1 /*
   2  * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "compiler/compilerDefinitions.inline.hpp"
  29 #include "gc/shared/barrierSetAssembler.hpp"
  30 #include "gc/shared/collectedHeap.hpp"
  31 #include "gc/shared/tlab_globals.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "interpreter/interpreterRuntime.hpp"
  34 #include "interpreter/interp_masm.hpp"
  35 #include "interpreter/templateTable.hpp"
  36 #include "memory/universe.hpp"
  37 #include "oops/methodData.hpp"
  38 #include "oops/method.inline.hpp"
  39 #include "oops/objArrayKlass.hpp"
  40 #include "oops/oop.inline.hpp"
  41 #include "oops/resolvedFieldEntry.hpp"
  42 #include "oops/resolvedIndyEntry.hpp"
  43 #include "oops/resolvedMethodEntry.hpp"
  44 #include "prims/jvmtiExport.hpp"
  45 #include "prims/methodHandles.hpp"
  46 #include "runtime/frame.inline.hpp"
  47 #include "runtime/sharedRuntime.hpp"
  48 #include "runtime/stubRoutines.hpp"
  49 #include "runtime/synchronizer.hpp"
  50 #include "utilities/powerOfTwo.hpp"
  51 
  52 #define __ _masm->
  53 
  54 // Address computation: local variables
  55 
  56 static inline Address iaddress(int n) {
  57   return Address(rlocals, Interpreter::local_offset_in_bytes(n));
  58 }
  59 
  60 static inline Address laddress(int n) {
  61   return iaddress(n + 1);
  62 }
  63 
  64 static inline Address faddress(int n) {
  65   return iaddress(n);
  66 }
  67 
  68 static inline Address daddress(int n) {
  69   return laddress(n);
  70 }
  71 
  72 static inline Address aaddress(int n) {
  73   return iaddress(n);
  74 }
  75 
  76 static inline Address iaddress(Register r) {
  77   return Address(rlocals, r, Address::lsl(3));
  78 }
  79 
  80 static inline Address laddress(Register r, Register scratch,
  81                                InterpreterMacroAssembler* _masm) {
  82   __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
  83   return Address(scratch, Interpreter::local_offset_in_bytes(1));
  84 }
  85 
  86 static inline Address faddress(Register r) {
  87   return iaddress(r);
  88 }
  89 
  90 static inline Address daddress(Register r, Register scratch,
  91                                InterpreterMacroAssembler* _masm) {
  92   return laddress(r, scratch, _masm);
  93 }
  94 
  95 static inline Address aaddress(Register r) {
  96   return iaddress(r);
  97 }
  98 
  99 static inline Address at_rsp() {
 100   return Address(esp, 0);
 101 }
 102 
 103 // At top of Java expression stack which may be different than esp().  It
 104 // isn't for category 1 objects.
 105 static inline Address at_tos   () {
 106   return Address(esp,  Interpreter::expr_offset_in_bytes(0));
 107 }
 108 
 109 static inline Address at_tos_p1() {
 110   return Address(esp,  Interpreter::expr_offset_in_bytes(1));
 111 }
 112 
 113 static inline Address at_tos_p2() {
 114   return Address(esp,  Interpreter::expr_offset_in_bytes(2));
 115 }
 116 
 117 static inline Address at_tos_p3() {
 118   return Address(esp,  Interpreter::expr_offset_in_bytes(3));
 119 }
 120 
 121 static inline Address at_tos_p4() {
 122   return Address(esp,  Interpreter::expr_offset_in_bytes(4));
 123 }
 124 
 125 static inline Address at_tos_p5() {
 126   return Address(esp,  Interpreter::expr_offset_in_bytes(5));
 127 }
 128 
 129 // Condition conversion
 130 static Assembler::Condition j_not(TemplateTable::Condition cc) {
 131   switch (cc) {
 132   case TemplateTable::equal        : return Assembler::NE;
 133   case TemplateTable::not_equal    : return Assembler::EQ;
 134   case TemplateTable::less         : return Assembler::GE;
 135   case TemplateTable::less_equal   : return Assembler::GT;
 136   case TemplateTable::greater      : return Assembler::LE;
 137   case TemplateTable::greater_equal: return Assembler::LT;
 138   }
 139   ShouldNotReachHere();
 140   return Assembler::EQ;
 141 }
 142 
 143 
 144 // Miscellaneous helper routines
 145 // Store an oop (or null) at the Address described by obj.
 146 // If val == noreg this means store a null
 147 static void do_oop_store(InterpreterMacroAssembler* _masm,
 148                          Address dst,
 149                          Register val,
 150                          DecoratorSet decorators) {
 151   assert(val == noreg || val == r0, "parameter is just for looks");
 152   __ store_heap_oop(dst, val, r10, r11, r3, decorators);
 153 }
 154 
 155 static void do_oop_load(InterpreterMacroAssembler* _masm,
 156                         Address src,
 157                         Register dst,
 158                         DecoratorSet decorators) {
 159   __ load_heap_oop(dst, src, r10, r11, decorators);
 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_vputfield:
 176   case Bytecodes::_fast_aputfield:
 177   case Bytecodes::_fast_bputfield:
 178   case Bytecodes::_fast_zputfield:
 179   case Bytecodes::_fast_cputfield:
 180   case Bytecodes::_fast_dputfield:
 181   case Bytecodes::_fast_fputfield:
 182   case Bytecodes::_fast_iputfield:
 183   case Bytecodes::_fast_lputfield:
 184   case Bytecodes::_fast_sputfield:
 185     {
 186       // We skip bytecode quickening for putfield instructions when
 187       // the put_code written to the constant pool cache is zero.
 188       // This is required so that every execution of this instruction
 189       // calls out to InterpreterRuntime::resolve_get_put to do
 190       // additional, required work.
 191       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 192       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 193       __ load_field_entry(temp_reg, bc_reg);
 194       if (byte_no == f1_byte) {
 195         __ lea(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::get_code_offset())));
 196       } else {
 197         __ lea(temp_reg, Address(temp_reg, in_bytes(ResolvedFieldEntry::put_code_offset())));
 198       }
 199       // Load-acquire the bytecode to match store-release in ResolvedFieldEntry::fill_in()
 200       __ ldarb(temp_reg, temp_reg);
 201       __ movw(bc_reg, bc);
 202       __ cbzw(temp_reg, L_patch_done);  // don't patch
 203     }
 204     break;
 205   default:
 206     assert(byte_no == -1, "sanity");
 207     // the pair bytecodes have already done the load.
 208     if (load_bc_into_bc_reg) {
 209       __ movw(bc_reg, bc);
 210     }
 211   }
 212 
 213   if (JvmtiExport::can_post_breakpoint()) {
 214     Label L_fast_patch;
 215     // if a breakpoint is present we can't rewrite the stream directly
 216     __ load_unsigned_byte(temp_reg, at_bcp(0));
 217     __ cmpw(temp_reg, Bytecodes::_breakpoint);
 218     __ br(Assembler::NE, L_fast_patch);
 219     // Let breakpoint table handling rewrite to quicker bytecode
 220     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
 221     __ b(L_patch_done);
 222     __ bind(L_fast_patch);
 223   }
 224 
 225 #ifdef ASSERT
 226   Label L_okay;
 227   __ load_unsigned_byte(temp_reg, at_bcp(0));
 228   __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
 229   __ br(Assembler::EQ, L_okay);
 230   __ cmpw(temp_reg, bc_reg);
 231   __ br(Assembler::EQ, L_okay);
 232   __ stop("patching the wrong bytecode");
 233   __ bind(L_okay);
 234 #endif
 235 
 236   // patch bytecode
 237   __ strb(bc_reg, at_bcp(0));
 238   __ bind(L_patch_done);
 239 }
 240 
 241 
 242 // Individual instructions
 243 
 244 void TemplateTable::nop() {
 245   transition(vtos, vtos);
 246   // nothing to do
 247 }
 248 
 249 void TemplateTable::shouldnotreachhere() {
 250   transition(vtos, vtos);
 251   __ stop("shouldnotreachhere bytecode");
 252 }
 253 
 254 void TemplateTable::aconst_null()
 255 {
 256   transition(vtos, atos);
 257   __ mov(r0, 0);
 258 }
 259 
 260 void TemplateTable::iconst(int value)
 261 {
 262   transition(vtos, itos);
 263   __ mov(r0, value);
 264 }
 265 
 266 void TemplateTable::lconst(int value)
 267 {
 268   __ mov(r0, value);
 269 }
 270 
 271 void TemplateTable::fconst(int value)
 272 {
 273   transition(vtos, ftos);
 274   switch (value) {
 275   case 0:
 276     __ fmovs(v0, 0.0);
 277     break;
 278   case 1:
 279     __ fmovs(v0, 1.0);
 280     break;
 281   case 2:
 282     __ fmovs(v0, 2.0);
 283     break;
 284   default:
 285     ShouldNotReachHere();
 286     break;
 287   }
 288 }
 289 
 290 void TemplateTable::dconst(int value)
 291 {
 292   transition(vtos, dtos);
 293   switch (value) {
 294   case 0:
 295     __ fmovd(v0, 0.0);
 296     break;
 297   case 1:
 298     __ fmovd(v0, 1.0);
 299     break;
 300   case 2:
 301     __ fmovd(v0, 2.0);
 302     break;
 303   default:
 304     ShouldNotReachHere();
 305     break;
 306   }
 307 }
 308 
 309 void TemplateTable::bipush()
 310 {
 311   transition(vtos, itos);
 312   __ load_signed_byte32(r0, at_bcp(1));
 313 }
 314 
 315 void TemplateTable::sipush()
 316 {
 317   transition(vtos, itos);
 318   __ load_unsigned_short(r0, at_bcp(1));
 319   __ revw(r0, r0);
 320   __ asrw(r0, r0, 16);
 321 }
 322 
 323 void TemplateTable::ldc(LdcType type)
 324 {
 325   transition(vtos, vtos);
 326   Label call_ldc, notFloat, notClass, notInt, Done;
 327 
 328   if (is_ldc_wide(type)) {
 329     __ get_unsigned_2_byte_index_at_bcp(r1, 1);
 330   } else {
 331     __ load_unsigned_byte(r1, at_bcp(1));
 332   }
 333   __ get_cpool_and_tags(r2, r0);
 334 
 335   const int base_offset = ConstantPool::header_size() * wordSize;
 336   const int tags_offset = Array<u1>::base_offset_in_bytes();
 337 
 338   // get type
 339   __ add(r3, r1, tags_offset);
 340   __ lea(r3, Address(r0, r3));
 341   __ ldarb(r3, r3);
 342 
 343   // unresolved class - get the resolved class
 344   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClass);
 345   __ br(Assembler::EQ, call_ldc);
 346 
 347   // unresolved class in error state - call into runtime to throw the error
 348   // from the first resolution attempt
 349   __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClassInError);
 350   __ br(Assembler::EQ, call_ldc);
 351 
 352   // resolved class - need to call vm to get java mirror of the class
 353   __ cmp(r3, (u1)JVM_CONSTANT_Class);
 354   __ br(Assembler::NE, notClass);
 355 
 356   __ bind(call_ldc);
 357   __ mov(c_rarg1, is_ldc_wide(type) ? 1 : 0);
 358   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
 359   __ push_ptr(r0);
 360   __ verify_oop(r0);
 361   __ b(Done);
 362 
 363   __ bind(notClass);
 364   __ cmp(r3, (u1)JVM_CONSTANT_Float);
 365   __ br(Assembler::NE, notFloat);
 366   // ftos
 367   __ adds(r1, r2, r1, Assembler::LSL, 3);
 368   __ ldrs(v0, Address(r1, base_offset));
 369   __ push_f();
 370   __ b(Done);
 371 
 372   __ bind(notFloat);
 373 
 374   __ cmp(r3, (u1)JVM_CONSTANT_Integer);
 375   __ br(Assembler::NE, notInt);
 376 
 377   // itos
 378   __ adds(r1, r2, r1, Assembler::LSL, 3);
 379   __ ldrw(r0, Address(r1, base_offset));
 380   __ push_i(r0);
 381   __ b(Done);
 382 
 383   __ bind(notInt);
 384   condy_helper(Done);
 385 
 386   __ bind(Done);
 387 }
 388 
 389 // Fast path for caching oop constants.
 390 void TemplateTable::fast_aldc(LdcType type)
 391 {
 392   transition(vtos, atos);
 393 
 394   Register result = r0;
 395   Register tmp = r1;
 396   Register rarg = r2;
 397 
 398   int index_size = is_ldc_wide(type) ? sizeof(u2) : sizeof(u1);
 399 
 400   Label resolved;
 401 
 402   // We are resolved if the resolved reference cache entry contains a
 403   // non-null object (String, MethodType, etc.)
 404   assert_different_registers(result, tmp);
 405   __ get_cache_index_at_bcp(tmp, 1, index_size);
 406   __ load_resolved_reference_at_index(result, tmp);
 407   __ cbnz(result, resolved);
 408 
 409   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 410 
 411   // first time invocation - must resolve first
 412   __ mov(rarg, (int)bytecode());
 413   __ call_VM(result, entry, rarg);
 414 
 415   __ bind(resolved);
 416 
 417   { // Check for the null sentinel.
 418     // If we just called the VM, it already did the mapping for us,
 419     // but it's harmless to retry.
 420     Label notNull;
 421 
 422     // Stash null_sentinel address to get its value later
 423     __ movptr(rarg, (uintptr_t)Universe::the_null_sentinel_addr());
 424     __ ldr(tmp, Address(rarg));
 425     __ resolve_oop_handle(tmp, r5, rscratch2);
 426     __ cmpoop(result, tmp);
 427     __ br(Assembler::NE, notNull);
 428     __ mov(result, 0);  // null object reference
 429     __ bind(notNull);
 430   }
 431 
 432   if (VerifyOops) {
 433     // Safe to call with 0 result
 434     __ verify_oop(result);
 435   }
 436 }
 437 
 438 void TemplateTable::ldc2_w()
 439 {
 440   transition(vtos, vtos);
 441   Label notDouble, notLong, Done;
 442   __ get_unsigned_2_byte_index_at_bcp(r0, 1);
 443 
 444   __ get_cpool_and_tags(r1, r2);
 445   const int base_offset = ConstantPool::header_size() * wordSize;
 446   const int tags_offset = Array<u1>::base_offset_in_bytes();
 447 
 448   // get type
 449   __ lea(r2, Address(r2, r0, Address::lsl(0)));
 450   __ load_unsigned_byte(r2, Address(r2, tags_offset));
 451   __ cmpw(r2, (int)JVM_CONSTANT_Double);
 452   __ br(Assembler::NE, notDouble);
 453 
 454   // dtos
 455   __ lea (r2, Address(r1, r0, Address::lsl(3)));
 456   __ ldrd(v0, Address(r2, base_offset));
 457   __ push_d();
 458   __ b(Done);
 459 
 460   __ bind(notDouble);
 461   __ cmpw(r2, (int)JVM_CONSTANT_Long);
 462   __ br(Assembler::NE, notLong);
 463 
 464   // ltos
 465   __ lea(r0, Address(r1, r0, Address::lsl(3)));
 466   __ ldr(r0, Address(r0, base_offset));
 467   __ push_l();
 468   __ b(Done);
 469 
 470   __ bind(notLong);
 471   condy_helper(Done);
 472 
 473   __ bind(Done);
 474 }
 475 
 476 void TemplateTable::condy_helper(Label& Done)
 477 {
 478   Register obj = r0;
 479   Register rarg = r1;
 480   Register flags = r2;
 481   Register off = r3;
 482 
 483   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 484 
 485   __ mov(rarg, (int) bytecode());
 486   __ call_VM(obj, entry, rarg);
 487 
 488   __ get_vm_result_2(flags, rthread);
 489 
 490   // VMr = obj = base address to find primitive value to push
 491   // VMr2 = flags = (tos, off) using format of CPCE::_flags
 492   __ mov(off, flags);
 493   __ andw(off, off, ConstantPoolCache::field_index_mask);
 494 
 495   const Address field(obj, off);
 496 
 497   // What sort of thing are we loading?
 498   // x86 uses a shift and mask or wings it with a shift plus assert
 499   // the mask is not needed. aarch64 just uses bitfield extract
 500   __ ubfxw(flags, flags, ConstantPoolCache::tos_state_shift,
 501            ConstantPoolCache::tos_state_bits);
 502 
 503   switch (bytecode()) {
 504     case Bytecodes::_ldc:
 505     case Bytecodes::_ldc_w:
 506       {
 507         // tos in (itos, ftos, stos, btos, ctos, ztos)
 508         Label notInt, notFloat, notShort, notByte, notChar, notBool;
 509         __ cmpw(flags, itos);
 510         __ br(Assembler::NE, notInt);
 511         // itos
 512         __ ldrw(r0, field);
 513         __ push(itos);
 514         __ b(Done);
 515 
 516         __ bind(notInt);
 517         __ cmpw(flags, ftos);
 518         __ br(Assembler::NE, notFloat);
 519         // ftos
 520         __ load_float(field);
 521         __ push(ftos);
 522         __ b(Done);
 523 
 524         __ bind(notFloat);
 525         __ cmpw(flags, stos);
 526         __ br(Assembler::NE, notShort);
 527         // stos
 528         __ load_signed_short(r0, field);
 529         __ push(stos);
 530         __ b(Done);
 531 
 532         __ bind(notShort);
 533         __ cmpw(flags, btos);
 534         __ br(Assembler::NE, notByte);
 535         // btos
 536         __ load_signed_byte(r0, field);
 537         __ push(btos);
 538         __ b(Done);
 539 
 540         __ bind(notByte);
 541         __ cmpw(flags, ctos);
 542         __ br(Assembler::NE, notChar);
 543         // ctos
 544         __ load_unsigned_short(r0, field);
 545         __ push(ctos);
 546         __ b(Done);
 547 
 548         __ bind(notChar);
 549         __ cmpw(flags, ztos);
 550         __ br(Assembler::NE, notBool);
 551         // ztos
 552         __ load_signed_byte(r0, field);
 553         __ push(ztos);
 554         __ b(Done);
 555 
 556         __ bind(notBool);
 557         break;
 558       }
 559 
 560     case Bytecodes::_ldc2_w:
 561       {
 562         Label notLong, notDouble;
 563         __ cmpw(flags, ltos);
 564         __ br(Assembler::NE, notLong);
 565         // ltos
 566         __ ldr(r0, field);
 567         __ push(ltos);
 568         __ b(Done);
 569 
 570         __ bind(notLong);
 571         __ cmpw(flags, dtos);
 572         __ br(Assembler::NE, notDouble);
 573         // dtos
 574         __ load_double(field);
 575         __ push(dtos);
 576         __ b(Done);
 577 
 578        __ bind(notDouble);
 579         break;
 580       }
 581 
 582     default:
 583       ShouldNotReachHere();
 584     }
 585 
 586     __ stop("bad ldc/condy");
 587 }
 588 
 589 void TemplateTable::locals_index(Register reg, int offset)
 590 {
 591   __ ldrb(reg, at_bcp(offset));
 592   __ neg(reg, reg);
 593 }
 594 
 595 void TemplateTable::iload() {
 596   iload_internal();
 597 }
 598 
 599 void TemplateTable::nofast_iload() {
 600   iload_internal(may_not_rewrite);
 601 }
 602 
 603 void TemplateTable::iload_internal(RewriteControl rc) {
 604   transition(vtos, itos);
 605   if (RewriteFrequentPairs && rc == may_rewrite) {
 606     Label rewrite, done;
 607     Register bc = r4;
 608 
 609     // get next bytecode
 610     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
 611 
 612     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 613     // last two iloads in a pair.  Comparing against fast_iload means that
 614     // the next bytecode is neither an iload or a caload, and therefore
 615     // an iload pair.
