguments_type(r3, method, r4, true);
3214   // FIXME -- this looks completely redundant. is it?
3215   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3216   __ jump_from_interpreted(method, r3);
3217 }
3218 
3219 void TemplateTable::invokevirtual(int byte_no)
3220 {
3221   transition(vtos, vtos);
3222   assert(byte_no == f2_byte, "use this argument");
3223 
3224   prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3225 
3226   // rmethod: index (actually a Method*)
3227   // r2: receiver
3228   // r3: flags
3229 
3230   invokevirtual_helper(rmethod, r2, r3);
3231 }
3232 
3233 void TemplateTable::invokespecial(int byte_no)
3234 {
3235   transition(vtos, vtos);
3236   assert(byte_no == f1_byte, "use this argument");
3237 
3238   prepare_invoke(byte_no, rmethod, noreg,  // get f1 Method*
3239                  r2);  // get receiver also for null check
3240   __ verify_oop(r2);
3241   __ null_check(r2);
3242   // do the call
3243   __ profile_call(r0);
3244   __ profile_arguments_type(r0, rmethod, rbcp, false);
3245   __ jump_from_interpreted(rmethod, r0);
3246 }
3247 
3248 void TemplateTable::invokestatic(int byte_no)
3249 {
3250   transition(vtos, vtos);
3251   assert(byte_no == f1_byte, "use this argument");
3252 
3253   prepare_invoke(byte_no, rmethod);  // get f1 Method*
3254   // do the call
3255   __ profile_call(r0);
3256   __ profile_arguments_type(r0, rmethod, r4, false);
3257   __ jump_from_interpreted(rmethod, r0);
3258 }
3259 
3260 void TemplateTable::fast_invokevfinal(int byte_no)
3261 {
3262   __ call_Unimplemented();
3263 }
3264 
3265 void TemplateTable::invokeinterface(int byte_no) {
3266   transition(vtos, vtos);
3267   assert(byte_no == f1_byte, "use this argument");
3268 
3269   prepare_invoke(byte_no, r0, rmethod,  // get f1 Klass*, f2 Method*
3270                  r2, r3); // recv, flags
3271 
3272   // r0: interface klass (from f1)
3273   // rmethod: method (from f2)
3274   // r2: receiver
3275   // r3: flags
3276 
3277   // Special case of invokeinterface called for virtual method of
3278   // java.lang.Object.  See cpCacheOop.cpp for details.
3279   // This code isn't produced by javac, but could be produced by
3280   // another compliant java compiler.
3281   Label notMethod;
3282   __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3283 
3284   invokevirtual_helper(rmethod, r2, r3);
3285   __ bind(notMethod);
3286 
3287   // Get receiver klass into r3 - also a null check
3288   __ restore_locals();
3289   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3290   __ load_klass(r3, r2);
3291 
3292   Label no_such_interface, no_such_method;
3293 
3294   // Receiver subtype check against REFC.
3295   // Superklass in r0. Subklass in r3. Blows rscratch2, r13.
3296   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3297                              r3, r0, noreg,
3298                              // outputs: scan temp. reg, scan temp. reg
3299                              rscratch2, r13,
3300                              no_such_interface,
3301                              /*return_method=*/false);
3302 
3303   // profile this call
3304   __ profile_virtual_call(r3, r13, r19);
3305 
3306   // Get declaring interface class from method, and itable index
3307   __ ldr(r0, Address(rmethod, Method::const_offset()));
3308   __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
3309   __ ldr(r0, Address(r0, ConstantPool::pool_holder_offset_in_bytes()));
3310   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3311   __ subw(rmethod, rmethod, Method::itable_index_max);
3312   __ negw(rmethod, rmethod);
3313 
3314   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3315                              r3, r0, rmethod,
3316                              // outputs: method, scan temp. reg
3317                              rmethod, r13,
3318                              no_such_interface);
3319 
3320   // rmethod,: methodOop to call
3321   // r2: receiver
3322   // Check for abstract method error
3323   // Note: This should be done more efficiently via a throw_abstract_method_error
3324   //       interpreter entry point and a conditional jump to it in case of a null
3325   //       method.
