1 /*
   2  * Copyright (c) 2003, 2025, 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 "asm/macroAssembler.inline.hpp"
  27 #include "compiler/disassembler.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 __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _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_metadata(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 (UseArrayFlattening) {
 822     Label is_flat_array, done;
 823 
 824     __ test_flat_array_oop(r0, rscratch1 /*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::flat_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 (UseArrayFlattening) {
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   // Clobbers: r10, r11, r3
1177   do_oop_store(_masm, element_address, r0, IS_ARRAY);
1178   __ b(done);
1179 
1180   // Have a null in r0, r3=array, r2=index.  Store null at ary[idx]
1181   __ bind(is_null);
1182   if (EnableValhalla) {
1183     Label is_null_into_value_array_npe, store_null;
1184 
1185     if (UseArrayFlattening) {
1186       __ test_flat_array_oop(r3, rscratch1, is_flat_array);
1187     }
1188 
1189     // No way to store null in a null-free array
1190     __ test_null_free_array_oop(r3, rscratch1, is_null_into_value_array_npe);
1191     __ b(store_null);
1192 
1193     __ bind(is_null_into_value_array_npe);
1194     __ b(ExternalAddress(Interpreter::_throw_NullPointerException_entry));
1195 
1196     __ bind(store_null);
1197   }
1198 
1199   // Store a null
1200   // Clobbers: r10, r11, r3
1201   do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1202   __ b(done);
1203 
1204   if (UseArrayFlattening) {
1205      Label is_type_ok;
1206     __ bind(is_flat_array); // Store non-null value to flat
1207 
1208     __ ldr(r0, at_tos());    // value
1209     __ ldr(r3, at_tos_p1()); // index
1210     __ ldr(r2, at_tos_p2()); // array
1211     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::flat_array_store), r0, r2, r3);
1212   }
1213 
1214   // Pop stack arguments
1215   __ bind(done);
1216   __ add(esp, esp, 3 * Interpreter::stackElementSize);
1217 }
1218 
1219 void TemplateTable::bastore()
1220 {
1221   transition(itos, vtos);
1222   __ pop_i(r1);
1223   __ pop_ptr(r3);
1224   // r0: value
1225   // r1: index
1226   // r3: array
1227   index_check(r3, r1); // prefer index in r1
1228 
1229   // Need to check whether array is boolean or byte
1230   // since both types share the bastore bytecode.
1231   __ load_klass(r2, r3);
1232   __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1233   int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1234   Label L_skip;
1235   __ tbz(r2, diffbit_index, L_skip);
1236   __ andw(r0, r0, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1237   __ bind(L_skip);
1238 
1239   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
1240   __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(0)), r0, noreg, noreg, noreg);
1241 }
1242 
1243 void TemplateTable::castore()
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   __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
1253   __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(1)), r0, noreg, noreg, noreg);
1254 }
1255 
1256 void TemplateTable::sastore()
1257 {
1258   castore();
1259 }
1260 
1261 void TemplateTable::istore(int n)
1262 {
1263   transition(itos, vtos);
1264   __ str(r0, iaddress(n));
1265 }
1266 
1267 void TemplateTable::lstore(int n)
1268 {
1269   transition(ltos, vtos);
1270   __ str(r0, laddress(n));
1271 }
1272 
1273 void TemplateTable::fstore(int n)
1274 {
1275   transition(ftos, vtos);
1276   __ strs(v0, faddress(n));
1277 }
1278 
1279 void TemplateTable::dstore(int n)
1280 {
1281   transition(dtos, vtos);
1282   __ strd(v0, daddress(n));
1283 }
1284 
1285 void TemplateTable::astore(int n)
1286 {
1287   transition(vtos, vtos);
1288   __ pop_ptr(r0);
1289   __ str(r0, iaddress(n));
1290 }
1291 
1292 void TemplateTable::pop()
1293 {
1294   transition(vtos, vtos);
1295   __ add(esp, esp, Interpreter::stackElementSize);
1296 }
1297 
1298 void TemplateTable::pop2()
1299 {
1300   transition(vtos, vtos);
1301   __ add(esp, esp, 2 * Interpreter::stackElementSize);
1302 }
1303 
1304 void TemplateTable::dup()
1305 {
1306   transition(vtos, vtos);
1307   __ ldr(r0, Address(esp, 0));
1308   __ push(r0);
1309   // stack: ..., a, a
1310 }
1311 
1312 void TemplateTable::dup_x1()
1313 {
1314   transition(vtos, vtos);
1315   // stack: ..., a, b
1316   __ ldr(r0, at_tos());  // load b
1317   __ ldr(r2, at_tos_p1());  // load a
1318   __ str(r0, at_tos_p1());  // store b
1319   __ str(r2, at_tos());  // store a
1320   __ push(r0);                  // push b
1321   // stack: ..., b, a, b
1322 }
1323 
1324 void TemplateTable::dup_x2()
1325 {
1326   transition(vtos, vtos);
1327   // stack: ..., a, b, c
1328   __ ldr(r0, at_tos());  // load c
1329   __ ldr(r2, at_tos_p2());  // load a
1330   __ str(r0, at_tos_p2());  // store c in a
1331   __ push(r0);      // push c
1332   // stack: ..., c, b, c, c
1333   __ ldr(r0, at_tos_p2());  // load b
1334   __ str(r2, at_tos_p2());  // store a in b
1335   // stack: ..., c, a, c, c
1336   __ str(r0, at_tos_p1());  // store b in c
1337   // stack: ..., c, a, b, c
1338 }
1339 
1340 void TemplateTable::dup2()
1341 {
1342   transition(vtos, vtos);
1343   // stack: ..., a, b
1344   __ ldr(r0, at_tos_p1());  // load a
1345   __ push(r0);                  // push a
1346   __ ldr(r0, at_tos_p1());  // load b
1347   __ push(r0);                  // push b
1348   // stack: ..., a, b, a, b
1349 }
1350 
1351 void TemplateTable::dup2_x1()
1352 {
1353   transition(vtos, vtos);
1354   // stack: ..., a, b, c
1355   __ ldr(r2, at_tos());  // load c
1356   __ ldr(r0, at_tos_p1());  // load b
1357   __ push(r0);                  // push b
1358   __ push(r2);                  // push c
1359   // stack: ..., a, b, c, b, c
1360   __ str(r2, at_tos_p3());  // store c in b
1361   // stack: ..., a, c, c, b, c
1362   __ ldr(r2, at_tos_p4());  // load a
1363   __ str(r2, at_tos_p2());  // store a in 2nd c
1364   // stack: ..., a, c, a, b, c
1365   __ str(r0, at_tos_p4());  // store b in a
1366   // stack: ..., b, c, a, b, c
1367 }
1368 
1369 void TemplateTable::dup2_x2()
1370 {
1371   transition(vtos, vtos);
1372   // stack: ..., a, b, c, d
1373   __ ldr(r2, at_tos());  // load d
1374   __ ldr(r0, at_tos_p1());  // load c
1375   __ push(r0)            ;      // push c
1376   __ push(r2);                  // push d
1377   // stack: ..., a, b, c, d, c, d