 616     __ cmpw(r1, Bytecodes::_iload);
 617     __ br(Assembler::EQ, done);
 618 
 619     // if _fast_iload rewrite to _fast_iload2
 620     __ cmpw(r1, Bytecodes::_fast_iload);
 621     __ movw(bc, Bytecodes::_fast_iload2);
 622     __ br(Assembler::EQ, rewrite);
 623 
 624     // if _caload rewrite to _fast_icaload
 625     __ cmpw(r1, Bytecodes::_caload);
 626     __ movw(bc, Bytecodes::_fast_icaload);
 627     __ br(Assembler::EQ, rewrite);
 628 
 629     // else rewrite to _fast_iload
 630     __ movw(bc, Bytecodes::_fast_iload);
 631 
 632     // rewrite
 633     // bc: new bytecode
 634     __ bind(rewrite);
 635     patch_bytecode(Bytecodes::_iload, bc, r1, false);
 636     __ bind(done);
 637 
 638   }
 639 
 640   // do iload, get the local value into tos
 641   locals_index(r1);
 642   __ ldr(r0, iaddress(r1));
 643 
 644 }
 645 
 646 void TemplateTable::fast_iload2()
 647 {
 648   transition(vtos, itos);
 649   locals_index(r1);
 650   __ ldr(r0, iaddress(r1));
 651   __ push(itos);
 652   locals_index(r1, 3);
 653   __ ldr(r0, iaddress(r1));
 654 }
 655 
 656 void TemplateTable::fast_iload()
 657 {
 658   transition(vtos, itos);
 659   locals_index(r1);
 660   __ ldr(r0, iaddress(r1));
 661 }
 662 
 663 void TemplateTable::lload()
 664 {
 665   transition(vtos, ltos);
 666   __ ldrb(r1, at_bcp(1));
 667   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 668   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 669 }
 670 
 671 void TemplateTable::fload()
 672 {
 673   transition(vtos, ftos);
 674   locals_index(r1);
 675   // n.b. we use ldrd here because this is a 64 bit slot
 676   // this is comparable to the iload case
 677   __ ldrd(v0, faddress(r1));
 678 }
 679 
 680 void TemplateTable::dload()
 681 {
 682   transition(vtos, dtos);
 683   __ ldrb(r1, at_bcp(1));
 684   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 685   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 686 }
 687 
 688 void TemplateTable::aload()
 689 {
 690   transition(vtos, atos);
 691   locals_index(r1);
 692   __ ldr(r0, iaddress(r1));
 693 }
 694 
 695 void TemplateTable::locals_index_wide(Register reg) {
 696   __ ldrh(reg, at_bcp(2));
 697   __ rev16w(reg, reg);
 698   __ neg(reg, reg);
 699 }
 700 
 701 void TemplateTable::wide_iload() {
 702   transition(vtos, itos);
 703   locals_index_wide(r1);
 704   __ ldr(r0, iaddress(r1));
 705 }
 706 
 707 void TemplateTable::wide_lload()
 708 {
 709   transition(vtos, ltos);
 710   __ ldrh(r1, at_bcp(2));
 711   __ rev16w(r1, r1);
 712   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 713   __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 714 }
 715 
 716 void TemplateTable::wide_fload()
 717 {
 718   transition(vtos, ftos);
 719   locals_index_wide(r1);
 720   // n.b. we use ldrd here because this is a 64 bit slot
 721   // this is comparable to the iload case
 722   __ ldrd(v0, faddress(r1));
 723 }
 724 
 725 void TemplateTable::wide_dload()
 726 {
 727   transition(vtos, dtos);
 728   __ ldrh(r1, at_bcp(2));
 729   __ rev16w(r1, r1);
 730   __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
 731   __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
 732 }
 733 
 734 void TemplateTable::wide_aload()
 735 {
 736   transition(vtos, atos);
 737   locals_index_wide(r1);
 738   __ ldr(r0, aaddress(r1));
 739 }
 740 
 741 void TemplateTable::index_check(Register array, Register index)
 742 {
 743   // destroys r1, rscratch1
 744   // sign extend index for use by indexed load
 745   // __ movl2ptr(index, index);
 746   // check index
 747   Register length = rscratch1;
 748   __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
 749   __ cmpw(index, length);
 750   if (index != r1) {
 751     // ??? convention: move aberrant index into r1 for exception message
 752     assert(r1 != array, "different registers");
 753     __ mov(r1, index);
 754   }
 755   Label ok;
 756   __ br(Assembler::LO, ok);
 757   // ??? convention: move array into r3 for exception message
 758    __ mov(r3, array);
 759    __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
 760    __ br(rscratch1);
 761   __ bind(ok);
 762 }
 763 
 764 void TemplateTable::iaload()
 765 {
 766   transition(itos, itos);
 767   __ mov(r1, r0);
 768   __ pop_ptr(r0);
 769   // r0: array
 770   // r1: index
 771   index_check(r0, r1); // leaves index in r1, kills rscratch1
 772   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
 773   __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 774 }
 775 
 776 void TemplateTable::laload()
 777 {
 778   transition(itos, ltos);
 779   __ mov(r1, r0);
 780   __ pop_ptr(r0);
 781   // r0: array
 782   // r1: index
 783   index_check(r0, r1); // leaves index in r1, kills rscratch1
 784   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
 785   __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 786 }
 787 
 788 void TemplateTable::faload()
 789 {
 790   transition(itos, ftos);
 791   __ mov(r1, r0);
 792   __ pop_ptr(r0);
 793   // r0: array
 794   // r1: index
 795   index_check(r0, r1); // leaves index in r1, kills rscratch1
 796   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
 797   __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
 798 }
 799 
 800 void TemplateTable::daload()
 801 {
 802   transition(itos, dtos);
 803   __ mov(r1, r0);
 804   __ pop_ptr(r0);
 805   // r0: array
 806   // r1: index
 807   index_check(r0, r1); // leaves index in r1, kills rscratch1
 808   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
 809   __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
 810 }
 811 
 812 void TemplateTable::aaload()
 813 {
 814   transition(itos, atos);
 815   __ mov(r1, r0);
 816   __ pop_ptr(r0);
 817   // r0: array
 818   // r1: index
 819   index_check(r0, r1); // leaves index in r1, kills rscratch1
 820   __ profile_array_type<ArrayLoadData>(r2, r0, r4);
 821   if (UseFlatArray) {
 822     Label is_flat_array, done;
 823 
 824     __ test_flat_array_oop(r0, r8 /*temp*/, is_flat_array);
 825     __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 826     do_oop_load(_masm, Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)), r0, IS_ARRAY);
 827 
 828     __ b(done);
 829     __ bind(is_flat_array);
 830     __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), r0, r1);
 831     // Ensure the stores to copy the inline field contents are visible
 832     // before any subsequent store that publishes this reference.
 833     __ membar(Assembler::StoreStore);
 834     __ bind(done);
 835   } else {
 836     __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 837     do_oop_load(_masm, Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)), r0, IS_ARRAY);
 838   }
 839   __ profile_element_type(r2, r0, r4);
 840 }
 841 
 842 void TemplateTable::baload()
 843 {
 844   transition(itos, itos);
 845   __ mov(r1, r0);
 846   __ pop_ptr(r0);
 847   // r0: array
 848   // r1: index
 849   index_check(r0, r1); // leaves index in r1, kills rscratch1
 850   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
 851   __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(0)), noreg, noreg);
 852 }
 853 
 854 void TemplateTable::caload()
 855 {
 856   transition(itos, itos);
 857   __ mov(r1, r0);
 858   __ pop_ptr(r0);
 859   // r0: array
 860   // r1: index
 861   index_check(r0, r1); // leaves index in r1, kills rscratch1
 862   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 863   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 864 }
 865 
 866 // iload followed by caload frequent pair
 867 void TemplateTable::fast_icaload()
 868 {
 869   transition(vtos, itos);
 870   // load index out of locals
 871   locals_index(r2);
 872   __ ldr(r1, iaddress(r2));
 873 
 874   __ pop_ptr(r0);
 875 
 876   // r0: array
 877   // r1: index
 878   index_check(r0, r1); // leaves index in r1, kills rscratch1
 879   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
 880   __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 881 }
 882 
 883 void TemplateTable::saload()
 884 {
 885   transition(itos, itos);
 886   __ mov(r1, r0);
 887   __ pop_ptr(r0);
 888   // r0: array
 889   // r1: index
 890   index_check(r0, r1); // leaves index in r1, kills rscratch1
 891   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_SHORT) >> 1);
 892   __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
 893 }
 894 
 895 void TemplateTable::iload(int n)
 896 {
 897   transition(vtos, itos);
 898   __ ldr(r0, iaddress(n));
 899 }
 900 
 901 void TemplateTable::lload(int n)
 902 {
 903   transition(vtos, ltos);
 904   __ ldr(r0, laddress(n));
 905 }
 906 
 907 void TemplateTable::fload(int n)
 908 {
 909   transition(vtos, ftos);
 910   __ ldrs(v0, faddress(n));
 911 }
 912 
 913 void TemplateTable::dload(int n)
 914 {
 915   transition(vtos, dtos);
 916   __ ldrd(v0, daddress(n));
 917 }
 918 
 919 void TemplateTable::aload(int n)
 920 {
 921   transition(vtos, atos);
 922   __ ldr(r0, iaddress(n));
 923 }
 924 
 925 void TemplateTable::aload_0() {
 926   aload_0_internal();
 927 }
 928 
 929 void TemplateTable::nofast_aload_0() {
 930   aload_0_internal(may_not_rewrite);
 931 }
 932 
 933 void TemplateTable::aload_0_internal(RewriteControl rc) {
 934   // According to bytecode histograms, the pairs:
 935   //
 936   // _aload_0, _fast_igetfield
 937   // _aload_0, _fast_agetfield
 938   // _aload_0, _fast_fgetfield
 939   //
 940   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 941   // _aload_0 bytecode checks if the next bytecode is either
 942   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 943   // rewrites the current bytecode into a pair bytecode; otherwise it
 944   // rewrites the current bytecode into _fast_aload_0 that doesn't do
 945   // the pair check anymore.
 946   //
 947   // Note: If the next bytecode is _getfield, the rewrite must be
 948   //       delayed, otherwise we may miss an opportunity for a pair.
 949   //
 950   // Also rewrite frequent pairs
 951   //   aload_0, aload_1
 952   //   aload_0, iload_1
 953   // These bytecodes with a small amount of code are most profitable
 954   // to rewrite
 955   if (RewriteFrequentPairs && rc == may_rewrite) {
 956     Label rewrite, done;
 957     const Register bc = r4;
 958 
 959     // get next bytecode
 960     __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
 961 
 962     // if _getfield then wait with rewrite
 963     __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
 964     __ br(Assembler::EQ, done);
 965 
 966     // if _igetfield then rewrite to _fast_iaccess_0
 967     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 968     __ cmpw(r1, Bytecodes::_fast_igetfield);
 969     __ movw(bc, Bytecodes::_fast_iaccess_0);
 970     __ br(Assembler::EQ, rewrite);
 971 
 972     // if _agetfield then rewrite to _fast_aaccess_0
 973     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 974     __ cmpw(r1, Bytecodes::_fast_agetfield);
 975     __ movw(bc, Bytecodes::_fast_aaccess_0);
 976     __ br(Assembler::EQ, rewrite);
 977 
 978     // if _fgetfield then rewrite to _fast_faccess_0
 979     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
 980     __ cmpw(r1, Bytecodes::_fast_fgetfield);
 981     __ movw(bc, Bytecodes::_fast_faccess_0);
 982     __ br(Assembler::EQ, rewrite);
 983 
 984     // else rewrite to _fast_aload0
 985     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
 986     __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
 987 
 988     // rewrite
 989     // bc: new bytecode
 990     __ bind(rewrite);
 991     patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
 992 
 993     __ bind(done);
 994   }
 995 
 996   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 997   aload(0);
 998 }
 999 
1000 void TemplateTable::istore()
1001 {
1002   transition(itos, vtos);
1003   locals_index(r1);
1004   // FIXME: We're being very pernickerty here storing a jint in a
1005   // local with strw, which costs an extra instruction over what we'd
1006   // be able to do with a simple str.  We should just store the whole
1007   // word.
1008   __ lea(rscratch1, iaddress(r1));
1009   __ strw(r0, Address(rscratch1));
1010 }
1011 
1012 void TemplateTable::lstore()
1013 {
1014   transition(ltos, vtos);
1015   locals_index(r1);
1016   __ str(r0, laddress(r1, rscratch1, _masm));
1017 }
1018 
1019 void TemplateTable::fstore() {
1020   transition(ftos, vtos);
1021   locals_index(r1);
1022   __ lea(rscratch1, iaddress(r1));
1023   __ strs(v0, Address(rscratch1));
1024 }
1025 
1026 void TemplateTable::dstore() {
1027   transition(dtos, vtos);
1028   locals_index(r1);
1029   __ strd(v0, daddress(r1, rscratch1, _masm));
1030 }
1031 
1032 void TemplateTable::astore()
1033 {
1034   transition(vtos, vtos);
1035   __ pop_ptr(r0);
1036   locals_index(r1);
1037   __ str(r0, aaddress(r1));
1038 }
1039 
1040 void TemplateTable::wide_istore() {
1041   transition(vtos, vtos);
1042   __ pop_i();
1043   locals_index_wide(r1);
1044   __ lea(rscratch1, iaddress(r1));
1045   __ strw(r0, Address(rscratch1));
1046 }
1047 
1048 void TemplateTable::wide_lstore() {
1049   transition(vtos, vtos);
1050   __ pop_l();
1051   locals_index_wide(r1);
1052   __ str(r0, laddress(r1, rscratch1, _masm));
1053 }
1054 
1055 void TemplateTable::wide_fstore() {
1056   transition(vtos, vtos);
1057   __ pop_f();
1058   locals_index_wide(r1);
1059   __ lea(rscratch1, faddress(r1));
1060   __ strs(v0, rscratch1);
1061 }
1062 
1063 void TemplateTable::wide_dstore() {
1064   transition(vtos, vtos);
1065   __ pop_d();
1066   locals_index_wide(r1);
1067   __ strd(v0, daddress(r1, rscratch1, _masm));
1068 }
1069 
1070 void TemplateTable::wide_astore() {
1071   transition(vtos, vtos);
1072   __ pop_ptr(r0);
1073   locals_index_wide(r1);
1074   __ str(r0, aaddress(r1));
1075 }
1076 
1077 void TemplateTable::iastore() {
1078   transition(itos, vtos);
1079   __ pop_i(r1);
1080   __ pop_ptr(r3);
1081   // r0: value
1082   // r1: index
1083   // r3: array
1084   index_check(r3, r1); // prefer index in r1
1085   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
1086   __ access_store_at(T_INT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), r0, noreg, noreg, noreg);
1087 }
1088 
1089 void TemplateTable::lastore() {
1090   transition(ltos, vtos);
1091   __ pop_i(r1);
1092   __ pop_ptr(r3);
1093   // r0: value
1094   // r1: index
1095   // r3: array
1096   index_check(r3, r1); // prefer index in r1
1097   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
1098   __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), r0, noreg, noreg, noreg);
1099 }
1100 
1101 void TemplateTable::fastore() {
1102   transition(ftos, vtos);
1103   __ pop_i(r1);
1104   __ pop_ptr(r3);
1105   // v0: value
1106   // r1:  index
1107   // r3:  array
1108   index_check(r3, r1); // prefer index in r1
1109   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
1110   __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), noreg /* ftos */, noreg, noreg, noreg);
1111 }
1112 
1113 void TemplateTable::dastore() {
1114   transition(dtos, vtos);
1115   __ pop_i(r1);
1116   __ pop_ptr(r3);
1117   // v0: value
1118   // r1:  index
1119   // r3:  array
1120   index_check(r3, r1); // prefer index in r1
1121   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
1122   __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), noreg /* dtos */, noreg, noreg, noreg);
1123 }
1124 
1125 void TemplateTable::aastore() {
1126   Label is_null, is_flat_array, ok_is_subtype, done;
1127   transition(vtos, vtos);
1128   // stack: ..., array, index, value
1129   __ ldr(r0, at_tos());    // value
1130   __ ldr(r2, at_tos_p1()); // index
1131   __ ldr(r3, at_tos_p2()); // array
1132 
1133   index_check(r3, r2);     // kills r1
1134 
1135   __ profile_array_type<ArrayStoreData>(r4, r3, r5);
1136   __ profile_multiple_element_types(r4, r0, r5, r6);
1137 
1138   __ add(r4, r2, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
1139   Address element_address(r3, r4, Address::uxtw(LogBytesPerHeapOop));
1140   // Be careful not to clobber r4 below
1141 
1142   // do array store check - check for null value first
1143   __ cbz(r0, is_null);
1144 
1145   // Move array class to r5
1146   __ load_klass(r5, r3);
1147 
1148   if (UseFlatArray) {
1149     __ ldrw(r6, Address(r5, Klass::layout_helper_offset()));
1150     __ test_flat_array_layout(r6, is_flat_array);
1151   }
1152 
1153   // Move subklass into r1
1154   __ load_klass(r1, r0);
1155 
1156   // Move array element superklass into r0
1157   __ ldr(r0, Address(r5, ObjArrayKlass::element_klass_offset()));
1158   // Compress array + index*oopSize + 12 into a single register.  Frees r2.
