rev 49827 : 8201593: Print array length in ArrayIndexOutOfBoundsException.

   1 /*
   2  * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.hpp"
  28 #include "gc/shared/barrierSetAssembler.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/interp_masm.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "memory/universe.hpp"
  34 #include "oops/methodData.hpp"
  35 #include "oops/method.hpp"
  36 #include "oops/objArrayKlass.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "prims/methodHandles.hpp"
  39 #include "runtime/frame.inline.hpp"
  40 #include "runtime/sharedRuntime.hpp"
  41 #include "runtime/stubRoutines.hpp"
  42 #include "runtime/synchronizer.hpp"
  43 
  44 #define __ _masm->
  45 
  46 // Platform-dependent initialization
  47 
  48 void TemplateTable::pd_initialize() {
  49   // No aarch64 specific initialization
  50 }
  51 
  52 // Address computation: local variables
  53 
  54 static inline Address iaddress(int n) {
  55   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  56 }
  57 
  58 static inline Address laddress(int n) {
  59   return iaddress(n + 1);
  60 }
  61 
  62 static inline Address faddress(int n) {
  63   return iaddress(n);
  64 }
  65 
  66 static inline Address daddress(int n) {
  67   return laddress(n);
  68 }
  69 
  70 static inline Address aaddress(int n) {
  71   return iaddress(n);
  72 }
  73 
  74 static inline Address iaddress(Register r) {
  75   return Address(rlocals, r, Address::lsl(3));
  76 }
  77 
  78 static inline Address laddress(Register r, Register scratch,
  79                                InterpreterMacroAssembler* _masm) {
  80   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  81   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  82 }
  83 
  84 static inline Address faddress(Register r) {
  85   return iaddress(r);
  86 }
  87 
  88 static inline Address daddress(Register r, Register scratch,
  89                                InterpreterMacroAssembler* _masm) {
  90   return laddress(r, scratch, _masm);
  91 }
  92 
  93 static inline Address aaddress(Register r) {
  94   return iaddress(r);
  95 }
  96 
  97 static inline Address at_rsp() {
  98   return Address(esp, 0);
  99 }
 100 
 101 // At top of Java expression stack which may be different than esp().  It
 102 // isn't for category 1 objects.
 103 static inline Address at_tos   () {
 104   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 105 }
 106 
 107 static inline Address at_tos_p1() {
 108   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 109 }
 110 
 111 static inline Address at_tos_p2() {
 112   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 113 }
 114 
 115 static inline Address at_tos_p3() {
 116   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 117 }
 118 
 119 static inline Address at_tos_p4() {
 120   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 121 }
 122 
 123 static inline Address at_tos_p5() {
 124   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 125 }
 126 
 127 // Condition conversion
 128 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 129   switch (cc) {
 130   case TemplateTable::equal        : return Assembler::NE;
 131   case TemplateTable::not_equal    : return Assembler::EQ;
 132   case TemplateTable::less         : return Assembler::GE;
 133   case TemplateTable::less_equal   : return Assembler::GT;
 134   case TemplateTable::greater      : return Assembler::LE;
 135   case TemplateTable::greater_equal: return Assembler::LT;
 136   }
 137   ShouldNotReachHere();
 138   return Assembler::EQ;
 139 }
 140 
 141 
 142 // Miscelaneous helper routines
 143 // Store an oop (or NULL) at the Address described by obj.
 144 // If val == noreg this means store a NULL
 145 static void do_oop_store(InterpreterMacroAssembler* _masm,
 146                          Address dst,
 147                          Register val,
 148                          DecoratorSet decorators) {
 149   assert(val == noreg || val == r0, "parameter is just for looks");
 150   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
 151   bs->store_at(_masm, decorators, T_OBJECT, dst, val, /*tmp1*/ r10, /*tmp2*/ r1);
 152 }
 153 
 154 static void do_oop_load(InterpreterMacroAssembler* _masm,
 155                         Address src,
 156                         Register dst,
 157                         DecoratorSet decorators) {
 158   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
 159   bs->load_at(_masm, decorators, T_OBJECT, dst, src, /*tmp1*/ r10, /*tmp_thread*/ r1);
 160 }
 161 
 162 Address TemplateTable::at_bcp(int offset) {
 163   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 164   return Address(rbcp, offset);
 165 }
 166 
 167 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 168                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 169                                    int byte_no)
 170 {
 171   if (!RewriteBytecodes)  return;
 172   Label L_patch_done;
 173 
 174   switch (bc) {
 175   case Bytecodes::_fast_aputfield:
 176   case Bytecodes::_fast_bputfield:
 177   case Bytecodes::_fast_zputfield:
 178   case Bytecodes::_fast_cputfield:
 179   case Bytecodes::_fast_dputfield:
 180   case Bytecodes::_fast_fputfield:
 181   case Bytecodes::_fast_iputfield:
 182   case Bytecodes::_fast_lputfield:
 183   case Bytecodes::_fast_sputfield:
 184     {
 185       // We skip bytecode quickening for putfield instructions when
 186       // the put_code written to the constant pool cache is zero.
 187       // This is required so that every execution of this instruction
 188       // calls out to InterpreterRuntime::resolve_get_put to do
 189       // additional, required work.
 190       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 191       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 192       __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
 193       __ movw(bc_reg, bc);
 194       __ cbzw(temp_reg, L_patch_done);  // don't patch
 195     }
 196     break;
 197   default:
 198     assert(byte_no == -1, "sanity");
 199     // the pair bytecodes have already done the load.
 200     if (load_bc_into_bc_reg) {
 201       __ movw(bc_reg, bc);
 202     }
 203   }
 204 
 205   if (JvmtiExport::can_post_breakpoint()) {
 206     Label L_fast_patch;
 207     // if a breakpoint is present we can't rewrite the stream directly
 208     __ load_unsigned_byte(temp_reg, at_bcp(0));
 209     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 210     __ br(Assembler::NE, L_fast_patch);
 211     // Let breakpoint table handling rewrite to quicker bytecode
 212     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 213     __ b(L_patch_done);
 214     __ bind(L_fast_patch);
 215   }
 216 
 217 #ifdef ASSERT
 218   Label L_okay;
 219   __ load_unsigned_byte(temp_reg, at_bcp(0));
 220   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 221   __ br(Assembler::EQ, L_okay);
 222   __ cmpw(temp_reg, bc_reg);
 223   __ br(Assembler::EQ, L_okay);
 224   __ stop("patching the wrong bytecode");
 225   __ bind(L_okay);
 226 #endif
 227 
 228   // patch bytecode
 229   __ strb(bc_reg, at_bcp(0));
 230   __ bind(L_patch_done);
 231 }
 232 
 233 
 234 // Individual instructions
 235 
 236 void TemplateTable::nop() {
 237   transition(vtos, vtos);
 238   // nothing to do
 239 }
 240 
 241 void TemplateTable::shouldnotreachhere() {
 242   transition(vtos, vtos);
 243   __ stop("shouldnotreachhere bytecode");
 244 }
 245 
 246 void TemplateTable::aconst_null()
 247 {
 248   transition(vtos, atos);
 249   __ mov(r0, 0);
 250 }
 251 
 252 void TemplateTable::iconst(int value)
 253 {
 254   transition(vtos, itos);
 255   __ mov(r0, value);
 256 }
 257 
 258 void TemplateTable::lconst(int value)
 259 {
 260   __ mov(r0, value);
 261 }
 262 
 263 void TemplateTable::fconst(int value)
 264 {
 265   transition(vtos, ftos);
 266   switch (value) {
 267   case 0:
 268     __ fmovs(v0, zr);
 269     break;
 270   case 1:
 271     __ fmovs(v0, 1.0);
 272     break;
 273   case 2:
 274     __ fmovs(v0, 2.0);
 275     break;
 276   default:
 277     ShouldNotReachHere();
 278     break;
 279   }
 280 }
 281 
 282 void TemplateTable::dconst(int value)
 283 {
 284   transition(vtos, dtos);
 285   switch (value) {
 286   case 0:
 287     __ fmovd(v0, zr);
 288     break;
 289   case 1:
 290     __ fmovd(v0, 1.0);
 291     break;
 292   case 2:
 293     __ fmovd(v0, 2.0);
 294     break;
 295   default:
 296     ShouldNotReachHere();
 297     break;
 298   }
 299 }
 300 
 301 void TemplateTable::bipush()
 302 {
 303   transition(vtos, itos);
 304   __ load_signed_byte32(r0, at_bcp(1));
 305 }
 306 
 307 void TemplateTable::sipush()
 308 {
 309   transition(vtos, itos);
 310   __ load_unsigned_short(r0, at_bcp(1));
 311   __ revw(r0, r0);
 312   __ asrw(r0, r0, 16);
 313 }
 314 
 315 void TemplateTable::ldc(bool wide)
 316 {
 317   transition(vtos, vtos);
 318   Label call_ldc, notFloat, notClass, notInt, Done;
 319 
 320   if (wide) {
 321     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 322   } else {
 323     __ load_unsigned_byte(r1, at_bcp(1));
 324   }
 325   __ get_cpool_and_tags(r2, r0);
 326 
 327   const int base_offset = ConstantPool::header_size() * wordSize;
 328   const int tags_offset = Array<u1>::base_offset_in_bytes();
 329 
 330   // get type
 331   __ add(r3, r1, tags_offset);
 332   __ lea(r3, Address(r0, r3));
 333   __ ldarb(r3, r3);
 334 
 335   // unresolved class - get the resolved class
 336   __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
 337   __ br(Assembler::EQ, call_ldc);
 338 
 339   // unresolved class in error state - call into runtime to throw the error
 340   // from the first resolution attempt
 341   __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
 342   __ br(Assembler::EQ, call_ldc);
 343 
 344   // resolved class - need to call vm to get java mirror of the class
 345   __ cmp(r3, JVM_CONSTANT_Class);
 346   __ br(Assembler::NE, notClass);
 347 
 348   __ bind(call_ldc);
 349   __ mov(c_rarg1, wide);
 350   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 351   __ push_ptr(r0);
 352   __ verify_oop(r0);
 353   __ b(Done);
 354 
 355   __ bind(notClass);
 356   __ cmp(r3, JVM_CONSTANT_Float);
 357   __ br(Assembler::NE, notFloat);
 358   // ftos
 359   __ adds(r1, r2, r1, Assembler::LSL, 3);
 360   __ ldrs(v0, Address(r1, base_offset));
 361   __ push_f();
 362   __ b(Done);
 363 
 364   __ bind(notFloat);
 365 
 366   __ cmp(r3, JVM_CONSTANT_Integer);
 367   __ br(Assembler::NE, notInt);
 368 
 369   // itos
 370   __ adds(r1, r2, r1, Assembler::LSL, 3);
 371   __ ldrw(r0, Address(r1, base_offset));
 372   __ push_i(r0);
 373   __ b(Done);
 374 
 375   __ bind(notInt);
 376   condy_helper(Done);
 377 
 378   __ bind(Done);
 379 }
 380 
 381 // Fast path for caching oop constants.
 382 void TemplateTable::fast_aldc(bool wide)
 383 {
 384   transition(vtos, atos);
 385 
 386   Register result = r0;
 387   Register tmp = r1;
 388   Register rarg = r2;
 389 
 390   int index_size = wide ? sizeof(u2) : sizeof(u1);
 391 
 392   Label resolved;
 393 
 394   // We are resolved if the resolved reference cache entry contains a
 395   // non-null object (String, MethodType, etc.)
 396   assert_different_registers(result, tmp);
 397   __ get_cache_index_at_bcp(tmp, 1, index_size);
 398   __ load_resolved_reference_at_index(result, tmp);
 399   __ cbnz(result, resolved);
 400 
 401   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 402 
 403   // first time invocation - must resolve first
 404   __ mov(rarg, (int)bytecode());
 405   __ call_VM(result, entry, rarg);
 406 
 407   __ bind(resolved);
 408 
 409   { // Check for the null sentinel.
 410     // If we just called the VM, that already did the mapping for us,
 411     // but it's harmless to retry.
 412     Label notNull;
 413 
 414     // Stash null_sentinel address to get its value later
 415     __ movptr(rarg, (uintptr_t)Universe::the_null_sentinel_addr());
 416     __ ldr(tmp, Address(rarg));
 417     __ cmp(result, tmp);
 418     __ br(Assembler::NE, notNull);
 419     __ mov(result, 0);  // NULL object reference
 420     __ bind(notNull);
 421   }
 422 
 423   if (VerifyOops) {
 424     // Safe to call with 0 result
 425     __ verify_oop(result);
 426   }
 427 }
 428 
 429 void TemplateTable::ldc2_w()
 430 {
 431   transition(vtos, vtos);
 432   Label notDouble, notLong, Done;
 433   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 434 
 435   __ get_cpool_and_tags(r1, r2);
 436   const int base_offset = ConstantPool::header_size() * wordSize;
 437   const int tags_offset = Array<u1>::base_offset_in_bytes();
 438 
 439   // get type
 440   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 441   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 442   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 443   __ br(Assembler::NE, notDouble);
 444 
 445   // dtos
 446   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 447   __ ldrd(v0, Address(r2, base_offset));
 448   __ push_d();
 449   __ b(Done);
 450 
 451   __ bind(notDouble);
 452   __ cmpw(r2, (int)JVM_CONSTANT_Long);
 453   __ br(Assembler::NE, notLong);
 454 
 455   // ltos
 456   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 457   __ ldr(r0, Address(r0, base_offset));
 458   __ push_l();
 459   __ b(Done);
 460 
 461   __ bind(notLong);
 462   condy_helper(Done);
 463 
 464   __ bind(Done);
 465 }
 466 
 467 void TemplateTable::condy_helper(Label& Done)
 468 {
 469   Register obj = r0;
 470   Register rarg = r1;
 471   Register flags = r2;
 472   Register off = r3;
 473 
 474   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 475 
 476   __ mov(rarg, (int) bytecode());
 477   __ call_VM(obj, entry, rarg);
 478 
 479   __ get_vm_result_2(flags, rthread);
 480 
 481   // VMr = obj = base address to find primitive value to push
 482   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 483   __ mov(off, flags);
 484   __ andw(off, off, ConstantPoolCacheEntry::field_index_mask);
 485 
 486   const Address field(obj, off);
 487 
 488   // What sort of thing are we loading?
 489   // x86 uses a shift and mask or wings it with a shift plus assert
 490   // the mask is not needed. aarch64 just uses bitfield extract
 491   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,
 492            ConstantPoolCacheEntry::tos_state_bits);
 493 
 494   switch (bytecode()) {
 495     case Bytecodes::_ldc:
 496     case Bytecodes::_ldc_w:
 497       {
 498         // tos in (itos, ftos, stos, btos, ctos, ztos)
 499         Label notInt, notFloat, notShort, notByte, notChar, notBool;
 500         __ cmpw(flags, itos);
 501         __ br(Assembler::NE, notInt);
 502         // itos
 503         __ ldrw(r0, field);
 504         __ push(itos);
 505         __ b(Done);
 506 
 507         __ bind(notInt);
 508         __ cmpw(flags, ftos);
 509         __ br(Assembler::NE, notFloat);
 510         // ftos
 511         __ load_float(field);
 512         __ push(ftos);
 513         __ b(Done);
 514 
 515         __ bind(notFloat);
 516         __ cmpw(flags, stos);
 517         __ br(Assembler::NE, notShort);
 518         // stos
 519         __ load_signed_short(r0, field);
 520         __ push(stos);
 521         __ b(Done);
 522 
 523         __ bind(notShort);
 524         __ cmpw(flags, btos);
 525         __ br(Assembler::NE, notByte);
 526         // btos
 527         __ load_signed_byte(r0, field);
 528         __ push(btos);
 529         __ b(Done);
 530 
 531         __ bind(notByte);
 532         __ cmpw(flags, ctos);
 533         __ br(Assembler::NE, notChar);
 534         // ctos
 535         __ load_unsigned_short(r0, field);
 536         __ push(ctos);
 537         __ b(Done);
 538 
 539         __ bind(notChar);
 540         __ cmpw(flags, ztos);
 541         __ br(Assembler::NE, notBool);
 542         // ztos
 543         __ load_signed_byte(r0, field);
 544         __ push(ztos);
 545         __ b(Done);
 546 
 547         __ bind(notBool);
 548         break;
 549       }
 550 
 551     case Bytecodes::_ldc2_w:
 552       {
 553         Label notLong, notDouble;
 554         __ cmpw(flags, ltos);
 555         __ br(Assembler::NE, notLong);
 556         // ltos
 557         __ ldr(r0, field);
 558         __ push(ltos);
 559         __ b(Done);
 560 
 561         __ bind(notLong);
 562         __ cmpw(flags, dtos);
 563         __ br(Assembler::NE, notDouble);
 564         // dtos
 565         __ load_double(field);
 566         __ push(dtos);
 567         __ b(Done);
 568 
 569        __ bind(notDouble);
 570         break;
 571       }
 572 
 573     default:
 574       ShouldNotReachHere();
 575     }
 576 
 577     __ stop("bad ldc/condy");
 578 }
 579 
 580 void TemplateTable::locals_index(Register reg, int offset)
 581 {
 582   __ ldrb(reg, at_bcp(offset));
 583   __ neg(reg, reg);
 584 }
 585 
 586 void TemplateTable::iload() {
 587   iload_internal();
 588 }
 589 
 590 void TemplateTable::nofast_iload() {
 591   iload_internal(may_not_rewrite);
 592 }
 593 
 594 void TemplateTable::iload_internal(RewriteControl rc) {
 595   transition(vtos, itos);
 596   if (RewriteFrequentPairs && rc == may_rewrite) {
 597     Label rewrite, done;
 598     Register bc = r4;
 599 
 600     // get next bytecode
 601     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 602 
 603     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 604     // last two iloads in a pair.  Comparing against fast_iload means that
 605     // the next bytecode is neither an iload or a caload, and therefore
 606     // an iload pair.