1378   __ ldr(r0, at_tos_p4());  // load b
1379   __ str(r0, at_tos_p2());  // store b in d
1380   __ str(r2, at_tos_p4());  // store d in b
1381   // stack: ..., a, d, c, b, c, d
1382   __ ldr(r2, at_tos_p5());  // load a
1383   __ ldr(r0, at_tos_p3());  // load c
1384   __ str(r2, at_tos_p3());  // store a in c
1385   __ str(r0, at_tos_p5());  // store c in a
1386   // stack: ..., c, d, a, b, c, d
1387 }
1388 
1389 void TemplateTable::swap()
1390 {
1391   transition(vtos, vtos);
1392   // stack: ..., a, b
1393   __ ldr(r2, at_tos_p1());  // load a
1394   __ ldr(r0, at_tos());  // load b
1395   __ str(r2, at_tos());  // store a in b
1396   __ str(r0, at_tos_p1());  // store b in a
1397   // stack: ..., b, a
1398 }
1399 
1400 void TemplateTable::iop2(Operation op)
1401 {
1402   transition(itos, itos);
1403   // r0 <== r1 op r0
1404   __ pop_i(r1);
1405   switch (op) {
1406   case add  : __ addw(r0, r1, r0); break;
1407   case sub  : __ subw(r0, r1, r0); break;
1408   case mul  : __ mulw(r0, r1, r0); break;
1409   case _and : __ andw(r0, r1, r0); break;
1410   case _or  : __ orrw(r0, r1, r0); break;
1411   case _xor : __ eorw(r0, r1, r0); break;
1412   case shl  : __ lslvw(r0, r1, r0); break;
1413   case shr  : __ asrvw(r0, r1, r0); break;
1414   case ushr : __ lsrvw(r0, r1, r0);break;
1415   default   : ShouldNotReachHere();
1416   }
1417 }
1418 
1419 void TemplateTable::lop2(Operation op)
1420 {
1421   transition(ltos, ltos);
1422   // r0 <== r1 op r0
1423   __ pop_l(r1);
1424   switch (op) {
1425   case add  : __ add(r0, r1, r0); break;
1426   case sub  : __ sub(r0, r1, r0); break;
1427   case mul  : __ mul(r0, r1, r0); break;
1428   case _and : __ andr(r0, r1, r0); break;
1429   case _or  : __ orr(r0, r1, r0); break;
1430   case _xor : __ eor(r0, r1, r0); break;
1431   default   : ShouldNotReachHere();
1432   }
1433 }
1434 
1435 void TemplateTable::idiv()
1436 {
1437   transition(itos, itos);
1438   // explicitly check for div0
1439   Label no_div0;
1440   __ cbnzw(r0, no_div0);
1441   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1442   __ br(rscratch1);
1443   __ bind(no_div0);
1444   __ pop_i(r1);
1445   // r0 <== r1 idiv r0
1446   __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1447 }
1448 
1449 void TemplateTable::irem()
1450 {
1451   transition(itos, itos);
1452   // explicitly check for div0
1453   Label no_div0;
1454   __ cbnzw(r0, no_div0);
1455   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1456   __ br(rscratch1);
1457   __ bind(no_div0);
1458   __ pop_i(r1);
1459   // r0 <== r1 irem r0
1460   __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1461 }
1462 
1463 void TemplateTable::lmul()
1464 {
1465   transition(ltos, ltos);
1466   __ pop_l(r1);
1467   __ mul(r0, r0, r1);
1468 }
1469 
1470 void TemplateTable::ldiv()
1471 {
1472   transition(ltos, ltos);
1473   // explicitly check for div0
1474   Label no_div0;
1475   __ cbnz(r0, no_div0);
1476   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1477   __ br(rscratch1);
1478   __ bind(no_div0);
1479   __ pop_l(r1);
1480   // r0 <== r1 ldiv r0
1481   __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1482 }
1483 
1484 void TemplateTable::lrem()
1485 {
1486   transition(ltos, ltos);
1487   // explicitly check for div0
1488   Label no_div0;
1489   __ cbnz(r0, no_div0);
1490   __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1491   __ br(rscratch1);
1492   __ bind(no_div0);
1493   __ pop_l(r1);
1494   // r0 <== r1 lrem r0
1495   __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1496 }
1497 
1498 void TemplateTable::lshl()
1499 {
1500   transition(itos, ltos);
1501   // shift count is in r0
1502   __ pop_l(r1);
1503   __ lslv(r0, r1, r0);
1504 }
1505 
1506 void TemplateTable::lshr()
1507 {
1508   transition(itos, ltos);
1509   // shift count is in r0
1510   __ pop_l(r1);
1511   __ asrv(r0, r1, r0);
1512 }
1513 
1514 void TemplateTable::lushr()
1515 {
1516   transition(itos, ltos);
1517   // shift count is in r0
1518   __ pop_l(r1);
1519   __ lsrv(r0, r1, r0);
1520 }
1521 
1522 void TemplateTable::fop2(Operation op)
1523 {
1524   transition(ftos, ftos);
1525   switch (op) {
1526   case add:
1527     // n.b. use ldrd because this is a 64 bit slot
1528     __ pop_f(v1);
1529     __ fadds(v0, v1, v0);
1530     break;
1531   case sub:
1532     __ pop_f(v1);
1533     __ fsubs(v0, v1, v0);
1534     break;
1535   case mul:
1536     __ pop_f(v1);
1537     __ fmuls(v0, v1, v0);
1538     break;
1539   case div:
1540     __ pop_f(v1);
1541     __ fdivs(v0, v1, v0);
1542     break;
1543   case rem:
1544     __ fmovs(v1, v0);
1545     __ pop_f(v0);
1546     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1547     break;
1548   default:
1549     ShouldNotReachHere();
1550     break;
1551   }
1552 }
1553 
1554 void TemplateTable::dop2(Operation op)
1555 {
1556   transition(dtos, dtos);
1557   switch (op) {
1558   case add:
1559     // n.b. use ldrd because this is a 64 bit slot
1560     __ pop_d(v1);
1561     __ faddd(v0, v1, v0);
1562     break;
1563   case sub:
1564     __ pop_d(v1);
1565     __ fsubd(v0, v1, v0);
1566     break;
1567   case mul:
1568     __ pop_d(v1);
1569     __ fmuld(v0, v1, v0);
1570     break;
1571   case div:
1572     __ pop_d(v1);
1573     __ fdivd(v0, v1, v0);
1574     break;
1575   case rem:
1576     __ fmovd(v1, v0);
1577     __ pop_d(v0);
1578     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1579     break;
1580   default:
1581     ShouldNotReachHere();
1582     break;
1583   }
1584 }
1585 
1586 void TemplateTable::ineg()
1587 {
1588   transition(itos, itos);
1589   __ negw(r0, r0);
1590 
1591 }
1592 
1593 void TemplateTable::lneg()
1594 {
1595   transition(ltos, ltos);
1596   __ neg(r0, r0);
1597 }
1598 
1599 void TemplateTable::fneg()
1600 {
1601   transition(ftos, ftos);
1602   __ fnegs(v0, v0);
1603 }
1604 
1605 void TemplateTable::dneg()
1606 {
1607   transition(dtos, dtos);
1608   __ fnegd(v0, v0);
1609 }
1610 
1611 void TemplateTable::iinc()
1612 {
1613   transition(vtos, vtos);
1614   __ load_signed_byte(r1, at_bcp(2)); // get constant
1615   locals_index(r2);
1616   __ ldr(r0, iaddress(r2));
1617   __ addw(r0, r0, r1);
1618   __ str(r0, iaddress(r2));
1619 }
1620 
1621 void TemplateTable::wide_iinc()
1622 {
1623   transition(vtos, vtos);
1624   // __ mov(r1, zr);
1625   __ ldrw(r1, at_bcp(2)); // get constant and index
1626   __ rev16(r1, r1);
1627   __ ubfx(r2, r1, 0, 16);
1628   __ neg(r2, r2);
1629   __ sbfx(r1, r1, 16, 16);