1159 
1160   // Generate subtype check.  Blows r2, r5
1161   // Superklass in r0.  Subklass in r1.
1162 
1163   // is "r1 <: r0" ? (value subclass <: array element superclass)
1164   __ gen_subtype_check(r1, ok_is_subtype, false);
1165 
1166   // Come here on failure
1167   // object is at TOS
1168   __ b(Interpreter::_throw_ArrayStoreException_entry);
1169 
1170   // Come here on success
1171   __ bind(ok_is_subtype);
1172 
1173   // Get the value we will store
1174   __ ldr(r0, at_tos());
1175   // Now store using the appropriate barrier
1176   do_oop_store(_masm, element_address, r0, IS_ARRAY);
1177   __ b(done);
1178 
1179   // Have a null in r0, r3=array, r2=index.  Store null at ary[idx]
1180   __ bind(is_null);
1181   if (EnableValhalla) {
1182     Label is_null_into_value_array_npe, store_null;
1183 
1184     // No way to store null in flat null-free array
1185     __ test_null_free_array_oop(r3, r8, is_null_into_value_array_npe);
1186     __ b(store_null);
1187 
1188     __ bind(is_null_into_value_array_npe);
1189     __ b(ExternalAddress(Interpreter::_throw_NullPointerException_entry));
1190 
1191     __ bind(store_null);
1192   }
1193 
1194   // Store a null
1195   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1196   __ b(done);
1197 
1198   if (UseFlatArray) {
1199      Label is_type_ok;
1200     __ bind(is_flat_array); // Store non-null value to flat
1201 
1202     // Simplistic type check...
1203     // r0 - value, r2 - index, r3 - array.
1204 
1205     // Profile the not-null value's klass.
1206     // Load value class
1207      __ load_klass(r1, r0);
1208 
1209     // Move element klass into r7
1210      __ ldr(r7, Address(r5, ArrayKlass::element_klass_offset()));
1211 
1212     // flat value array needs exact type match
1213     // is "r1 == r7" (value subclass == array element superclass)
1214 
1215      __ cmp(r7, r1);
1216      __ br(Assembler::EQ, is_type_ok);
1217 
1218      __ b(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1219 
1220      __ bind(is_type_ok);
1221     // r1: value's klass
1222     // r3: array
1223     // r5: array klass
1224     __ test_klass_is_empty_inline_type(r1, r7, done);
1225 
1226     // calc dst for copy
1227     __ ldrw(r7, at_tos_p1()); // index
1228     __ data_for_value_array_index(r3, r5, r7, r7);
1229 
1230     // ...and src for copy
1231     __ ldr(r6, at_tos());  // value
1232     __ data_for_oop(r6, r6, r1);
1233 
1234     __ mov(r4, r1);  // Shuffle arguments to avoid conflict with c_rarg1
1235     __ access_value_copy(IN_HEAP, r6, r7, r4);
1236   }
1237 
1238   // Pop stack arguments
1239   __ bind(done);
1240   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1241 }
1242 
1243 void TemplateTable::bastore()
1244 {
1245   transition(itos, vtos);
1246   __ pop_i(r1);
1247   __ pop_ptr(r3);
1248   // r0: value
1249   // r1: index
1250   // r3: array
1251   index_check(r3, r1); // prefer index in r1
1252 
1253   // Need to check whether array is boolean or byte
1254   // since both types share the bastore bytecode.
1255   __ load_klass(r2, r3);
1256   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1257   int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1258   Label L_skip;
1259   __ tbz(r2, diffbit_index, L_skip);
1260   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1261   __ bind(L_skip);
1262 
1263   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
1264   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(0)), r0, noreg, noreg, noreg);
1265 }
1266 
1267 void TemplateTable::castore()
1268 {
1269   transition(itos, vtos);
1270   __ pop_i(r1);
1271   __ pop_ptr(r3);
1272   // r0: value
1273   // r1: index
1274   // r3: array
1275   index_check(r3, r1); // prefer index in r1
1276   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
1277   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(1)), r0, noreg, noreg, noreg);
1278 }
1279 
1280 void TemplateTable::sastore()
1281 {
1282   castore();
1283 }
1284 
1285 void TemplateTable::istore(int n)
1286 {
1287   transition(itos, vtos);
1288   __ str(r0, iaddress(n));
1289 }
1290 
1291 void TemplateTable::lstore(int n)
1292 {
1293   transition(ltos, vtos);
1294   __ str(r0, laddress(n));
1295 }
1296 
1297 void TemplateTable::fstore(int n)
1298 {
1299   transition(ftos, vtos);
1300   __ strs(v0, faddress(n));
1301 }
1302 
1303 void TemplateTable::dstore(int n)
1304 {
1305   transition(dtos, vtos);
1306   __ strd(v0, daddress(n));
1307 }
1308 
1309 void TemplateTable::astore(int n)
1310 {
1311   transition(vtos, vtos);
1312   __ pop_ptr(r0);
1313   __ str(r0, iaddress(n));
1314 }
1315 
1316 void TemplateTable::pop()
1317 {
1318   transition(vtos, vtos);
1319   __ add(esp, esp, Interpreter::stackElementSize);
1320 }
1321 
1322 void TemplateTable::pop2()
1323 {
1324   transition(vtos, vtos);
1325   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1326 }
1327 
1328 void TemplateTable::dup()
1329 {
1330   transition(vtos, vtos);
1331   __ ldr(r0, Address(esp, 0));
1332   __ push(r0);
1333   // stack: ..., a, a
1334 }
1335 
1336 void TemplateTable::dup_x1()
1337 {
1338   transition(vtos, vtos);
1339   // stack: ..., a, b
1340   __ ldr(r0, at_tos());  // load b
1341   __ ldr(r2, at_tos_p1());  // load a
1342   __ str(r0, at_tos_p1());  // store b
1343   __ str(r2, at_tos());  // store a
1344   __ push(r0);                  // push b
1345   // stack: ..., b, a, b
1346 }
1347 
1348 void TemplateTable::dup_x2()
1349 {
1350   transition(vtos, vtos);
1351   // stack: ..., a, b, c
1352   __ ldr(r0, at_tos());  // load c
1353   __ ldr(r2, at_tos_p2());  // load a
1354   __ str(r0, at_tos_p2());  // store c in a
1355   __ push(r0);      // push c
1356   // stack: ..., c, b, c, c
1357   __ ldr(r0, at_tos_p2());  // load b
1358   __ str(r2, at_tos_p2());  // store a in b
1359   // stack: ..., c, a, c, c
1360   __ str(r0, at_tos_p1());  // store b in c
1361   // stack: ..., c, a, b, c
1362 }
1363 
1364 void TemplateTable::dup2()
1365 {
1366   transition(vtos, vtos);
1367   // stack: ..., a, b
1368   __ ldr(r0, at_tos_p1());  // load a
1369   __ push(r0);                  // push a
1370   __ ldr(r0, at_tos_p1());  // load b
1371   __ push(r0);                  // push b
1372   // stack: ..., a, b, a, b
1373 }
1374 
1375 void TemplateTable::dup2_x1()
1376 {
1377   transition(vtos, vtos);
1378   // stack: ..., a, b, c
1379   __ ldr(r2, at_tos());  // load c
1380   __ ldr(r0, at_tos_p1());  // load b
1381   __ push(r0);                  // push b
1382   __ push(r2);                  // push c
1383   // stack: ..., a, b, c, b, c
1384   __ str(r2, at_tos_p3());  // store c in b
1385   // stack: ..., a, c, c, b, c
1386   __ ldr(r2, at_tos_p4());  // load a
1387   __ str(r2, at_tos_p2());  // store a in 2nd c
1388   // stack: ..., a, c, a, b, c
1389   __ str(r0, at_tos_p4());  // store b in a
1390   // stack: ..., b, c, a, b, c
1391 }
1392 
1393 void TemplateTable::dup2_x2()
1394 {
1395   transition(vtos, vtos);
1396   // stack: ..., a, b, c, d
1397   __ ldr(r2, at_tos());  // load d
1398   __ ldr(r0, at_tos_p1());  // load c
1399   __ push(r0)            ;      // push c
1400   __ push(r2);                  // push d
1401   // stack: ..., a, b, c, d, c, d
1402   __ ldr(r0, at_tos_p4());  // load b
1403   __ str(r0, at_tos_p2());  // store b in d
1404   __ str(r2, at_tos_p4());  // store d in b
1405   // stack: ..., a, d, c, b, c, d
1406   __ ldr(r2, at_tos_p5());  // load a
1407   __ ldr(r0, at_tos_p3());  // load c
1408   __ str(r2, at_tos_p3());  // store a in c
1409   __ str(r0, at_tos_p5());  // store c in a
1410   // stack: ..., c, d, a, b, c, d
1411 }
1412 
1413 void TemplateTable::swap()
1414 {
1415   transition(vtos, vtos);
1416   // stack: ..., a, b
1417   __ ldr(r2, at_tos_p1());  // load a
1418   __ ldr(r0, at_tos());  // load b
1419   __ str(r2, at_tos());  // store a in b
1420   __ str(r0, at_tos_p1());  // store b in a
1421   // stack: ..., b, a
1422 }
1423 
1424 void TemplateTable::iop2(Operation op)
1425 {
1426   transition(itos, itos);
1427   // r0 <== r1 op r0
1428   __ pop_i(r1);
1429   switch (op) {
1430   case add  : __ addw(r0, r1, r0); break;
1431   case sub  : __ subw(r0, r1, r0); break;
1432   case mul  : __ mulw(r0, r1, r0); break;
1433   case _and : __ andw(r0, r1, r0); break;
1434   case _or  : __ orrw(r0, r1, r0); break;
1435   case _xor : __ eorw(r0, r1, r0); break;
1436   case shl  : __ lslvw(r0, r1, r0); break;
1437   case shr  : __ asrvw(r0, r1, r0); break;
1438   case ushr : __ lsrvw(r0, r1, r0);break;
1439   default   : ShouldNotReachHere();
1440   }
1441 }
1442 
1443 void TemplateTable::lop2(Operation op)
1444 {
1445   transition(ltos, ltos);
1446   // r0 <== r1 op r0
1447   __ pop_l(r1);
1448   switch (op) {
1449   case add  : __ add(r0, r1, r0); break;
1450   case sub  : __ sub(r0, r1, r0); break;
1451   case mul  : __ mul(r0, r1, r0); break;
1452   case _and : __ andr(r0, r1, r0); break;
1453   case _or  : __ orr(r0, r1, r0); break;
1454   case _xor : __ eor(r0, r1, r0); break;
1455   default   : ShouldNotReachHere();
1456   }
1457 }
1458 
1459 void TemplateTable::idiv()
1460 {
1461   transition(itos, itos);
1462   // explicitly check for div0
1463   Label no_div0;
1464   __ cbnzw(r0, no_div0);
1465   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1466   __ br(rscratch1);
1467   __ bind(no_div0);
1468   __ pop_i(r1);
1469   // r0 <== r1 idiv r0
1470   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1471 }
1472 
1473 void TemplateTable::irem()
1474 {
1475   transition(itos, itos);
1476   // explicitly check for div0
1477   Label no_div0;
1478   __ cbnzw(r0, no_div0);
1479   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1480   __ br(rscratch1);
1481   __ bind(no_div0);
1482   __ pop_i(r1);
1483   // r0 <== r1 irem r0
1484   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1485 }
1486 
1487 void TemplateTable::lmul()
1488 {
1489   transition(ltos, ltos);
1490   __ pop_l(r1);
1491   __ mul(r0, r0, r1);
1492 }
1493 
1494 void TemplateTable::ldiv()
1495 {
1496   transition(ltos, ltos);
1497   // explicitly check for div0
1498   Label no_div0;
1499   __ cbnz(r0, no_div0);
1500   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1501   __ br(rscratch1);
1502   __ bind(no_div0);
1503   __ pop_l(r1);
1504   // r0 <== r1 ldiv r0
1505   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1506 }
1507 
1508 void TemplateTable::lrem()
1509 {
1510   transition(ltos, ltos);
1511   // explicitly check for div0
1512   Label no_div0;
1513   __ cbnz(r0, no_div0);
1514   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1515   __ br(rscratch1);
1516   __ bind(no_div0);
1517   __ pop_l(r1);
1518   // r0 <== r1 lrem r0
1519   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1520 }
1521 
1522 void TemplateTable::lshl()
1523 {
1524   transition(itos, ltos);
1525   // shift count is in r0
1526   __ pop_l(r1);
1527   __ lslv(r0, r1, r0);
1528 }
1529 
1530 void TemplateTable::lshr()
1531 {
1532   transition(itos, ltos);
1533   // shift count is in r0
1534   __ pop_l(r1);
1535   __ asrv(r0, r1, r0);
1536 }
1537 
1538 void TemplateTable::lushr()
1539 {
1540   transition(itos, ltos);
1541   // shift count is in r0
1542   __ pop_l(r1);
1543   __ lsrv(r0, r1, r0);
1544 }
1545 
1546 void TemplateTable::fop2(Operation op)
1547 {
1548   transition(ftos, ftos);
1549   switch (op) {
1550   case add:
1551     // n.b. use ldrd because this is a 64 bit slot
1552     __ pop_f(v1);
1553     __ fadds(v0, v1, v0);
1554     break;
1555   case sub:
1556     __ pop_f(v1);
1557     __ fsubs(v0, v1, v0);
1558     break;
1559   case mul:
1560     __ pop_f(v1);
1561     __ fmuls(v0, v1, v0);
1562     break;
1563   case div:
1564     __ pop_f(v1);
1565     __ fdivs(v0, v1, v0);
1566     break;
1567   case rem:
1568     __ fmovs(v1, v0);
1569     __ pop_f(v0);
1570     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1571     break;
1572   default:
1573     ShouldNotReachHere();
1574     break;
1575   }
1576 }
1577 
1578 void TemplateTable::dop2(Operation op)
1579 {
1580   transition(dtos, dtos);
1581   switch (op) {
1582   case add:
1583     // n.b. use ldrd because this is a 64 bit slot
1584     __ pop_d(v1);
1585     __ faddd(v0, v1, v0);
1586     break;
1587   case sub:
1588     __ pop_d(v1);
1589     __ fsubd(v0, v1, v0);
1590     break;
1591   case mul:
1592     __ pop_d(v1);
1593     __ fmuld(v0, v1, v0);
1594     break;
1595   case div:
1596     __ pop_d(v1);
1597     __ fdivd(v0, v1, v0);
1598     break;
1599   case rem:
1600     __ fmovd(v1, v0);
1601     __ pop_d(v0);
1602     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1603     break;
1604   default:
1605     ShouldNotReachHere();
1606     break;
1607   }
1608 }
1609 
1610 void TemplateTable::ineg()
1611 {
1612   transition(itos, itos);
1613   __ negw(r0, r0);
1614 
1615 }
1616 
1617 void TemplateTable::lneg()
1618 {
1619   transition(ltos, ltos);
1620   __ neg(r0, r0);
1621 }
1622 
1623 void TemplateTable::fneg()
1624 {
1625   transition(ftos, ftos);
1626   __ fnegs(v0, v0);
1627 }
1628 
1629 void TemplateTable::dneg()
1630 {
1631   transition(dtos, dtos);
1632   __ fnegd(v0, v0);
1633 }
1634 
1635 void TemplateTable::iinc()
1636 {
1637   transition(vtos, vtos);
1638   __ load_signed_byte(r1, at_bcp(2)); // get constant
1639   locals_index(r2);
1640   __ ldr(r0, iaddress(r2));
1641   __ addw(r0, r0, r1);
1642   __ str(r0, iaddress(r2));
1643 }
1644 
1645 void TemplateTable::wide_iinc()
1646 {
1647   transition(vtos, vtos);
1648   // __ mov(r1, zr);
1649   __ ldrw(r1, at_bcp(2)); // get constant and index
1650   __ rev16(r1, r1);
1651   __ ubfx(r2, r1, 0, 16);
1652   __ neg(r2, r2);
1653   __ sbfx(r1, r1, 16, 16);
1654   __ ldr(r0, iaddress(r2));
1655   __ addw(r0, r0, r1);
1656   __ str(r0, iaddress(r2));
1657 }
1658 
1659 void TemplateTable::convert()
1660 {
1661   // Checking
1662 #ifdef ASSERT
1663   {
1664     TosState tos_in  = ilgl;
1665     TosState tos_out = ilgl;
1666     switch (bytecode()) {
1667     case Bytecodes::_i2l: // fall through
1668     case Bytecodes::_i2f: // fall through
1669     case Bytecodes::_i2d: // fall through
1670     case Bytecodes::_i2b: // fall through
1671     case Bytecodes::_i2c: // fall through
1672     case Bytecodes::_i2s: tos_in = itos; break;
1673     case Bytecodes::_l2i: // fall through
1674     case Bytecodes::_l2f: // fall through
1675     case Bytecodes::_l2d: tos_in = ltos; break;
1676     case Bytecodes::_f2i: // fall through
1677     case Bytecodes::_f2l: // fall through
1678     case Bytecodes::_f2d: tos_in = ftos; break;
1679     case Bytecodes::_d2i: // fall through
1680     case Bytecodes::_d2l: // fall through
1681     case Bytecodes::_d2f: tos_in = dtos; break;
1682     default             : ShouldNotReachHere();
1683     }
1684     switch (bytecode()) {
1685     case Bytecodes::_l2i: // fall through
1686     case Bytecodes::_f2i: // fall through
1687     case Bytecodes::_d2i: // fall through
1688     case Bytecodes::_i2b: // fall through
1689     case Bytecodes::_i2c: // fall through
1690     case Bytecodes::_i2s: tos_out = itos; break;
1691     case Bytecodes::_i2l: // fall through
1692     case Bytecodes::_f2l: // fall through
1693     case Bytecodes::_d2l: tos_out = ltos; break;
1694     case Bytecodes::_i2f: // fall through
1695     case Bytecodes::_l2f: // fall through
1696     case Bytecodes::_d2f: tos_out = ftos; break;
1697     case Bytecodes::_i2d: // fall through
1698     case Bytecodes::_l2d: // fall through
1699     case Bytecodes::_f2d: tos_out = dtos; break;
1700     default             : ShouldNotReachHere();
1701     }
1702     transition(tos_in, tos_out);
1703   }
1704 #endif // ASSERT
1705   // static const int64_t is_nan = 0x8000000000000000L;
1706 
1707   // Conversion
1708   switch (bytecode()) {
1709   case Bytecodes::_i2l:
1710     __ sxtw(r0, r0);
1711     break;
1712   case Bytecodes::_i2f:
1713     __ scvtfws(v0, r0);
1714     break;
1715   case Bytecodes::_i2d:
1716     __ scvtfwd(v0, r0);
1717     break;
1718   case Bytecodes::_i2b:
1719     __ sxtbw(r0, r0);
1720     break;
1721   case Bytecodes::_i2c:
1722     __ uxthw(r0, r0);
1723     break;
1724   case Bytecodes::_i2s:
1725     __ sxthw(r0, r0);
1726     break;
1727   case Bytecodes::_l2i:
1728     __ uxtw(r0, r0);
1729     break;
1730   case Bytecodes::_l2f:
1731     __ scvtfs(v0, r0);
1732     break;
1733   case Bytecodes::_l2d:
1734     __ scvtfd(v0, r0);
1735     break;
1736   case Bytecodes::_f2i:
1737   {
1738     Label L_Okay;
1739     __ clear_fpsr();
1740     __ fcvtzsw(r0, v0);
1741     __ get_fpsr(r1);
1742     __ cbzw(r1, L_Okay);
1743     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i));
1744     __ bind(L_Okay);
1745   }
1746     break;
1747   case Bytecodes::_f2l:
1748   {
1749     Label L_Okay;
1750     __ clear_fpsr();
1751     __ fcvtzs(r0, v0);
1752     __ get_fpsr(r1);
1753     __ cbzw(r1, L_Okay);
1754     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1755     __ bind(L_Okay);
1756   }
1757     break;
1758   case Bytecodes::_f2d:
1759     __ fcvts(v0, v0);
1760     break;
1761   case Bytecodes::_d2i:
1762   {
1763     Label L_Okay;
1764     __ clear_fpsr();
1765     __ fcvtzdw(r0, v0);
1766     __ get_fpsr(r1);
1767     __ cbzw(r1, L_Okay);
1768     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1769     __ bind(L_Okay);
1770   }
1771     break;
1772   case Bytecodes::_d2l:
1773   {
1774     Label L_Okay;
1775     __ clear_fpsr();
1776     __ fcvtzd(r0, v0);
1777     __ get_fpsr(r1);
1778     __ cbzw(r1, L_Okay);
1779     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1780     __ bind(L_Okay);
1781   }
1782     break;
1783   case Bytecodes::_d2f:
1784     __ fcvtd(v0, v0);
1785     break;
1786   default:
1787     ShouldNotReachHere();
1788   }
1789 }
1790 
1791 void TemplateTable::lcmp()
1792 {
1793   transition(ltos, itos);
1794   Label done;
1795   __ pop_l(r1);
1796   __ cmp(r1, r0);
1797   __ mov(r0, (uint64_t)-1L);
1798   __ br(Assembler::LT, done);
1799   // __ mov(r0, 1UL);
1800   // __ csel(r0, r0, zr, Assembler::NE);
1801   // and here is a faster way
1802   __ csinc(r0, zr, zr, Assembler::EQ);
1803   __ bind(done);
1804 }
1805 
1806 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1807 {
1808   Label done;
1809   if (is_float) {
1810     // XXX get rid of pop here, use ... reg, mem32
1811     __ pop_f(v1);
1812     __ fcmps(v1, v0);
1813   } else {
1814     // XXX get rid of pop here, use ... reg, mem64
1815     __ pop_d(v1);
1816     __ fcmpd(v1, v0);
1817   }
1818   if (unordered_result < 0) {
1819     // we want -1 for unordered or less than, 0 for equal and 1 for
1820     // greater than.