 607     __ cmpw(r1, Bytecodes::_iload);
 608     __ br(Assembler::EQ, done);
 609 
 610     // if _fast_iload rewrite to _fast_iload2
 611     __ cmpw(r1, Bytecodes::_fast_iload);
 612     __ movw(bc, Bytecodes::_fast_iload2);
 613     __ br(Assembler::EQ, rewrite);
 614 
 615     // if _caload rewrite to _fast_icaload
 616     __ cmpw(r1, Bytecodes::_caload);
 617     __ movw(bc, Bytecodes::_fast_icaload);
 618     __ br(Assembler::EQ, rewrite);
 619 
 620     // else rewrite to _fast_iload
 621     __ movw(bc, Bytecodes::_fast_iload);
 622 
 623     // rewrite
 624     // bc: new bytecode
 625     __ bind(rewrite);
 626     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 627     __ bind(done);
 628 
 629   }
 630 
 631   // do iload, get the local value into tos
 632   locals_index(r1);
 633   __ ldr(r0, iaddress(r1));
 634 
 635 }
 636 
 637 void TemplateTable::fast_iload2()
 638 {
 639   transition(vtos, itos);
 640   locals_index(r1);
 641   __ ldr(r0, iaddress(r1));
 642   __ push(itos);
 643   locals_index(r1, 3);
 644   __ ldr(r0, iaddress(r1));
 645 }
 646 
 647 void TemplateTable::fast_iload()
 648 {
 649   transition(vtos, itos);
 650   locals_index(r1);
 651   __ ldr(r0, iaddress(r1));
 652 }
 653 
 654 void TemplateTable::lload()
 655 {
 656   transition(vtos, ltos);
 657   __ ldrb(r1, at_bcp(1));
 658   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 659   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 660 }
 661 
 662 void TemplateTable::fload()
 663 {
 664   transition(vtos, ftos);
 665   locals_index(r1);
 666   // n.b. we use ldrd here because this is a 64 bit slot
 667   // this is comparable to the iload case
 668   __ ldrd(v0, faddress(r1));
 669 }
 670 
 671 void TemplateTable::dload()
 672 {
 673   transition(vtos, dtos);
 674   __ ldrb(r1, at_bcp(1));
 675   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 676   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 677 }
 678 
 679 void TemplateTable::aload()
 680 {
 681   transition(vtos, atos);
 682   locals_index(r1);
 683   __ ldr(r0, iaddress(r1));
 684 }
 685 
 686 void TemplateTable::locals_index_wide(Register reg) {
 687   __ ldrh(reg, at_bcp(2));
 688   __ rev16w(reg, reg);
 689   __ neg(reg, reg);
 690 }
 691 
 692 void TemplateTable::wide_iload() {
 693   transition(vtos, itos);
 694   locals_index_wide(r1);
 695   __ ldr(r0, iaddress(r1));
 696 }
 697 
 698 void TemplateTable::wide_lload()
 699 {
 700   transition(vtos, ltos);
 701   __ ldrh(r1, at_bcp(2));
 702   __ rev16w(r1, r1);
 703   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 704   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 705 }
 706 
 707 void TemplateTable::wide_fload()
 708 {
 709   transition(vtos, ftos);
 710   locals_index_wide(r1);
 711   // n.b. we use ldrd here because this is a 64 bit slot
 712   // this is comparable to the iload case
 713   __ ldrd(v0, faddress(r1));
 714 }
 715 
 716 void TemplateTable::wide_dload()
 717 {
 718   transition(vtos, dtos);
 719   __ ldrh(r1, at_bcp(2));
 720   __ rev16w(r1, r1);
 721   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 722   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 723 }
 724 
 725 void TemplateTable::wide_aload()
 726 {
 727   transition(vtos, atos);
 728   locals_index_wide(r1);
 729   __ ldr(r0, aaddress(r1));
 730 }
 731 
 732 void TemplateTable::index_check(Register array, Register index)
 733 {
 734   // destroys r1, rscratch1
 735   // check array
 736   __ null_check(array, arrayOopDesc::length_offset_in_bytes());
 737   // sign extend index for use by indexed load
 738   // __ movl2ptr(index, index);
 739   // check index
 740   Register length = rscratch1;
 741   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 742   __ cmpw(index, length);
 743   if (index != r1) {
 744     // ??? convention: move aberrant index into r1 for exception message
 745     assert(r1 != array, "different registers");
 746     __ mov(r1, index);
 747   }
 748   Label ok;
 749   __ br(Assembler::LO, ok);
 750     // ??? convention: move array into r3 for exception message
 751   __ mov(r3, array);
 752   __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 753   __ br(rscratch1);
 754   __ bind(ok);
 755 }
 756 
 757 void TemplateTable::iaload()
 758 {
 759   transition(itos, itos);
 760   __ mov(r1, r0);
 761   __ pop_ptr(r0);
 762   // r0: array
 763   // r1: index
 764   index_check(r0, r1); // leaves index in r1, kills rscratch1
 765   __ lea(r1, Address(r0, r1, Address::uxtw(2)));
 766   __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
 767 }
 768 
 769 void TemplateTable::laload()
 770 {
 771   transition(itos, ltos);
 772   __ mov(r1, r0);
 773   __ pop_ptr(r0);
 774   // r0: array
 775   // r1: index
 776   index_check(r0, r1); // leaves index in r1, kills rscratch1
 777   __ lea(r1, Address(r0, r1, Address::uxtw(3)));
 778   __ ldr(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_LONG)));
 779 }
 780 
 781 void TemplateTable::faload()
 782 {
 783   transition(itos, ftos);
 784   __ mov(r1, r0);
 785   __ pop_ptr(r0);
 786   // r0: array
 787   // r1: index
 788   index_check(r0, r1); // leaves index in r1, kills rscratch1
 789   __ lea(r1,  Address(r0, r1, Address::uxtw(2)));
 790   __ ldrs(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
 791 }
 792 
 793 void TemplateTable::daload()
 794 {
 795   transition(itos, dtos);
 796   __ mov(r1, r0);
 797   __ pop_ptr(r0);
 798   // r0: array
 799   // r1: index
 800   index_check(r0, r1); // leaves index in r1, kills rscratch1
 801   __ lea(r1,  Address(r0, r1, Address::uxtw(3)));
 802   __ ldrd(v0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
 803 }
 804 
 805 void TemplateTable::aaload()
 806 {
 807   transition(itos, atos);
 808   __ mov(r1, r0);
 809   __ pop_ptr(r0);
 810   // r0: array
 811   // r1: index
 812   index_check(r0, r1); // leaves index in r1, kills rscratch1
 813   int s = (UseCompressedOops ? 2 : 3);
 814   __ lea(r1, Address(r0, r1, Address::uxtw(s)));
 815   do_oop_load(_masm,
 816               Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
 817               r0,
 818               IN_HEAP | IN_HEAP_ARRAY);
 819 }
 820 
 821 void TemplateTable::baload()
 822 {
 823   transition(itos, itos);
 824   __ mov(r1, r0);
 825   __ pop_ptr(r0);
 826   // r0: array
 827   // r1: index
 828   index_check(r0, r1); // leaves index in r1, kills rscratch1
 829   __ lea(r1,  Address(r0, r1, Address::uxtw(0)));
 830   __ load_signed_byte(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_BYTE)));
 831 }
 832 
 833 void TemplateTable::caload()
 834 {
 835   transition(itos, itos);
 836   __ mov(r1, r0);
 837   __ pop_ptr(r0);
 838   // r0: array
 839   // r1: index
 840   index_check(r0, r1); // leaves index in r1, kills rscratch1
 841   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 842   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 843 }
 844 
 845 // iload followed by caload frequent pair
 846 void TemplateTable::fast_icaload()
 847 {
 848   transition(vtos, itos);
 849   // load index out of locals
 850   locals_index(r2);
 851   __ ldr(r1, iaddress(r2));
 852 
 853   __ pop_ptr(r0);
 854 
 855   // r0: array
 856   // r1: index
 857   index_check(r0, r1); // leaves index in r1, kills rscratch1
 858   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 859   __ load_unsigned_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_CHAR)));
 860 }
 861 
 862 void TemplateTable::saload()
 863 {
 864   transition(itos, itos);
 865   __ mov(r1, r0);
 866   __ pop_ptr(r0);
 867   // r0: array
 868   // r1: index
 869   index_check(r0, r1); // leaves index in r1, kills rscratch1
 870   __ lea(r1,  Address(r0, r1, Address::uxtw(1)));
 871   __ load_signed_short(r0, Address(r1,  arrayOopDesc::base_offset_in_bytes(T_SHORT)));
 872 }
 873 
 874 void TemplateTable::iload(int n)
 875 {
 876   transition(vtos, itos);
 877   __ ldr(r0, iaddress(n));
 878 }
 879 
 880 void TemplateTable::lload(int n)
 881 {
 882   transition(vtos, ltos);
 883   __ ldr(r0, laddress(n));
 884 }
 885 
 886 void TemplateTable::fload(int n)
 887 {
 888   transition(vtos, ftos);
 889   __ ldrs(v0, faddress(n));
 890 }
 891 
 892 void TemplateTable::dload(int n)
 893 {
 894   transition(vtos, dtos);
 895   __ ldrd(v0, daddress(n));
 896 }
 897 
 898 void TemplateTable::aload(int n)
 899 {
 900   transition(vtos, atos);
 901   __ ldr(r0, iaddress(n));
 902 }
 903 
 904 void TemplateTable::aload_0() {
 905   aload_0_internal();
 906 }
 907 
 908 void TemplateTable::nofast_aload_0() {
 909   aload_0_internal(may_not_rewrite);
 910 }
 911 
 912 void TemplateTable::aload_0_internal(RewriteControl rc) {
 913   // According to bytecode histograms, the pairs:
 914   //
 915   // _aload_0, _fast_igetfield
 916   // _aload_0, _fast_agetfield
 917   // _aload_0, _fast_fgetfield
 918   //
 919   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 920   // _aload_0 bytecode checks if the next bytecode is either
 921   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 922   // rewrites the current bytecode into a pair bytecode; otherwise it
 923   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 924   // the pair check anymore.
 925   //
 926   // Note: If the next bytecode is _getfield, the rewrite must be
 927   //       delayed, otherwise we may miss an opportunity for a pair.
 928   //
 929   // Also rewrite frequent pairs
 930   //   aload_0, aload_1
 931   //   aload_0, iload_1
 932   // These bytecodes with a small amount of code are most profitable
 933   // to rewrite
 934   if (RewriteFrequentPairs && rc == may_rewrite) {
 935     Label rewrite, done;
 936     const Register bc = r4;
 937 
 938     // get next bytecode
 939     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 940 
 941     // if _getfield then wait with rewrite
 942     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 943     __ br(Assembler::EQ, done);
 944 
 945     // if _igetfield then rewrite to _fast_iaccess_0
 946     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 947     __ cmpw(r1, Bytecodes::_fast_igetfield);
 948     __ movw(bc, Bytecodes::_fast_iaccess_0);
 949     __ br(Assembler::EQ, rewrite);
 950 
 951     // if _agetfield then rewrite to _fast_aaccess_0
 952     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 953     __ cmpw(r1, Bytecodes::_fast_agetfield);
 954     __ movw(bc, Bytecodes::_fast_aaccess_0);
 955     __ br(Assembler::EQ, rewrite);
 956 
 957     // if _fgetfield then rewrite to _fast_faccess_0
 958     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 959     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 960     __ movw(bc, Bytecodes::_fast_faccess_0);
 961     __ br(Assembler::EQ, rewrite);
 962 
 963     // else rewrite to _fast_aload0
 964     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 965     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 966 
 967     // rewrite
 968     // bc: new bytecode
 969     __ bind(rewrite);
 970     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 971 
 972     __ bind(done);
 973   }
 974 
 975   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 976   aload(0);
 977 }
 978 
 979 void TemplateTable::istore()
 980 {
 981   transition(itos, vtos);
 982   locals_index(r1);
 983   // FIXME: We're being very pernickerty here storing a jint in a
 984   // local with strw, which costs an extra instruction over what we'd
 985   // be able to do with a simple str.  We should just store the whole
 986   // word.