3326   __ cbz(rmethod, no_such_method);
3327 
3328   __ profile_arguments_type(r3, rmethod, r13, true);
3329 
3330   // do the call
3331   // r2: receiver
3332   // rmethod,: methodOop
3333   __ jump_from_interpreted(rmethod, r3);
3334   __ should_not_reach_here();
3335 
3336   // exception handling code follows...
3337   // note: must restore interpreter registers to canonical
3338   //       state for exception handling to work correctly!
3339 
3340   __ bind(no_such_method);
3341   // throw exception
3342   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3343   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3344   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3345   // the call_VM checks for exception, so we should never return here.
3346   __ should_not_reach_here();
3347 
3348   __ bind(no_such_interface);
3349   // throw exception
3350   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3351   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3352   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3353                    InterpreterRuntime::throw_IncompatibleClassChangeError));
3354   // the call_VM checks for exception, so we should never return here.
3355   __ should_not_reach_here();
3356   return;
3357 }
3358 
3359 void TemplateTable::invokehandle(int byte_no) {
3360   transition(vtos, vtos);
3361   assert(byte_no == f1_byte, "use this argument");
3362 
3363   if (!EnableInvokeDynamic) {
3364     // rewriter does not generate this bytecode
3365     __ should_not_reach_here();
3366     return;
3367   }
3368 
3369   prepare_invoke(byte_no, rmethod, r0, r2);
3370   __ verify_method_ptr(r2);
3371   __ verify_oop(r2);
3372   __ null_check(r2);
3373 
3374   // FIXME: profile the LambdaForm also
3375 
3376   // r13 is safe to use here as a scratch reg because it is about to
3377   // be clobbered by jump_from_interpreted().
3378   __ profile_final_call(r13);
3379   __ profile_arguments_type(r13, rmethod, r4, true);
3380 
3381   __ jump_from_interpreted(rmethod, r0);
3382 }
3383 
3384 void TemplateTable::invokedynamic(int byte_no) {
3385   transition(vtos, vtos);
3386   assert(byte_no == f1_byte, "use this argument");
3387 
3388   if (!EnableInvokeDynamic) {
3389     // We should not encounter this bytecode if !EnableInvokeDynamic.
3390     // The verifier will stop it.  However, if we get past the verifier,
3391     // this will stop the thread in a reasonable way, without crashing the JVM.
3392     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3393                      InterpreterRuntime::throw_IncompatibleClassChangeError));
3394     // the call_VM checks for exception, so we should never return here.
3395     __ should_not_reach_here();
3396     return;
3397   }
3398 
3399   prepare_invoke(byte_no, rmethod, r0);
3400 
3401   // r0: CallSite object (from cpool->resolved_references[])
3402   // rmethod: MH.linkToCallSite method (from f2)
3403 
3404   // Note:  r0_callsite is already pushed by prepare_invoke
3405 
3406   // %%% should make a type profile for any invokedynamic that takes a ref argument
3407   // profile this call
3408   __ profile_call(rbcp);
3409   __ profile_arguments_type(r3, rmethod, r13, false);
3410 
3411   __ verify_oop(r0);
3412 
3413   __ jump_from_interpreted(rmethod, r0);
3414 }
3415 
3416 
3417 //-----------------------------------------------------------------------------
3418 // Allocation
3419 
3420 void TemplateTable::_new() {
3421   transition(vtos, atos);
3422 
3423   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3424   Label slow_case;
3425   Label done;
3426   Label initialize_header;
3427   Label initialize_object; // including clearing the fields
3428   Label allocate_shared;
3429 
3430   __ get_cpool_and_tags(r4, r0);
3431   // Make sure the class we're about to instantiate has been resolved.