1821     __ mov(r0, (uint64_t)-1L);
1822     // for FP LT tests less than or unordered
1823     __ br(Assembler::LT, done);
1824     // install 0 for EQ otherwise 1
1825     __ csinc(r0, zr, zr, Assembler::EQ);
1826   } else {
1827     // we want -1 for less than, 0 for equal and 1 for unordered or
1828     // greater than.
1829     __ mov(r0, 1L);
1830     // for FP HI tests greater than or unordered
1831     __ br(Assembler::HI, done);
1832     // install 0 for EQ otherwise ~0
1833     __ csinv(r0, zr, zr, Assembler::EQ);
1834 
1835   }
1836   __ bind(done);
1837 }
1838 
1839 void TemplateTable::branch(bool is_jsr, bool is_wide)
1840 {
1841   __ profile_taken_branch(r0, r1);
1842   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1843                              InvocationCounter::counter_offset();
1844   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1845                               InvocationCounter::counter_offset();
1846 
1847   // load branch displacement
1848   if (!is_wide) {
1849     __ ldrh(r2, at_bcp(1));
1850     __ rev16(r2, r2);
1851     // sign extend the 16 bit value in r2
1852     __ sbfm(r2, r2, 0, 15);
1853   } else {
1854     __ ldrw(r2, at_bcp(1));
1855     __ revw(r2, r2);
1856     // sign extend the 32 bit value in r2
1857     __ sbfm(r2, r2, 0, 31);
1858   }
1859 
1860   // Handle all the JSR stuff here, then exit.
1861   // It's much shorter and cleaner than intermingling with the non-JSR
1862   // normal-branch stuff occurring below.
1863 
1864   if (is_jsr) {
1865     // Pre-load the next target bytecode into rscratch1
1866     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1867     // compute return address as bci
1868     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1869     __ add(rscratch2, rscratch2,
1870            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1871     __ sub(r1, rbcp, rscratch2);
1872     __ push_i(r1);
1873     // Adjust the bcp by the 16-bit displacement in r2
1874     __ add(rbcp, rbcp, r2);
1875     __ dispatch_only(vtos, /*generate_poll*/true);
1876     return;
1877   }
1878 
1879   // Normal (non-jsr) branch handling
1880 
1881   // Adjust the bcp by the displacement in r2
1882   __ add(rbcp, rbcp, r2);
1883 
1884   assert(UseLoopCounter || !UseOnStackReplacement,
1885          "on-stack-replacement requires loop counters");
1886   Label backedge_counter_overflow;
1887   Label dispatch;
1888   if (UseLoopCounter) {
1889     // increment backedge counter for backward branches
1890     // r0: MDO
1891     // w1: MDO bumped taken-count
1892     // r2: target offset
1893     __ cmp(r2, zr);
1894     __ br(Assembler::GT, dispatch); // count only if backward branch
1895 
1896     // ECN: FIXME: This code smells
1897     // check if MethodCounters exists
1898     Label has_counters;
1899     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1900     __ cbnz(rscratch1, has_counters);
1901     __ push(r0);
1902     __ push(r1);
1903     __ push(r2);
1904     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1905             InterpreterRuntime::build_method_counters), rmethod);
1906     __ pop(r2);
1907     __ pop(r1);
1908     __ pop(r0);
1909     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1910     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1911     __ bind(has_counters);
1912 
1913     Label no_mdo;
1914     int increment = InvocationCounter::count_increment;
1915     if (ProfileInterpreter) {
1916       // Are we profiling?
1917       __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1918       __ cbz(r1, no_mdo);
1919       // Increment the MDO backedge counter
1920       const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1921                                          in_bytes(InvocationCounter::counter_offset()));
1922       const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1923       __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1924                                  r0, rscratch1, false, Assembler::EQ,
1925                                  UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1926       __ b(dispatch);
1927     }
1928     __ bind(no_mdo);
1929     // Increment backedge counter in MethodCounters*
1930     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1931     const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1932     __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1933                                r0, rscratch2, false, Assembler::EQ,
1934                                UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1935     __ bind(dispatch);
1936   }
1937 
1938   // Pre-load the next target bytecode into rscratch1
1939   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1940 
1941   // continue with the bytecode @ target
1942   // rscratch1: target bytecode
1943   // rbcp: target bcp
1944   __ dispatch_only(vtos, /*generate_poll*/true);
1945 
1946   if (UseLoopCounter && UseOnStackReplacement) {
1947     // invocation counter overflow
1948     __ bind(backedge_counter_overflow);
1949     __ neg(r2, r2);
1950     __ add(r2, r2, rbcp);     // branch bcp
1951     // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1952     __ call_VM(noreg,
1953                CAST_FROM_FN_PTR(address,
1954                                 InterpreterRuntime::frequency_counter_overflow),
1955                r2);
1956     __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1957 
1958     // r0: osr nmethod (osr ok) or null (osr not possible)
1959     // w1: target bytecode
1960     // r2: scratch
1961     __ cbz(r0, dispatch);     // test result -- no osr if null
1962     // nmethod may have been invalidated (VM may block upon call_VM return)
1963     __ ldrb(r2, Address(r0, nmethod::state_offset()));
1964     if (nmethod::in_use != 0)
1965       __ sub(r2, r2, nmethod::in_use);
1966     __ cbnz(r2, dispatch);
1967 
1968     // We have the address of an on stack replacement routine in r0
1969     // We need to prepare to execute the OSR method. First we must
1970     // migrate the locals and monitors off of the stack.
1971 
1972     __ mov(r19, r0);                             // save the nmethod
1973 
1974     call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1975 
1976     // r0 is OSR buffer, move it to expected parameter location
1977     __ mov(j_rarg0, r0);
1978 
1979     // remove activation
1980     // get sender esp
1981     __ ldr(esp,
1982         Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1983     // remove frame anchor
1984     __ leave();
1985     // Ensure compiled code always sees stack at proper alignment
1986     __ andr(sp, esp, -16);
1987 
1988     // and begin the OSR nmethod
1989     __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1990     __ br(rscratch1);
1991   }
1992 }
1993 
1994 
1995 void TemplateTable::if_0cmp(Condition cc)
1996 {
1997   transition(itos, vtos);
1998   // assume branch is more often taken than not (loops use backward branches)
1999   Label not_taken;
2000   if (cc == equal)
2001     __ cbnzw(r0, not_taken);
2002   else if (cc == not_equal)
2003     __ cbzw(r0, not_taken);
2004   else {
2005     __ andsw(zr, r0, r0);
2006     __ br(j_not(cc), not_taken);
2007   }
2008 
2009   branch(false, false);
2010   __ bind(not_taken);
2011   __ profile_not_taken_branch(r0);
2012 }
2013 
2014 void TemplateTable::if_icmp(Condition cc)
2015 {
2016   transition(itos, vtos);
2017   // assume branch is more often taken than not (loops use backward branches)
2018   Label not_taken;
2019   __ pop_i(r1);
2020   __ cmpw(r1, r0, Assembler::LSL);
2021   __ br(j_not(cc), not_taken);
2022   branch(false, false);
2023   __ bind(not_taken);
2024   __ profile_not_taken_branch(r0);
2025 }
2026 
2027 void TemplateTable::if_nullcmp(Condition cc)
2028 {
2029   transition(atos, vtos);
2030   // assume branch is more often taken than not (loops use backward branches)
2031   Label not_taken;
2032   if (cc == equal)
2033     __ cbnz(r0, not_taken);
2034   else
2035     __ cbz(r0, not_taken);
2036   branch(false, false);
2037   __ bind(not_taken);
2038   __ profile_not_taken_branch(r0);
2039 }
2040 
2041 void TemplateTable::if_acmp(Condition cc) {
2042   transition(atos, vtos);
2043   // assume branch is more often taken than not (loops use backward branches)
2044   Label taken, not_taken;
2045   __ pop_ptr(r1);
2046 
2047   __ profile_acmp(r2, r1, r0, r4);
2048 
2049   Register is_inline_type_mask = rscratch1;
2050   __ mov(is_inline_type_mask, markWord::inline_type_pattern);
2051 
2052   if (EnableValhalla) {
2053     __ cmp(r1, r0);
2054     __ br(Assembler::EQ, (cc == equal) ? taken : not_taken);
2055 
2056     // might be substitutable, test if either r0 or r1 is null
2057     __ andr(r2, r0, r1);
2058     __ cbz(r2, (cc == equal) ? not_taken : taken);
2059 
2060     // and both are values ?
2061     __ ldr(r2, Address(r1, oopDesc::mark_offset_in_bytes()));
2062     __ andr(r2, r2, is_inline_type_mask);
2063     __ ldr(r4, Address(r0, oopDesc::mark_offset_in_bytes()));
2064     __ andr(r4, r4, is_inline_type_mask);
2065     __ andr(r2, r2, r4);
2066     __ cmp(r2,  is_inline_type_mask);
2067     __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2068 
2069     // same value klass ?
2070     __ load_metadata(r2, r1);
2071     __ load_metadata(r4, r0);
2072     __ cmp(r2, r4);
2073     __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2074 
2075     // Know both are the same type, let's test for substitutability...
2076     if (cc == equal) {
2077       invoke_is_substitutable(r0, r1, taken, not_taken);
2078     } else {
2079       invoke_is_substitutable(r0, r1, not_taken, taken);
2080     }
2081     __ stop("Not reachable");
2082   }
2083 
2084   __ cmpoop(r1, r0);
2085   __ br(j_not(cc), not_taken);
2086   __ bind(taken);
2087   branch(false, false);
2088   __ bind(not_taken);
2089   __ profile_not_taken_branch(r0, true);
2090 }
2091 
2092 void TemplateTable::invoke_is_substitutable(Register aobj, Register bobj,
2093                                             Label& is_subst, Label& not_subst) {
2094 
2095   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::is_substitutable), aobj, bobj);
2096   // Restored... r0 answer, jmp to outcome...
2097   __ cbz(r0, not_subst);
2098   __ b(is_subst);
2099 }
2100 
2101 
2102 void TemplateTable::ret() {
2103   transition(vtos, vtos);
2104   locals_index(r1);
2105   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2106   __ profile_ret(r1, r2);
2107   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2108   __ lea(rbcp, Address(rbcp, r1));
2109   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2110   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2111 }
2112 
2113 void TemplateTable::wide_ret() {
2114   transition(vtos, vtos);
2115   locals_index_wide(r1);
2116   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2117   __ profile_ret(r1, r2);
2118   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2119   __ lea(rbcp, Address(rbcp, r1));
2120   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2121   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2122 }
2123 
2124 
2125 void TemplateTable::tableswitch() {
2126   Label default_case, continue_execution;
2127   transition(itos, vtos);
2128   // align rbcp
2129   __ lea(r1, at_bcp(BytesPerInt));
2130   __ andr(r1, r1, -BytesPerInt);
2131   // load lo & hi
2132   __ ldrw(r2, Address(r1, BytesPerInt));
2133   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2134   __ rev32(r2, r2);
2135   __ rev32(r3, r3);
2136   // check against lo & hi
2137   __ cmpw(r0, r2);
2138   __ br(Assembler::LT, default_case);
2139   __ cmpw(r0, r3);
2140   __ br(Assembler::GT, default_case);
2141   // lookup dispatch offset
2142   __ subw(r0, r0, r2);
2143   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2144   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2145   __ profile_switch_case(r0, r1, r2);
2146   // continue execution
2147   __ bind(continue_execution);
2148   __ rev32(r3, r3);
2149   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2150   __ add(rbcp, rbcp, r3, ext::sxtw);
2151   __ dispatch_only(vtos, /*generate_poll*/true);
2152   // handle default
2153   __ bind(default_case);
2154   __ profile_switch_default(r0);
2155   __ ldrw(r3, Address(r1, 0));
2156   __ b(continue_execution);
2157 }
2158 
2159 void TemplateTable::lookupswitch() {
2160   transition(itos, itos);
2161   __ stop("lookupswitch bytecode should have been rewritten");
2162 }
2163 
2164 void TemplateTable::fast_linearswitch() {
2165   transition(itos, vtos);
2166   Label loop_entry, loop, found, continue_execution;
2167   // bswap r0 so we can avoid bswapping the table entries
2168   __ rev32(r0, r0);
2169   // align rbcp
2170   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2171                                     // this instruction (change offsets
2172                                     // below)
2173   __ andr(r19, r19, -BytesPerInt);
2174   // set counter
2175   __ ldrw(r1, Address(r19, BytesPerInt));
2176   __ rev32(r1, r1);
2177   __ b(loop_entry);
2178   // table search
2179   __ bind(loop);
2180   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2181   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2182   __ cmpw(r0, rscratch1);
2183   __ br(Assembler::EQ, found);
2184   __ bind(loop_entry);
2185   __ subs(r1, r1, 1);
2186   __ br(Assembler::PL, loop);
2187   // default case
2188   __ profile_switch_default(r0);
2189   __ ldrw(r3, Address(r19, 0));
2190   __ b(continue_execution);
2191   // entry found -> get offset
2192   __ bind(found);
2193   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2194   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2195   __ profile_switch_case(r1, r0, r19);
2196   // continue execution
2197   __ bind(continue_execution);
2198   __ rev32(r3, r3);
2199   __ add(rbcp, rbcp, r3, ext::sxtw);
2200   __ ldrb(rscratch1, Address(rbcp, 0));
2201   __ dispatch_only(vtos, /*generate_poll*/true);
2202 }
2203 
2204 void TemplateTable::fast_binaryswitch() {
2205   transition(itos, vtos);
2206   // Implementation using the following core algorithm:
2207   //
2208   // int binary_search(int key, LookupswitchPair* array, int n) {
2209   //   // Binary search according to "Methodik des Programmierens" by
2210   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2211   //   int i = 0;
2212   //   int j = n;
2213   //   while (i+1 < j) {
2214   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2215   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2216   //     // where a stands for the array and assuming that the (inexisting)
2217   //     // element a[n] is infinitely big.