1630   __ ldr(r0, iaddress(r2));
1631   __ addw(r0, r0, r1);
1632   __ str(r0, iaddress(r2));
1633 }
1634 
1635 void TemplateTable::convert()
1636 {
1637   // Checking
1638 #ifdef ASSERT
1639   {
1640     TosState tos_in  = ilgl;
1641     TosState tos_out = ilgl;
1642     switch (bytecode()) {
1643     case Bytecodes::_i2l: // fall through
1644     case Bytecodes::_i2f: // fall through
1645     case Bytecodes::_i2d: // fall through
1646     case Bytecodes::_i2b: // fall through
1647     case Bytecodes::_i2c: // fall through
1648     case Bytecodes::_i2s: tos_in = itos; break;
1649     case Bytecodes::_l2i: // fall through
1650     case Bytecodes::_l2f: // fall through
1651     case Bytecodes::_l2d: tos_in = ltos; break;
1652     case Bytecodes::_f2i: // fall through
1653     case Bytecodes::_f2l: // fall through
1654     case Bytecodes::_f2d: tos_in = ftos; break;
1655     case Bytecodes::_d2i: // fall through
1656     case Bytecodes::_d2l: // fall through
1657     case Bytecodes::_d2f: tos_in = dtos; break;
1658     default             : ShouldNotReachHere();
1659     }
1660     switch (bytecode()) {
1661     case Bytecodes::_l2i: // fall through
1662     case Bytecodes::_f2i: // fall through
1663     case Bytecodes::_d2i: // fall through
1664     case Bytecodes::_i2b: // fall through
1665     case Bytecodes::_i2c: // fall through
1666     case Bytecodes::_i2s: tos_out = itos; break;
1667     case Bytecodes::_i2l: // fall through
1668     case Bytecodes::_f2l: // fall through
1669     case Bytecodes::_d2l: tos_out = ltos; break;
1670     case Bytecodes::_i2f: // fall through
1671     case Bytecodes::_l2f: // fall through
1672     case Bytecodes::_d2f: tos_out = ftos; break;
1673     case Bytecodes::_i2d: // fall through
1674     case Bytecodes::_l2d: // fall through
1675     case Bytecodes::_f2d: tos_out = dtos; break;
1676     default             : ShouldNotReachHere();
1677     }
1678     transition(tos_in, tos_out);
1679   }
1680 #endif // ASSERT
1681   // static const int64_t is_nan = 0x8000000000000000L;
1682 
1683   // Conversion
1684   switch (bytecode()) {
1685   case Bytecodes::_i2l:
1686     __ sxtw(r0, r0);
1687     break;
1688   case Bytecodes::_i2f:
1689     __ scvtfws(v0, r0);
1690     break;
1691   case Bytecodes::_i2d:
1692     __ scvtfwd(v0, r0);
1693     break;
1694   case Bytecodes::_i2b:
1695     __ sxtbw(r0, r0);
1696     break;
1697   case Bytecodes::_i2c:
1698     __ uxthw(r0, r0);
1699     break;
1700   case Bytecodes::_i2s:
1701     __ sxthw(r0, r0);
1702     break;
1703   case Bytecodes::_l2i:
1704     __ uxtw(r0, r0);
1705     break;
1706   case Bytecodes::_l2f:
1707     __ scvtfs(v0, r0);
1708     break;
1709   case Bytecodes::_l2d:
1710     __ scvtfd(v0, r0);
1711     break;
1712   case Bytecodes::_f2i:
1713   {
1714     Label L_Okay;
1715     __ clear_fpsr();
1716     __ fcvtzsw(r0, v0);
1717     __ get_fpsr(r1);
1718     __ cbzw(r1, L_Okay);
1719     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i));
1720     __ bind(L_Okay);
1721   }
1722     break;
1723   case Bytecodes::_f2l:
1724   {
1725     Label L_Okay;
1726     __ clear_fpsr();
1727     __ fcvtzs(r0, v0);
1728     __ get_fpsr(r1);
1729     __ cbzw(r1, L_Okay);
1730     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1731     __ bind(L_Okay);
1732   }
1733     break;
1734   case Bytecodes::_f2d:
1735     __ fcvts(v0, v0);
1736     break;
1737   case Bytecodes::_d2i:
1738   {
1739     Label L_Okay;
1740     __ clear_fpsr();
1741     __ fcvtzdw(r0, v0);
1742     __ get_fpsr(r1);
1743     __ cbzw(r1, L_Okay);
1744     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1745     __ bind(L_Okay);
1746   }
1747     break;
1748   case Bytecodes::_d2l:
1749   {
1750     Label L_Okay;
1751     __ clear_fpsr();
1752     __ fcvtzd(r0, v0);
1753     __ get_fpsr(r1);
1754     __ cbzw(r1, L_Okay);
1755     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1756     __ bind(L_Okay);
1757   }
1758     break;
1759   case Bytecodes::_d2f:
1760     __ fcvtd(v0, v0);
1761     break;
1762   default:
1763     ShouldNotReachHere();
1764   }
1765 }
1766 
1767 void TemplateTable::lcmp()
1768 {
1769   transition(ltos, itos);
1770   Label done;
1771   __ pop_l(r1);
1772   __ cmp(r1, r0);
1773   __ mov(r0, (uint64_t)-1L);
1774   __ br(Assembler::LT, done);
1775   // __ mov(r0, 1UL);
1776   // __ csel(r0, r0, zr, Assembler::NE);
1777   // and here is a faster way
1778   __ csinc(r0, zr, zr, Assembler::EQ);
1779   __ bind(done);
1780 }
1781 
1782 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1783 {
1784   Label done;
1785   if (is_float) {
1786     // XXX get rid of pop here, use ... reg, mem32
1787     __ pop_f(v1);
1788     __ fcmps(v1, v0);
1789   } else {
1790     // XXX get rid of pop here, use ... reg, mem64
1791     __ pop_d(v1);
1792     __ fcmpd(v1, v0);
1793   }
1794   if (unordered_result < 0) {
1795     // we want -1 for unordered or less than, 0 for equal and 1 for
1796     // greater than.
1797     __ mov(r0, (uint64_t)-1L);
1798     // for FP LT tests less than or unordered
1799     __ br(Assembler::LT, done);
1800     // install 0 for EQ otherwise 1
1801     __ csinc(r0, zr, zr, Assembler::EQ);
1802   } else {
1803     // we want -1 for less than, 0 for equal and 1 for unordered or
1804     // greater than.
1805     __ mov(r0, 1L);
1806     // for FP HI tests greater than or unordered
1807     __ br(Assembler::HI, done);
1808     // install 0 for EQ otherwise ~0
1809     __ csinv(r0, zr, zr, Assembler::EQ);
1810 
1811   }
1812   __ bind(done);
1813 }
1814 
1815 void TemplateTable::branch(bool is_jsr, bool is_wide)
1816 {
1817   __ profile_taken_branch(r0);
1818   const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1819                              InvocationCounter::counter_offset();
1820   const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1821                               InvocationCounter::counter_offset();
1822 
1823   // load branch displacement
1824   if (!is_wide) {
1825     __ ldrh(r2, at_bcp(1));
1826     __ rev16(r2, r2);
1827     // sign extend the 16 bit value in r2
1828     __ sbfm(r2, r2, 0, 15);
1829   } else {
1830     __ ldrw(r2, at_bcp(1));
1831     __ revw(r2, r2);
1832     // sign extend the 32 bit value in r2
1833     __ sbfm(r2, r2, 0, 31);
1834   }
1835 
1836   // Handle all the JSR stuff here, then exit.
1837   // It's much shorter and cleaner than intermingling with the non-JSR
1838   // normal-branch stuff occurring below.