 987   __ lea(rscratch1, iaddress(r1));
 988   __ strw(r0, Address(rscratch1));
 989 }
 990 
 991 void TemplateTable::lstore()
 992 {
 993   transition(ltos, vtos);
 994   locals_index(r1);
 995   __ str(r0, laddress(r1, rscratch1, _masm));
 996 }
 997 
 998 void TemplateTable::fstore() {
 999   transition(ftos, vtos);
1000   locals_index(r1);
1001   __ lea(rscratch1, iaddress(r1));
1002   __ strs(v0, Address(rscratch1));
1003 }
1004 
1005 void TemplateTable::dstore() {
1006   transition(dtos, vtos);
1007   locals_index(r1);
1008   __ strd(v0, daddress(r1, rscratch1, _masm));
1009 }
1010 
1011 void TemplateTable::astore()
1012 {
1013   transition(vtos, vtos);
1014   __ pop_ptr(r0);
1015   locals_index(r1);
1016   __ str(r0, aaddress(r1));
1017 }
1018 
1019 void TemplateTable::wide_istore() {
1020   transition(vtos, vtos);
1021   __ pop_i();
1022   locals_index_wide(r1);
1023   __ lea(rscratch1, iaddress(r1));
1024   __ strw(r0, Address(rscratch1));
1025 }
1026 
1027 void TemplateTable::wide_lstore() {
1028   transition(vtos, vtos);
1029   __ pop_l();
1030   locals_index_wide(r1);
1031   __ str(r0, laddress(r1, rscratch1, _masm));
1032 }
1033 
1034 void TemplateTable::wide_fstore() {
1035   transition(vtos, vtos);
1036   __ pop_f();
1037   locals_index_wide(r1);
1038   __ lea(rscratch1, faddress(r1));
1039   __ strs(v0, rscratch1);
1040 }
1041 
1042 void TemplateTable::wide_dstore() {
1043   transition(vtos, vtos);
1044   __ pop_d();
1045   locals_index_wide(r1);
1046   __ strd(v0, daddress(r1, rscratch1, _masm));
1047 }
1048 
1049 void TemplateTable::wide_astore() {
1050   transition(vtos, vtos);
1051   __ pop_ptr(r0);
1052   locals_index_wide(r1);
1053   __ str(r0, aaddress(r1));
1054 }
1055 
1056 void TemplateTable::iastore() {
1057   transition(itos, vtos);
1058   __ pop_i(r1);
1059   __ pop_ptr(r3);
1060   // r0: value
1061   // r1: index
1062   // r3: array
1063   index_check(r3, r1); // prefer index in r1
1064   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1065   __ strw(r0, Address(rscratch1,
1066                       arrayOopDesc::base_offset_in_bytes(T_INT)));
1067 }
1068 
1069 void TemplateTable::lastore() {
1070   transition(ltos, vtos);
1071   __ pop_i(r1);
1072   __ pop_ptr(r3);
1073   // r0: value
1074   // r1: index
1075   // r3: array
1076   index_check(r3, r1); // prefer index in r1
1077   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1078   __ str(r0, Address(rscratch1,
1079                       arrayOopDesc::base_offset_in_bytes(T_LONG)));
1080 }
1081 
1082 void TemplateTable::fastore() {
1083   transition(ftos, vtos);
1084   __ pop_i(r1);
1085   __ pop_ptr(r3);
1086   // v0: value
1087   // r1:  index
1088   // r3:  array
1089   index_check(r3, r1); // prefer index in r1
1090   __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1091   __ strs(v0, Address(rscratch1,
1092                       arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1093 }
1094 
1095 void TemplateTable::dastore() {
1096   transition(dtos, vtos);
1097   __ pop_i(r1);
1098   __ pop_ptr(r3);
1099   // v0: value
1100   // r1:  index
1101   // r3:  array
1102   index_check(r3, r1); // prefer index in r1
1103   __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1104   __ strd(v0, Address(rscratch1,
1105                       arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1106 }
1107 
1108 void TemplateTable::aastore() {
1109   Label is_null, ok_is_subtype, done;
1110   transition(vtos, vtos);
1111   // stack: ..., array, index, value
1112   __ ldr(r0, at_tos());    // value
1113   __ ldr(r2, at_tos_p1()); // index
1114   __ ldr(r3, at_tos_p2()); // array
1115 
1116   Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1117 
1118   index_check(r3, r2);     // kills r1
1119   __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1120 
1121   // do array store check - check for NULL value first
1122   __ cbz(r0, is_null);
1123 
1124   // Move subklass into r1
1125   __ load_klass(r1, r0);
1126   // Move superklass into r0
1127   __ load_klass(r0, r3);
1128   __ ldr(r0, Address(r0,
1129                      ObjArrayKlass::element_klass_offset()));
1130   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1131 
1132   // Generate subtype check.  Blows r2, r5
1133   // Superklass in r0.  Subklass in r1.
1134   __ gen_subtype_check(r1, ok_is_subtype);
1135 
1136   // Come here on failure
1137   // object is at TOS
1138   __ b(Interpreter::_throw_ArrayStoreException_entry);
1139 
1140   // Come here on success
1141   __ bind(ok_is_subtype);
1142 
1143   // Get the value we will store
1144   __ ldr(r0, at_tos());
1145   // Now store using the appropriate barrier
1146   do_oop_store(_masm, element_address, r0, IN_HEAP | IN_HEAP_ARRAY);
1147   __ b(done);
1148 
1149   // Have a NULL in r0, r3=array, r2=index.  Store NULL at ary[idx]
1150   __ bind(is_null);
1151   __ profile_null_seen(r2);
1152 
1153   // Store a NULL
1154   do_oop_store(_masm, element_address, noreg, IN_HEAP | IN_HEAP_ARRAY);
1155 
1156   // Pop stack arguments
1157   __ bind(done);
1158   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1159 }
1160 
1161 void TemplateTable::bastore()
1162 {
1163   transition(itos, vtos);
1164   __ pop_i(r1);
1165   __ pop_ptr(r3);
1166   // r0: value
1167   // r1: index
1168   // r3: array
1169   index_check(r3, r1); // prefer index in r1
1170 
1171   // Need to check whether array is boolean or byte
1172   // since both types share the bastore bytecode.
1173   __ load_klass(r2, r3);
1174   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1175   int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1176   Label L_skip;
1177   __ tbz(r2, diffbit_index, L_skip);
1178   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1179   __ bind(L_skip);
1180 
1181   __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1182   __ strb(r0, Address(rscratch1,
1183                       arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1184 }
1185 
1186 void TemplateTable::castore()
1187 {
1188   transition(itos, vtos);
1189   __ pop_i(r1);
1190   __ pop_ptr(r3);
1191   // r0: value
1192   // r1: index
1193   // r3: array
1194   index_check(r3, r1); // prefer index in r1
1195   __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1196   __ strh(r0, Address(rscratch1,
1197                       arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1198 }
1199 
1200 void TemplateTable::sastore()
1201 {
1202   castore();
1203 }
1204 
1205 void TemplateTable::istore(int n)
1206 {
1207   transition(itos, vtos);
1208   __ str(r0, iaddress(n));
1209 }
1210 
1211 void TemplateTable::lstore(int n)
1212 {
1213   transition(ltos, vtos);
1214   __ str(r0, laddress(n));
1215 }
1216 
1217 void TemplateTable::fstore(int n)
1218 {
1219   transition(ftos, vtos);
1220   __ strs(v0, faddress(n));
1221 }
1222 
1223 void TemplateTable::dstore(int n)
1224 {
1225   transition(dtos, vtos);
1226   __ strd(v0, daddress(n));
1227 }
1228 
1229 void TemplateTable::astore(int n)
1230 {
1231   transition(vtos, vtos);
1232   __ pop_ptr(r0);
1233   __ str(r0, iaddress(n));
1234 }
1235 
1236 void TemplateTable::pop()
1237 {
1238   transition(vtos, vtos);
1239   __ add(esp, esp, Interpreter::stackElementSize);
1240 }
1241 
1242 void TemplateTable::pop2()
1243 {
1244   transition(vtos, vtos);
1245   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1246 }
1247 
1248 void TemplateTable::dup()
1249 {
1250   transition(vtos, vtos);
1251   __ ldr(r0, Address(esp, 0));
1252   __ push(r0);
1253   // stack: ..., a, a
1254 }
1255 
1256 void TemplateTable::dup_x1()
1257 {
1258   transition(vtos, vtos);
1259   // stack: ..., a, b
1260   __ ldr(r0, at_tos());  // load b
1261   __ ldr(r2, at_tos_p1());  // load a
1262   __ str(r0, at_tos_p1());  // store b
1263   __ str(r2, at_tos());  // store a
1264   __ push(r0);                  // push b
1265   // stack: ..., b, a, b
1266 }
1267 
1268 void TemplateTable::dup_x2()
1269 {
1270   transition(vtos, vtos);
1271   // stack: ..., a, b, c
1272   __ ldr(r0, at_tos());  // load c
1273   __ ldr(r2, at_tos_p2());  // load a
1274   __ str(r0, at_tos_p2());  // store c in a
1275   __ push(r0);      // push c
1276   // stack: ..., c, b, c, c
1277   __ ldr(r0, at_tos_p2());  // load b
1278   __ str(r2, at_tos_p2());  // store a in b
1279   // stack: ..., c, a, c, c
1280   __ str(r0, at_tos_p1());  // store b in c
1281   // stack: ..., c, a, b, c
1282 }
1283 
1284 void TemplateTable::dup2()
1285 {
1286   transition(vtos, vtos);
1287   // stack: ..., a, b
1288   __ ldr(r0, at_tos_p1());  // load a
1289   __ push(r0);                  // push a
1290   __ ldr(r0, at_tos_p1());  // load b
1291   __ push(r0);                  // push b
1292   // stack: ..., a, b, a, b
1293 }
1294 
1295 void TemplateTable::dup2_x1()
1296 {
1297   transition(vtos, vtos);
1298   // stack: ..., a, b, c
1299   __ ldr(r2, at_tos());  // load c
1300   __ ldr(r0, at_tos_p1());  // load b
1301   __ push(r0);                  // push b
1302   __ push(r2);                  // push c
1303   // stack: ..., a, b, c, b, c
1304   __ str(r2, at_tos_p3());  // store c in b
1305   // stack: ..., a, c, c, b, c
1306   __ ldr(r2, at_tos_p4());  // load a
1307   __ str(r2, at_tos_p2());  // store a in 2nd c
1308   // stack: ..., a, c, a, b, c
1309   __ str(r0, at_tos_p4());  // store b in a
1310   // stack: ..., b, c, a, b, c
1311 }
1312 
1313 void TemplateTable::dup2_x2()
1314 {
1315   transition(vtos, vtos);
1316   // stack: ..., a, b, c, d
1317   __ ldr(r2, at_tos());  // load d
1318   __ ldr(r0, at_tos_p1());  // load c
1319   __ push(r0)            ;      // push c
1320   __ push(r2);                  // push d
1321   // stack: ..., a, b, c, d, c, d
1322   __ ldr(r0, at_tos_p4());  // load b
1323   __ str(r0, at_tos_p2());  // store b in d
1324   __ str(r2, at_tos_p4());  // store d in b
1325   // stack: ..., a, d, c, b, c, d
1326   __ ldr(r2, at_tos_p5());  // load a
1327   __ ldr(r0, at_tos_p3());  // load c
1328   __ str(r2, at_tos_p3());  // store a in c
1329   __ str(r0, at_tos_p5());  // store c in a
1330   // stack: ..., c, d, a, b, c, d
1331 }
1332 
1333 void TemplateTable::swap()
1334 {
1335   transition(vtos, vtos);
1336   // stack: ..., a, b
1337   __ ldr(r2, at_tos_p1());  // load a
1338   __ ldr(r0, at_tos());  // load b
1339   __ str(r2, at_tos());  // store a in b
1340   __ str(r0, at_tos_p1());  // store b in a
1341   // stack: ..., b, a
1342 }
1343 
1344 void TemplateTable::iop2(Operation op)
1345 {
1346   transition(itos, itos);
1347   // r0 <== r1 op r0
1348   __ pop_i(r1);
1349   switch (op) {
1350   case add  : __ addw(r0, r1, r0); break;
1351   case sub  : __ subw(r0, r1, r0); break;
1352   case mul  : __ mulw(r0, r1, r0); break;
1353   case _and : __ andw(r0, r1, r0); break;
1354   case _or  : __ orrw(r0, r1, r0); break;
1355   case _xor : __ eorw(r0, r1, r0); break;
1356   case shl  : __ lslvw(r0, r1, r0); break;
1357   case shr  : __ asrvw(r0, r1, r0); break;
1358   case ushr : __ lsrvw(r0, r1, r0);break;
1359   default   : ShouldNotReachHere();
1360   }
1361 }
1362 
1363 void TemplateTable::lop2(Operation op)
1364 {
1365   transition(ltos, ltos);
1366   // r0 <== r1 op r0
1367   __ pop_l(r1);
1368   switch (op) {
1369   case add  : __ add(r0, r1, r0); break;
1370   case sub  : __ sub(r0, r1, r0); break;
1371   case mul  : __ mul(r0, r1, r0); break;
1372   case _and : __ andr(r0, r1, r0); break;
1373   case _or  : __ orr(r0, r1, r0); break;
1374   case _xor : __ eor(r0, r1, r0); break;
1375   default   : ShouldNotReachHere();
1376   }
1377 }
1378 
1379 void TemplateTable::idiv()
1380 {
1381   transition(itos, itos);
1382   // explicitly check for div0
1383   Label no_div0;
1384   __ cbnzw(r0, no_div0);
1385   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1386   __ br(rscratch1);
1387   __ bind(no_div0);
1388   __ pop_i(r1);
1389   // r0 <== r1 idiv r0
1390   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1391 }
1392 
1393 void TemplateTable::irem()
1394 {
1395   transition(itos, itos);
1396   // explicitly check for div0
1397   Label no_div0;
1398   __ cbnzw(r0, no_div0);
1399   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1400   __ br(rscratch1);
1401   __ bind(no_div0);
1402   __ pop_i(r1);
1403   // r0 <== r1 irem r0
1404   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1405 }
1406 
1407 void TemplateTable::lmul()
1408 {
1409   transition(ltos, ltos);
1410   __ pop_l(r1);
1411   __ mul(r0, r0, r1);
1412 }
1413 
1414 void TemplateTable::ldiv()
1415 {
1416   transition(ltos, ltos);
1417   // explicitly check for div0
1418   Label no_div0;
1419   __ cbnz(r0, no_div0);
1420   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1421   __ br(rscratch1);
1422   __ bind(no_div0);
1423   __ pop_l(r1);
1424   // r0 <== r1 ldiv r0
1425   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1426 }
1427 
1428 void TemplateTable::lrem()
1429 {
1430   transition(ltos, ltos);
1431   // explicitly check for div0
1432   Label no_div0;
1433   __ cbnz(r0, no_div0);
1434   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1435   __ br(rscratch1);
1436   __ bind(no_div0);
1437   __ pop_l(r1);
1438   // r0 <== r1 lrem r0
1439   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1440 }
1441 
1442 void TemplateTable::lshl()
1443 {
1444   transition(itos, ltos);
1445   // shift count is in r0
1446   __ pop_l(r1);
1447   __ lslv(r0, r1, r0);
1448 }
1449 
1450 void TemplateTable::lshr()
1451 {
1452   transition(itos, ltos);
1453   // shift count is in r0
1454   __ pop_l(r1);
1455   __ asrv(r0, r1, r0);
1456 }
1457 
1458 void TemplateTable::lushr()
1459 {
1460   transition(itos, ltos);
1461   // shift count is in r0
1462   __ pop_l(r1);
1463   __ lsrv(r0, r1, r0);
1464 }
1465 
1466 void TemplateTable::fop2(Operation op)
1467 {
1468   transition(ftos, ftos);
1469   switch (op) {
1470   case add:
1471     // n.b. use ldrd because this is a 64 bit slot
1472     __ pop_f(v1);
1473     __ fadds(v0, v1, v0);
1474     break;
1475   case sub:
1476     __ pop_f(v1);
1477     __ fsubs(v0, v1, v0);
1478     break;
1479   case mul:
1480     __ pop_f(v1);
1481     __ fmuls(v0, v1, v0);
1482     break;
1483   case div:
1484     __ pop_f(v1);
1485     __ fdivs(v0, v1, v0);
1486     break;
1487   case rem:
1488     __ fmovs(v1, v0);
1489     __ pop_f(v0);
1490     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1491                          0, 2, MacroAssembler::ret_type_float);
1492     break;
1493   default:
1494     ShouldNotReachHere();
1495     break;
1496   }
1497 }
1498 
1499 void TemplateTable::dop2(Operation op)
1500 {
1501   transition(dtos, dtos);
1502   switch (op) {
1503   case add:
1504     // n.b. use ldrd because this is a 64 bit slot
1505     __ pop_d(v1);
1506     __ faddd(v0, v1, v0);
1507     break;
1508   case sub:
1509     __ pop_d(v1);
1510     __ fsubd(v0, v1, v0);
1511     break;
1512   case mul:
1513     __ pop_d(v1);
1514     __ fmuld(v0, v1, v0);
1515     break;
1516   case div:
1517     __ pop_d(v1);
1518     __ fdivd(v0, v1, v0);
1519     break;
1520   case rem:
1521     __ fmovd(v1, v0);
1522     __ pop_d(v0);
1523     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1524                          0, 2, MacroAssembler::ret_type_double);
1525     break;
1526   default:
1527     ShouldNotReachHere();
1528     break;
1529   }
1530 }
1531 
1532 void TemplateTable::ineg()
1533 {
1534   transition(itos, itos);
1535   __ negw(r0, r0);
1536 
1537 }
1538 
1539 void TemplateTable::lneg()
1540 {
1541   transition(ltos, ltos);
1542   __ neg(r0, r0);
1543 }
1544 
1545 void TemplateTable::fneg()
1546 {
1547   transition(ftos, ftos);
1548   __ fnegs(v0, v0);
1549 }
1550 
1551 void TemplateTable::dneg()
1552 {
1553   transition(dtos, dtos);
1554   __ fnegd(v0, v0);
1555 }
1556 
1557 void TemplateTable::iinc()
1558 {
1559   transition(vtos, vtos);
1560   __ load_signed_byte(r1, at_bcp(2)); // get constant
1561   locals_index(r2);
1562   __ ldr(r0, iaddress(r2));
1563   __ addw(r0, r0, r1);
1564   __ str(r0, iaddress(r2));
1565 }
1566 
1567 void TemplateTable::wide_iinc()
1568 {
1569   transition(vtos, vtos);
1570   // __ mov(r1, zr);
1571   __ ldrw(r1, at_bcp(2)); // get constant and index
1572   __ rev16(r1, r1);
1573   __ ubfx(r2, r1, 0, 16);
1574   __ neg(r2, r2);
1575   __ sbfx(r1, r1, 16, 16);
1576   __ ldr(r0, iaddress(r2));
1577   __ addw(r0, r0, r1);
1578   __ str(r0, iaddress(r2));
1579 }
1580 
1581 void TemplateTable::convert()
1582 {
1583   // Checking
1584 #ifdef ASSERT
1585   {
1586     TosState tos_in  = ilgl;
1587     TosState tos_out = ilgl;
1588     switch (bytecode()) {
1589     case Bytecodes::_i2l: // fall through
1590     case Bytecodes::_i2f: // fall through
1591     case Bytecodes::_i2d: // fall through
1592     case Bytecodes::_i2b: // fall through
1593     case Bytecodes::_i2c: // fall through