2218   //     int h = (i + j) >> 1;
2219   //     // i < h < j
2220   //     if (key < array[h].fast_match()) {
2221   //       j = h;
2222   //     } else {
2223   //       i = h;
2224   //     }
2225   //   }
2226   //   // R: a[i] <= key < a[i+1] or Q
2227   //   // (i.e., if key is within array, i is the correct index)
2228   //   return i;
2229   // }
2230 
2231   // Register allocation
2232   const Register key   = r0; // already set (tosca)
2233   const Register array = r1;
2234   const Register i     = r2;
2235   const Register j     = r3;
2236   const Register h     = rscratch1;
2237   const Register temp  = rscratch2;
2238 
2239   // Find array start
2240   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2241                                           // get rid of this
2242                                           // instruction (change
2243                                           // offsets below)
2244   __ andr(array, array, -BytesPerInt);
2245 
2246   // Initialize i & j
2247   __ mov(i, 0);                            // i = 0;
2248   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2249 
2250   // Convert j into native byteordering
2251   __ rev32(j, j);
2252 
2253   // And start
2254   Label entry;
2255   __ b(entry);
2256 
2257   // binary search loop
2258   {
2259     Label loop;
2260     __ bind(loop);
2261     // int h = (i + j) >> 1;
2262     __ addw(h, i, j);                           // h = i + j;
2263     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2264     // if (key < array[h].fast_match()) {
2265     //   j = h;
2266     // } else {
2267     //   i = h;
2268     // }
2269     // Convert array[h].match to native byte-ordering before compare
2270     __ ldr(temp, Address(array, h, Address::lsl(3)));
2271     __ rev32(temp, temp);
2272     __ cmpw(key, temp);
2273     // j = h if (key <  array[h].fast_match())
2274     __ csel(j, h, j, Assembler::LT);
2275     // i = h if (key >= array[h].fast_match())
2276     __ csel(i, h, i, Assembler::GE);
2277     // while (i+1 < j)
2278     __ bind(entry);
2279     __ addw(h, i, 1);          // i+1
2280     __ cmpw(h, j);             // i+1 < j
2281     __ br(Assembler::LT, loop);
2282   }
2283 
2284   // end of binary search, result index is i (must check again!)
2285   Label default_case;
2286   // Convert array[i].match to native byte-ordering before compare
2287   __ ldr(temp, Address(array, i, Address::lsl(3)));
2288   __ rev32(temp, temp);
2289   __ cmpw(key, temp);
2290   __ br(Assembler::NE, default_case);
2291 
2292   // entry found -> j = offset
2293   __ add(j, array, i, ext::uxtx, 3);
2294   __ ldrw(j, Address(j, BytesPerInt));
2295   __ profile_switch_case(i, key, array);
2296   __ rev32(j, j);
2297   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2298   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2299   __ dispatch_only(vtos, /*generate_poll*/true);
2300 
2301   // default case -> j = default offset
2302   __ bind(default_case);
2303   __ profile_switch_default(i);
2304   __ ldrw(j, Address(array, -2 * BytesPerInt));
2305   __ rev32(j, j);
2306   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2307   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2308   __ dispatch_only(vtos, /*generate_poll*/true);
2309 }
2310 
2311 
2312 void TemplateTable::_return(TosState state)
2313 {
2314   transition(state, state);
2315   assert(_desc->calls_vm(),
2316          "inconsistent calls_vm information"); // call in remove_activation
2317 
2318   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2319     assert(state == vtos, "only valid state");
2320 
2321     __ ldr(c_rarg1, aaddress(0));
2322     __ load_klass(r3, c_rarg1);
2323     __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2324     Label skip_register_finalizer;
2325     __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2326 
2327     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2328 
2329     __ bind(skip_register_finalizer);
2330   }
2331 
2332   // Issue a StoreStore barrier after all stores but before return
2333   // from any constructor for any class with a final field.  We don't
2334   // know if this is a finalizer, so we always do so.
2335   if (_desc->bytecode() == Bytecodes::_return)
2336     __ membar(MacroAssembler::StoreStore);
2337 
2338   if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2339     Label no_safepoint;
2340     __ ldr(rscratch1, Address(rthread, JavaThread::polling_word_offset()));
2341     __ tbz(rscratch1, log2i_exact(SafepointMechanism::poll_bit()), no_safepoint);
2342     __ push(state);
2343     __ push_cont_fastpath(rthread);
2344     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint));
2345     __ pop_cont_fastpath(rthread);
2346     __ pop(state);
2347     __ bind(no_safepoint);
2348   }
2349 
2350   // Narrow result if state is itos but result type is smaller.
2351   // Need to narrow in the return bytecode rather than in generate_return_entry
2352   // since compiled code callers expect the result to already be narrowed.
2353   if (state == itos) {
2354     __ narrow(r0);
2355   }
2356 
2357   __ remove_activation(state);
2358   __ ret(lr);
2359 }
2360 
2361 // ----------------------------------------------------------------------------
2362 // Volatile variables demand their effects be made known to all CPU's
2363 // in order.  Store buffers on most chips allow reads & writes to
2364 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2365 // without some kind of memory barrier (i.e., it's not sufficient that
2366 // the interpreter does not reorder volatile references, the hardware
2367 // also must not reorder them).
2368 //
2369 // According to the new Java Memory Model (JMM):
2370 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2371 //     writes act as acquire & release, so:
2372 // (2) A read cannot let unrelated NON-volatile memory refs that
2373 //     happen after the read float up to before the read.  It's OK for
2374 //     non-volatile memory refs that happen before the volatile read to
2375 //     float down below it.
2376 // (3) Similar a volatile write cannot let unrelated NON-volatile
2377 //     memory refs that happen BEFORE the write float down to after the
2378 //     write.  It's OK for non-volatile memory refs that happen after the
2379 //     volatile write to float up before it.
2380 //
2381 // We only put in barriers around volatile refs (they are expensive),
2382 // not _between_ memory refs (that would require us to track the
2383 // flavor of the previous memory refs).  Requirements (2) and (3)
2384 // require some barriers before volatile stores and after volatile
2385 // loads.  These nearly cover requirement (1) but miss the
2386 // volatile-store-volatile-load case.  This final case is placed after
2387 // volatile-stores although it could just as well go before
2388 // volatile-loads.
2389 
2390 void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
2391                                             Register Rcache,
2392                                             Register index) {
2393   const Register temp = r19;
2394   assert_different_registers(Rcache, index, temp);
2395   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2396 
2397   Label resolved, clinit_barrier_slow;
2398 
2399   Bytecodes::Code code = bytecode();
2400   __ load_method_entry(Rcache, index);
2401   switch(byte_no) {
2402     case f1_byte:
2403       __ lea(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::bytecode1_offset())));
2404       break;
2405     case f2_byte:
2406       __ lea(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::bytecode2_offset())));
2407       break;
2408   }
2409   // Load-acquire the bytecode to match store-release in InterpreterRuntime
2410   __ ldarb(temp, temp);
2411   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2412   __ br(Assembler::EQ, resolved);
2413 
2414   // resolve first time through
2415   // Class initialization barrier slow path lands here as well.
2416   __ bind(clinit_barrier_slow);
2417   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2418   __ mov(temp, (int) code);
2419   __ call_VM(noreg, entry, temp);
2420 
2421   // Update registers with resolved info
2422   __ load_method_entry(Rcache, index);
2423   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2424   // so all clients ofthis method must be modified accordingly
2425   __ bind(resolved);
2426 
2427   // Class initialization barrier for static methods
2428   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2429     __ ldr(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::method_offset())));
2430     __ load_method_holder(temp, temp);
2431     __ clinit_barrier(temp, rscratch1, nullptr, &clinit_barrier_slow);
2432   }
2433 }
2434 
2435 void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
2436                                             Register Rcache,
2437                                             Register index) {
2438   const Register temp = r19;
2439   assert_different_registers(Rcache, index, temp);
2440 
2441   Label resolved;
2442 
2443   Bytecodes::Code code = bytecode();
2444   switch (code) {
2445   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2446   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2447   default: break;
2448   }
2449 
2450   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2451   __ load_field_entry(Rcache, index);
2452   if (byte_no == f1_byte) {
2453     __ lea(temp, Address(Rcache, in_bytes(ResolvedFieldEntry::get_code_offset())));
2454   } else {
2455     __ lea(temp, Address(Rcache, in_bytes(ResolvedFieldEntry::put_code_offset())));
2456   }
2457   // Load-acquire the bytecode to match store-release in ResolvedFieldEntry::fill_in()
2458   __ ldarb(temp, temp);
2459   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2460   __ br(Assembler::EQ, resolved);
2461 
2462   // resolve first time through
2463   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2464   __ mov(temp, (int) code);
2465   __ call_VM(noreg, entry, temp);
2466 
2467   // Update registers with resolved info
2468   __ load_field_entry(Rcache, index);
2469   __ bind(resolved);
2470 }
2471 
2472 void TemplateTable::load_resolved_field_entry(Register obj,
2473                                               Register cache,
2474                                               Register tos_state,
2475                                               Register offset,
2476                                               Register flags,
2477                                               bool is_static = false) {
2478   assert_different_registers(cache, tos_state, flags, offset);
2479 
2480   // Field offset
2481   __ load_sized_value(offset, Address(cache, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
2482 
2483   // Flags
2484   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedFieldEntry::flags_offset())));
2485 
2486   // TOS state
2487   if (tos_state != noreg) {
2488     __ load_unsigned_byte(tos_state, Address(cache, in_bytes(ResolvedFieldEntry::type_offset())));
2489   }
2490 
2491   // Klass overwrite register
2492   if (is_static) {
2493     __ ldr(obj, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2494     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2495     __ ldr(obj, Address(obj, mirror_offset));
2496     __ resolve_oop_handle(obj, r5, rscratch2);
2497   }
2498 }
2499 
2500 void TemplateTable::load_resolved_method_entry_special_or_static(Register cache,
2501                                                                  Register method,
2502                                                                  Register flags) {
2503 
2504   // setup registers
2505   const Register index = flags;
2506   assert_different_registers(method, cache, flags);
2507 
2508   // determine constant pool cache field offsets
2509   resolve_cache_and_index_for_method(f1_byte, cache, index);
2510   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2511   __ ldr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2512 }
2513 
2514 void TemplateTable::load_resolved_method_entry_handle(Register cache,
2515                                                       Register method,
2516                                                       Register ref_index,
2517                                                       Register flags) {
2518   // setup registers
2519   const Register index = ref_index;
2520   assert_different_registers(method, flags);
2521   assert_different_registers(method, cache, index);
2522 
2523   // determine constant pool cache field offsets
2524   resolve_cache_and_index_for_method(f1_byte, cache, index);
2525   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2526 
2527   // maybe push appendix to arguments (just before return address)
2528   Label L_no_push;
2529   __ tbz(flags, ResolvedMethodEntry::has_appendix_shift, L_no_push);
2530   // invokehandle uses an index into the resolved references array
2531   __ load_unsigned_short(ref_index, Address(cache, in_bytes(ResolvedMethodEntry::resolved_references_index_offset())));
2532   // Push the appendix as a trailing parameter.
2533   // This must be done before we get the receiver,
2534   // since the parameter_size includes it.
2535   Register appendix = method;
2536   __ load_resolved_reference_at_index(appendix, ref_index);
2537   __ push(appendix);  // push appendix (MethodType, CallSite, etc.)
2538   __ bind(L_no_push);
2539 
2540   __ ldr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2541 }
2542 
2543 void TemplateTable::load_resolved_method_entry_interface(Register cache,
2544                                                          Register klass,
2545                                                          Register method_or_table_index,
2546                                                          Register flags) {
2547   // setup registers
2548   const Register index = method_or_table_index;
2549   assert_different_registers(method_or_table_index, cache, flags);
2550 
2551   // determine constant pool cache field offsets
2552   resolve_cache_and_index_for_method(f1_byte, cache, index);
2553   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2554 
2555   // Invokeinterface can behave in different ways:
2556   // If calling a method from java.lang.Object, the forced virtual flag is true so the invocation will
2557   // behave like an invokevirtual call. The state of the virtual final flag will determine whether a method or
2558   // vtable index is placed in the register.
2559   // Otherwise, the registers will be populated with the klass and method.
2560 
2561   Label NotVirtual; Label NotVFinal; Label Done;
2562   __ tbz(flags, ResolvedMethodEntry::is_forced_virtual_shift, NotVirtual);
2563   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, NotVFinal);
2564   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2565   __ b(Done);
2566 
2567   __ bind(NotVFinal);
2568   __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2569   __ b(Done);
2570 
2571   __ bind(NotVirtual);
2572   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2573   __ ldr(klass, Address(cache, in_bytes(ResolvedMethodEntry::klass_offset())));
2574   __ bind(Done);
2575 }
2576 
2577 void TemplateTable::load_resolved_method_entry_virtual(Register cache,
2578                                                        Register method_or_table_index,
2579                                                        Register flags) {
2580   // setup registers
2581   const Register index = flags;
2582   assert_different_registers(method_or_table_index, cache, flags);
2583 
2584   // determine constant pool cache field offsets
2585   resolve_cache_and_index_for_method(f2_byte, cache, index);
2586   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2587 
2588   // method_or_table_index can either be an itable index or a method depending on the virtual final flag
2589   Label NotVFinal; Label Done;
2590   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, NotVFinal);
2591   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2592   __ b(Done);
2593 
2594   __ bind(NotVFinal);
2595   __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2596   __ bind(Done);
2597 }
2598 
2599 // The rmethod register is input and overwritten to be the adapter method for the
2600 // indy call. Link Register (lr) is set to the return address for the adapter and
2601 // an appendix may be pushed to the stack. Registers r0-r3 are clobbered
2602 void TemplateTable::load_invokedynamic_entry(Register method) {
2603   // setup registers
2604   const Register appendix = r0;
2605   const Register cache = r2;
2606   const Register index = r3;
2607   assert_different_registers(method, appendix, cache, index, rcpool);
2608 
2609   __ save_bcp();
2610 
2611   Label resolved;
2612 
2613   __ load_resolved_indy_entry(cache, index);
2614   // Load-acquire the adapter method to match store-release in ResolvedIndyEntry::fill_in()
2615   __ lea(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2616   __ ldar(method, method);
2617 
2618   // Compare the method to zero
2619   __ cbnz(method, resolved);
2620 
2621   Bytecodes::Code code = bytecode();
2622 
2623   // Call to the interpreter runtime to resolve invokedynamic
2624   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2625   __ mov(method, code); // this is essentially Bytecodes::_invokedynamic
2626   __ call_VM(noreg, entry, method);
2627   // Update registers with resolved info
2628   __ load_resolved_indy_entry(cache, index);
2629   // Load-acquire the adapter method to match store-release in ResolvedIndyEntry::fill_in()
2630   __ lea(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2631   __ ldar(method, method);
2632 
2633 #ifdef ASSERT
2634   __ cbnz(method, resolved);
2635   __ stop("Should be resolved by now");
2636 #endif // ASSERT
2637   __ bind(resolved);
2638 
2639   Label L_no_push;
2640   // Check if there is an appendix
2641   __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::flags_offset())));
2642   __ tbz(index, ResolvedIndyEntry::has_appendix_shift, L_no_push);
2643 
2644   // Get appendix
2645   __ load_unsigned_short(index, Address(cache, in_bytes(ResolvedIndyEntry::resolved_references_index_offset())));
2646   // Push the appendix as a trailing parameter
2647   // since the parameter_size includes it.
2648   __ push(method);
2649   __ mov(method, index);
2650   __ load_resolved_reference_at_index(appendix, method);
2651   __ verify_oop(appendix);
2652   __ pop(method);
2653   __ push(appendix);  // push appendix (MethodType, CallSite, etc.)
2654   __ bind(L_no_push);
2655 
2656   // compute return type
2657   __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::result_type_offset())));
2658   // load return address
2659   // Return address is loaded into link register(lr) and not pushed to the stack
2660   // like x86
2661   {
2662     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2663     __ mov(rscratch1, table_addr);
2664     __ ldr(lr, Address(rscratch1, index, Address::lsl(3)));
2665   }
2666 }
2667 
2668 // The registers cache and index expected to be set before call.
2669 // Correct values of the cache and index registers are preserved.
2670 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2671                                             bool is_static, bool has_tos) {
2672   // do the JVMTI work here to avoid disturbing the register state below
2673   // We use c_rarg registers here because we want to use the register used in
2674   // the call to the VM
2675   if (JvmtiExport::can_post_field_access()) {
2676     // Check to see if a field access watch has been set before we
2677     // take the time to call into the VM.
2678     Label L1;
2679     assert_different_registers(cache, index, r0);
2680     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2681     __ ldrw(r0, Address(rscratch1));
2682     __ cbzw(r0, L1);
2683 
2684     __ load_field_entry(c_rarg2, index);
2685 
2686     if (is_static) {
2687       __ mov(c_rarg1, zr); // null object reference
2688     } else {
2689       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2690       __ verify_oop(c_rarg1);
2691     }
2692     // c_rarg1: object pointer or null
2693     // c_rarg2: cache entry pointer
2694     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2695                                        InterpreterRuntime::post_field_access),
2696                c_rarg1, c_rarg2);
2697     __ load_field_entry(cache, index);
2698     __ bind(L1);
2699   }
2700 }
2701 
2702 void TemplateTable::pop_and_check_object(Register r)
2703 {
2704   __ pop_ptr(r);
2705   __ null_check(r);  // for field access must check obj.