1839 
1840   if (is_jsr) {
1841     // Pre-load the next target bytecode into rscratch1
1842     __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1843     // compute return address as bci
1844     __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1845     __ add(rscratch2, rscratch2,
1846            in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1847     __ sub(r1, rbcp, rscratch2);
1848     __ push_i(r1);
1849     // Adjust the bcp by the 16-bit displacement in r2
1850     __ add(rbcp, rbcp, r2);
1851     __ dispatch_only(vtos, /*generate_poll*/true);
1852     return;
1853   }
1854 
1855   // Normal (non-jsr) branch handling
1856 
1857   // Adjust the bcp by the displacement in r2
1858   __ add(rbcp, rbcp, r2);
1859 
1860   assert(UseLoopCounter || !UseOnStackReplacement,
1861          "on-stack-replacement requires loop counters");
1862   Label backedge_counter_overflow;
1863   Label dispatch;
1864   if (UseLoopCounter) {
1865     // increment backedge counter for backward branches
1866     // r0: MDO
1867     // r2: target offset
1868     __ cmp(r2, zr);
1869     __ br(Assembler::GT, dispatch); // count only if backward branch
1870 
1871     // ECN: FIXME: This code smells
1872     // check if MethodCounters exists
1873     Label has_counters;
1874     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1875     __ cbnz(rscratch1, has_counters);
1876     __ push(r0);
1877     __ push(r2);
1878     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1879             InterpreterRuntime::build_method_counters), rmethod);
1880     __ pop(r2);
1881     __ pop(r0);
1882     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1883     __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1884     __ bind(has_counters);
1885 
1886     Label no_mdo;
1887     int increment = InvocationCounter::count_increment;
1888     if (ProfileInterpreter) {
1889       // Are we profiling?
1890       __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1891       __ cbz(r1, no_mdo);
1892       // Increment the MDO backedge counter
1893       const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1894                                          in_bytes(InvocationCounter::counter_offset()));
1895       const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1896       __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1897                                  r0, rscratch1, false, Assembler::EQ,
1898                                  UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1899       __ b(dispatch);
1900     }
1901     __ bind(no_mdo);
1902     // Increment backedge counter in MethodCounters*
1903     __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1904     const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1905     __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1906                                r0, rscratch2, false, Assembler::EQ,
1907                                UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1908     __ bind(dispatch);
1909   }
1910 
1911   // Pre-load the next target bytecode into rscratch1
1912   __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1913 
1914   // continue with the bytecode @ target
1915   // rscratch1: target bytecode
1916   // rbcp: target bcp
1917   __ dispatch_only(vtos, /*generate_poll*/true);
1918 
1919   if (UseLoopCounter && UseOnStackReplacement) {
1920     // invocation counter overflow
1921     __ bind(backedge_counter_overflow);
1922     __ neg(r2, r2);
1923     __ add(r2, r2, rbcp);     // branch bcp
1924     // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1925     __ call_VM(noreg,
1926                CAST_FROM_FN_PTR(address,
1927                                 InterpreterRuntime::frequency_counter_overflow),
1928                r2);
1929     __ load_unsigned_byte(r1, Address(rbcp, 0));  // restore target bytecode
1930 
1931     // r0: osr nmethod (osr ok) or null (osr not possible)
1932     // w1: target bytecode
1933     // r2: scratch
1934     __ cbz(r0, dispatch);     // test result -- no osr if null
1935     // nmethod may have been invalidated (VM may block upon call_VM return)
1936     __ ldrb(r2, Address(r0, nmethod::state_offset()));
1937     if (nmethod::in_use != 0)
1938       __ sub(r2, r2, nmethod::in_use);
1939     __ cbnz(r2, dispatch);
1940 
1941     // We have the address of an on stack replacement routine in r0
1942     // We need to prepare to execute the OSR method. First we must
1943     // migrate the locals and monitors off of the stack.
1944 
1945     __ mov(r19, r0);                             // save the nmethod
1946 
1947     JFR_ONLY(__ enter_jfr_critical_section();)
1948 
1949     call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1950 
1951     // r0 is OSR buffer, move it to expected parameter location
1952     __ mov(j_rarg0, r0);
1953 
1954     // remove activation
1955     // get sender esp
1956     __ ldr(esp,
1957         Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1958     // remove frame anchor
1959     __ leave();
1960 
1961     JFR_ONLY(__ leave_jfr_critical_section();)
1962 
1963     // Ensure compiled code always sees stack at proper alignment
1964     __ andr(sp, esp, -16);
1965 
1966     // and begin the OSR nmethod
1967     __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1968     __ br(rscratch1);
1969   }
1970 }
1971 
1972 
1973 void TemplateTable::if_0cmp(Condition cc)
1974 {
1975   transition(itos, vtos);
1976   // assume branch is more often taken than not (loops use backward branches)
1977   Label not_taken;
1978   if (cc == equal)
1979     __ cbnzw(r0, not_taken);
1980   else if (cc == not_equal)
1981     __ cbzw(r0, not_taken);
1982   else {
1983     __ andsw(zr, r0, r0);
1984     __ br(j_not(cc), not_taken);
1985   }
1986 
1987   branch(false, false);
1988   __ bind(not_taken);
1989   __ profile_not_taken_branch(r0);
1990 }
1991 
1992 void TemplateTable::if_icmp(Condition cc)
1993 {
1994   transition(itos, vtos);
1995   // assume branch is more often taken than not (loops use backward branches)
1996   Label not_taken;
1997   __ pop_i(r1);
1998   __ cmpw(r1, r0, Assembler::LSL);
1999   __ br(j_not(cc), not_taken);
2000   branch(false, false);
2001   __ bind(not_taken);
2002   __ profile_not_taken_branch(r0);
2003 }
2004 
2005 void TemplateTable::if_nullcmp(Condition cc)
2006 {
2007   transition(atos, vtos);
2008   // assume branch is more often taken than not (loops use backward branches)
2009   Label not_taken;
2010   if (cc == equal)
2011     __ cbnz(r0, not_taken);
2012   else
2013     __ cbz(r0, not_taken);
2014   branch(false, false);
2015   __ bind(not_taken);
2016   __ profile_not_taken_branch(r0);
2017 }
2018 
2019 void TemplateTable::if_acmp(Condition cc) {
2020   transition(atos, vtos);
2021   // assume branch is more often taken than not (loops use backward branches)
2022   Label taken, not_taken;
2023   __ pop_ptr(r1);
2024 
2025   __ profile_acmp(r2, r1, r0, r4);
2026 
2027   Register is_inline_type_mask = rscratch1;
2028   __ mov(is_inline_type_mask, markWord::inline_type_pattern);
2029 
2030   if (EnableValhalla) {
2031     __ cmp(r1, r0);
2032     __ br(Assembler::EQ, (cc == equal) ? taken : not_taken);
2033 
2034     // might be substitutable, test if either r0 or r1 is null
2035     __ andr(r2, r0, r1);
2036     __ cbz(r2, (cc == equal) ? not_taken : taken);
2037 
2038     // and both are values ?
2039     __ ldr(r2, Address(r1, oopDesc::mark_offset_in_bytes()));
2040     __ andr(r2, r2, is_inline_type_mask);
2041     __ ldr(r4, Address(r0, oopDesc::mark_offset_in_bytes()));
2042     __ andr(r4, r4, is_inline_type_mask);
2043     __ andr(r2, r2, r4);
2044     __ cmp(r2,  is_inline_type_mask);
2045     __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2046 
2047     // same value klass ?
2048     __ load_metadata(r2, r1);
2049     __ load_metadata(r4, r0);
2050     __ cmp(r2, r4);
2051     __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2052 
2053     // Know both are the same type, let's test for substitutability...
2054     if (cc == equal) {
2055       invoke_is_substitutable(r0, r1, taken, not_taken);
2056     } else {
2057       invoke_is_substitutable(r0, r1, not_taken, taken);
2058     }
2059     __ stop("Not reachable");
2060   }
2061 
2062   __ cmpoop(r1, r0);
2063   __ br(j_not(cc), not_taken);
2064   __ bind(taken);
2065   branch(false, false);
2066   __ bind(not_taken);
2067   __ profile_not_taken_branch(r0, true);
2068 }
2069 
2070 void TemplateTable::invoke_is_substitutable(Register aobj, Register bobj,
2071                                             Label& is_subst, Label& not_subst) {
2072 
2073   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::is_substitutable), aobj, bobj);
2074   // Restored... r0 answer, jmp to outcome...