1594     case Bytecodes::_i2s: tos_in = itos; break;
1595     case Bytecodes::_l2i: // fall through
1596     case Bytecodes::_l2f: // fall through
1597     case Bytecodes::_l2d: tos_in = ltos; break;
1598     case Bytecodes::_f2i: // fall through
1599     case Bytecodes::_f2l: // fall through
1600     case Bytecodes::_f2d: tos_in = ftos; break;
1601     case Bytecodes::_d2i: // fall through
1602     case Bytecodes::_d2l: // fall through
1603     case Bytecodes::_d2f: tos_in = dtos; break;
1604     default             : ShouldNotReachHere();
1605     }
1606     switch (bytecode()) {
1607     case Bytecodes::_l2i: // fall through
1608     case Bytecodes::_f2i: // fall through
1609     case Bytecodes::_d2i: // fall through
1610     case Bytecodes::_i2b: // fall through
1611     case Bytecodes::_i2c: // fall through
1612     case Bytecodes::_i2s: tos_out = itos; break;
1613     case Bytecodes::_i2l: // fall through
1614     case Bytecodes::_f2l: // fall through
1615     case Bytecodes::_d2l: tos_out = ltos; break;
1616     case Bytecodes::_i2f: // fall through
1617     case Bytecodes::_l2f: // fall through
1618     case Bytecodes::_d2f: tos_out = ftos; break;
1619     case Bytecodes::_i2d: // fall through
1620     case Bytecodes::_l2d: // fall through
1621     case Bytecodes::_f2d: tos_out = dtos; break;
1622     default             : ShouldNotReachHere();
1623     }
1624     transition(tos_in, tos_out);
1625   }
1626 #endif // ASSERT
1627   // static const int64_t is_nan = 0x8000000000000000L;
1628 
1629   // Conversion
1630   switch (bytecode()) {
1631   case Bytecodes::_i2l:
1632     __ sxtw(r0, r0);
1633     break;
1634   case Bytecodes::_i2f:
1635     __ scvtfws(v0, r0);
1636     break;
1637   case Bytecodes::_i2d:
1638     __ scvtfwd(v0, r0);
1639     break;
1640   case Bytecodes::_i2b:
1641     __ sxtbw(r0, r0);
1642     break;
1643   case Bytecodes::_i2c:
1644     __ uxthw(r0, r0);
1645     break;
1646   case Bytecodes::_i2s:
1647     __ sxthw(r0, r0);
1648     break;
1649   case Bytecodes::_l2i:
1650     __ uxtw(r0, r0);
1651     break;
1652   case Bytecodes::_l2f:
1653     __ scvtfs(v0, r0);
1654     break;
1655   case Bytecodes::_l2d:
1656     __ scvtfd(v0, r0);
1657     break;
1658   case Bytecodes::_f2i:
1659   {
1660     Label L_Okay;
1661     __ clear_fpsr();
1662     __ fcvtzsw(r0, v0);
1663     __ get_fpsr(r1);
1664     __ cbzw(r1, L_Okay);
1665     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1666                          0, 1, MacroAssembler::ret_type_integral);
1667     __ bind(L_Okay);
1668   }
1669     break;
1670   case Bytecodes::_f2l:
1671   {
1672     Label L_Okay;
1673     __ clear_fpsr();
1674     __ fcvtzs(r0, v0);
1675     __ get_fpsr(r1);
1676     __ cbzw(r1, L_Okay);
1677     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1678                          0, 1, MacroAssembler::ret_type_integral);
1679     __ bind(L_Okay);
1680   }
1681     break;
1682   case Bytecodes::_f2d:
1683     __ fcvts(v0, v0);
1684     break;
1685   case Bytecodes::_d2i:
1686   {
1687     Label L_Okay;
1688     __ clear_fpsr();
1689     __ fcvtzdw(r0, v0);
1690     __ get_fpsr(r1);
1691     __ cbzw(r1, L_Okay);
1692     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1693                          0, 1, MacroAssembler::ret_type_integral);
1694     __ bind(L_Okay);
1695   }
1696     break;
1697   case Bytecodes::_d2l:
1698   {
1699     Label L_Okay;
1700     __ clear_fpsr();
1701     __ fcvtzd(r0, v0);
1702     __ get_fpsr(r1);
1703     __ cbzw(r1, L_Okay);
1704     __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1705                          0, 1, MacroAssembler::ret_type_integral);
1706     __ bind(L_Okay);
1707   }
1708     break;
1709   case Bytecodes::_d2f:
1710     __ fcvtd(v0, v0);
1711     break;
1712   default:
1713     ShouldNotReachHere();
1714   }
1715 }
1716 
1717 void TemplateTable::lcmp()
1718 {
1719   transition(ltos, itos);
1720   Label done;
1721   __ pop_l(r1);
1722   __ cmp(r1, r0);
1723   __ mov(r0, (u_int64_t)-1L);
1724   __ br(Assembler::LT, done);
1725   // __ mov(r0, 1UL);
1726   // __ csel(r0, r0, zr, Assembler::NE);
1727   // and here is a faster way
1728   __ csinc(r0, zr, zr, Assembler::EQ);
1729   __ bind(done);
1730 }
1731 
1732 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1733 {
1734   Label done;
1735   if (is_float) {
1736     // XXX get rid of pop here, use ... reg, mem32
1737     __ pop_f(v1);
1738     __ fcmps(v1, v0);
1739   } else {
1740     // XXX get rid of pop here, use ... reg, mem64
1741     __ pop_d(v1);
1742     __ fcmpd(v1, v0);
1743   }
1744   if (unordered_result < 0) {
1745     // we want -1 for unordered or less than, 0 for equal and 1 for
1746     // greater than.
1747     __ mov(r0, (u_int64_t)-1L);
1748     // for FP LT tests less than or unordered
1749     __ br(Assembler::LT, done);
1750     // install 0 for EQ otherwise 1
1751     __ csinc(r0, zr, zr, Assembler::EQ);
1752   } else {
1753     // we want -1 for less than, 0 for equal and 1 for unordered or
1754     // greater than.
1755     __ mov(r0, 1L);
1756     // for FP HI tests greater than or unordered
1757     __ br(Assembler::HI, done);
1758     // install 0 for EQ otherwise ~0
1759     __ csinv(r0, zr, zr, Assembler::EQ);
1760 
1761   }
1762   __ bind(done);
1763 }
1764 
1765 void TemplateTable::branch(bool is_jsr, bool is_wide)
1766 {
1767   // We might be moving to a safepoint.  The thread which calls
1768   // Interpreter::notice_safepoints() will effectively flush its cache
1769   // when it makes a system call, but we need to do something to
1770   // ensure that we see the changed dispatch table.
1771   __ membar(MacroAssembler::LoadLoad);
1772 
1773   __ profile_taken_branch(r0, r1);
1774   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1775                              InvocationCounter::counter_offset();
1776   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1777                               InvocationCounter::counter_offset();
1778 
1779   // load branch displacement
1780   if (!is_wide) {
1781     __ ldrh(r2, at_bcp(1));
1782     __ rev16(r2, r2);
1783     // sign extend the 16 bit value in r2
1784     __ sbfm(r2, r2, 0, 15);
1785   } else {
1786     __ ldrw(r2, at_bcp(1));
1787     __ revw(r2, r2);
1788     // sign extend the 32 bit value in r2
1789     __ sbfm(r2, r2, 0, 31);
1790   }
1791 
1792   // Handle all the JSR stuff here, then exit.
1793   // It's much shorter and cleaner than intermingling with the non-JSR
1794   // normal-branch stuff occurring below.
1795 
1796   if (is_jsr) {
1797     // Pre-load the next target bytecode into rscratch1
1798     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1799     // compute return address as bci
1800     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1801     __ add(rscratch2, rscratch2,
1802            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1803     __ sub(r1, rbcp, rscratch2);
1804     __ push_i(r1);
1805     // Adjust the bcp by the 16-bit displacement in r2
1806     __ add(rbcp, rbcp, r2);
1807     __ dispatch_only(vtos, /*generate_poll*/true);
1808     return;
1809   }
1810 
1811   // Normal (non-jsr) branch handling
1812 
1813   // Adjust the bcp by the displacement in r2
1814   __ add(rbcp, rbcp, r2);
1815 
1816   assert(UseLoopCounter || !UseOnStackReplacement,
1817          "on-stack-replacement requires loop counters");
1818   Label backedge_counter_overflow;
1819   Label profile_method;
1820   Label dispatch;
1821   if (UseLoopCounter) {
1822     // increment backedge counter for backward branches
1823     // r0: MDO
1824     // w1: MDO bumped taken-count
1825     // r2: target offset
1826     __ cmp(r2, zr);
1827     __ br(Assembler::GT, dispatch); // count only if backward branch
1828 
1829     // ECN: FIXME: This code smells
1830     // check if MethodCounters exists
1831     Label has_counters;
1832     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1833     __ cbnz(rscratch1, has_counters);
1834     __ push(r0);
1835     __ push(r1);
1836     __ push(r2);
1837     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1838             InterpreterRuntime::build_method_counters), rmethod);
1839     __ pop(r2);
1840     __ pop(r1);
1841     __ pop(r0);
1842     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1843     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1844     __ bind(has_counters);
1845 
1846     if (TieredCompilation) {
1847       Label no_mdo;
1848       int increment = InvocationCounter::count_increment;
1849       if (ProfileInterpreter) {
1850         // Are we profiling?
1851         __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1852         __ cbz(r1, no_mdo);
1853         // Increment the MDO backedge counter
1854         const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1855                                            in_bytes(InvocationCounter::counter_offset()));
1856         const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1857         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1858                                    r0, rscratch1, false, Assembler::EQ,
1859                                    UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1860         __ b(dispatch);
1861       }
1862       __ bind(no_mdo);
1863       // Increment backedge counter in MethodCounters*
1864       __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1865       const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1866       __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1867                                  r0, rscratch2, false, Assembler::EQ,
1868                                  UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1869     } else { // not TieredCompilation
1870       // increment counter
1871       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1872       __ ldrw(r0, Address(rscratch2, be_offset));        // load backedge counter
1873       __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1874       __ strw(rscratch1, Address(rscratch2, be_offset));        // store counter
1875 
1876       __ ldrw(r0, Address(rscratch2, inv_offset));    // load invocation counter
1877       __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1878       __ addw(r0, r0, rscratch1);        // add both counters
1879 
1880       if (ProfileInterpreter) {
1881         // Test to see if we should create a method data oop
1882         __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1883         __ cmpw(r0, rscratch1);
1884         __ br(Assembler::LT, dispatch);
1885 
1886         // if no method data exists, go to profile method
1887         __ test_method_data_pointer(r0, profile_method);
1888 
1889         if (UseOnStackReplacement) {
1890           // check for overflow against w1 which is the MDO taken count
1891           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1892           __ cmpw(r1, rscratch1);
1893           __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1894 
1895           // When ProfileInterpreter is on, the backedge_count comes
1896           // from the MethodData*, which value does not get reset on
1897           // the call to frequency_counter_overflow().  To avoid
1898           // excessive calls to the overflow routine while the method is
1899           // being compiled, add a second test to make sure the overflow
1900           // function is called only once every overflow_frequency.
1901           const int overflow_frequency = 1024;
1902           __ andsw(r1, r1, overflow_frequency - 1);
1903           __ br(Assembler::EQ, backedge_counter_overflow);
1904 
1905         }
1906       } else {
1907         if (UseOnStackReplacement) {
1908           // check for overflow against w0, which is the sum of the
1909           // counters
1910           __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1911           __ cmpw(r0, rscratch1);
1912           __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1913         }
1914       }
1915     }
1916     __ bind(dispatch);
1917   }
1918 
1919   // Pre-load the next target bytecode into rscratch1
1920   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1921 
1922   // continue with the bytecode @ target
1923   // rscratch1: target bytecode
1924   // rbcp: target bcp
1925   __ dispatch_only(vtos, /*generate_poll*/true);
1926 
1927   if (UseLoopCounter) {
1928     if (ProfileInterpreter) {
1929       // Out-of-line code to allocate method data oop.
1930       __ bind(profile_method);
1931       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1932       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1933       __ set_method_data_pointer_for_bcp();
1934       __ b(dispatch);
1935     }
1936 
1937     if (UseOnStackReplacement) {
1938       // invocation counter overflow
1939       __ bind(backedge_counter_overflow);
1940       __ neg(r2, r2);
1941       __ add(r2, r2, rbcp);     // branch bcp
1942       // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1943       __ call_VM(noreg,
1944                  CAST_FROM_FN_PTR(address,
1945                                   InterpreterRuntime::frequency_counter_overflow),
1946                  r2);
1947       __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1948 
1949       // r0: osr nmethod (osr ok) or NULL (osr not possible)
1950       // w1: target bytecode
1951       // r2: scratch
1952       __ cbz(r0, dispatch);     // test result -- no osr if null
1953       // nmethod may have been invalidated (VM may block upon call_VM return)
1954       __ ldrb(r2, Address(r0, nmethod::state_offset()));
1955       if (nmethod::in_use != 0)
1956         __ sub(r2, r2, nmethod::in_use);
1957       __ cbnz(r2, dispatch);
1958 
1959       // We have the address of an on stack replacement routine in r0
1960       // We need to prepare to execute the OSR method. First we must
1961       // migrate the locals and monitors off of the stack.
1962 
1963       __ mov(r19, r0);                             // save the nmethod
1964 
1965       call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1966 
1967       // r0 is OSR buffer, move it to expected parameter location
1968       __ mov(j_rarg0, r0);
1969 
1970       // remove activation
1971       // get sender esp
1972       __ ldr(esp,
1973           Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1974       // remove frame anchor
1975       __ leave();
1976       // Ensure compiled code always sees stack at proper alignment
1977       __ andr(sp, esp, -16);
1978 
1979       // and begin the OSR nmethod
1980       __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1981       __ br(rscratch1);
1982     }
1983   }
1984 }
1985 
1986 
1987 void TemplateTable::if_0cmp(Condition cc)
1988 {
1989   transition(itos, vtos);
1990   // assume branch is more often taken than not (loops use backward branches)
1991   Label not_taken;
1992   if (cc == equal)
1993     __ cbnzw(r0, not_taken);
1994   else if (cc == not_equal)
1995     __ cbzw(r0, not_taken);
1996   else {
1997     __ andsw(zr, r0, r0);
1998     __ br(j_not(cc), not_taken);
1999   }
2000 
2001   branch(false, false);
2002   __ bind(not_taken);
2003   __ profile_not_taken_branch(r0);
2004 }
2005 
2006 void TemplateTable::if_icmp(Condition cc)
2007 {
2008   transition(itos, vtos);
2009   // assume branch is more often taken than not (loops use backward branches)
2010   Label not_taken;
2011   __ pop_i(r1);
2012   __ cmpw(r1, r0, Assembler::LSL);
2013   __ br(j_not(cc), not_taken);
2014   branch(false, false);
2015   __ bind(not_taken);
2016   __ profile_not_taken_branch(r0);
2017 }
2018 
2019 void TemplateTable::if_nullcmp(Condition cc)
2020 {
2021   transition(atos, vtos);
2022   // assume branch is more often taken than not (loops use backward branches)
2023   Label not_taken;
2024   if (cc == equal)
2025     __ cbnz(r0, not_taken);
2026   else
2027     __ cbz(r0, not_taken);
2028   branch(false, false);
2029   __ bind(not_taken);
2030   __ profile_not_taken_branch(r0);
2031 }
2032 
2033 void TemplateTable::if_acmp(Condition cc)
2034 {
2035   transition(atos, vtos);
2036   // assume branch is more often taken than not (loops use backward branches)
2037   Label not_taken;
2038   __ pop_ptr(r1);
2039   __ cmp(r1, r0);
2040   __ br(j_not(cc), not_taken);
2041   branch(false, false);
2042   __ bind(not_taken);
2043   __ profile_not_taken_branch(r0);
2044 }
2045 
2046 void TemplateTable::ret() {
2047   transition(vtos, vtos);
2048   // We might be moving to a safepoint.  The thread which calls
2049   // Interpreter::notice_safepoints() will effectively flush its cache
2050   // when it makes a system call, but we need to do something to
2051   // ensure that we see the changed dispatch table.