2706   __ verify_oop(r);
2707 }
2708 
2709 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2710 {
2711   const Register cache     = r2;
2712   const Register obj       = r4;
2713   const Register klass     = r5;
2714   const Register inline_klass = r7;
2715   const Register field_index = r23;
2716   const Register index     = r3;
2717   const Register tos_state = r3;
2718   const Register off       = r19;
2719   const Register flags     = r6;
2720   const Register bc        = r4; // uses same reg as obj, so don't mix them
2721 
2722   resolve_cache_and_index_for_field(byte_no, cache, index);
2723   jvmti_post_field_access(cache, index, is_static, false);
2724 
2725   // Valhalla extras
2726   __ load_unsigned_short(field_index, Address(cache, in_bytes(ResolvedFieldEntry::field_index_offset())));
2727   __ ldr(klass, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2728 
2729   load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
2730 
2731   if (!is_static) {
2732     // obj is on the stack
2733     pop_and_check_object(obj);
2734   }
2735 
2736   // 8179954: We need to make sure that the code generated for
2737   // volatile accesses forms a sequentially-consistent set of
2738   // operations when combined with STLR and LDAR.  Without a leading
2739   // membar it's possible for a simple Dekker test to fail if loads
2740   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2741   // the stores in one method and we interpret the loads in another.
2742   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()){
2743     Label notVolatile;
2744     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
2745     __ membar(MacroAssembler::AnyAny);
2746     __ bind(notVolatile);
2747   }
2748 
2749   const Address field(obj, off);
2750 
2751   Label Done, notByte, notBool, notInt, notShort, notChar,
2752               notLong, notFloat, notObj, notDouble;
2753 
2754   assert(btos == 0, "change code, btos != 0");
2755   __ cbnz(tos_state, notByte);
2756 
2757   // Don't rewrite getstatic, only getfield
2758   if (is_static) rc = may_not_rewrite;
2759 
2760   // btos
2761   __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
2762   __ push(btos);
2763   // Rewrite bytecode to be faster
2764   if (rc == may_rewrite) {
2765     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2766   }
2767   __ b(Done);
2768 
2769   __ bind(notByte);
2770   __ cmp(tos_state, (u1)ztos);
2771   __ br(Assembler::NE, notBool);
2772 
2773   // ztos (same code as btos)
2774   __ access_load_at(T_BOOLEAN, IN_HEAP, r0, field, noreg, noreg);
2775   __ push(ztos);
2776   // Rewrite bytecode to be faster
2777   if (rc == may_rewrite) {
2778     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2779     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2780   }
2781   __ b(Done);
2782 
2783   __ bind(notBool);
2784   __ cmp(tos_state, (u1)atos);
2785   __ br(Assembler::NE, notObj);
2786   // atos
2787   if (!EnableValhalla) {
2788     do_oop_load(_masm, field, r0, IN_HEAP);
2789     __ push(atos);
2790     if (rc == may_rewrite) {
2791       patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2792     }
2793     __ b(Done);
2794   } else { // Valhalla
2795     if (is_static) {
2796       __ load_heap_oop(r0, field, rscratch1, rscratch2);
2797       Label is_null_free_inline_type, uninitialized;
2798       // Issue below if the static field has not been initialized yet
2799       __ test_field_is_null_free_inline_type(flags, noreg /*temp*/, is_null_free_inline_type);
2800         // field is not a null free inline type
2801         __ push(atos);
2802         __ b(Done);
2803       // field is a null free inline type, must not return null even if uninitialized
2804       __ bind(is_null_free_inline_type);
2805         __ cbz(r0, uninitialized);
2806           __ push(atos);
2807           __ b(Done);
2808         __ bind(uninitialized);
2809           Label slow_case, finish;
2810           __ ldrb(rscratch1, Address(klass, InstanceKlass::init_state_offset()));
2811           __ cmp(rscratch1, (u1)InstanceKlass::fully_initialized);
2812           __ br(Assembler::NE, slow_case);
2813           __ get_default_value_oop(klass, off /* temp */, r0);
2814         __ b(finish);
2815         __ bind(slow_case);
2816           __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_inline_type_field), obj, cache);
2817           __ bind(finish);
2818           __ verify_oop(r0);
2819           __ push(atos);
2820           __ b(Done);
2821     } else {
2822       Label is_flat, nonnull, is_inline_type, has_null_marker, rewrite_inline;
2823       __ test_field_is_null_free_inline_type(flags, noreg /*temp*/, is_inline_type);
2824       __ test_field_has_null_marker(flags, noreg /*temp*/, has_null_marker);
2825         // Non-inline field case
2826         __ load_heap_oop(r0, field, rscratch1, rscratch2);
2827         __ push(atos);
2828         if (rc == may_rewrite) {
2829           patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2830         }
2831         __ b(Done);
2832       __ bind(is_inline_type);
2833         __ test_field_is_flat(flags, noreg /* temp */, is_flat);
2834          // field is not flat
2835           __ load_heap_oop(r0, field, rscratch1, rscratch2);
2836           __ cbnz(r0, nonnull);
2837             __ get_inline_type_field_klass(klass, field_index, inline_klass);
2838             __ get_default_value_oop(inline_klass, klass /* temp */, r0);
2839           __ bind(nonnull);
2840           __ verify_oop(r0);
2841           __ push(atos);
2842           __ b(rewrite_inline);
2843         __ bind(is_flat);
2844         // field is flat
2845           __ mov(r0, obj);
2846           __ read_flat_field(klass, field_index, off, inline_klass /* temp */, r0);
2847           __ verify_oop(r0);
2848           __ push(atos);
2849           __ b(rewrite_inline);
2850         __ bind(has_null_marker);
2851           call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), obj, cache);
2852           __ verify_oop(r0);
2853           __ push(atos);
2854       __ bind(rewrite_inline);
2855       if (rc == may_rewrite) {
2856         patch_bytecode(Bytecodes::_fast_vgetfield, bc, r1);
2857       }
2858       __ b(Done);
2859     }
2860   }
2861 
2862   __ bind(notObj);
2863   __ cmp(tos_state, (u1)itos);
2864   __ br(Assembler::NE, notInt);
2865   // itos
2866   __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
2867   __ push(itos);
2868   // Rewrite bytecode to be faster
2869   if (rc == may_rewrite) {
2870     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2871   }
2872   __ b(Done);
2873 
2874   __ bind(notInt);
2875   __ cmp(tos_state, (u1)ctos);
2876   __ br(Assembler::NE, notChar);
2877   // ctos
2878   __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
2879   __ push(ctos);
2880   // Rewrite bytecode to be faster
2881   if (rc == may_rewrite) {
2882     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2883   }
2884   __ b(Done);
2885 
2886   __ bind(notChar);
2887   __ cmp(tos_state, (u1)stos);
2888   __ br(Assembler::NE, notShort);
2889   // stos
2890   __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
2891   __ push(stos);
2892   // Rewrite bytecode to be faster
2893   if (rc == may_rewrite) {
2894     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2895   }
2896   __ b(Done);
2897 
2898   __ bind(notShort);
2899   __ cmp(tos_state, (u1)ltos);
2900   __ br(Assembler::NE, notLong);
2901   // ltos
2902   __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
2903   __ push(ltos);
2904   // Rewrite bytecode to be faster
2905   if (rc == may_rewrite) {
2906     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2907   }
2908   __ b(Done);
2909 
2910   __ bind(notLong);
2911   __ cmp(tos_state, (u1)ftos);
2912   __ br(Assembler::NE, notFloat);
2913   // ftos
2914   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2915   __ push(ftos);
2916   // Rewrite bytecode to be faster
2917   if (rc == may_rewrite) {
2918     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2919   }
2920   __ b(Done);
2921 
2922   __ bind(notFloat);
2923 #ifdef ASSERT
2924   __ cmp(tos_state, (u1)dtos);
2925   __ br(Assembler::NE, notDouble);
2926 #endif
2927   // dtos
2928   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2929   __ push(dtos);
2930   // Rewrite bytecode to be faster
2931   if (rc == may_rewrite) {
2932     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2933   }
2934 #ifdef ASSERT
2935   __ b(Done);
2936 
2937   __ bind(notDouble);
2938   __ stop("Bad state");
2939 #endif
2940 
2941   __ bind(Done);
2942 
2943   Label notVolatile;
2944   __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
2945   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2946   __ bind(notVolatile);
2947 }
2948 
2949 
2950 void TemplateTable::getfield(int byte_no)
2951 {
2952   getfield_or_static(byte_no, false);
2953 }
2954 
2955 void TemplateTable::nofast_getfield(int byte_no) {
2956   getfield_or_static(byte_no, false, may_not_rewrite);
2957 }
2958 
2959 void TemplateTable::getstatic(int byte_no)
2960 {
2961   getfield_or_static(byte_no, true);
2962 }
2963 
2964 // The registers cache and index expected to be set before call.
2965 // The function may destroy various registers, just not the cache and index registers.
2966 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2967   transition(vtos, vtos);
2968 
2969   if (JvmtiExport::can_post_field_modification()) {
2970     // Check to see if a field modification watch has been set before
2971     // we take the time to call into the VM.
2972     Label L1;
2973     assert_different_registers(cache, index, r0);
2974     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2975     __ ldrw(r0, Address(rscratch1));
2976     __ cbz(r0, L1);
2977 
2978     __ mov(c_rarg2, cache);
2979 
2980     if (is_static) {
2981       // Life is simple.  Null out the object pointer.
2982       __ mov(c_rarg1, zr);
2983     } else {
2984       // Life is harder. The stack holds the value on top, followed by
2985       // the object.  We don't know the size of the value, though; it
2986       // could be one or two words depending on its type. As a result,
2987       // we must find the type to determine where the object is.
2988       __ load_unsigned_byte(c_rarg3, Address(c_rarg2, in_bytes(ResolvedFieldEntry::type_offset())));
2989       Label nope2, done, ok;
2990       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2991       __ cmpw(c_rarg3, ltos);
2992       __ br(Assembler::EQ, ok);
2993       __ cmpw(c_rarg3, dtos);
2994       __ br(Assembler::NE, nope2);
2995       __ bind(ok);
2996       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2997       __ bind(nope2);
2998     }
2999     // object (tos)
3000     __ mov(c_rarg3, esp);
3001     // c_rarg1: object pointer set up above (null if static)
3002     // c_rarg2: cache entry pointer
3003     // c_rarg3: jvalue object on the stack
3004     __ call_VM(noreg,
3005                CAST_FROM_FN_PTR(address,
3006                                 InterpreterRuntime::post_field_modification),
3007                c_rarg1, c_rarg2, c_rarg3);
3008     __ load_field_entry(cache, index);
3009     __ bind(L1);
3010   }
3011 }
3012 
3013 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3014   transition(vtos, vtos);
3015 
3016   const Register cache     = r2;
3017   const Register index     = r3;
3018   const Register tos_state = r3;
3019   const Register obj       = r2;
3020   const Register off       = r19;
3021   const Register flags     = r6;
3022   const Register bc        = r4;
3023   const Register inline_klass = r5;
3024 
3025   resolve_cache_and_index_for_field(byte_no, cache, index);
3026   jvmti_post_field_mod(cache, index, is_static);
3027   load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
3028 
3029   Label Done;
3030   {
3031     Label notVolatile;
3032     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3033     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
3034     __ bind(notVolatile);
3035   }
3036 
3037   // field address
3038   const Address field(obj, off);
3039 
3040   Label notByte, notBool, notInt, notShort, notChar,
3041         notLong, notFloat, notObj, notDouble;
3042 
3043   assert(btos == 0, "change code, btos != 0");
3044   __ cbnz(tos_state, notByte);
3045 
3046   // Don't rewrite putstatic, only putfield
3047   if (is_static) rc = may_not_rewrite;
3048 
3049   // btos
3050   {
3051     __ pop(btos);
3052     if (!is_static) pop_and_check_object(obj);
3053     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
3054     if (rc == may_rewrite) {
3055       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
3056     }
3057     __ b(Done);
3058   }
3059 
3060   __ bind(notByte);
3061   __ cmp(tos_state, (u1)ztos);
3062   __ br(Assembler::NE, notBool);
3063 
3064   // ztos
3065   {
3066     __ pop(ztos);
3067     if (!is_static) pop_and_check_object(obj);
3068     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
3069     if (rc == may_rewrite) {
3070       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
3071     }
3072     __ b(Done);
3073   }
3074 
3075   __ bind(notBool);
3076   __ cmp(tos_state, (u1)atos);
3077   __ br(Assembler::NE, notObj);
3078 
3079   // atos
3080   {
3081      if (!EnableValhalla) {
3082       __ pop(atos);
3083       if (!is_static) pop_and_check_object(obj);
3084       // Store into the field
3085       do_oop_store(_masm, field, r0, IN_HEAP);
3086       if (rc == may_rewrite) {
3087         patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
3088       }
3089       __ b(Done);
3090      } else { // Valhalla
3091       __ pop(atos);
3092       if (is_static) {
3093         Label is_inline_type;
3094          __ test_field_is_not_null_free_inline_type(flags, noreg /* temp */, is_inline_type);
3095          __ null_check(r0);
3096          __ bind(is_inline_type);
3097          do_oop_store(_masm, field, r0, IN_HEAP);
3098          __ b(Done);
3099       } else {
3100         Label is_inline_type, is_flat, has_null_marker, rewrite_not_inline, rewrite_inline;
3101         __ test_field_is_null_free_inline_type(flags, noreg /*temp*/, is_inline_type);
3102         __ test_field_has_null_marker(flags, noreg /*temp*/, has_null_marker);
3103         // Not an inline type
3104         pop_and_check_object(obj);
3105         // Store into the field
3106         do_oop_store(_masm, field, r0, IN_HEAP);
3107         __ bind(rewrite_not_inline);
3108         if (rc == may_rewrite) {
3109           patch_bytecode(Bytecodes::_fast_aputfield, bc, r19, true, byte_no);
3110         }
3111         __ b(Done);
3112         // Implementation of the inline type semantic
3113         __ bind(is_inline_type);
3114         __ null_check(r0);
3115         __ test_field_is_flat(flags, noreg /*temp*/, is_flat);
3116         // field is not flat
3117         pop_and_check_object(obj);
3118         // Store into the field
3119         do_oop_store(_masm, field, r0, IN_HEAP);
3120         __ b(rewrite_inline);
3121         __ bind(is_flat);
3122         // field is flat
3123         pop_and_check_object(obj);
3124         assert_different_registers(r0, inline_klass, obj, off);
3125         __ load_klass(inline_klass, r0);
3126         __ data_for_oop(r0, r0, inline_klass);
3127         __ add(obj, obj, off);
3128         __ access_value_copy(IN_HEAP, r0, obj, inline_klass);
3129         __ b(rewrite_inline);
3130         __ bind(has_null_marker);
3131         assert_different_registers(r0, cache, r19);
3132         pop_and_check_object(r19);
3133         __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), r19, r0, cache);
3134         __ bind(rewrite_inline);
3135         if (rc == may_rewrite) {
3136           patch_bytecode(Bytecodes::_fast_vputfield, bc, r19, true, byte_no);
3137         }
3138         __ b(Done);
3139       }
3140      }  // Valhalla
3141   }
3142 
3143   __ bind(notObj);
3144   __ cmp(tos_state, (u1)itos);
3145   __ br(Assembler::NE, notInt);
3146 
3147   // itos
3148   {
3149     __ pop(itos);
3150     if (!is_static) pop_and_check_object(obj);
3151     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
3152     if (rc == may_rewrite) {
3153       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
3154     }
3155     __ b(Done);
3156   }
3157 
3158   __ bind(notInt);
3159   __ cmp(tos_state, (u1)ctos);
3160   __ br(Assembler::NE, notChar);
3161 
3162   // ctos
3163   {
3164     __ pop(ctos);
3165     if (!is_static) pop_and_check_object(obj);
3166     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
3167     if (rc == may_rewrite) {
3168       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
3169     }
3170     __ b(Done);
3171   }
3172 
3173   __ bind(notChar);
3174   __ cmp(tos_state, (u1)stos);
3175   __ br(Assembler::NE, notShort);
3176 
3177   // stos
3178   {
3179     __ pop(stos);
3180     if (!is_static) pop_and_check_object(obj);
3181     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
3182     if (rc == may_rewrite) {
3183       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
3184     }
3185     __ b(Done);
3186   }
3187 
3188   __ bind(notShort);
3189   __ cmp(tos_state, (u1)ltos);
3190   __ br(Assembler::NE, notLong);
3191 
3192   // ltos
3193   {
3194     __ pop(ltos);
3195     if (!is_static) pop_and_check_object(obj);
3196     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
3197     if (rc == may_rewrite) {
3198       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
3199     }
3200     __ b(Done);
3201   }
3202 
3203   __ bind(notLong);
3204   __ cmp(tos_state, (u1)ftos);
3205   __ br(Assembler::NE, notFloat);
3206 
3207   // ftos
3208   {
3209     __ pop(ftos);
3210     if (!is_static) pop_and_check_object(obj);
3211     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3212     if (rc == may_rewrite) {
3213       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
3214     }
3215     __ b(Done);
3216   }
3217 
3218   __ bind(notFloat);
3219 #ifdef ASSERT
3220   __ cmp(tos_state, (u1)dtos);
3221   __ br(Assembler::NE, notDouble);
3222 #endif
3223 
3224   // dtos
3225   {
3226     __ pop(dtos);
3227     if (!is_static) pop_and_check_object(obj);
3228     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
3229     if (rc == may_rewrite) {
3230       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
3231     }
3232   }
3233 
3234 #ifdef ASSERT
3235   __ b(Done);
3236 
3237   __ bind(notDouble);
3238   __ stop("Bad state");
3239 #endif
3240 
3241   __ bind(Done);
3242 
3243   {
3244     Label notVolatile;
3245     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3246     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3247     __ bind(notVolatile);
3248   }
3249 }
3250 
3251 void TemplateTable::putfield(int byte_no)
3252 {
3253   putfield_or_static(byte_no, false);
3254 }
3255 
3256 void TemplateTable::nofast_putfield(int byte_no) {
3257   putfield_or_static(byte_no, false, may_not_rewrite);
3258 }
3259 
3260 void TemplateTable::putstatic(int byte_no) {
3261   putfield_or_static(byte_no, true);
3262 }
3263 
3264 void TemplateTable::jvmti_post_fast_field_mod() {
3265   if (JvmtiExport::can_post_field_modification()) {
3266     // Check to see if a field modification watch has been set before
3267     // we take the time to call into the VM.