2075   __ cbz(r0, not_subst);
2076   __ b(is_subst);
2077 }
2078 
2079 
2080 void TemplateTable::ret() {
2081   transition(vtos, vtos);
2082   locals_index(r1);
2083   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2084   __ profile_ret(r1, r2);
2085   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2086   __ lea(rbcp, Address(rbcp, r1));
2087   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2088   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2089 }
2090 
2091 void TemplateTable::wide_ret() {
2092   transition(vtos, vtos);
2093   locals_index_wide(r1);
2094   __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2095   __ profile_ret(r1, r2);
2096   __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2097   __ lea(rbcp, Address(rbcp, r1));
2098   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2099   __ dispatch_next(vtos, 0, /*generate_poll*/true);
2100 }
2101 
2102 
2103 void TemplateTable::tableswitch() {
2104   Label default_case, continue_execution;
2105   transition(itos, vtos);
2106   // align rbcp
2107   __ lea(r1, at_bcp(BytesPerInt));
2108   __ andr(r1, r1, -BytesPerInt);
2109   // load lo & hi
2110   __ ldrw(r2, Address(r1, BytesPerInt));
2111   __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2112   __ rev32(r2, r2);
2113   __ rev32(r3, r3);
2114   // check against lo & hi
2115   __ cmpw(r0, r2);
2116   __ br(Assembler::LT, default_case);
2117   __ cmpw(r0, r3);
2118   __ br(Assembler::GT, default_case);
2119   // lookup dispatch offset
2120   __ subw(r0, r0, r2);
2121   __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2122   __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2123   __ profile_switch_case(r0, r1, r2);
2124   // continue execution
2125   __ bind(continue_execution);
2126   __ rev32(r3, r3);
2127   __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2128   __ add(rbcp, rbcp, r3, ext::sxtw);
2129   __ dispatch_only(vtos, /*generate_poll*/true);
2130   // handle default
2131   __ bind(default_case);
2132   __ profile_switch_default(r0);
2133   __ ldrw(r3, Address(r1, 0));
2134   __ b(continue_execution);
2135 }
2136 
2137 void TemplateTable::lookupswitch() {
2138   transition(itos, itos);
2139   __ stop("lookupswitch bytecode should have been rewritten");
2140 }
2141 
2142 void TemplateTable::fast_linearswitch() {
2143   transition(itos, vtos);
2144   Label loop_entry, loop, found, continue_execution;
2145   // bswap r0 so we can avoid bswapping the table entries
2146   __ rev32(r0, r0);
2147   // align rbcp
2148   __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2149                                     // this instruction (change offsets
2150                                     // below)
2151   __ andr(r19, r19, -BytesPerInt);
2152   // set counter
2153   __ ldrw(r1, Address(r19, BytesPerInt));
2154   __ rev32(r1, r1);
2155   __ b(loop_entry);
2156   // table search
2157   __ bind(loop);
2158   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2159   __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2160   __ cmpw(r0, rscratch1);
2161   __ br(Assembler::EQ, found);
2162   __ bind(loop_entry);
2163   __ subs(r1, r1, 1);
2164   __ br(Assembler::PL, loop);
2165   // default case
2166   __ profile_switch_default(r0);
2167   __ ldrw(r3, Address(r19, 0));
2168   __ b(continue_execution);
2169   // entry found -> get offset
2170   __ bind(found);
2171   __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2172   __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2173   __ profile_switch_case(r1, r0, r19);
2174   // continue execution
2175   __ bind(continue_execution);
2176   __ rev32(r3, r3);
2177   __ add(rbcp, rbcp, r3, ext::sxtw);
2178   __ ldrb(rscratch1, Address(rbcp, 0));
2179   __ dispatch_only(vtos, /*generate_poll*/true);
2180 }
2181 
2182 void TemplateTable::fast_binaryswitch() {
2183   transition(itos, vtos);
2184   // Implementation using the following core algorithm:
2185   //
2186   // int binary_search(int key, LookupswitchPair* array, int n) {
2187   //   // Binary search according to "Methodik des Programmierens" by
2188   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2189   //   int i = 0;
2190   //   int j = n;
2191   //   while (i+1 < j) {
2192   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2193   //     // with      Q: for all i: 0 <= i < n: key < a[i]
2194   //     // where a stands for the array and assuming that the (inexisting)
2195   //     // element a[n] is infinitely big.
2196   //     int h = (i + j) >> 1;
2197   //     // i < h < j
2198   //     if (key < array[h].fast_match()) {
2199   //       j = h;
2200   //     } else {
2201   //       i = h;
2202   //     }
2203   //   }
2204   //   // R: a[i] <= key < a[i+1] or Q
2205   //   // (i.e., if key is within array, i is the correct index)
2206   //   return i;
2207   // }
2208 
2209   // Register allocation
2210   const Register key   = r0; // already set (tosca)
2211   const Register array = r1;
2212   const Register i     = r2;
2213   const Register j     = r3;
2214   const Register h     = rscratch1;
2215   const Register temp  = rscratch2;
2216 
2217   // Find array start
2218   __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2219                                           // get rid of this
2220                                           // instruction (change
2221                                           // offsets below)
2222   __ andr(array, array, -BytesPerInt);
2223 
2224   // Initialize i & j
2225   __ mov(i, 0);                            // i = 0;
2226   __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2227 
2228   // Convert j into native byteordering
2229   __ rev32(j, j);
2230 
2231   // And start
2232   Label entry;
2233   __ b(entry);
2234 
2235   // binary search loop
2236   {
2237     Label loop;
2238     __ bind(loop);
2239     // int h = (i + j) >> 1;
2240     __ addw(h, i, j);                           // h = i + j;
2241     __ lsrw(h, h, 1);                                   // h = (i + j) >> 1;
2242     // if (key < array[h].fast_match()) {
2243     //   j = h;
2244     // } else {
2245     //   i = h;
2246     // }
2247     // Convert array[h].match to native byte-ordering before compare
2248     __ ldr(temp, Address(array, h, Address::lsl(3)));
2249     __ rev32(temp, temp);
2250     __ cmpw(key, temp);
2251     // j = h if (key <  array[h].fast_match())
2252     __ csel(j, h, j, Assembler::LT);
2253     // i = h if (key >= array[h].fast_match())
2254     __ csel(i, h, i, Assembler::GE);
2255     // while (i+1 < j)
2256     __ bind(entry);
2257     __ addw(h, i, 1);          // i+1
2258     __ cmpw(h, j);             // i+1 < j
2259     __ br(Assembler::LT, loop);
2260   }
2261 
2262   // end of binary search, result index is i (must check again!)
2263   Label default_case;
2264   // Convert array[i].match to native byte-ordering before compare
2265   __ ldr(temp, Address(array, i, Address::lsl(3)));
2266   __ rev32(temp, temp);
2267   __ cmpw(key, temp);
2268   __ br(Assembler::NE, default_case);
2269 
2270   // entry found -> j = offset
2271   __ add(j, array, i, ext::uxtx, 3);
2272   __ ldrw(j, Address(j, BytesPerInt));
2273   __ profile_switch_case(i, key, array);
2274   __ rev32(j, j);
2275   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2276   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2277   __ dispatch_only(vtos, /*generate_poll*/true);
2278 
2279   // default case -> j = default offset
2280   __ bind(default_case);
2281   __ profile_switch_default(i);
2282   __ ldrw(j, Address(array, -2 * BytesPerInt));
2283   __ rev32(j, j);
2284   __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2285   __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2286   __ dispatch_only(vtos, /*generate_poll*/true);
2287 }
2288 
2289 
2290 void TemplateTable::_return(TosState state)
2291 {
2292   transition(state, state);
2293   assert(_desc->calls_vm(),
2294          "inconsistent calls_vm information"); // call in remove_activation
2295 
2296   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2297     assert(state == vtos, "only valid state");
2298 
2299     __ ldr(c_rarg1, aaddress(0));
2300     __ load_klass(r3, c_rarg1);
2301     __ ldrb(r3, Address(r3, Klass::misc_flags_offset()));
2302     Label skip_register_finalizer;
2303     __ tbz(r3, exact_log2(KlassFlags::_misc_has_finalizer), skip_register_finalizer);
2304 
2305     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2306 
2307     __ bind(skip_register_finalizer);
2308   }
2309 
2310   // Issue a StoreStore barrier after all stores but before return
2311   // from any constructor for any class with a final field.  We don't
2312   // know if this is a finalizer, so we always do so.