2052   __ membar(MacroAssembler::LoadLoad);
2053 
2054   locals_index(r1);
2055   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2056   __ profile_ret(r1, r2);
2057   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2058   __ lea(rbcp, Address(rbcp, r1));
2059   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2060   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2061 }
2062 
2063 void TemplateTable::wide_ret() {
2064   transition(vtos, vtos);
2065   locals_index_wide(r1);
2066   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2067   __ profile_ret(r1, r2);
2068   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2069   __ lea(rbcp, Address(rbcp, r1));
2070   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2071   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2072 }
2073 
2074 
2075 void TemplateTable::tableswitch() {
2076   Label default_case, continue_execution;
2077   transition(itos, vtos);
2078   // align rbcp
2079   __ lea(r1, at_bcp(BytesPerInt));
2080   __ andr(r1, r1, -BytesPerInt);
2081   // load lo & hi
2082   __ ldrw(r2, Address(r1, BytesPerInt));
2083   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2084   __ rev32(r2, r2);
2085   __ rev32(r3, r3);
2086   // check against lo & hi
2087   __ cmpw(r0, r2);
2088   __ br(Assembler::LT, default_case);
2089   __ cmpw(r0, r3);
2090   __ br(Assembler::GT, default_case);
2091   // lookup dispatch offset
2092   __ subw(r0, r0, r2);
2093   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2094   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2095   __ profile_switch_case(r0, r1, r2);
2096   // continue execution
2097   __ bind(continue_execution);
2098   __ rev32(r3, r3);
2099   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2100   __ add(rbcp, rbcp, r3, ext::sxtw);
2101   __ dispatch_only(vtos, /*generate_poll*/true);
2102   // handle default
2103   __ bind(default_case);
2104   __ profile_switch_default(r0);
2105   __ ldrw(r3, Address(r1, 0));
2106   __ b(continue_execution);
2107 }
2108 
2109 void TemplateTable::lookupswitch() {
2110   transition(itos, itos);
2111   __ stop("lookupswitch bytecode should have been rewritten");
2112 }
2113 
2114 void TemplateTable::fast_linearswitch() {
2115   transition(itos, vtos);
2116   Label loop_entry, loop, found, continue_execution;
2117   // bswap r0 so we can avoid bswapping the table entries
2118   __ rev32(r0, r0);
2119   // align rbcp
2120   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2121                                     // this instruction (change offsets
2122                                     // below)
2123   __ andr(r19, r19, -BytesPerInt);
2124   // set counter
2125   __ ldrw(r1, Address(r19, BytesPerInt));
2126   __ rev32(r1, r1);
2127   __ b(loop_entry);
2128   // table search
2129   __ bind(loop);
2130   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2131   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2132   __ cmpw(r0, rscratch1);
2133   __ br(Assembler::EQ, found);
2134   __ bind(loop_entry);
2135   __ subs(r1, r1, 1);
2136   __ br(Assembler::PL, loop);
2137   // default case
2138   __ profile_switch_default(r0);
2139   __ ldrw(r3, Address(r19, 0));
2140   __ b(continue_execution);
2141   // entry found -> get offset
2142   __ bind(found);
2143   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2144   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2145   __ profile_switch_case(r1, r0, r19);
2146   // continue execution
2147   __ bind(continue_execution);
2148   __ rev32(r3, r3);
2149   __ add(rbcp, rbcp, r3, ext::sxtw);
2150   __ ldrb(rscratch1, Address(rbcp, 0));
2151   __ dispatch_only(vtos, /*generate_poll*/true);
2152 }
2153 
2154 void TemplateTable::fast_binaryswitch() {
2155   transition(itos, vtos);
2156   // Implementation using the following core algorithm:
2157   //
2158   // int binary_search(int key, LookupswitchPair* array, int n) {
2159   //   // Binary search according to "Methodik des Programmierens" by
2160   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2161   //   int i = 0;
2162   //   int j = n;
2163   //   while (i+1 < j) {
2164   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2165   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2166   //     // where a stands for the array and assuming that the (inexisting)
2167   //     // element a[n] is infinitely big.
2168   //     int h = (i + j) >> 1;
2169   //     // i < h < j
2170   //     if (key < array[h].fast_match()) {
2171   //       j = h;
2172   //     } else {
2173   //       i = h;
2174   //     }
2175   //   }
2176   //   // R: a[i] <= key < a[i+1] or Q
2177   //   // (i.e., if key is within array, i is the correct index)
2178   //   return i;
2179   // }
2180 
2181   // Register allocation
2182   const Register key   = r0; // already set (tosca)
2183   const Register array = r1;
2184   const Register i     = r2;
2185   const Register j     = r3;
2186   const Register h     = rscratch1;
2187   const Register temp  = rscratch2;
2188 
2189   // Find array start
2190   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2191                                           // get rid of this
2192                                           // instruction (change
2193                                           // offsets below)
2194   __ andr(array, array, -BytesPerInt);
2195 
2196   // Initialize i & j
2197   __ mov(i, 0);                            // i = 0;
2198   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2199 
2200   // Convert j into native byteordering
2201   __ rev32(j, j);
2202 
2203   // And start
2204   Label entry;
2205   __ b(entry);
2206 
2207   // binary search loop
2208   {
2209     Label loop;
2210     __ bind(loop);
2211     // int h = (i + j) >> 1;
2212     __ addw(h, i, j);                           // h = i + j;
2213     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2214     // if (key < array[h].fast_match()) {
2215     //   j = h;
2216     // } else {
2217     //   i = h;
2218     // }
2219     // Convert array[h].match to native byte-ordering before compare
2220     __ ldr(temp, Address(array, h, Address::lsl(3)));
2221     __ rev32(temp, temp);
2222     __ cmpw(key, temp);
2223     // j = h if (key <  array[h].fast_match())
2224     __ csel(j, h, j, Assembler::LT);
2225     // i = h if (key >= array[h].fast_match())
2226     __ csel(i, h, i, Assembler::GE);
2227     // while (i+1 < j)
2228     __ bind(entry);
2229     __ addw(h, i, 1);          // i+1
2230     __ cmpw(h, j);             // i+1 < j
2231     __ br(Assembler::LT, loop);
2232   }
2233 
2234   // end of binary search, result index is i (must check again!)
2235   Label default_case;
2236   // Convert array[i].match to native byte-ordering before compare
2237   __ ldr(temp, Address(array, i, Address::lsl(3)));
2238   __ rev32(temp, temp);
2239   __ cmpw(key, temp);
2240   __ br(Assembler::NE, default_case);
2241 
2242   // entry found -> j = offset
2243   __ add(j, array, i, ext::uxtx, 3);
2244   __ ldrw(j, Address(j, BytesPerInt));
2245   __ profile_switch_case(i, key, array);
2246   __ rev32(j, j);
2247   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2248   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2249   __ dispatch_only(vtos, /*generate_poll*/true);
2250 
2251   // default case -> j = default offset
2252   __ bind(default_case);
2253   __ profile_switch_default(i);
2254   __ ldrw(j, Address(array, -2 * BytesPerInt));
2255   __ rev32(j, j);
2256   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2257   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2258   __ dispatch_only(vtos, /*generate_poll*/true);
2259 }
2260 
2261 
2262 void TemplateTable::_return(TosState state)
2263 {
2264   transition(state, state);
2265   assert(_desc->calls_vm(),
2266          "inconsistent calls_vm information"); // call in remove_activation
2267 
2268   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2269     assert(state == vtos, "only valid state");
2270 
2271     __ ldr(c_rarg1, aaddress(0));
2272     __ load_klass(r3, c_rarg1);
2273     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2274     Label skip_register_finalizer;
2275     __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2276 
2277     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2278 
2279     __ bind(skip_register_finalizer);
2280   }
2281 
2282   // Issue a StoreStore barrier after all stores but before return
2283   // from any constructor for any class with a final field.  We don't
2284   // know if this is a finalizer, so we always do so.
2285   if (_desc->bytecode() == Bytecodes::_return)
2286     __ membar(MacroAssembler::StoreStore);
2287 
2288   // Narrow result if state is itos but result type is smaller.
2289   // Need to narrow in the return bytecode rather than in generate_return_entry
2290   // since compiled code callers expect the result to already be narrowed.
2291   if (state == itos) {
2292     __ narrow(r0);
2293   }
2294 
2295   __ remove_activation(state);
2296   __ ret(lr);
2297 }
2298 
2299 // ----------------------------------------------------------------------------
2300 // Volatile variables demand their effects be made known to all CPU's
2301 // in order.  Store buffers on most chips allow reads & writes to
2302 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2303 // without some kind of memory barrier (i.e., it's not sufficient that
2304 // the interpreter does not reorder volatile references, the hardware
2305 // also must not reorder them).
2306 //
2307 // According to the new Java Memory Model (JMM):
2308 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2309 //     writes act as aquire & release, so:
2310 // (2) A read cannot let unrelated NON-volatile memory refs that
2311 //     happen after the read float up to before the read.  It's OK for
2312 //     non-volatile memory refs that happen before the volatile read to
2313 //     float down below it.
2314 // (3) Similar a volatile write cannot let unrelated NON-volatile
2315 //     memory refs that happen BEFORE the write float down to after the
2316 //     write.  It's OK for non-volatile memory refs that happen after the
2317 //     volatile write to float up before it.
2318 //
2319 // We only put in barriers around volatile refs (they are expensive),
2320 // not _between_ memory refs (that would require us to track the
2321 // flavor of the previous memory refs).  Requirements (2) and (3)
2322 // require some barriers before volatile stores and after volatile
2323 // loads.  These nearly cover requirement (1) but miss the
2324 // volatile-store-volatile-load case.  This final case is placed after
2325 // volatile-stores although it could just as well go before
2326 // volatile-loads.
2327 
2328 void TemplateTable::resolve_cache_and_index(int byte_no,
2329                                             Register Rcache,
2330                                             Register index,
2331                                             size_t index_size) {
2332   const Register temp = r19;
2333   assert_different_registers(Rcache, index, temp);
2334 
2335   Label resolved;
2336 
2337   Bytecodes::Code code = bytecode();
2338   switch (code) {
2339   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2340   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2341   }
2342 
2343   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2344   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2345   __ cmp(temp, (int) code);  // have we resolved this bytecode?
2346   __ br(Assembler::EQ, resolved);
2347 
2348   // resolve first time through
2349   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2350   __ mov(temp, (int) code);
2351   __ call_VM(noreg, entry, temp);
2352 
2353   // Update registers with resolved info
2354   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2355   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2356   // so all clients ofthis method must be modified accordingly
2357   __ bind(resolved);
2358 }
2359 
2360 // The Rcache and index registers must be set before call
2361 // n.b unlike x86 cache already includes the index offset
2362 void TemplateTable::load_field_cp_cache_entry(Register obj,
2363                                               Register cache,
2364                                               Register index,
2365                                               Register off,
2366                                               Register flags,
2367                                               bool is_static = false) {
2368   assert_different_registers(cache, index, flags, off);
2369 
2370   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2371   // Field offset
2372   __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2373                                           ConstantPoolCacheEntry::f2_offset())));
2374   // Flags
2375   __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2376                                            ConstantPoolCacheEntry::flags_offset())));
2377 
2378   // klass overwrite register
2379   if (is_static) {
2380     __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2381                                         ConstantPoolCacheEntry::f1_offset())));
2382     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2383     __ ldr(obj, Address(obj, mirror_offset));
2384     __ resolve_oop_handle(obj);
2385   }
2386 }
2387 
2388 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2389                                                Register method,
2390                                                Register itable_index,
2391                                                Register flags,
2392                                                bool is_invokevirtual,
2393                                                bool is_invokevfinal, /*unused*/
2394                                                bool is_invokedynamic) {
2395   // setup registers
2396   const Register cache = rscratch2;
2397   const Register index = r4;
2398   assert_different_registers(method, flags);
2399   assert_different_registers(method, cache, index);
2400   assert_different_registers(itable_index, flags);
2401   assert_different_registers(itable_index, cache, index);
2402   // determine constant pool cache field offsets
2403   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2404   const int method_offset = in_bytes(
2405     ConstantPoolCache::base_offset() +
2406       (is_invokevirtual
2407        ? ConstantPoolCacheEntry::f2_offset()
2408        : ConstantPoolCacheEntry::f1_offset()));
2409   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2410                                     ConstantPoolCacheEntry::flags_offset());
2411   // access constant pool cache fields
2412   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2413                                     ConstantPoolCacheEntry::f2_offset());
2414 
2415   size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2416   resolve_cache_and_index(byte_no, cache, index, index_size);
2417   __ ldr(method, Address(cache, method_offset));
2418 
2419   if (itable_index != noreg) {
2420     __ ldr(itable_index, Address(cache, index_offset));
2421   }
2422   __ ldrw(flags, Address(cache, flags_offset));
2423 }
2424 
2425 
2426 // The registers cache and index expected to be set before call.
2427 // Correct values of the cache and index registers are preserved.
2428 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2429                                             bool is_static, bool has_tos) {
2430   // do the JVMTI work here to avoid disturbing the register state below
2431   // We use c_rarg registers here because we want to use the register used in
2432   // the call to the VM
2433   if (JvmtiExport::can_post_field_access()) {
2434     // Check to see if a field access watch has been set before we
2435     // take the time to call into the VM.
2436     Label L1;
2437     assert_different_registers(cache, index, r0);
2438     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2439     __ ldrw(r0, Address(rscratch1));
2440     __ cbzw(r0, L1);
2441 
2442     __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2443     __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2444 
2445     if (is_static) {
2446       __ mov(c_rarg1, zr); // NULL object reference
2447     } else {
2448       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2449       __ verify_oop(c_rarg1);
2450     }
2451     // c_rarg1: object pointer or NULL
2452     // c_rarg2: cache entry pointer
2453     // c_rarg3: jvalue object on the stack
2454     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2455                                        InterpreterRuntime::post_field_access),
2456                c_rarg1, c_rarg2, c_rarg3);
2457     __ get_cache_and_index_at_bcp(cache, index, 1);
2458     __ bind(L1);
2459   }
2460 }
2461 
2462 void TemplateTable::pop_and_check_object(Register r)
2463 {
2464   __ pop_ptr(r);
2465   __ null_check(r);  // for field access must check obj.
2466   __ verify_oop(r);
2467 }
2468 
2469 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2470 {
2471   const Register cache = r2;
2472   const Register index = r3;
2473   const Register obj   = r4;
2474   const Register off   = r19;
2475   const Register flags = r0;
2476   const Register raw_flags = r6;
2477   const Register bc    = r4; // uses same reg as obj, so don't mix them
2478 
2479   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2480   jvmti_post_field_access(cache, index, is_static, false);
2481   load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2482 
2483   if (!is_static) {
2484     // obj is on the stack
2485     pop_and_check_object(obj);
2486   }
2487 
2488   // 8179954: We need to make sure that the code generated for
2489   // volatile accesses forms a sequentially-consistent set of
2490   // operations when combined with STLR and LDAR.  Without a leading
2491   // membar it's possible for a simple Dekker test to fail if loads
2492   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2493   // the stores in one method and we interpret the loads in another.