3268     Label L2;
3269     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3270     __ ldrw(c_rarg3, Address(rscratch1));
3271     __ cbzw(c_rarg3, L2);
3272     __ pop_ptr(r19);                  // copy the object pointer from tos
3273     __ verify_oop(r19);
3274     __ push_ptr(r19);                 // put the object pointer back on tos
3275     // Save tos values before call_VM() clobbers them. Since we have
3276     // to do it for every data type, we use the saved values as the
3277     // jvalue object.
3278     switch (bytecode()) {          // load values into the jvalue object
3279     case Bytecodes::_fast_vputfield: //fall through
3280     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
3281     case Bytecodes::_fast_bputfield: // fall through
3282     case Bytecodes::_fast_zputfield: // fall through
3283     case Bytecodes::_fast_sputfield: // fall through
3284     case Bytecodes::_fast_cputfield: // fall through
3285     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
3286     case Bytecodes::_fast_dputfield: __ push_d(); break;
3287     case Bytecodes::_fast_fputfield: __ push_f(); break;
3288     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
3289 
3290     default:
3291       ShouldNotReachHere();
3292     }
3293     __ mov(c_rarg3, esp);             // points to jvalue on the stack
3294     // access constant pool cache entry
3295     __ load_field_entry(c_rarg2, r0);
3296     __ verify_oop(r19);
3297     // r19: object pointer copied above
3298     // c_rarg2: cache entry pointer
3299     // c_rarg3: jvalue object on the stack
3300     __ call_VM(noreg,
3301                CAST_FROM_FN_PTR(address,
3302                                 InterpreterRuntime::post_field_modification),
3303                r19, c_rarg2, c_rarg3);
3304 
3305     switch (bytecode()) {             // restore tos values
3306     case Bytecodes::_fast_vputfield: //fall through
3307     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
3308     case Bytecodes::_fast_bputfield: // fall through
3309     case Bytecodes::_fast_zputfield: // fall through
3310     case Bytecodes::_fast_sputfield: // fall through
3311     case Bytecodes::_fast_cputfield: // fall through
3312     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
3313     case Bytecodes::_fast_dputfield: __ pop_d(); break;
3314     case Bytecodes::_fast_fputfield: __ pop_f(); break;
3315     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
3316     default: break;
3317     }
3318     __ bind(L2);
3319   }
3320 }
3321 
3322 void TemplateTable::fast_storefield(TosState state)
3323 {
3324   transition(state, vtos);
3325 
3326   ByteSize base = ConstantPoolCache::base_offset();
3327 
3328   jvmti_post_fast_field_mod();
3329 
3330   // access constant pool cache
3331   __ load_field_entry(r2, r1);
3332 
3333   // R1: field offset, R2: field holder, R3: flags
3334   load_resolved_field_entry(r2, r2, noreg, r1, r3);
3335 
3336   {
3337     Label notVolatile;
3338     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3339     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
3340     __ bind(notVolatile);
3341   }
3342 
3343   Label notVolatile;
3344 
3345   // Get object from stack
3346   pop_and_check_object(r2);
3347 
3348   // field address
3349   const Address field(r2, r1);
3350 
3351   // access field
3352   switch (bytecode()) {
3353   case Bytecodes::_fast_vputfield:
3354    {
3355       Label is_flat, has_null_marker, done;
3356       __ test_field_has_null_marker(r3, noreg /* temp */, has_null_marker);
3357       __ null_check(r0);
3358       __ test_field_is_flat(r3, noreg /* temp */, is_flat);
3359       // field is not flat
3360       do_oop_store(_masm, field, r0, IN_HEAP);
3361       __ b(done);
3362       __ bind(is_flat);
3363       // field is flat
3364       __ load_klass(r4, r0);
3365       __ data_for_oop(r0, r0, r4);
3366       __ lea(rscratch1, field);
3367       __ access_value_copy(IN_HEAP, r0, rscratch1, r4);
3368       __ b(done);
3369       __ bind(has_null_marker);
3370       __ load_field_entry(r4, r1);
3371       __ mov(r1, r2);
3372       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_nullable_flat_field), r1, r0, r4);
3373       __ bind(done);
3374     }
3375     break;
3376   case Bytecodes::_fast_aputfield:
3377     do_oop_store(_masm, field, r0, IN_HEAP);
3378     break;
3379   case Bytecodes::_fast_lputfield:
3380     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
3381     break;
3382   case Bytecodes::_fast_iputfield:
3383     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
3384     break;
3385   case Bytecodes::_fast_zputfield:
3386     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
3387     break;
3388   case Bytecodes::_fast_bputfield:
3389     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
3390     break;
3391   case Bytecodes::_fast_sputfield:
3392     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
3393     break;
3394   case Bytecodes::_fast_cputfield:
3395     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
3396     break;
3397   case Bytecodes::_fast_fputfield:
3398     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3399     break;
3400   case Bytecodes::_fast_dputfield:
3401     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
3402     break;
3403   default:
3404     ShouldNotReachHere();
3405   }
3406 
3407   {
3408     Label notVolatile;
3409     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3410     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3411     __ bind(notVolatile);
3412   }
3413 }
3414 
3415 
3416 void TemplateTable::fast_accessfield(TosState state)
3417 {
3418   transition(atos, state);
3419   // Do the JVMTI work here to avoid disturbing the register state below
3420   if (JvmtiExport::can_post_field_access()) {
3421     // Check to see if a field access watch has been set before we
3422     // take the time to call into the VM.
3423     Label L1;
3424     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3425     __ ldrw(r2, Address(rscratch1));
3426     __ cbzw(r2, L1);
3427     // access constant pool cache entry
3428     __ load_field_entry(c_rarg2, rscratch2);
3429     __ verify_oop(r0);
3430     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
3431     __ mov(c_rarg1, r0);
3432     // c_rarg1: object pointer copied above
3433     // c_rarg2: cache entry pointer
3434     __ call_VM(noreg,
3435                CAST_FROM_FN_PTR(address,
3436                                 InterpreterRuntime::post_field_access),
3437                c_rarg1, c_rarg2);
3438     __ pop_ptr(r0); // restore object pointer
3439     __ bind(L1);
3440   }
3441 
3442   // access constant pool cache
3443   __ load_field_entry(r2, r1);
3444 
3445   __ load_sized_value(r1, Address(r2, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3446   __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3447 
3448   // r0: object
3449   __ verify_oop(r0);
3450   __ null_check(r0);
3451   const Address field(r0, r1);
3452 
3453   // 8179954: We need to make sure that the code generated for
3454   // volatile accesses forms a sequentially-consistent set of
3455   // operations when combined with STLR and LDAR.  Without a leading
3456   // membar it's possible for a simple Dekker test to fail if loads
3457   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3458   // the stores in one method and we interpret the loads in another.
3459   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3460     Label notVolatile;
3461     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3462     __ membar(MacroAssembler::AnyAny);
3463     __ bind(notVolatile);
3464   }
3465 
3466   // access field
3467   switch (bytecode()) {
3468   case Bytecodes::_fast_vgetfield:
3469     {
3470       Register index = r4, klass = r5, inline_klass = r6, tmp = r7;
3471       Label is_flat, has_null_marker, nonnull, Done;
3472       __ test_field_has_null_marker(r3, noreg /*temp*/, has_null_marker);
3473       __ test_field_is_flat(r3, noreg /* temp */, is_flat);
3474         // field is not flat
3475         __ load_heap_oop(r0, field, rscratch1, rscratch2);
3476         __ cbnz(r0, nonnull);
3477           __ load_unsigned_short(index, Address(r2, in_bytes(ResolvedFieldEntry::field_index_offset())));
3478           __ ldr(klass, Address(r2, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3479           __ get_inline_type_field_klass(klass, index, inline_klass);
3480           __ get_default_value_oop(inline_klass, tmp /* temp */, r0);
3481         __ bind(nonnull);
3482         __ verify_oop(r0);
3483         __ b(Done);
3484       __ bind(is_flat);
3485       // field is flat
3486         __ load_unsigned_short(index, Address(r2, in_bytes(ResolvedFieldEntry::field_index_offset())));
3487         __ ldr(klass, Address(r2, in_bytes(ResolvedFieldEntry::field_holder_offset())));
3488         __ read_flat_field(klass, index, r1, tmp /* temp */, r0);
3489         __ verify_oop(r0);
3490         __ b(Done);
3491       __ bind(has_null_marker);
3492         call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_nullable_flat_field), r0, r2);
3493         __ verify_oop(r0);
3494       __ bind(Done);
3495     }
3496     break;
3497   case Bytecodes::_fast_agetfield:
3498     do_oop_load(_masm, field, r0, IN_HEAP);
3499     __ verify_oop(r0);
3500     break;
3501   case Bytecodes::_fast_lgetfield:
3502     __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
3503     break;
3504   case Bytecodes::_fast_igetfield:
3505     __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
3506     break;
3507   case Bytecodes::_fast_bgetfield:
3508     __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
3509     break;
3510   case Bytecodes::_fast_sgetfield:
3511     __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
3512     break;
3513   case Bytecodes::_fast_cgetfield:
3514     __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
3515     break;
3516   case Bytecodes::_fast_fgetfield:
3517     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3518     break;
3519   case Bytecodes::_fast_dgetfield:
3520     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3521     break;
3522   default:
3523     ShouldNotReachHere();
3524   }
3525   {
3526     Label notVolatile;
3527     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3528     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3529     __ bind(notVolatile);
3530   }
3531 }
3532 
3533 void TemplateTable::fast_xaccess(TosState state)
3534 {
3535   transition(vtos, state);
3536 
3537   // get receiver
3538   __ ldr(r0, aaddress(0));
3539   // access constant pool cache
3540   __ load_field_entry(r2, r3, 2);
3541   __ load_sized_value(r1, Address(r2, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3542 
3543   // 8179954: We need to make sure that the code generated for
3544   // volatile accesses forms a sequentially-consistent set of
3545   // operations when combined with STLR and LDAR.  Without a leading
3546   // membar it's possible for a simple Dekker test to fail if loads
3547   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3548   // the stores in one method and we interpret the loads in another.
3549   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3550     Label notVolatile;
3551     __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3552     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3553     __ membar(MacroAssembler::AnyAny);
3554     __ bind(notVolatile);
3555   }
3556 
3557   // make sure exception is reported in correct bcp range (getfield is
3558   // next instruction)
3559   __ increment(rbcp);
3560   __ null_check(r0);
3561   switch (state) {
3562   case itos:
3563     __ access_load_at(T_INT, IN_HEAP, r0, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3564     break;
3565   case atos:
3566     do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3567     __ verify_oop(r0);
3568     break;
3569   case ftos:
3570     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3571     break;
3572   default:
3573     ShouldNotReachHere();
3574   }
3575 
3576   {
3577     Label notVolatile;
3578     __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3579     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3580     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3581     __ bind(notVolatile);
3582   }
3583 
3584   __ decrement(rbcp);
3585 }
3586 
3587 
3588 
3589 //-----------------------------------------------------------------------------
3590 // Calls
3591 
3592 void TemplateTable::prepare_invoke(Register cache, Register recv) {
3593 
3594   Bytecodes::Code code = bytecode();
3595   const bool load_receiver       = (code != Bytecodes::_invokestatic) && (code != Bytecodes::_invokedynamic);
3596 
3597   // save 'interpreter return address'
3598   __ save_bcp();
3599 
3600   // Load TOS state for later
3601   __ load_unsigned_byte(rscratch2, Address(cache, in_bytes(ResolvedMethodEntry::type_offset())));
3602 
3603   // load receiver if needed (note: no return address pushed yet)
3604   if (load_receiver) {
3605     __ load_unsigned_short(recv, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
3606     __ add(rscratch1, esp, recv, ext::uxtx, 3);
3607     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3608     __ verify_oop(recv);
3609   }
3610 
3611   // load return address
3612   {
3613     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3614     __ mov(rscratch1, table_addr);
3615     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3616   }
3617 }
3618 
3619 
3620 void TemplateTable::invokevirtual_helper(Register index,
3621                                          Register recv,
3622                                          Register flags)
3623 {
3624   // Uses temporary registers r0, r3
3625   assert_different_registers(index, recv, r0, r3);
3626   // Test for an invoke of a final method
3627   Label notFinal;
3628   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, notFinal);
3629 
3630   const Register method = index;  // method must be rmethod
3631   assert(method == rmethod,
3632          "Method must be rmethod for interpreter calling convention");
3633 
3634   // do the call - the index is actually the method to call
3635   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3636 
3637   // It's final, need a null check here!
3638   __ null_check(recv);
3639 
3640   // profile this call
3641   __ profile_final_call(r0);
3642   __ profile_arguments_type(r0, method, r4, true);
3643 
3644   __ jump_from_interpreted(method, r0);
3645 
3646   __ bind(notFinal);
3647 
3648   // get receiver klass
3649   __ load_klass(r0, recv);
3650 
3651   // profile this call
3652   __ profile_virtual_call(r0, rlocals, r3);
3653 
3654   // get target Method & entry point
3655   __ lookup_virtual_method(r0, index, method);
3656   __ profile_arguments_type(r3, method, r4, true);
3657   // FIXME -- this looks completely redundant. is it?
3658   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3659   __ jump_from_interpreted(method, r3);
3660 }
3661 
3662 void TemplateTable::invokevirtual(int byte_no)
3663 {
3664   transition(vtos, vtos);
3665   assert(byte_no == f2_byte, "use this argument");
3666 
3667   load_resolved_method_entry_virtual(r2,      // ResolvedMethodEntry*
3668                                      rmethod, // Method* or itable index
3669                                      r3);     // flags
3670   prepare_invoke(r2, r2); // recv
3671 
3672   // rmethod: index (actually a Method*)
3673   // r2: receiver
3674   // r3: flags
3675 
3676   invokevirtual_helper(rmethod, r2, r3);
3677 }
3678 
3679 void TemplateTable::invokespecial(int byte_no)
3680 {
3681   transition(vtos, vtos);
3682   assert(byte_no == f1_byte, "use this argument");
3683 
3684   load_resolved_method_entry_special_or_static(r2,      // ResolvedMethodEntry*
3685                                                rmethod, // Method*
3686                                                r3);     // flags
3687   prepare_invoke(r2, r2);  // get receiver also for null check
3688   __ verify_oop(r2);
3689   __ null_check(r2);
3690   // do the call
3691   __ profile_call(r0);
3692   __ profile_arguments_type(r0, rmethod, rbcp, false);
3693   __ jump_from_interpreted(rmethod, r0);
3694 }
3695 
3696 void TemplateTable::invokestatic(int byte_no)
3697 {
3698   transition(vtos, vtos);
3699   assert(byte_no == f1_byte, "use this argument");
3700 
3701   load_resolved_method_entry_special_or_static(r2,      // ResolvedMethodEntry*
3702                                                rmethod, // Method*
3703                                                r3);     // flags
3704   prepare_invoke(r2, r2);  // get receiver also for null check
3705 
3706   // do the call
3707   __ profile_call(r0);
3708   __ profile_arguments_type(r0, rmethod, r4, false);
3709   __ jump_from_interpreted(rmethod, r0);
3710 }
3711 
3712 void TemplateTable::fast_invokevfinal(int byte_no)
3713 {
3714   __ call_Unimplemented();
3715 }
3716 
3717 void TemplateTable::invokeinterface(int byte_no) {
3718   transition(vtos, vtos);
3719   assert(byte_no == f1_byte, "use this argument");
3720 
3721   load_resolved_method_entry_interface(r2,      // ResolvedMethodEntry*
3722                                        r0,      // Klass*
3723                                        rmethod, // Method* or itable/vtable index
3724                                        r3);     // flags
3725   prepare_invoke(r2, r2); // receiver
3726 
3727   // r0: interface klass (from f1)
3728   // rmethod: method (from f2)
3729   // r2: receiver
3730   // r3: flags
3731 
3732   // First check for Object case, then private interface method,
3733   // then regular interface method.
3734 
3735   // Special case of invokeinterface called for virtual method of
3736   // java.lang.Object.  See cpCache.cpp for details.
3737   Label notObjectMethod;
3738   __ tbz(r3, ResolvedMethodEntry::is_forced_virtual_shift, notObjectMethod);
3739 
3740   invokevirtual_helper(rmethod, r2, r3);
3741   __ bind(notObjectMethod);
3742 
3743   Label no_such_interface;
3744 
3745   // Check for private method invocation - indicated by vfinal
3746   Label notVFinal;
3747   __ tbz(r3, ResolvedMethodEntry::is_vfinal_shift, notVFinal);
3748 
3749   // Get receiver klass into r3
3750   __ load_klass(r3, r2);
3751 
3752   Label subtype;
3753   __ check_klass_subtype(r3, r0, r4, subtype);
3754   // If we get here the typecheck failed
3755   __ b(no_such_interface);
3756   __ bind(subtype);
3757 
3758   __ profile_final_call(r0);
3759   __ profile_arguments_type(r0, rmethod, r4, true);
3760   __ jump_from_interpreted(rmethod, r0);
3761 
3762   __ bind(notVFinal);
3763 
3764   // Get receiver klass into r3
3765   __ restore_locals();
3766   __ load_klass(r3, r2);
3767 
3768   Label no_such_method;
3769 
3770   // Preserve method for throw_AbstractMethodErrorVerbose.