2313   if (_desc->bytecode() == Bytecodes::_return
2314       || _desc->bytecode() == Bytecodes::_return_register_finalizer)
2315     __ membar(MacroAssembler::StoreStore);
2316 
2317   if (_desc->bytecode() != Bytecodes::_return_register_finalizer) {
2318     Label no_safepoint;
2319     __ ldr(rscratch1, Address(rthread, JavaThread::polling_word_offset()));
2320     __ tbz(rscratch1, log2i_exact(SafepointMechanism::poll_bit()), no_safepoint);
2321     __ push(state);
2322     __ push_cont_fastpath(rthread);
2323     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint));
2324     __ pop_cont_fastpath(rthread);
2325     __ pop(state);
2326     __ bind(no_safepoint);
2327   }
2328 
2329   // Narrow result if state is itos but result type is smaller.
2330   // Need to narrow in the return bytecode rather than in generate_return_entry
2331   // since compiled code callers expect the result to already be narrowed.
2332   if (state == itos) {
2333     __ narrow(r0);
2334   }
2335 
2336   __ remove_activation(state);
2337   __ ret(lr);
2338 }
2339 
2340 // ----------------------------------------------------------------------------
2341 // Volatile variables demand their effects be made known to all CPU's
2342 // in order.  Store buffers on most chips allow reads & writes to
2343 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2344 // without some kind of memory barrier (i.e., it's not sufficient that
2345 // the interpreter does not reorder volatile references, the hardware
2346 // also must not reorder them).
2347 //
2348 // According to the new Java Memory Model (JMM):
2349 // (1) All volatiles are serialized wrt to each other.  ALSO reads &
2350 //     writes act as acquire & release, so:
2351 // (2) A read cannot let unrelated NON-volatile memory refs that
2352 //     happen after the read float up to before the read.  It's OK for
2353 //     non-volatile memory refs that happen before the volatile read to
2354 //     float down below it.
2355 // (3) Similar a volatile write cannot let unrelated NON-volatile
2356 //     memory refs that happen BEFORE the write float down to after the
2357 //     write.  It's OK for non-volatile memory refs that happen after the
2358 //     volatile write to float up before it.
2359 //
2360 // We only put in barriers around volatile refs (they are expensive),
2361 // not _between_ memory refs (that would require us to track the
2362 // flavor of the previous memory refs).  Requirements (2) and (3)
2363 // require some barriers before volatile stores and after volatile
2364 // loads.  These nearly cover requirement (1) but miss the
2365 // volatile-store-volatile-load case.  This final case is placed after
2366 // volatile-stores although it could just as well go before
2367 // volatile-loads.
2368 
2369 void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
2370                                             Register Rcache,
2371                                             Register index) {
2372   const Register temp = r19;
2373   assert_different_registers(Rcache, index, temp);
2374   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2375 
2376   Label resolved, clinit_barrier_slow;
2377 
2378   Bytecodes::Code code = bytecode();
2379   __ load_method_entry(Rcache, index);
2380   switch(byte_no) {
2381     case f1_byte:
2382       __ lea(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::bytecode1_offset())));
2383       break;
2384     case f2_byte:
2385       __ lea(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::bytecode2_offset())));
2386       break;
2387   }
2388   // Load-acquire the bytecode to match store-release in InterpreterRuntime
2389   __ ldarb(temp, temp);
2390   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2391   __ br(Assembler::EQ, resolved);
2392 
2393   // resolve first time through
2394   // Class initialization barrier slow path lands here as well.
2395   __ bind(clinit_barrier_slow);
2396   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2397   __ mov(temp, (int) code);
2398   __ call_VM(noreg, entry, temp);
2399 
2400   // Update registers with resolved info
2401   __ load_method_entry(Rcache, index);
2402   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2403   // so all clients ofthis method must be modified accordingly
2404   __ bind(resolved);
2405 
2406   // Class initialization barrier for static methods
2407   if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2408     __ ldr(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::method_offset())));
2409     __ load_method_holder(temp, temp);
2410     __ clinit_barrier(temp, rscratch1, nullptr, &clinit_barrier_slow);
2411   }
2412 }
2413 
2414 void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
2415                                             Register Rcache,
2416                                             Register index) {
2417   const Register temp = r19;
2418   assert_different_registers(Rcache, index, temp);
2419 
2420   Label resolved;
2421 
2422   Bytecodes::Code code = bytecode();
2423   switch (code) {
2424   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2425   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2426   default: break;
2427   }
2428 
2429   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2430   __ load_field_entry(Rcache, index);
2431   if (byte_no == f1_byte) {
2432     __ lea(temp, Address(Rcache, in_bytes(ResolvedFieldEntry::get_code_offset())));
2433   } else {
2434     __ lea(temp, Address(Rcache, in_bytes(ResolvedFieldEntry::put_code_offset())));
2435   }
2436   // Load-acquire the bytecode to match store-release in ResolvedFieldEntry::fill_in()
2437   __ ldarb(temp, temp);
2438   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
2439   __ br(Assembler::EQ, resolved);
2440 
2441   // resolve first time through
2442   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2443   __ mov(temp, (int) code);
2444   __ call_VM(noreg, entry, temp);
2445 
2446   // Update registers with resolved info
2447   __ load_field_entry(Rcache, index);
2448   __ bind(resolved);
2449 }
2450 
2451 void TemplateTable::load_resolved_field_entry(Register obj,
2452                                               Register cache,
2453                                               Register tos_state,
2454                                               Register offset,
2455                                               Register flags,
2456                                               bool is_static = false) {
2457   assert_different_registers(cache, tos_state, flags, offset);
2458 
2459   // Field offset
2460   __ load_sized_value(offset, Address(cache, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
2461 
2462   // Flags
2463   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedFieldEntry::flags_offset())));
2464 
2465   // TOS state
2466   if (tos_state != noreg) {
2467     __ load_unsigned_byte(tos_state, Address(cache, in_bytes(ResolvedFieldEntry::type_offset())));
2468   }
2469 
2470   // Klass overwrite register
2471   if (is_static) {
2472     __ ldr(obj, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2473     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2474     __ ldr(obj, Address(obj, mirror_offset));
2475     __ resolve_oop_handle(obj, r5, rscratch2);
2476   }
2477 }
2478 
2479 void TemplateTable::load_resolved_method_entry_special_or_static(Register cache,
2480                                                                  Register method,
2481                                                                  Register flags) {
2482 
2483   // setup registers
2484   const Register index = flags;
2485   assert_different_registers(method, cache, flags);
2486 
2487   // determine constant pool cache field offsets
2488   resolve_cache_and_index_for_method(f1_byte, cache, index);
2489   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2490   __ ldr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2491 }
2492 
2493 void TemplateTable::load_resolved_method_entry_handle(Register cache,
2494                                                       Register method,
2495                                                       Register ref_index,
2496                                                       Register flags) {
2497   // setup registers
2498   const Register index = ref_index;
2499   assert_different_registers(method, flags);
2500   assert_different_registers(method, cache, index);
2501 
2502   // determine constant pool cache field offsets
2503   resolve_cache_and_index_for_method(f1_byte, cache, index);
2504   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2505 
2506   // maybe push appendix to arguments (just before return address)
2507   Label L_no_push;
2508   __ tbz(flags, ResolvedMethodEntry::has_appendix_shift, L_no_push);
2509   // invokehandle uses an index into the resolved references array
2510   __ load_unsigned_short(ref_index, Address(cache, in_bytes(ResolvedMethodEntry::resolved_references_index_offset())));
2511   // Push the appendix as a trailing parameter.