3432   // This is done before loading InstanceKlass to be consistent with the order
3433   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3434   const int tags_offset = Array<u1>::base_offset_in_bytes();
3435   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3436   __ lea(rscratch1, Address(rscratch1, tags_offset));
3437   __ ldarb(rscratch1, rscratch1);
3438   __ cmp(rscratch1, JVM_CONSTANT_Class);
3439   __ br(Assembler::NE, slow_case);
3440 
3441   // get InstanceKlass
3442   __ lea(r4, Address(r4, r3, Address::lsl(3)));
3443   __ ldr(r4, Address(r4, sizeof(ConstantPool)));
3444 
3445   // make sure klass is initialized & doesn't have finalizer
3446   // make sure klass is fully initialized
3447   __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3448   __ cmp(rscratch1, InstanceKlass::fully_initialized);
3449   __ br(Assembler::NE, slow_case);
3450 
3451   // get instance_size in InstanceKlass (scaled to a count of bytes)
3452   __ ldrw(r3,
3453           Address(r4,
3454                   Klass::layout_helper_offset()));
3455   // test to see if it has a finalizer or is malformed in some way
3456   __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3457 
3458   // Allocate the instance
3459   // 1) Try to allocate in the TLAB
3460   // 2) if fail and the object is large allocate in the shared Eden
3461   // 3) if the above fails (or is not applicable), go to a slow case
3462   // (creates a new TLAB, etc.)
3463 
3464   const bool allow_shared_alloc =
3465     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3466 
3467   if (UseTLAB) {
3468     __ tlab_allocate(r0, r3, 0, noreg, r1,
3469                      allow_shared_alloc ? allocate_shared : slow_case);
3470 
3471     if (ZeroTLAB) {
3472       // the fields have been already cleared
3473       __ b(initialize_header);
3474     } else {
3475       // initialize both the header and fields
3476       __ b(initialize_object);
3477     }
3478   }
3479 
3480   // Allocation in the shared Eden, if allowed.
3481   //
3482   // r3: instance size in bytes
3483   if (allow_shared_alloc) {
3484     __ bind(allocate_shared);
3485 
3486     __ eden_allocate(r0, r3, 0, r10, slow_case);
3487     __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3488   }
3489 
3490   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3491     // The object is initialized before the header.  If the object size is
3492     // zero, go directly to the header initialization.
3493     __ bind(initialize_object);
3494     __ sub(r3, r3, sizeof(oopDesc));
3495     __ cbz(r3, initialize_header);
3496 
3497     // Initialize object fields
3498     {
3499       __ add(r2, r0, sizeof(oopDesc));
3500       Label loop;
3501       __ bind(loop);
3502       __ str(zr, Address(__ post(r2, BytesPerLong)));
3503       __ sub(r3, r3, BytesPerLong);
3504       __ cbnz(r3, loop);
3505     }
3506 
3507     // initialize object header only.
3508     __ bind(initialize_header);
3509     if (UseBiasedLocking) {
3510       __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3511     } else {
3512       __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3513     }
3514     __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3515     __ store_klass_gap(r0, zr);  // zero klass gap for compressed oops
3516     __ store_klass(r0, r4);      // store klass last
3517 
3518     {
3519       SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3520       // Trigger dtrace event for fastpath
3521       __ push(atos); // save the return value
3522       __ call_VM_leaf(
3523            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3524       __ pop(atos); // restore the return value
3525 
3526     }
3527     __ b(done);
3528   }
3529 
3530   // slow case
3531   __ bind(slow_case);
3532   __ get_constant_pool(c_rarg1);
3533   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3534   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3535   __ verify_oop(r0);
3536 
3537   // continue
3538   __ bind(done);
3539   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3540   __ membar(Assembler::StoreStore);
3541 }
3542 
3543 void TemplateTable::newarray() {
3544   transition(itos, atos);
3545   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3546   __ mov(c_rarg2, r0);