2494   if (! UseBarriersForVolatile) {
2495     Label notVolatile;
2496     __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2497     __ membar(MacroAssembler::AnyAny);
2498     __ bind(notVolatile);
2499   }
2500 
2501   const Address field(obj, off);
2502 
2503   Label Done, notByte, notBool, notInt, notShort, notChar,
2504               notLong, notFloat, notObj, notDouble;
2505 
2506   // x86 uses a shift and mask or wings it with a shift plus assert
2507   // the mask is not needed. aarch64 just uses bitfield extract
2508   __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2509            ConstantPoolCacheEntry::tos_state_bits);
2510 
2511   assert(btos == 0, "change code, btos != 0");
2512   __ cbnz(flags, notByte);
2513 
2514   // Don't rewrite getstatic, only getfield
2515   if (is_static) rc = may_not_rewrite;
2516 
2517   // btos
2518   __ load_signed_byte(r0, field);
2519   __ push(btos);
2520   // Rewrite bytecode to be faster
2521   if (rc == may_rewrite) {
2522     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2523   }
2524   __ b(Done);
2525 
2526   __ bind(notByte);
2527   __ cmp(flags, ztos);
2528   __ br(Assembler::NE, notBool);
2529 
2530   // ztos (same code as btos)
2531   __ ldrsb(r0, field);
2532   __ push(ztos);
2533   // Rewrite bytecode to be faster
2534   if (rc == may_rewrite) {
2535     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2536     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2537   }
2538   __ b(Done);
2539 
2540   __ bind(notBool);
2541   __ cmp(flags, atos);
2542   __ br(Assembler::NE, notObj);
2543   // atos
2544   do_oop_load(_masm, field, r0, IN_HEAP);
2545   __ push(atos);
2546   if (rc == may_rewrite) {
2547     patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2548   }
2549   __ b(Done);
2550 
2551   __ bind(notObj);
2552   __ cmp(flags, itos);
2553   __ br(Assembler::NE, notInt);
2554   // itos
2555   __ ldrw(r0, field);
2556   __ push(itos);
2557   // Rewrite bytecode to be faster
2558   if (rc == may_rewrite) {
2559     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2560   }
2561   __ b(Done);
2562 
2563   __ bind(notInt);
2564   __ cmp(flags, ctos);
2565   __ br(Assembler::NE, notChar);
2566   // ctos
2567   __ load_unsigned_short(r0, field);
2568   __ push(ctos);
2569   // Rewrite bytecode to be faster
2570   if (rc == may_rewrite) {
2571     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2572   }
2573   __ b(Done);
2574 
2575   __ bind(notChar);
2576   __ cmp(flags, stos);
2577   __ br(Assembler::NE, notShort);
2578   // stos
2579   __ load_signed_short(r0, field);
2580   __ push(stos);
2581   // Rewrite bytecode to be faster
2582   if (rc == may_rewrite) {
2583     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2584   }
2585   __ b(Done);
2586 
2587   __ bind(notShort);
2588   __ cmp(flags, ltos);
2589   __ br(Assembler::NE, notLong);
2590   // ltos
2591   __ ldr(r0, field);
2592   __ push(ltos);
2593   // Rewrite bytecode to be faster
2594   if (rc == may_rewrite) {
2595     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2596   }
2597   __ b(Done);
2598 
2599   __ bind(notLong);
2600   __ cmp(flags, ftos);
2601   __ br(Assembler::NE, notFloat);
2602   // ftos
2603   __ ldrs(v0, field);
2604   __ push(ftos);
2605   // Rewrite bytecode to be faster
2606   if (rc == may_rewrite) {
2607     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2608   }
2609   __ b(Done);
2610 
2611   __ bind(notFloat);
2612 #ifdef ASSERT
2613   __ cmp(flags, dtos);
2614   __ br(Assembler::NE, notDouble);
2615 #endif
2616   // dtos
2617   __ ldrd(v0, field);
2618   __ push(dtos);
2619   // Rewrite bytecode to be faster
2620   if (rc == may_rewrite) {
2621     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2622   }
2623 #ifdef ASSERT
2624   __ b(Done);
2625 
2626   __ bind(notDouble);
2627   __ stop("Bad state");
2628 #endif
2629 
2630   __ bind(Done);
2631 
2632   Label notVolatile;
2633   __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2634   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2635   __ bind(notVolatile);
2636 }
2637 
2638 
2639 void TemplateTable::getfield(int byte_no)
2640 {
2641   getfield_or_static(byte_no, false);
2642 }
2643 
2644 void TemplateTable::nofast_getfield(int byte_no) {
2645   getfield_or_static(byte_no, false, may_not_rewrite);
2646 }
2647 
2648 void TemplateTable::getstatic(int byte_no)
2649 {
2650   getfield_or_static(byte_no, true);
2651 }
2652 
2653 // The registers cache and index expected to be set before call.
2654 // The function may destroy various registers, just not the cache and index registers.
2655 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2656   transition(vtos, vtos);
2657 
2658   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2659 
2660   if (JvmtiExport::can_post_field_modification()) {
2661     // Check to see if a field modification watch has been set before
2662     // we take the time to call into the VM.
2663     Label L1;
2664     assert_different_registers(cache, index, r0);
2665     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2666     __ ldrw(r0, Address(rscratch1));
2667     __ cbz(r0, L1);
2668 
2669     __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2670 
2671     if (is_static) {
2672       // Life is simple.  Null out the object pointer.
2673       __ mov(c_rarg1, zr);
2674     } else {
2675       // Life is harder. The stack holds the value on top, followed by
2676       // the object.  We don't know the size of the value, though; it
2677       // could be one or two words depending on its type. As a result,
2678       // we must find the type to determine where the object is.
2679       __ ldrw(c_rarg3, Address(c_rarg2,
2680                                in_bytes(cp_base_offset +
2681                                         ConstantPoolCacheEntry::flags_offset())));
2682       __ lsr(c_rarg3, c_rarg3,
2683              ConstantPoolCacheEntry::tos_state_shift);
2684       ConstantPoolCacheEntry::verify_tos_state_shift();
2685       Label nope2, done, ok;
2686       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2687       __ cmpw(c_rarg3, ltos);
2688       __ br(Assembler::EQ, ok);
2689       __ cmpw(c_rarg3, dtos);
2690       __ br(Assembler::NE, nope2);
2691       __ bind(ok);
2692       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2693       __ bind(nope2);
2694     }
2695     // cache entry pointer
2696     __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2697     // object (tos)
2698     __ mov(c_rarg3, esp);
2699     // c_rarg1: object pointer set up above (NULL if static)
2700     // c_rarg2: cache entry pointer
2701     // c_rarg3: jvalue object on the stack
2702     __ call_VM(noreg,
2703                CAST_FROM_FN_PTR(address,
2704                                 InterpreterRuntime::post_field_modification),
2705                c_rarg1, c_rarg2, c_rarg3);
2706     __ get_cache_and_index_at_bcp(cache, index, 1);
2707     __ bind(L1);
2708   }
2709 }
2710 
2711 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2712   transition(vtos, vtos);
2713 
2714   const Register cache = r2;
2715   const Register index = r3;
2716   const Register obj   = r2;
2717   const Register off   = r19;
2718   const Register flags = r0;
2719   const Register bc    = r4;
2720 
2721   resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2722   jvmti_post_field_mod(cache, index, is_static);
2723   load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2724 
2725   Label Done;
2726   __ mov(r5, flags);
2727 
2728   {
2729     Label notVolatile;
2730     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2731     __ membar(MacroAssembler::StoreStore);
2732     __ bind(notVolatile);
2733   }
2734 
2735   // field address
2736   const Address field(obj, off);
2737 
2738   Label notByte, notBool, notInt, notShort, notChar,
2739         notLong, notFloat, notObj, notDouble;
2740 
2741   // x86 uses a shift and mask or wings it with a shift plus assert
2742   // the mask is not needed. aarch64 just uses bitfield extract
2743   __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
2744 
2745   assert(btos == 0, "change code, btos != 0");
2746   __ cbnz(flags, notByte);
2747 
2748   // Don't rewrite putstatic, only putfield
2749   if (is_static) rc = may_not_rewrite;
2750 
2751   // btos
2752   {
2753     __ pop(btos);
2754     if (!is_static) pop_and_check_object(obj);
2755     __ strb(r0, field);
2756     if (rc == may_rewrite) {
2757       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2758     }
2759     __ b(Done);
2760   }
2761 
2762   __ bind(notByte);
2763   __ cmp(flags, ztos);
2764   __ br(Assembler::NE, notBool);
2765 
2766   // ztos
2767   {
2768     __ pop(ztos);
2769     if (!is_static) pop_and_check_object(obj);
2770     __ andw(r0, r0, 0x1);
2771     __ strb(r0, field);
2772     if (rc == may_rewrite) {
2773       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2774     }
2775     __ b(Done);
2776   }
2777 
2778   __ bind(notBool);
2779   __ cmp(flags, atos);
2780   __ br(Assembler::NE, notObj);
2781 
2782   // atos
2783   {
2784     __ pop(atos);
2785     if (!is_static) pop_and_check_object(obj);
2786     // Store into the field
2787     do_oop_store(_masm, field, r0, IN_HEAP);
2788     if (rc == may_rewrite) {
2789       patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2790     }
2791     __ b(Done);
2792   }
2793 
2794   __ bind(notObj);
2795   __ cmp(flags, itos);
2796   __ br(Assembler::NE, notInt);
2797 
2798   // itos
2799   {
2800     __ pop(itos);
2801     if (!is_static) pop_and_check_object(obj);
2802     __ strw(r0, field);
2803     if (rc == may_rewrite) {
2804       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2805     }
2806     __ b(Done);
2807   }
2808 
2809   __ bind(notInt);
2810   __ cmp(flags, ctos);
2811   __ br(Assembler::NE, notChar);
2812 
2813   // ctos
2814   {
2815     __ pop(ctos);
2816     if (!is_static) pop_and_check_object(obj);
2817     __ strh(r0, field);
2818     if (rc == may_rewrite) {
2819       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2820     }
2821     __ b(Done);
2822   }
2823 
2824   __ bind(notChar);
2825   __ cmp(flags, stos);
2826   __ br(Assembler::NE, notShort);
2827 
2828   // stos
2829   {
2830     __ pop(stos);
2831     if (!is_static) pop_and_check_object(obj);
2832     __ strh(r0, field);
2833     if (rc == may_rewrite) {
2834       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2835     }
2836     __ b(Done);
2837   }
2838 
2839   __ bind(notShort);
2840   __ cmp(flags, ltos);
2841   __ br(Assembler::NE, notLong);
2842 
2843   // ltos
2844   {
2845     __ pop(ltos);
2846     if (!is_static) pop_and_check_object(obj);
2847     __ str(r0, field);
2848     if (rc == may_rewrite) {
2849       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2850     }
2851     __ b(Done);
2852   }
2853 
2854   __ bind(notLong);
2855   __ cmp(flags, ftos);
2856   __ br(Assembler::NE, notFloat);
2857 
2858   // ftos
2859   {
2860     __ pop(ftos);
2861     if (!is_static) pop_and_check_object(obj);
2862     __ strs(v0, field);
2863     if (rc == may_rewrite) {
2864       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2865     }
2866     __ b(Done);
2867   }
2868 
2869   __ bind(notFloat);
2870 #ifdef ASSERT
2871   __ cmp(flags, dtos);
2872   __ br(Assembler::NE, notDouble);
2873 #endif
2874 
2875   // dtos
2876   {
2877     __ pop(dtos);
2878     if (!is_static) pop_and_check_object(obj);
2879     __ strd(v0, field);
2880     if (rc == may_rewrite) {
2881       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2882     }
2883   }
2884 
2885 #ifdef ASSERT
2886   __ b(Done);
2887 
2888   __ bind(notDouble);
2889   __ stop("Bad state");
2890 #endif
2891 
2892   __ bind(Done);
2893 
2894   {
2895     Label notVolatile;
2896     __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2897     __ membar(MacroAssembler::StoreLoad);
2898     __ bind(notVolatile);
2899   }
2900 }
2901 
2902 void TemplateTable::putfield(int byte_no)
2903 {
2904   putfield_or_static(byte_no, false);
2905 }
2906 
2907 void TemplateTable::nofast_putfield(int byte_no) {
2908   putfield_or_static(byte_no, false, may_not_rewrite);
2909 }
2910 
2911 void TemplateTable::putstatic(int byte_no) {
2912   putfield_or_static(byte_no, true);
2913 }
2914 
2915 void TemplateTable::jvmti_post_fast_field_mod()
2916 {
2917   if (JvmtiExport::can_post_field_modification()) {
2918     // Check to see if a field modification watch has been set before
2919     // we take the time to call into the VM.
2920     Label L2;
2921     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2922     __ ldrw(c_rarg3, Address(rscratch1));
2923     __ cbzw(c_rarg3, L2);
2924     __ pop_ptr(r19);                  // copy the object pointer from tos
2925     __ verify_oop(r19);
2926     __ push_ptr(r19);                 // put the object pointer back on tos
2927     // Save tos values before call_VM() clobbers them. Since we have
2928     // to do it for every data type, we use the saved values as the
2929     // jvalue object.
2930     switch (bytecode()) {          // load values into the jvalue object
2931     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2932     case Bytecodes::_fast_bputfield: // fall through
2933     case Bytecodes::_fast_zputfield: // fall through
2934     case Bytecodes::_fast_sputfield: // fall through
2935     case Bytecodes::_fast_cputfield: // fall through
2936     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2937     case Bytecodes::_fast_dputfield: __ push_d(); break;
2938     case Bytecodes::_fast_fputfield: __ push_f(); break;
2939     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2940 
2941     default:
2942       ShouldNotReachHere();
2943     }
2944     __ mov(c_rarg3, esp);             // points to jvalue on the stack
2945     // access constant pool cache entry
2946     __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2947     __ verify_oop(r19);
2948     // r19: object pointer copied above
2949     // c_rarg2: cache entry pointer
2950     // c_rarg3: jvalue object on the stack
2951     __ call_VM(noreg,
2952                CAST_FROM_FN_PTR(address,
2953                                 InterpreterRuntime::post_field_modification),
2954                r19, c_rarg2, c_rarg3);
2955 
2956     switch (bytecode()) {             // restore tos values
2957     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2958     case Bytecodes::_fast_bputfield: // fall through
2959     case Bytecodes::_fast_zputfield: // fall through
2960     case Bytecodes::_fast_sputfield: // fall through
2961     case Bytecodes::_fast_cputfield: // fall through
2962     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2963     case Bytecodes::_fast_dputfield: __ pop_d(); break;
2964     case Bytecodes::_fast_fputfield: __ pop_f(); break;
2965     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2966     }
2967     __ bind(L2);
2968   }
2969 }
2970 
2971 void TemplateTable::fast_storefield(TosState state)
2972 {
2973   transition(state, vtos);
2974 
2975   ByteSize base = ConstantPoolCache::base_offset();
2976 
2977   jvmti_post_fast_field_mod();
2978 
2979   // access constant pool cache
2980   __ get_cache_and_index_at_bcp(r2, r1, 1);
2981 
2982   // test for volatile with r3
2983   __ ldrw(r3, Address(r2, in_bytes(base +
2984                                    ConstantPoolCacheEntry::flags_offset())));
2985 
2986   // replace index with field offset from cache entry
2987   __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2988 
2989   {
2990     Label notVolatile;
2991     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2992     __ membar(MacroAssembler::StoreStore);
2993     __ bind(notVolatile);
2994   }
2995 
2996   Label notVolatile;
2997 
2998   // Get object from stack
2999   pop_and_check_object(r2);
3000 
3001   // field address
3002   const Address field(r2, r1);
3003 
3004   // access field
3005   switch (bytecode()) {
3006   case Bytecodes::_fast_aputfield:
3007     do_oop_store(_masm, field, r0, IN_HEAP);
3008     break;
3009   case Bytecodes::_fast_lputfield:
3010     __ str(r0, field);
3011     break;
3012   case Bytecodes::_fast_iputfield:
3013     __ strw(r0, field);
3014     break;
3015   case Bytecodes::_fast_zputfield:
3016     __ andw(r0, r0, 0x1);  // boolean is true if LSB is 1
3017     // fall through to bputfield
3018   case Bytecodes::_fast_bputfield:
3019     __ strb(r0, field);
3020     break;
3021   case Bytecodes::_fast_sputfield:
3022     // fall through
3023   case Bytecodes::_fast_cputfield:
3024     __ strh(r0, field);
3025     break;
3026   case Bytecodes::_fast_fputfield:
3027     __ strs(v0, field);
3028     break;
3029   case Bytecodes::_fast_dputfield:
3030     __ strd(v0, field);
3031     break;
3032   default:
3033     ShouldNotReachHere();
3034   }
3035 
3036   {
3037     Label notVolatile;
3038     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3039     __ membar(MacroAssembler::StoreLoad);
3040     __ bind(notVolatile);
3041   }
3042 }
3043 
3044 
3045 void TemplateTable::fast_accessfield(TosState state)
3046 {
3047   transition(atos, state);
3048   // Do the JVMTI work here to avoid disturbing the register state below
3049   if (JvmtiExport::can_post_field_access()) {
3050     // Check to see if a field access watch has been set before we
3051     // take the time to call into the VM.