2512   // This must be done before we get the receiver,
2513   // since the parameter_size includes it.
2514   Register appendix = method;
2515   __ load_resolved_reference_at_index(appendix, ref_index);
2516   __ push(appendix);  // push appendix (MethodType, CallSite, etc.)
2517   __ bind(L_no_push);
2518 
2519   __ ldr(method, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2520 }
2521 
2522 void TemplateTable::load_resolved_method_entry_interface(Register cache,
2523                                                          Register klass,
2524                                                          Register method_or_table_index,
2525                                                          Register flags) {
2526   // setup registers
2527   const Register index = method_or_table_index;
2528   assert_different_registers(method_or_table_index, cache, flags);
2529 
2530   // determine constant pool cache field offsets
2531   resolve_cache_and_index_for_method(f1_byte, cache, index);
2532   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2533 
2534   // Invokeinterface can behave in different ways:
2535   // If calling a method from java.lang.Object, the forced virtual flag is true so the invocation will
2536   // behave like an invokevirtual call. The state of the virtual final flag will determine whether a method or
2537   // vtable index is placed in the register.
2538   // Otherwise, the registers will be populated with the klass and method.
2539 
2540   Label NotVirtual; Label NotVFinal; Label Done;
2541   __ tbz(flags, ResolvedMethodEntry::is_forced_virtual_shift, NotVirtual);
2542   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, NotVFinal);
2543   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2544   __ b(Done);
2545 
2546   __ bind(NotVFinal);
2547   __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2548   __ b(Done);
2549 
2550   __ bind(NotVirtual);
2551   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2552   __ ldr(klass, Address(cache, in_bytes(ResolvedMethodEntry::klass_offset())));
2553   __ bind(Done);
2554 }
2555 
2556 void TemplateTable::load_resolved_method_entry_virtual(Register cache,
2557                                                        Register method_or_table_index,
2558                                                        Register flags) {
2559   // setup registers
2560   const Register index = flags;
2561   assert_different_registers(method_or_table_index, cache, flags);
2562 
2563   // determine constant pool cache field offsets
2564   resolve_cache_and_index_for_method(f2_byte, cache, index);
2565   __ load_unsigned_byte(flags, Address(cache, in_bytes(ResolvedMethodEntry::flags_offset())));
2566 
2567   // method_or_table_index can either be an itable index or a method depending on the virtual final flag
2568   Label NotVFinal; Label Done;
2569   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, NotVFinal);
2570   __ ldr(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::method_offset())));
2571   __ b(Done);
2572 
2573   __ bind(NotVFinal);
2574   __ load_unsigned_short(method_or_table_index, Address(cache, in_bytes(ResolvedMethodEntry::table_index_offset())));
2575   __ bind(Done);
2576 }
2577 
2578 // The rmethod register is input and overwritten to be the adapter method for the
2579 // indy call. Link Register (lr) is set to the return address for the adapter and
2580 // an appendix may be pushed to the stack. Registers r0-r3 are clobbered
2581 void TemplateTable::load_invokedynamic_entry(Register method) {
2582   // setup registers
2583   const Register appendix = r0;
2584   const Register cache = r2;
2585   const Register index = r3;
2586   assert_different_registers(method, appendix, cache, index, rcpool);
2587 
2588   __ save_bcp();
2589 
2590   Label resolved;
2591 
2592   __ load_resolved_indy_entry(cache, index);
2593   // Load-acquire the adapter method to match store-release in ResolvedIndyEntry::fill_in()
2594   __ lea(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2595   __ ldar(method, method);
2596 
2597   // Compare the method to zero
2598   __ cbnz(method, resolved);
2599 
2600   Bytecodes::Code code = bytecode();
2601 
2602   // Call to the interpreter runtime to resolve invokedynamic
2603   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2604   __ mov(method, code); // this is essentially Bytecodes::_invokedynamic
2605   __ call_VM(noreg, entry, method);
2606   // Update registers with resolved info
2607   __ load_resolved_indy_entry(cache, index);
2608   // Load-acquire the adapter method to match store-release in ResolvedIndyEntry::fill_in()
2609   __ lea(method, Address(cache, in_bytes(ResolvedIndyEntry::method_offset())));
2610   __ ldar(method, method);
2611 
2612 #ifdef ASSERT
2613   __ cbnz(method, resolved);
2614   __ stop("Should be resolved by now");
2615 #endif // ASSERT
2616   __ bind(resolved);
2617 
2618   Label L_no_push;
2619   // Check if there is an appendix
2620   __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::flags_offset())));
2621   __ tbz(index, ResolvedIndyEntry::has_appendix_shift, L_no_push);
2622 
2623   // Get appendix
2624   __ load_unsigned_short(index, Address(cache, in_bytes(ResolvedIndyEntry::resolved_references_index_offset())));
2625   // Push the appendix as a trailing parameter
2626   // since the parameter_size includes it.
2627   __ push(method);
2628   __ mov(method, index);
2629   __ load_resolved_reference_at_index(appendix, method);
2630   __ verify_oop(appendix);
2631   __ pop(method);
2632   __ push(appendix);  // push appendix (MethodType, CallSite, etc.)
2633   __ bind(L_no_push);
2634 
2635   // compute return type
2636   __ load_unsigned_byte(index, Address(cache, in_bytes(ResolvedIndyEntry::result_type_offset())));
2637   // load return address
2638   // Return address is loaded into link register(lr) and not pushed to the stack
2639   // like x86
2640   {
2641     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2642     __ mov(rscratch1, table_addr);
2643     __ ldr(lr, Address(rscratch1, index, Address::lsl(3)));
2644   }
2645 }
2646 
2647 // The registers cache and index expected to be set before call.
2648 // Correct values of the cache and index registers are preserved.
2649 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2650                                             bool is_static, bool has_tos) {
2651   // do the JVMTI work here to avoid disturbing the register state below
2652   // We use c_rarg registers here because we want to use the register used in
2653   // the call to the VM
2654   if (JvmtiExport::can_post_field_access()) {
2655     // Check to see if a field access watch has been set before we
2656     // take the time to call into the VM.
2657     Label L1;
2658     assert_different_registers(cache, index, r0);
2659     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2660     __ ldrw(r0, Address(rscratch1));
2661     __ cbzw(r0, L1);
2662 
2663     __ load_field_entry(c_rarg2, index);
2664 
2665     if (is_static) {
2666       __ mov(c_rarg1, zr); // null object reference
2667     } else {
2668       __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2669       __ verify_oop(c_rarg1);
2670     }
2671     // c_rarg1: object pointer or null
2672     // c_rarg2: cache entry pointer
2673     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2674                                        InterpreterRuntime::post_field_access),
2675                c_rarg1, c_rarg2);
2676     __ load_field_entry(cache, index);
2677     __ bind(L1);
2678   }
2679 }
2680 
2681 void TemplateTable::pop_and_check_object(Register r)
2682 {
2683   __ pop_ptr(r);
2684   __ null_check(r);  // for field access must check obj.