3547   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3548           c_rarg1, c_rarg2);
3549   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3550   __ membar(Assembler::StoreStore);
3551 }
3552 
3553 void TemplateTable::anewarray() {
3554   transition(itos, atos);
3555   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3556   __ get_constant_pool(c_rarg1);
3557   __ mov(c_rarg3, r0);
3558   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3559           c_rarg1, c_rarg2, c_rarg3);
3560   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3561   __ membar(Assembler::StoreStore);
3562 }
3563 
3564 void TemplateTable::arraylength() {
3565   transition(atos, itos);
3566   __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3567   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3568 }
3569 
3570 void TemplateTable::checkcast()
3571 {
3572   transition(atos, atos);
3573   Label done, is_null, ok_is_subtype, quicked, resolved;
3574   __ cbz(r0, is_null);
3575 
3576   // Get cpool & tags index
3577   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3578   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3579   // See if bytecode has already been quicked
3580   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3581   __ lea(r1, Address(rscratch1, r19));
3582   __ ldarb(r1, r1);
3583   __ cmp(r1, JVM_CONSTANT_Class);
3584   __ br(Assembler::EQ, quicked);
3585 
3586   __ push(atos); // save receiver for result, and for GC
3587   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3588   // vm_result_2 has metadata result
3589   __ get_vm_result_2(r0, rthread);
3590   __ pop(r3); // restore receiver
3591   __ b(resolved);
3592 
3593   // Get superklass in r0 and subklass in r3
3594   __ bind(quicked);
3595   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3596   __ lea(r0, Address(r2, r19, Address::lsl(3)));
3597   __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3598 
3599   __ bind(resolved);
3600   __ load_klass(r19, r3);
3601 
3602   // Generate subtype check.  Blows r2, r5.  Object in r3.
3603   // Superklass in r0.  Subklass in r19.
3604   __ gen_subtype_check(r19, ok_is_subtype);
3605 
3606   // Come here on failure
3607   __ push(r3);
3608   // object is at TOS
3609   __ b(Interpreter::_throw_ClassCastException_entry);
3610 
3611   // Come here on success
3612   __ bind(ok_is_subtype);
3613   __ mov(r0, r3); // Restore object in r3
3614 
3615   // Collect counts on whether this test sees NULLs a lot or not.
3616   if (ProfileInterpreter) {
3617     __ b(done);
3618     __ bind(is_null);
3619     __ profile_null_seen(r2);
3620   } else {
3621     __ bind(is_null);   // same as 'done'
3622   }
3623   __ bind(done);
3624 }
3625 
3626 void TemplateTable::instanceof() {
3627   transition(atos, itos);
3628   Label done, is_null, ok_is_subtype, quicked, resolved;
3629   __ cbz(r0, is_null);
3630 
3631   // Get cpool & tags index
3632   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3633   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3634   // See if bytecode has already been quicked
3635   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3636   __ lea(r1, Address(rscratch1, r19));
3637   __ ldarb(r1, r1);
3638   __ cmp(r1, JVM_CONSTANT_Class);
3639   __ br(Assembler::EQ, quicked);
3640 
3641   __ push(atos); // save receiver for result, and for GC
3642   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3643   // vm_result_2 has metadata result
3644   __ get_vm_result_2(r0, rthread);
3645   __ pop(r3); // restore receiver
3646   __ verify_oop(r3);
3647   __ load_klass(r3, r3);
3648   __ b(resolved);
3649 
3650   // Get superklass in r0 and subklass in r3
3651   __ bind(quicked);
3652   __ load_klass(r3, r0);
3653   __ lea(r0, Address(r2, r19, Address::lsl(3)));
3654   __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3655 
3656   __ bind(resolved);
3657 
3658   // Generate subtype check.  Blows r2, r5
3659   // Superklass in r0.  Subklass in r3.
3660   __ gen_subtype_check(r3, ok_is_subtype);
3661 
3662   // Come here on failure
3663   __ mov(r0, 0);
3664   __ b(done);
3665   // Come here on success
3666   __ bind(ok_is_subtype);
3667   __ mov(r0, 1);
3668 
3669   // Collect counts on whether this test sees NULLs a lot or not.