3052     Label L1;
3053     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3054     __ ldrw(r2, Address(rscratch1));
3055     __ cbzw(r2, L1);
3056     // access constant pool cache entry
3057     __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
3058     __ verify_oop(r0);
3059     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
3060     __ mov(c_rarg1, r0);
3061     // c_rarg1: object pointer copied above
3062     // c_rarg2: cache entry pointer
3063     __ call_VM(noreg,
3064                CAST_FROM_FN_PTR(address,
3065                                 InterpreterRuntime::post_field_access),
3066                c_rarg1, c_rarg2);
3067     __ pop_ptr(r0); // restore object pointer
3068     __ bind(L1);
3069   }
3070 
3071   // access constant pool cache
3072   __ get_cache_and_index_at_bcp(r2, r1, 1);
3073   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3074                                   ConstantPoolCacheEntry::f2_offset())));
3075   __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3076                                    ConstantPoolCacheEntry::flags_offset())));
3077 
3078   // r0: object
3079   __ verify_oop(r0);
3080   __ null_check(r0);
3081   const Address field(r0, r1);
3082 
3083   // 8179954: We need to make sure that the code generated for
3084   // volatile accesses forms a sequentially-consistent set of
3085   // operations when combined with STLR and LDAR.  Without a leading
3086   // membar it's possible for a simple Dekker test to fail if loads
3087   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3088   // the stores in one method and we interpret the loads in another.
3089   if (! UseBarriersForVolatile) {
3090     Label notVolatile;
3091     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3092     __ membar(MacroAssembler::AnyAny);
3093     __ bind(notVolatile);
3094   }
3095 
3096   // access field
3097   switch (bytecode()) {
3098   case Bytecodes::_fast_agetfield:
3099     do_oop_load(_masm, field, r0, IN_HEAP);
3100     __ verify_oop(r0);
3101     break;
3102   case Bytecodes::_fast_lgetfield:
3103     __ ldr(r0, field);
3104     break;
3105   case Bytecodes::_fast_igetfield:
3106     __ ldrw(r0, field);
3107     break;
3108   case Bytecodes::_fast_bgetfield:
3109     __ load_signed_byte(r0, field);
3110     break;
3111   case Bytecodes::_fast_sgetfield:
3112     __ load_signed_short(r0, field);
3113     break;
3114   case Bytecodes::_fast_cgetfield:
3115     __ load_unsigned_short(r0, field);
3116     break;
3117   case Bytecodes::_fast_fgetfield:
3118     __ ldrs(v0, field);
3119     break;
3120   case Bytecodes::_fast_dgetfield:
3121     __ ldrd(v0, field);
3122     break;
3123   default:
3124     ShouldNotReachHere();
3125   }
3126   {
3127     Label notVolatile;
3128     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3129     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3130     __ bind(notVolatile);
3131   }
3132 }
3133 
3134 void TemplateTable::fast_xaccess(TosState state)
3135 {
3136   transition(vtos, state);
3137 
3138   // get receiver
3139   __ ldr(r0, aaddress(0));
3140   // access constant pool cache
3141   __ get_cache_and_index_at_bcp(r2, r3, 2);
3142   __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3143                                   ConstantPoolCacheEntry::f2_offset())));
3144 
3145   // 8179954: We need to make sure that the code generated for
3146   // volatile accesses forms a sequentially-consistent set of
3147   // operations when combined with STLR and LDAR.  Without a leading
3148   // membar it's possible for a simple Dekker test to fail if loads
3149   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3150   // the stores in one method and we interpret the loads in another.
3151   if (! UseBarriersForVolatile) {
3152     Label notVolatile;
3153     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3154                                      ConstantPoolCacheEntry::flags_offset())));
3155     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3156     __ membar(MacroAssembler::AnyAny);
3157     __ bind(notVolatile);
3158   }
3159 
3160   // make sure exception is reported in correct bcp range (getfield is
3161   // next instruction)
3162   __ increment(rbcp);
3163   __ null_check(r0);
3164   switch (state) {
3165   case itos:
3166     __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3167     break;
3168   case atos:
3169     do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3170     __ verify_oop(r0);
3171     break;
3172   case ftos:
3173     __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3174     break;
3175   default:
3176     ShouldNotReachHere();
3177   }
3178 
3179   {
3180     Label notVolatile;
3181     __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3182                                      ConstantPoolCacheEntry::flags_offset())));
3183     __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3184     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3185     __ bind(notVolatile);
3186   }
3187 
3188   __ decrement(rbcp);
3189 }
3190 
3191 
3192 
3193 //-----------------------------------------------------------------------------
3194 // Calls
3195 
3196 void TemplateTable::count_calls(Register method, Register temp)
3197 {
3198   __ call_Unimplemented();
3199 }
3200 
3201 void TemplateTable::prepare_invoke(int byte_no,
3202                                    Register method, // linked method (or i-klass)
3203                                    Register index,  // itable index, MethodType, etc.
3204                                    Register recv,   // if caller wants to see it
3205                                    Register flags   // if caller wants to test it
3206                                    ) {
3207   // determine flags
3208   Bytecodes::Code code = bytecode();
3209   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3210   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3211   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3212   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3213   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3214   const bool load_receiver       = (recv  != noreg);
3215   const bool save_flags          = (flags != noreg);
3216   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3217   assert(save_flags    == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3218   assert(flags == noreg || flags == r3, "");
3219   assert(recv  == noreg || recv  == r2, "");
3220 
3221   // setup registers & access constant pool cache
3222   if (recv  == noreg)  recv  = r2;
3223   if (flags == noreg)  flags = r3;
3224   assert_different_registers(method, index, recv, flags);
3225 
3226   // save 'interpreter return address'
3227   __ save_bcp();
3228 
3229   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3230 
3231   // maybe push appendix to arguments (just before return address)
3232   if (is_invokedynamic || is_invokehandle) {
3233     Label L_no_push;
3234     __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3235     // Push the appendix as a trailing parameter.
3236     // This must be done before we get the receiver,
3237     // since the parameter_size includes it.
3238     __ push(r19);
3239     __ mov(r19, index);
3240     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3241     __ load_resolved_reference_at_index(index, r19);
3242     __ pop(r19);
3243     __ push(index);  // push appendix (MethodType, CallSite, etc.)
3244     __ bind(L_no_push);
3245   }
3246 
3247   // load receiver if needed (note: no return address pushed yet)
3248   if (load_receiver) {
3249     __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3250     // FIXME -- is this actually correct? looks like it should be 2
3251     // const int no_return_pc_pushed_yet = -1;  // argument slot correction before we push return address
3252     // const int receiver_is_at_end      = -1;  // back off one slot to get receiver
3253     // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3254     // __ movptr(recv, recv_addr);
3255     __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3256     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3257     __ verify_oop(recv);
3258   }
3259 
3260   // compute return type
3261   // x86 uses a shift and mask or wings it with a shift plus assert
3262   // the mask is not needed. aarch64 just uses bitfield extract
3263   __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift,  ConstantPoolCacheEntry::tos_state_bits);
3264   // load return address
3265   {
3266     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3267     __ mov(rscratch1, table_addr);
3268     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3269   }
3270 }
3271 
3272 
3273 void TemplateTable::invokevirtual_helper(Register index,
3274                                          Register recv,
3275                                          Register flags)
3276 {
3277   // Uses temporary registers r0, r3
3278   assert_different_registers(index, recv, r0, r3);
3279   // Test for an invoke of a final method
3280   Label notFinal;
3281   __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3282 
3283   const Register method = index;  // method must be rmethod
3284   assert(method == rmethod,
3285          "methodOop must be rmethod for interpreter calling convention");
3286 
3287   // do the call - the index is actually the method to call
3288   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3289 
3290   // It's final, need a null check here!
3291   __ null_check(recv);
3292 
3293   // profile this call
3294   __ profile_final_call(r0);
3295   __ profile_arguments_type(r0, method, r4, true);
3296 
3297   __ jump_from_interpreted(method, r0);
3298 
3299   __ bind(notFinal);
3300 
3301   // get receiver klass
3302   __ null_check(recv, oopDesc::klass_offset_in_bytes());
3303   __ load_klass(r0, recv);
3304 
3305   // profile this call
3306   __ profile_virtual_call(r0, rlocals, r3);
3307 
3308   // get target methodOop & entry point
3309   __ lookup_virtual_method(r0, index, method);
3310   __ profile_arguments_type(r3, method, r4, true);
3311   // FIXME -- this looks completely redundant. is it?
3312   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3313   __ jump_from_interpreted(method, r3);
3314 }
3315 
3316 void TemplateTable::invokevirtual(int byte_no)
3317 {
3318   transition(vtos, vtos);
3319   assert(byte_no == f2_byte, "use this argument");
3320 
3321   prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3322 
3323   // rmethod: index (actually a Method*)
3324   // r2: receiver
3325   // r3: flags
3326 
3327   invokevirtual_helper(rmethod, r2, r3);
3328 }
3329 
3330 void TemplateTable::invokespecial(int byte_no)
3331 {
3332   transition(vtos, vtos);
3333   assert(byte_no == f1_byte, "use this argument");
3334 
3335   prepare_invoke(byte_no, rmethod, noreg,  // get f1 Method*
3336                  r2);  // get receiver also for null check
3337   __ verify_oop(r2);
3338   __ null_check(r2);
3339   // do the call
3340   __ profile_call(r0);
3341   __ profile_arguments_type(r0, rmethod, rbcp, false);
3342   __ jump_from_interpreted(rmethod, r0);
3343 }
3344 
3345 void TemplateTable::invokestatic(int byte_no)
3346 {
3347   transition(vtos, vtos);
3348   assert(byte_no == f1_byte, "use this argument");
3349 
3350   prepare_invoke(byte_no, rmethod);  // get f1 Method*
3351   // do the call
3352   __ profile_call(r0);
3353   __ profile_arguments_type(r0, rmethod, r4, false);
3354   __ jump_from_interpreted(rmethod, r0);
3355 }
3356 
3357 void TemplateTable::fast_invokevfinal(int byte_no)
3358 {
3359   __ call_Unimplemented();
3360 }
3361 
3362 void TemplateTable::invokeinterface(int byte_no) {
3363   transition(vtos, vtos);
3364   assert(byte_no == f1_byte, "use this argument");
3365 
3366   prepare_invoke(byte_no, r0, rmethod,  // get f1 Klass*, f2 Method*
3367                  r2, r3); // recv, flags
3368 
3369   // r0: interface klass (from f1)
3370   // rmethod: method (from f2)
3371   // r2: receiver
3372   // r3: flags
3373 
3374   // Special case of invokeinterface called for virtual method of
3375   // java.lang.Object.  See cpCacheOop.cpp for details.
3376   // This code isn't produced by javac, but could be produced by
3377   // another compliant java compiler.
3378   Label notMethod;
3379   __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3380 
3381   invokevirtual_helper(rmethod, r2, r3);
3382   __ bind(notMethod);
3383 
3384   // Get receiver klass into r3 - also a null check
3385   __ restore_locals();
3386   __ null_check(r2, oopDesc::klass_offset_in_bytes());
3387   __ load_klass(r3, r2);
3388 
3389   Label no_such_interface, no_such_method;
3390 
3391   // Preserve method for throw_AbstractMethodErrorVerbose.
3392   __ mov(r16, rmethod);
3393   // Receiver subtype check against REFC.
3394   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3395   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3396                              r3, r0, noreg,
3397                              // outputs: scan temp. reg, scan temp. reg
3398                              rscratch2, r13,
3399                              no_such_interface,
3400                              /*return_method=*/false);
3401 
3402   // profile this call
3403   __ profile_virtual_call(r3, r13, r19);
3404 
3405   // Get declaring interface class from method, and itable index
3406   __ ldr(r0, Address(rmethod, Method::const_offset()));
3407   __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
3408   __ ldr(r0, Address(r0, ConstantPool::pool_holder_offset_in_bytes()));
3409   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3410   __ subw(rmethod, rmethod, Method::itable_index_max);
3411   __ negw(rmethod, rmethod);
3412 
3413   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3414   __ mov(rlocals, r3);
3415   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3416                              rlocals, r0, rmethod,
3417                              // outputs: method, scan temp. reg
3418                              rmethod, r13,
3419                              no_such_interface);
3420 
3421   // rmethod,: methodOop to call
3422   // r2: receiver
3423   // Check for abstract method error
3424   // Note: This should be done more efficiently via a throw_abstract_method_error
3425   //       interpreter entry point and a conditional jump to it in case of a null
3426   //       method.
3427   __ cbz(rmethod, no_such_method);
3428 
3429   __ profile_arguments_type(r3, rmethod, r13, true);
3430 
3431   // do the call
3432   // r2: receiver
3433   // rmethod,: methodOop
3434   __ jump_from_interpreted(rmethod, r3);
3435   __ should_not_reach_here();
3436 
3437   // exception handling code follows...
3438   // note: must restore interpreter registers to canonical
3439   //       state for exception handling to work correctly!
3440 
3441   __ bind(no_such_method);
3442   // throw exception
3443   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3444   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3445   // Pass arguments for generating a verbose error message.
3446   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3447   // the call_VM checks for exception, so we should never return here.
3448   __ should_not_reach_here();
3449 
3450   __ bind(no_such_interface);
3451   // throw exception
3452   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3453   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3454   // Pass arguments for generating a verbose error message.
3455   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3456                    InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3457   // the call_VM checks for exception, so we should never return here.
3458   __ should_not_reach_here();
3459   return;
3460 }
3461 
3462 void TemplateTable::invokehandle(int byte_no) {
3463   transition(vtos, vtos);
3464   assert(byte_no == f1_byte, "use this argument");
3465 
3466   prepare_invoke(byte_no, rmethod, r0, r2);
3467   __ verify_method_ptr(r2);
3468   __ verify_oop(r2);
3469   __ null_check(r2);
3470 
3471   // FIXME: profile the LambdaForm also
3472 
3473   // r13 is safe to use here as a scratch reg because it is about to
3474   // be clobbered by jump_from_interpreted().