3771   __ mov(r16, rmethod);
3772   // Receiver subtype check against REFC.
3773   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3774   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3775                              r3, r0, noreg,
3776                              // outputs: scan temp. reg, scan temp. reg
3777                              rscratch2, r13,
3778                              no_such_interface,
3779                              /*return_method=*/false);
3780 
3781   // profile this call
3782   __ profile_virtual_call(r3, r13, r19);
3783 
3784   // Get declaring interface class from method, and itable index
3785 
3786   __ load_method_holder(r0, rmethod);
3787   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3788   __ subw(rmethod, rmethod, Method::itable_index_max);
3789   __ negw(rmethod, rmethod);
3790 
3791   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3792   __ mov(rlocals, r3);
3793   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3794                              rlocals, r0, rmethod,
3795                              // outputs: method, scan temp. reg
3796                              rmethod, r13,
3797                              no_such_interface);
3798 
3799   // rmethod,: Method to call
3800   // r2: receiver
3801   // Check for abstract method error
3802   // Note: This should be done more efficiently via a throw_abstract_method_error
3803   //       interpreter entry point and a conditional jump to it in case of a null
3804   //       method.
3805   __ cbz(rmethod, no_such_method);
3806 
3807   __ profile_arguments_type(r3, rmethod, r13, true);
3808 
3809   // do the call
3810   // r2: receiver
3811   // rmethod,: Method
3812   __ jump_from_interpreted(rmethod, r3);
3813   __ should_not_reach_here();
3814 
3815   // exception handling code follows...
3816   // note: must restore interpreter registers to canonical
3817   //       state for exception handling to work correctly!
3818 
3819   __ bind(no_such_method);
3820   // throw exception
3821   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3822   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3823   // Pass arguments for generating a verbose error message.
3824   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3825   // the call_VM checks for exception, so we should never return here.
3826   __ should_not_reach_here();
3827 
3828   __ bind(no_such_interface);
3829   // throw exception
3830   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3831   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3832   // Pass arguments for generating a verbose error message.
3833   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3834                    InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3835   // the call_VM checks for exception, so we should never return here.
3836   __ should_not_reach_here();
3837   return;
3838 }
3839 
3840 void TemplateTable::invokehandle(int byte_no) {
3841   transition(vtos, vtos);
3842   assert(byte_no == f1_byte, "use this argument");
3843 
3844   load_resolved_method_entry_handle(r2,      // ResolvedMethodEntry*
3845                                     rmethod, // Method*
3846                                     r0,      // Resolved reference
3847                                     r3);     // flags
3848   prepare_invoke(r2, r2);
3849 
3850   __ verify_method_ptr(r2);
3851   __ verify_oop(r2);
3852   __ null_check(r2);
3853 
3854   // FIXME: profile the LambdaForm also
3855 
3856   // r13 is safe to use here as a scratch reg because it is about to
3857   // be clobbered by jump_from_interpreted().
3858   __ profile_final_call(r13);
3859   __ profile_arguments_type(r13, rmethod, r4, true);
3860 
3861   __ jump_from_interpreted(rmethod, r0);
3862 }
3863 
3864 void TemplateTable::invokedynamic(int byte_no) {
3865   transition(vtos, vtos);
3866   assert(byte_no == f1_byte, "use this argument");
3867 
3868   load_invokedynamic_entry(rmethod);
3869 
3870   // r0: CallSite object (from cpool->resolved_references[])
3871   // rmethod: MH.linkToCallSite method
3872 
3873   // Note:  r0_callsite is already pushed
3874 
3875   // %%% should make a type profile for any invokedynamic that takes a ref argument
3876   // profile this call
3877   __ profile_call(rbcp);
3878   __ profile_arguments_type(r3, rmethod, r13, false);
3879 
3880   __ verify_oop(r0);
3881 
3882   __ jump_from_interpreted(rmethod, r0);
3883 }
3884 
3885 
3886 //-----------------------------------------------------------------------------
3887 // Allocation
3888 
3889 void TemplateTable::_new() {
3890   transition(vtos, atos);
3891 
3892   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3893   Label slow_case;
3894   Label done;
3895   Label initialize_header;
3896 
3897   __ get_cpool_and_tags(r4, r0);
3898   // Make sure the class we're about to instantiate has been resolved.
3899   // This is done before loading InstanceKlass to be consistent with the order
3900   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3901   const int tags_offset = Array<u1>::base_offset_in_bytes();
3902   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3903   __ lea(rscratch1, Address(rscratch1, tags_offset));
3904   __ ldarb(rscratch1, rscratch1);
3905   __ cmp(rscratch1, (u1)JVM_CONSTANT_Class);
3906   __ br(Assembler::NE, slow_case);
3907 
3908   // get InstanceKlass
3909   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3910 
3911   // make sure klass is initialized
3912   assert(VM_Version::supports_fast_class_init_checks(), "Optimization requires support for fast class initialization checks");
3913   __ clinit_barrier(r4, rscratch1, nullptr /*L_fast_path*/, &slow_case);
3914 
3915   __ allocate_instance(r4, r0, r3, r1, true, slow_case);
3916   __ b(done);
3917 
3918   // slow case
3919   __ bind(slow_case);
3920   __ get_constant_pool(c_rarg1);
3921   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3922   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3923   __ verify_oop(r0);
3924 
3925   // continue
3926   __ bind(done);
3927   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3928   __ membar(Assembler::StoreStore);
3929 }
3930 
3931 void TemplateTable::newarray() {
3932   transition(itos, atos);
3933   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3934   __ mov(c_rarg2, r0);
3935   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3936           c_rarg1, c_rarg2);
3937   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3938   __ membar(Assembler::StoreStore);
3939 }
3940 
3941 void TemplateTable::anewarray() {
3942   transition(itos, atos);
3943   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3944   __ get_constant_pool(c_rarg1);
3945   __ mov(c_rarg3, r0);
3946   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3947           c_rarg1, c_rarg2, c_rarg3);
3948   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3949   __ membar(Assembler::StoreStore);
3950 }
3951 
3952 void TemplateTable::arraylength() {
3953   transition(atos, itos);
3954   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3955 }
3956 
3957 void TemplateTable::checkcast()
3958 {
3959   transition(atos, atos);
3960   Label done, is_null, ok_is_subtype, quicked, resolved;
3961   __ cbz(r0, is_null);
3962 
3963   // Get cpool & tags index
3964   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3965   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3966   // See if bytecode has already been quicked
3967   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3968   __ lea(r1, Address(rscratch1, r19));
3969   __ ldarb(r1, r1);
3970   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3971   __ br(Assembler::EQ, quicked);
3972 
3973   __ push(atos); // save receiver for result, and for GC
3974   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3975   // vm_result_2 has metadata result
3976   __ get_vm_result_2(r0, rthread);
3977   __ pop(r3); // restore receiver
3978   __ b(resolved);
3979 
3980   // Get superklass in r0 and subklass in r3
3981   __ bind(quicked);
3982   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3983   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3984 
3985   __ bind(resolved);
3986   __ load_klass(r19, r3);
3987 
3988   // Generate subtype check.  Blows r2, r5.  Object in r3.
3989   // Superklass in r0.  Subklass in r19.
3990   __ gen_subtype_check(r19, ok_is_subtype);
3991 
3992   // Come here on failure
3993   __ push(r3);
3994   // object is at TOS
3995   __ b(Interpreter::_throw_ClassCastException_entry);
3996 
3997   // Come here on success
3998   __ bind(ok_is_subtype);
3999   __ mov(r0, r3); // Restore object in r3
4000 
4001   __ b(done);
4002   __ bind(is_null);
4003 
4004   // Collect counts on whether this test sees nulls a lot or not.
4005   if (ProfileInterpreter) {
4006     __ profile_null_seen(r2);
4007   }
4008 
4009   __ bind(done);
4010 }
4011 
4012 void TemplateTable::instanceof() {
4013   transition(atos, itos);
4014   Label done, is_null, ok_is_subtype, quicked, resolved;
4015   __ cbz(r0, is_null);
4016 
4017   // Get cpool & tags index
4018   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
4019   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
4020   // See if bytecode has already been quicked
4021   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
4022   __ lea(r1, Address(rscratch1, r19));
4023   __ ldarb(r1, r1);
4024   __ cmp(r1, (u1)JVM_CONSTANT_Class);
4025   __ br(Assembler::EQ, quicked);
4026 
4027   __ push(atos); // save receiver for result, and for GC
4028   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4029   // vm_result_2 has metadata result
4030   __ get_vm_result_2(r0, rthread);
4031   __ pop(r3); // restore receiver
4032   __ verify_oop(r3);
4033   __ load_klass(r3, r3);
4034   __ b(resolved);
4035 
4036   // Get superklass in r0 and subklass in r3
4037   __ bind(quicked);
4038   __ load_klass(r3, r0);
4039   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
4040 
4041   __ bind(resolved);
4042 
4043   // Generate subtype check.  Blows r2, r5
4044   // Superklass in r0.  Subklass in r3.
4045   __ gen_subtype_check(r3, ok_is_subtype);
4046 
4047   // Come here on failure
4048   __ mov(r0, 0);
4049   __ b(done);
4050   // Come here on success
4051   __ bind(ok_is_subtype);
4052   __ mov(r0, 1);
4053 
4054   // Collect counts on whether this test sees nulls a lot or not.
4055   if (ProfileInterpreter) {
4056     __ b(done);
4057     __ bind(is_null);
4058     __ profile_null_seen(r2);
4059   } else {
4060     __ bind(is_null);   // same as 'done'
4061   }
4062   __ bind(done);
4063   // r0 = 0: obj == nullptr or  obj is not an instanceof the specified klass
4064   // r0 = 1: obj != nullptr and obj is     an instanceof the specified klass
4065 }
4066 
4067 //-----------------------------------------------------------------------------
4068 // Breakpoints
4069 void TemplateTable::_breakpoint() {
4070   // Note: We get here even if we are single stepping..
4071   // jbug inists on setting breakpoints at every bytecode
4072   // even if we are in single step mode.
4073 
4074   transition(vtos, vtos);
4075 
4076   // get the unpatched byte code
4077   __ get_method(c_rarg1);
4078   __ call_VM(noreg,
4079              CAST_FROM_FN_PTR(address,
4080                               InterpreterRuntime::get_original_bytecode_at),
4081              c_rarg1, rbcp);
4082   __ mov(r19, r0);
4083 
4084   // post the breakpoint event
4085   __ call_VM(noreg,
4086              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4087              rmethod, rbcp);
4088 
4089   // complete the execution of original bytecode
4090   __ mov(rscratch1, r19);
4091   __ dispatch_only_normal(vtos);
4092 }
4093 
4094 //-----------------------------------------------------------------------------
4095 // Exceptions
4096 
4097 void TemplateTable::athrow() {
4098   transition(atos, vtos);
4099   __ null_check(r0);
4100   __ b(Interpreter::throw_exception_entry());
4101 }
4102 
4103 //-----------------------------------------------------------------------------
4104 // Synchronization
4105 //
4106 // Note: monitorenter & exit are symmetric routines; which is reflected
4107 //       in the assembly code structure as well
4108 //
4109 // Stack layout:
4110 //
4111 // [expressions  ] <--- esp               = expression stack top
4112 // ..
4113 // [expressions  ]
4114 // [monitor entry] <--- monitor block top = expression stack bot
4115 // ..
4116 // [monitor entry]
4117 // [frame data   ] <--- monitor block bot
4118 // ...
4119 // [saved rfp    ] <--- rfp
4120 void TemplateTable::monitorenter()
4121 {
4122   transition(atos, vtos);
4123 
4124   // check for null object
4125   __ null_check(r0);
4126 
4127   Label is_inline_type;
4128   __ ldr(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
4129   __ test_markword_is_inline_type(rscratch1, is_inline_type);
4130 
4131   const Address monitor_block_top(
4132         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4133   const Address monitor_block_bot(
4134         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4135   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4136 
4137   Label allocated;
4138 
4139   // initialize entry pointer
4140   __ mov(c_rarg1, zr); // points to free slot or null
4141 
4142   // find a free slot in the monitor block (result in c_rarg1)
4143   {
4144     Label entry, loop, exit;
4145     __ ldr(c_rarg3, monitor_block_top); // derelativize pointer
4146     __ lea(c_rarg3, Address(rfp, c_rarg3, Address::lsl(Interpreter::logStackElementSize)));
4147     // c_rarg3 points to current entry, starting with top-most entry
4148 
4149     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4150 
4151     __ b(entry);
4152 
4153     __ bind(loop);
4154     // check if current entry is used
4155     // if not used then remember entry in c_rarg1
4156     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset()));
4157     __ cmp(zr, rscratch1);
4158     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
4159     // check if current entry is for same object
4160     __ cmp(r0, rscratch1);
4161     // if same object then stop searching
4162     __ br(Assembler::EQ, exit);
4163     // otherwise advance to next entry
4164     __ add(c_rarg3, c_rarg3, entry_size);
4165     __ bind(entry);
4166     // check if bottom reached
4167     __ cmp(c_rarg3, c_rarg2);
4168     // if not at bottom then check this entry
4169     __ br(Assembler::NE, loop);
4170     __ bind(exit);
4171   }
4172 
4173   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
4174                             // if found, continue with that on
4175 
4176   // allocate one if there's no free slot
4177   {
4178     Label entry, loop;
4179     // 1. compute new pointers            // rsp: old expression stack top
4180 
4181     __ check_extended_sp();
4182     __ sub(sp, sp, entry_size);           // make room for the monitor
4183     __ sub(rscratch1, sp, rfp);
4184     __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
4185     __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
4186 
4187     __ ldr(c_rarg1, monitor_block_bot);   // derelativize pointer
4188     __ lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
4189     // c_rarg1 points to the old expression stack bottom
4190 
4191     __ sub(esp, esp, entry_size);         // move expression stack top
4192     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
4193     __ mov(c_rarg3, esp);                 // set start value for copy loop
4194     __ sub(rscratch1, c_rarg1, rfp);      // relativize pointer
4195     __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
4196     __ str(rscratch1, monitor_block_bot);  // set new monitor block bottom
4197 
4198     __ b(entry);
4199     // 2. move expression stack contents
4200     __ bind(loop);
4201     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
4202                                                    // word from old location
4203     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
4204     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
4205     __ bind(entry);
4206     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
4207     __ br(Assembler::NE, loop);      // if not at bottom then
4208                                      // copy next word
4209   }
4210 
4211   // call run-time routine
4212   // c_rarg1: points to monitor entry
4213   __ bind(allocated);
4214 
4215   // Increment bcp to point to the next bytecode, so exception
4216   // handling for async. exceptions work correctly.
4217   // The object has already been popped from the stack, so the
4218   // expression stack looks correct.
4219   __ increment(rbcp);
4220 
4221   // store object
4222   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
4223   __ lock_object(c_rarg1);
4224 
4225   // check to make sure this monitor doesn't cause stack overflow after locking
4226   __ save_bcp();  // in case of exception
4227   __ generate_stack_overflow_check(0);
4228 
4229   // The bcp has already been incremented. Just need to dispatch to
4230   // next instruction.
4231   __ dispatch_next(vtos);
4232 
4233   __ bind(is_inline_type);
4234   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4235                     InterpreterRuntime::throw_identity_exception), r0);
4236   __ should_not_reach_here();
4237 }
4238 
4239 
4240 void TemplateTable::monitorexit()
4241 {
4242   transition(atos, vtos);
4243 
4244   // check for null object
4245   __ null_check(r0);
4246 
4247   const int is_inline_type_mask = markWord::inline_type_pattern;
4248   Label has_identity;
4249   __ ldr(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
4250   __ mov(rscratch2, is_inline_type_mask);
4251   __ andr(rscratch1, rscratch1, rscratch2);
4252   __ cmp(rscratch1, rscratch2);
4253   __ br(Assembler::NE, has_identity);
4254   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4255                      InterpreterRuntime::throw_illegal_monitor_state_exception));
4256   __ should_not_reach_here();
4257   __ bind(has_identity);
4258 
4259   const Address monitor_block_top(
4260         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4261   const Address monitor_block_bot(
4262         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4263   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4264 
4265   Label found;
4266 
4267   // find matching slot
4268   {
4269     Label entry, loop;
4270     __ ldr(c_rarg1, monitor_block_top); // derelativize pointer
4271     __ lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
4272     // c_rarg1 points to current entry, starting with top-most entry
4273 
4274     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4275                                         // of monitor block
4276     __ b(entry);
4277 
4278     __ bind(loop);
4279     // check if current entry is for same object
4280     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
4281     __ cmp(r0, rscratch1);
4282     // if same object then stop searching
4283     __ br(Assembler::EQ, found);
4284     // otherwise advance to next entry
4285     __ add(c_rarg1, c_rarg1, entry_size);
4286     __ bind(entry);
4287     // check if bottom reached
4288     __ cmp(c_rarg1, c_rarg2);
4289     // if not at bottom then check this entry
4290     __ br(Assembler::NE, loop);
4291   }
4292 
4293   // error handling. Unlocking was not block-structured
4294   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4295                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4296   __ should_not_reach_here();
4297 
4298   // call run-time routine
4299   __ bind(found);
4300   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
4301   __ unlock_object(c_rarg1);
4302   __ pop_ptr(r0); // discard object
4303 }
4304 
4305 
4306 // Wide instructions
4307 void TemplateTable::wide()
4308 {
4309   __ load_unsigned_byte(r19, at_bcp(1));
4310   __ mov(rscratch1, (address)Interpreter::_wentry_point);
4311   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
4312   __ br(rscratch1);
4313 }
4314 
4315 
4316 // Multi arrays
4317 void TemplateTable::multianewarray() {
4318   transition(vtos, atos);
4319   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
4320   // last dim is on top of stack; we want address of first one:
4321   // first_addr = last_addr + (ndims - 1) * wordSize
4322   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
4323   __ sub(c_rarg1, c_rarg1, wordSize);
4324   call_VM(r0,
4325           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
4326           c_rarg1);
4327   __ load_unsigned_byte(r1, at_bcp(3));
4328   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
4329 }