3670   if (ProfileInterpreter) {
3671     __ b(done);
3672     __ bind(is_null);
3673     __ profile_null_seen(r2);
3674   } else {
3675     __ bind(is_null);   // same as 'done'
3676   }
3677   __ bind(done);
3678   // r0 = 0: obj == NULL or  obj is not an instanceof the specified klass
3679   // r0 = 1: obj != NULL and obj is     an instanceof the specified klass
3680 }
3681 
3682 //-----------------------------------------------------------------------------
3683 // Breakpoints
3684 void TemplateTable::_breakpoint() {
3685   // Note: We get here even if we are single stepping..
3686   // jbug inists on setting breakpoints at every bytecode
3687   // even if we are in single step mode.
3688 
3689   transition(vtos, vtos);
3690 
3691   // get the unpatched byte code
3692   __ get_method(c_rarg1);
3693   __ call_VM(noreg,
3694              CAST_FROM_FN_PTR(address,
3695                               InterpreterRuntime::get_original_bytecode_at),
3696              c_rarg1, rbcp);
3697   __ mov(r19, r0);
3698 
3699   // post the breakpoint event
3700   __ call_VM(noreg,
3701              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3702              rmethod, rbcp);
3703 
3704   // complete the execution of original bytecode
3705   __ mov(rscratch1, r19);
3706   __ dispatch_only_normal(vtos);
3707 }
3708 
3709 //-----------------------------------------------------------------------------
3710 // Exceptions
3711 
3712 void TemplateTable::athrow() {
3713   transition(atos, vtos);
3714   __ null_check(r0);
3715   __ b(Interpreter::throw_exception_entry());
3716 }
3717 
3718 //-----------------------------------------------------------------------------
3719 // Synchronization
3720 //
3721 // Note: monitorenter & exit are symmetric routines; which is reflected
3722 //       in the assembly code structure as well
3723 //
3724 // Stack layout:
3725 //
3726 // [expressions  ] <--- esp               = expression stack top
3727 // ..
3728 // [expressions  ]
3729 // [monitor entry] <--- monitor block top = expression stack bot
3730 // ..
3731 // [monitor entry]
3732 // [frame data   ] <--- monitor block bot
3733 // ...
3734 // [saved rbp    ] <--- rbp
3735 void TemplateTable::monitorenter()
3736 {
3737   transition(atos, vtos);
3738 
3739   // check for NULL object
3740   __ null_check(r0);
3741 
3742   // We need to preemptively evacuate the object, because we later compare
3743   // it to objects in the BasicObjectLock list, and we might get false negatives
3744   // if another thread evacuates the object in the meantime. See acmp.
3745   oopDesc::bs()->interpreter_write_barrier(_masm, r0);
3746 
3747   const Address monitor_block_top(
3748         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3749   const Address monitor_block_bot(
3750         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3751   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3752 
3753   Label allocated;
3754 
3755   // initialize entry pointer
3756   __ mov(c_rarg1, zr); // points to free slot or NULL
3757 
3758   // find a free slot in the monitor block (result in c_rarg1)
3759   {
3760     Label entry, loop, exit;
3761     __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3762                                         // starting with top-most entry
3763     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3764 
3765     __ b(entry);
3766 
3767     __ bind(loop);
3768     // check if current entry is used
3769     // if not used then remember entry in c_rarg1
3770     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3771     __ cmp(zr, rscratch1);
3772     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3773     // check if current entry is for same object
3774     __ cmp(r0, rscratch1);
3775     // if same object then stop searching
3776     __ br(Assembler::EQ, exit);
3777     // otherwise advance to next entry
3778     __ add(c_rarg3, c_rarg3, entry_size);
3779     __ bind(entry);
3780     // check if bottom reached
3781     __ cmp(c_rarg3, c_rarg2);
3782     // if not at bottom then check this entry
3783     __ br(Assembler::NE, loop);
3784     __ bind(exit);
3785   }
3786 
3787   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3788                             // if found, continue with that on
3789 
3790   // allocate one if there's no free slot
3791   {
3792     Label entry, loop;
3793     // 1. compute new pointers            // rsp: old expression stack top
3794     __ ldr(c_rarg1, monitor_block_bot);   // c_rarg1: old expression stack bottom
3795     __ sub(esp, esp, entry_size);         // move expression stack top
3796     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3797     __ mov(c_rarg3, esp);                 // set start value for copy loop
3798     __ str(c_rarg1, monitor_block_bot);   // set new monitor block bottom
3799 
3800     __ sub(sp, sp, entry_size);           // make room for the monitor
3801 
3802     __ b(entry);
3803     // 2. move expression stack contents
3804     __ bind(loop);
3805     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3806                                                    // word from old location
3807     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
3808     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
3809     __ bind(entry);
3810     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
3811     __ br(Assembler::NE, loop);      // if not at bottom then
3812                                      // copy next word
3813   }
3814 
3815   // call run-time routine
3816   // c_rarg1: points to monitor entry
3817   __ bind(allocated);
3818 
3819   // Increment bcp to point to the next bytecode, so exception
3820   // handling for async. exceptions work correctly.