3475   __ profile_final_call(r13);
3476   __ profile_arguments_type(r13, rmethod, r4, true);
3477 
3478   __ jump_from_interpreted(rmethod, r0);
3479 }
3480 
3481 void TemplateTable::invokedynamic(int byte_no) {
3482   transition(vtos, vtos);
3483   assert(byte_no == f1_byte, "use this argument");
3484 
3485   prepare_invoke(byte_no, rmethod, r0);
3486 
3487   // r0: CallSite object (from cpool->resolved_references[])
3488   // rmethod: MH.linkToCallSite method (from f2)
3489 
3490   // Note:  r0_callsite is already pushed by prepare_invoke
3491 
3492   // %%% should make a type profile for any invokedynamic that takes a ref argument
3493   // profile this call
3494   __ profile_call(rbcp);
3495   __ profile_arguments_type(r3, rmethod, r13, false);
3496 
3497   __ verify_oop(r0);
3498 
3499   __ jump_from_interpreted(rmethod, r0);
3500 }
3501 
3502 
3503 //-----------------------------------------------------------------------------
3504 // Allocation
3505 
3506 void TemplateTable::_new() {
3507   transition(vtos, atos);
3508 
3509   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3510   Label slow_case;
3511   Label done;
3512   Label initialize_header;
3513   Label initialize_object; // including clearing the fields
3514 
3515   __ get_cpool_and_tags(r4, r0);
3516   // Make sure the class we're about to instantiate has been resolved.
3517   // This is done before loading InstanceKlass to be consistent with the order
3518   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3519   const int tags_offset = Array<u1>::base_offset_in_bytes();
3520   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3521   __ lea(rscratch1, Address(rscratch1, tags_offset));
3522   __ ldarb(rscratch1, rscratch1);
3523   __ cmp(rscratch1, JVM_CONSTANT_Class);
3524   __ br(Assembler::NE, slow_case);
3525 
3526   // get InstanceKlass
3527   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3528 
3529   // make sure klass is initialized & doesn't have finalizer
3530   // make sure klass is fully initialized
3531   __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3532   __ cmp(rscratch1, InstanceKlass::fully_initialized);
3533   __ br(Assembler::NE, slow_case);
3534 
3535   // get instance_size in InstanceKlass (scaled to a count of bytes)
3536   __ ldrw(r3,
3537           Address(r4,
3538                   Klass::layout_helper_offset()));
3539   // test to see if it has a finalizer or is malformed in some way
3540   __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3541 
3542   // Allocate the instance:
3543   //  If TLAB is enabled:
3544   //    Try to allocate in the TLAB.
3545   //    If fails, go to the slow path.
3546   //  Else If inline contiguous allocations are enabled:
3547   //    Try to allocate in eden.
3548   //    If fails due to heap end, go to slow path.
3549   //
3550   //  If TLAB is enabled OR inline contiguous is enabled:
3551   //    Initialize the allocation.
3552   //    Exit.
3553   //
3554   //  Go to slow path.
3555   const bool allow_shared_alloc =
3556     Universe::heap()->supports_inline_contig_alloc();
3557 
3558   if (UseTLAB) {
3559     __ tlab_allocate(r0, r3, 0, noreg, r1, slow_case);
3560 
3561     if (ZeroTLAB) {
3562       // the fields have been already cleared
3563       __ b(initialize_header);
3564     } else {
3565       // initialize both the header and fields
3566       __ b(initialize_object);
3567     }
3568   } else {
3569     // Allocation in the shared Eden, if allowed.
3570     //
3571     // r3: instance size in bytes
3572     if (allow_shared_alloc) {
3573       __ eden_allocate(r0, r3, 0, r10, slow_case);
3574       __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3575     }
3576   }
3577 
3578   // If UseTLAB or allow_shared_alloc are true, the object is created above and
3579   // there is an initialize need. Otherwise, skip and go to the slow path.
3580   if (UseTLAB || allow_shared_alloc) {
3581     // The object is initialized before the header.  If the object size is
3582     // zero, go directly to the header initialization.
3583     __ bind(initialize_object);
3584     __ sub(r3, r3, sizeof(oopDesc));
3585     __ cbz(r3, initialize_header);
3586 
3587     // Initialize object fields
3588     {
3589       __ add(r2, r0, sizeof(oopDesc));
3590       Label loop;
3591       __ bind(loop);
3592       __ str(zr, Address(__ post(r2, BytesPerLong)));
3593       __ sub(r3, r3, BytesPerLong);
3594       __ cbnz(r3, loop);
3595     }
3596 
3597     // initialize object header only.
3598     __ bind(initialize_header);
3599     if (UseBiasedLocking) {
3600       __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3601     } else {
3602       __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3603     }
3604     __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3605     __ store_klass_gap(r0, zr);  // zero klass gap for compressed oops
3606     __ store_klass(r0, r4);      // store klass last
3607 
3608     {
3609       SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3610       // Trigger dtrace event for fastpath
3611       __ push(atos); // save the return value
3612       __ call_VM_leaf(
3613            CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3614       __ pop(atos); // restore the return value
3615 
3616     }
3617     __ b(done);
3618   }
3619 
3620   // slow case
3621   __ bind(slow_case);
3622   __ get_constant_pool(c_rarg1);
3623   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3624   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3625   __ verify_oop(r0);
3626 
3627   // continue
3628   __ bind(done);
3629   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3630   __ membar(Assembler::StoreStore);
3631 }
3632 
3633 void TemplateTable::newarray() {
3634   transition(itos, atos);
3635   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3636   __ mov(c_rarg2, r0);
3637   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3638           c_rarg1, c_rarg2);
3639   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3640   __ membar(Assembler::StoreStore);
3641 }
3642 
3643 void TemplateTable::anewarray() {
3644   transition(itos, atos);
3645   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3646   __ get_constant_pool(c_rarg1);
3647   __ mov(c_rarg3, r0);
3648   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3649           c_rarg1, c_rarg2, c_rarg3);
3650   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3651   __ membar(Assembler::StoreStore);
3652 }
3653 
3654 void TemplateTable::arraylength() {
3655   transition(atos, itos);
3656   __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3657   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3658 }
3659 
3660 void TemplateTable::checkcast()
3661 {
3662   transition(atos, atos);
3663   Label done, is_null, ok_is_subtype, quicked, resolved;
3664   __ cbz(r0, is_null);
3665 
3666   // Get cpool & tags index
3667   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3668   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3669   // See if bytecode has already been quicked
3670   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3671   __ lea(r1, Address(rscratch1, r19));
3672   __ ldarb(r1, r1);
3673   __ cmp(r1, JVM_CONSTANT_Class);
3674   __ br(Assembler::EQ, quicked);
3675 
3676   __ push(atos); // save receiver for result, and for GC
3677   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3678   // vm_result_2 has metadata result
3679   __ get_vm_result_2(r0, rthread);
3680   __ pop(r3); // restore receiver
3681   __ b(resolved);
3682 
3683   // Get superklass in r0 and subklass in r3
3684   __ bind(quicked);
3685   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3686   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3687 
3688   __ bind(resolved);
3689   __ load_klass(r19, r3);
3690 
3691   // Generate subtype check.  Blows r2, r5.  Object in r3.
3692   // Superklass in r0.  Subklass in r19.
3693   __ gen_subtype_check(r19, ok_is_subtype);
3694 
3695   // Come here on failure
3696   __ push(r3);
3697   // object is at TOS
3698   __ b(Interpreter::_throw_ClassCastException_entry);
3699 
3700   // Come here on success
3701   __ bind(ok_is_subtype);
3702   __ mov(r0, r3); // Restore object in r3
3703 
3704   // Collect counts on whether this test sees NULLs a lot or not.
3705   if (ProfileInterpreter) {
3706     __ b(done);
3707     __ bind(is_null);
3708     __ profile_null_seen(r2);
3709   } else {
3710     __ bind(is_null);   // same as 'done'
3711   }
3712   __ bind(done);
3713 }
3714 
3715 void TemplateTable::instanceof() {
3716   transition(atos, itos);
3717   Label done, is_null, ok_is_subtype, quicked, resolved;
3718   __ cbz(r0, is_null);
3719 
3720   // Get cpool & tags index
3721   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3722   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3723   // See if bytecode has already been quicked
3724   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3725   __ lea(r1, Address(rscratch1, r19));
3726   __ ldarb(r1, r1);
3727   __ cmp(r1, JVM_CONSTANT_Class);
3728   __ br(Assembler::EQ, quicked);
3729 
3730   __ push(atos); // save receiver for result, and for GC
3731   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3732   // vm_result_2 has metadata result
3733   __ get_vm_result_2(r0, rthread);
3734   __ pop(r3); // restore receiver
3735   __ verify_oop(r3);
3736   __ load_klass(r3, r3);
3737   __ b(resolved);
3738 
3739   // Get superklass in r0 and subklass in r3
3740   __ bind(quicked);
3741   __ load_klass(r3, r0);
3742   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3743 
3744   __ bind(resolved);
3745 
3746   // Generate subtype check.  Blows r2, r5
3747   // Superklass in r0.  Subklass in r3.
3748   __ gen_subtype_check(r3, ok_is_subtype);
3749 
3750   // Come here on failure
3751   __ mov(r0, 0);
3752   __ b(done);
3753   // Come here on success
3754   __ bind(ok_is_subtype);
3755   __ mov(r0, 1);
3756 
3757   // Collect counts on whether this test sees NULLs a lot or not.
3758   if (ProfileInterpreter) {
3759     __ b(done);
3760     __ bind(is_null);
3761     __ profile_null_seen(r2);
3762   } else {
3763     __ bind(is_null);   // same as 'done'
3764   }
3765   __ bind(done);
3766   // r0 = 0: obj == NULL or  obj is not an instanceof the specified klass
3767   // r0 = 1: obj != NULL and obj is     an instanceof the specified klass
3768 }
3769 
3770 //-----------------------------------------------------------------------------
3771 // Breakpoints
3772 void TemplateTable::_breakpoint() {
3773   // Note: We get here even if we are single stepping..
3774   // jbug inists on setting breakpoints at every bytecode
3775   // even if we are in single step mode.
3776 
3777   transition(vtos, vtos);
3778 
3779   // get the unpatched byte code
3780   __ get_method(c_rarg1);
3781   __ call_VM(noreg,
3782              CAST_FROM_FN_PTR(address,
3783                               InterpreterRuntime::get_original_bytecode_at),
3784              c_rarg1, rbcp);
3785   __ mov(r19, r0);
3786 
3787   // post the breakpoint event
3788   __ call_VM(noreg,
3789              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3790              rmethod, rbcp);
3791 
3792   // complete the execution of original bytecode
3793   __ mov(rscratch1, r19);
3794   __ dispatch_only_normal(vtos);
3795 }
3796 
3797 //-----------------------------------------------------------------------------
3798 // Exceptions
3799 
3800 void TemplateTable::athrow() {
3801   transition(atos, vtos);
3802   __ null_check(r0);
3803   __ b(Interpreter::throw_exception_entry());
3804 }
3805 
3806 //-----------------------------------------------------------------------------
3807 // Synchronization
3808 //
3809 // Note: monitorenter & exit are symmetric routines; which is reflected
3810 //       in the assembly code structure as well
3811 //
3812 // Stack layout:
3813 //
3814 // [expressions  ] <--- esp               = expression stack top
3815 // ..
3816 // [expressions  ]
3817 // [monitor entry] <--- monitor block top = expression stack bot
3818 // ..
3819 // [monitor entry]
3820 // [frame data   ] <--- monitor block bot
3821 // ...
3822 // [saved rbp    ] <--- rbp
3823 void TemplateTable::monitorenter()
3824 {
3825   transition(atos, vtos);
3826 
3827   // check for NULL object
3828   __ null_check(r0);
3829 
3830   const Address monitor_block_top(
3831         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3832   const Address monitor_block_bot(
3833         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3834   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3835 
3836   Label allocated;
3837 
3838   // initialize entry pointer
3839   __ mov(c_rarg1, zr); // points to free slot or NULL
3840 
3841   // find a free slot in the monitor block (result in c_rarg1)
3842   {
3843     Label entry, loop, exit;
3844     __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3845                                         // starting with top-most entry
3846     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3847 
3848     __ b(entry);
3849 
3850     __ bind(loop);
3851     // check if current entry is used
3852     // if not used then remember entry in c_rarg1
3853     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3854     __ cmp(zr, rscratch1);
3855     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3856     // check if current entry is for same object
3857     __ cmp(r0, rscratch1);
3858     // if same object then stop searching
3859     __ br(Assembler::EQ, exit);
3860     // otherwise advance to next entry
3861     __ add(c_rarg3, c_rarg3, entry_size);
3862     __ bind(entry);
3863     // check if bottom reached
3864     __ cmp(c_rarg3, c_rarg2);
3865     // if not at bottom then check this entry
3866     __ br(Assembler::NE, loop);
3867     __ bind(exit);
3868   }
3869 
3870   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3871                             // if found, continue with that on
3872 
3873   // allocate one if there's no free slot
3874   {
3875     Label entry, loop;
3876     // 1. compute new pointers            // rsp: old expression stack top
3877     __ ldr(c_rarg1, monitor_block_bot);   // c_rarg1: old expression stack bottom
3878     __ sub(esp, esp, entry_size);         // move expression stack top
3879     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3880     __ mov(c_rarg3, esp);                 // set start value for copy loop
3881     __ str(c_rarg1, monitor_block_bot);   // set new monitor block bottom
3882 
3883     __ sub(sp, sp, entry_size);           // make room for the monitor
3884 
3885     __ b(entry);
3886     // 2. move expression stack contents
3887     __ bind(loop);
3888     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3889                                                    // word from old location
3890     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
3891     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
3892     __ bind(entry);
3893     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
3894     __ br(Assembler::NE, loop);      // if not at bottom then
3895                                      // copy next word
3896   }
3897 
3898   // call run-time routine
3899   // c_rarg1: points to monitor entry
3900   __ bind(allocated);
3901 
3902   // Increment bcp to point to the next bytecode, so exception
3903   // handling for async. exceptions work correctly.
3904   // The object has already been poped from the stack, so the
3905   // expression stack looks correct.
3906   __ increment(rbcp);
3907 
3908   // store object
3909   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3910   __ lock_object(c_rarg1);
3911 
3912   // check to make sure this monitor doesn't cause stack overflow after locking
3913   __ save_bcp();  // in case of exception
3914   __ generate_stack_overflow_check(0);
3915 
3916   // The bcp has already been incremented. Just need to dispatch to
3917   // next instruction.
3918   __ dispatch_next(vtos);
3919 }
3920 
3921 
3922 void TemplateTable::monitorexit()
3923 {
3924   transition(atos, vtos);
3925 
3926   // check for NULL object
3927   __ null_check(r0);
3928 
3929   const Address monitor_block_top(
3930         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3931   const Address monitor_block_bot(
3932         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3933   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3934 
3935   Label found;
3936 
3937   // find matching slot
3938   {
3939     Label entry, loop;
3940     __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3941                                         // starting with top-most entry
3942     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3943                                         // of monitor block
3944     __ b(entry);
3945 
3946     __ bind(loop);
3947     // check if current entry is for same object
3948     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3949     __ cmp(r0, rscratch1);
3950     // if same object then stop searching
3951     __ br(Assembler::EQ, found);
3952     // otherwise advance to next entry
3953     __ add(c_rarg1, c_rarg1, entry_size);
3954     __ bind(entry);
3955     // check if bottom reached
3956     __ cmp(c_rarg1, c_rarg2);
3957     // if not at bottom then check this entry
3958     __ br(Assembler::NE, loop);
3959   }
3960 
3961   // error handling. Unlocking was not block-structured
3962   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3963                    InterpreterRuntime::throw_illegal_monitor_state_exception));
3964   __ should_not_reach_here();
3965 
3966   // call run-time routine
3967   __ bind(found);
3968   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3969   __ unlock_object(c_rarg1);
3970   __ pop_ptr(r0); // discard object
3971 }
3972 
3973 
3974 // Wide instructions
3975 void TemplateTable::wide()
3976 {
3977   __ load_unsigned_byte(r19, at_bcp(1));
3978   __ mov(rscratch1, (address)Interpreter::_wentry_point);
3979   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3980   __ br(rscratch1);
3981 }
3982 
3983 
3984 // Multi arrays
3985 void TemplateTable::multianewarray() {
3986   transition(vtos, atos);
3987   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3988   // last dim is on top of stack; we want address of first one:
3989   // first_addr = last_addr + (ndims - 1) * wordSize
3990   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3991   __ sub(c_rarg1, c_rarg1, wordSize);
3992   call_VM(r0,
3993           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3994           c_rarg1);
3995   __ load_unsigned_byte(r1, at_bcp(3));
3996   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3997 }
--- EOF ---