3821   // The object has already been poped from the stack, so the
3822   // expression stack looks correct.
3823   __ increment(rbcp);
3824 
3825   // store object
3826   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3827   __ lock_object(c_rarg1);
3828 
3829   // check to make sure this monitor doesn't cause stack overflow after locking
3830   __ save_bcp();  // in case of exception
3831   __ generate_stack_overflow_check(0);
3832 
3833   // The bcp has already been incremented. Just need to dispatch to
3834   // next instruction.
3835   __ dispatch_next(vtos);
3836 }
3837 
3838 
3839 void TemplateTable::monitorexit()
3840 {
3841   transition(atos, vtos);
3842 
3843   // check for NULL object
3844   __ null_check(r0);
3845 
3846   // We need to preemptively evacuate the object, because we later compare
3847   // it to objects in the BasicObjectLock list, and we might get false negatives
3848   // if another thread evacuates the object in the meantime. See acmp.
3849   oopDesc::bs()->interpreter_write_barrier(_masm, r0);
3850 
3851   const Address monitor_block_top(
3852         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3853   const Address monitor_block_bot(
3854         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3855   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3856 
3857   Label found;
3858 
3859   // find matching slot
3860   {
3861     Label entry, loop;
3862     __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3863                                         // starting with top-most entry
3864     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3865                                         // of monitor block
3866     __ b(entry);
3867 
3868     __ bind(loop);
3869     // check if current entry is for same object
3870     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3871     __ cmp(r0, rscratch1);
3872     // if same object then stop searching
3873     __ br(Assembler::EQ, found);
3874     // otherwise advance to next entry
3875     __ add(c_rarg1, c_rarg1, entry_size);
3876     __ bind(entry);
3877     // check if bottom reached
3878     __ cmp(c_rarg1, c_rarg2);
3879     // if not at bottom then check this entry
3880     __ br(Assembler::NE, loop);
3881   }
3882 
3883   // error handling. Unlocking was not block-structured
3884   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3885                    InterpreterRuntime::throw_illegal_monitor_state_exception));
3886   __ should_not_reach_here();
3887 
3888   // call run-time routine
3889   __ bind(found);
3890   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3891   __ unlock_object(c_rarg1);
3892   __ pop_ptr(r0); // discard object
3893 }
3894 
3895 
3896 // Wide instructions
3897 void TemplateTable::wide()
3898 {
3899   __ load_unsigned_byte(r19, at_bcp(1));
3900   __ mov(rscratch1, (address)Interpreter::_wentry_point);
3901   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3902   __ br(rscratch1);
3903 }
3904 
3905 
3906 // Multi arrays
3907 void TemplateTable::multianewarray() {
3908   transition(vtos, atos);
3909   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3910   // last dim is on top of stack; we want address of first one:
3911   // first_addr = last_addr + (ndims - 1) * wordSize
3912   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3913   __ sub(c_rarg1, c_rarg1, wordSize);
3914   call_VM(r0,
3915           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3916           c_rarg1);
3917   __ load_unsigned_byte(r1, at_bcp(3));
3918   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3919 }
3920 #endif // !CC_INTERP