2685   __ verify_oop(r);
2686 }
2687 
2688 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2689 {
2690   const Register cache     = r2;
2691   const Register obj       = r4;
2692   const Register klass     = r5;
2693   const Register inline_klass = r7;
2694   const Register field_index = r23;
2695   const Register index     = r3;
2696   const Register tos_state = r3;
2697   const Register off       = r19;
2698   const Register flags     = r6;
2699   const Register bc        = r4; // uses same reg as obj, so don't mix them
2700 
2701   resolve_cache_and_index_for_field(byte_no, cache, index);
2702   jvmti_post_field_access(cache, index, is_static, false);
2703 
2704   // Valhalla extras
2705   __ load_unsigned_short(field_index, Address(cache, in_bytes(ResolvedFieldEntry::field_index_offset())));
2706   __ ldr(klass, Address(cache, ResolvedFieldEntry::field_holder_offset()));
2707 
2708   load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
2709 
2710   if (!is_static) {
2711     // obj is on the stack
2712     pop_and_check_object(obj);
2713   }
2714 
2715   // 8179954: We need to make sure that the code generated for
2716   // volatile accesses forms a sequentially-consistent set of
2717   // operations when combined with STLR and LDAR.  Without a leading
2718   // membar it's possible for a simple Dekker test to fail if loads
2719   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
2720   // the stores in one method and we interpret the loads in another.
2721   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()){
2722     Label notVolatile;
2723     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
2724     __ membar(MacroAssembler::AnyAny);
2725     __ bind(notVolatile);
2726   }
2727 
2728   const Address field(obj, off);
2729 
2730   Label Done, notByte, notBool, notInt, notShort, notChar,
2731               notLong, notFloat, notObj, notDouble;
2732 
2733   assert(btos == 0, "change code, btos != 0");
2734   __ cbnz(tos_state, notByte);
2735 
2736   // Don't rewrite getstatic, only getfield
2737   if (is_static) rc = may_not_rewrite;
2738 
2739   // btos
2740   __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
2741   __ push(btos);
2742   // Rewrite bytecode to be faster
2743   if (rc == may_rewrite) {
2744     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2745   }
2746   __ b(Done);
2747 
2748   __ bind(notByte);
2749   __ cmp(tos_state, (u1)ztos);
2750   __ br(Assembler::NE, notBool);
2751 
2752   // ztos (same code as btos)
2753   __ access_load_at(T_BOOLEAN, IN_HEAP, r0, field, noreg, noreg);
2754   __ push(ztos);
2755   // Rewrite bytecode to be faster
2756   if (rc == may_rewrite) {
2757     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2758     patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2759   }
2760   __ b(Done);
2761 
2762   __ bind(notBool);
2763   __ cmp(tos_state, (u1)atos);
2764   __ br(Assembler::NE, notObj);
2765   // atos
2766   if (!EnableValhalla) {
2767     do_oop_load(_masm, field, r0, IN_HEAP);
2768     __ push(atos);
2769     if (rc == may_rewrite) {
2770       patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2771     }
2772     __ b(Done);
2773   } else { // Valhalla
2774     if (is_static) {
2775       __ load_heap_oop(r0, field, rscratch1, rscratch2);
2776       __ push(atos);
2777       __ b(Done);
2778     } else {
2779       Label is_flat, rewrite_inline;
2780       __ test_field_is_flat(flags, noreg /*temp*/, is_flat);
2781       __ load_heap_oop(r0, field, rscratch1, rscratch2);
2782       __ push(atos);
2783       if (rc == may_rewrite) {
2784         patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2785       }
2786       __ b(Done);
2787       __ bind(is_flat);
2788         // field is flat (null-free or nullable with a null-marker)
2789         __ mov(r0, obj);
2790         __ read_flat_field(cache, field_index, off, inline_klass /* temp */, r0);
2791         __ verify_oop(r0);
2792         __ push(atos);
2793       __ bind(rewrite_inline);
2794       if (rc == may_rewrite) {
2795         patch_bytecode(Bytecodes::_fast_vgetfield, bc, r1);
2796       }
2797       __ b(Done);
2798     }
2799   }
2800 
2801   __ bind(notObj);
2802   __ cmp(tos_state, (u1)itos);
2803   __ br(Assembler::NE, notInt);
2804   // itos
2805   __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
2806   __ push(itos);
2807   // Rewrite bytecode to be faster
2808   if (rc == may_rewrite) {
2809     patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2810   }
2811   __ b(Done);
2812 
2813   __ bind(notInt);
2814   __ cmp(tos_state, (u1)ctos);
2815   __ br(Assembler::NE, notChar);
2816   // ctos
2817   __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
2818   __ push(ctos);
2819   // Rewrite bytecode to be faster
2820   if (rc == may_rewrite) {
2821     patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2822   }
2823   __ b(Done);
2824 
2825   __ bind(notChar);
2826   __ cmp(tos_state, (u1)stos);
2827   __ br(Assembler::NE, notShort);
2828   // stos
2829   __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
2830   __ push(stos);
2831   // Rewrite bytecode to be faster
2832   if (rc == may_rewrite) {
2833     patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2834   }
2835   __ b(Done);
2836 
2837   __ bind(notShort);
2838   __ cmp(tos_state, (u1)ltos);
2839   __ br(Assembler::NE, notLong);
2840   // ltos
2841   __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
2842   __ push(ltos);
2843   // Rewrite bytecode to be faster
2844   if (rc == may_rewrite) {
2845     patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2846   }
2847   __ b(Done);
2848 
2849   __ bind(notLong);
2850   __ cmp(tos_state, (u1)ftos);
2851   __ br(Assembler::NE, notFloat);
2852   // ftos
2853   __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2854   __ push(ftos);
2855   // Rewrite bytecode to be faster
2856   if (rc == may_rewrite) {
2857     patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2858   }
2859   __ b(Done);
2860 
2861   __ bind(notFloat);
2862 #ifdef ASSERT
2863   __ cmp(tos_state, (u1)dtos);
2864   __ br(Assembler::NE, notDouble);
2865 #endif
2866   // dtos
2867   __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2868   __ push(dtos);
2869   // Rewrite bytecode to be faster
2870   if (rc == may_rewrite) {
2871     patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2872   }
2873 #ifdef ASSERT
2874   __ b(Done);
2875 
2876   __ bind(notDouble);
2877   __ stop("Bad state");
2878 #endif
2879 
2880   __ bind(Done);
2881 
2882   Label notVolatile;
2883   __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
2884   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2885   __ bind(notVolatile);
2886 }
2887 
2888 
2889 void TemplateTable::getfield(int byte_no)
2890 {
2891   getfield_or_static(byte_no, false);
2892 }
2893 
2894 void TemplateTable::nofast_getfield(int byte_no) {
2895   getfield_or_static(byte_no, false, may_not_rewrite);
2896 }
2897 
2898 void TemplateTable::getstatic(int byte_no)
2899 {
2900   getfield_or_static(byte_no, true);
2901 }
2902 
2903 // The registers cache and index expected to be set before call.
2904 // The function may destroy various registers, just not the cache and index registers.
2905 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2906   transition(vtos, vtos);
2907 
2908   if (JvmtiExport::can_post_field_modification()) {
2909     // Check to see if a field modification watch has been set before
2910     // we take the time to call into the VM.
2911     Label L1;
2912     assert_different_registers(cache, index, r0);
2913     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2914     __ ldrw(r0, Address(rscratch1));
2915     __ cbz(r0, L1);
2916 
2917     __ mov(c_rarg2, cache);
2918 
2919     if (is_static) {
2920       // Life is simple.  Null out the object pointer.
2921       __ mov(c_rarg1, zr);
2922     } else {
2923       // Life is harder. The stack holds the value on top, followed by
2924       // the object.  We don't know the size of the value, though; it
2925       // could be one or two words depending on its type. As a result,
2926       // we must find the type to determine where the object is.
2927       __ load_unsigned_byte(c_rarg3, Address(c_rarg2, in_bytes(ResolvedFieldEntry::type_offset())));
2928       Label nope2, done, ok;
2929       __ ldr(c_rarg1, at_tos_p1());  // initially assume a one word jvalue
2930       __ cmpw(c_rarg3, ltos);
2931       __ br(Assembler::EQ, ok);
2932       __ cmpw(c_rarg3, dtos);
2933       __ br(Assembler::NE, nope2);
2934       __ bind(ok);
2935       __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2936       __ bind(nope2);
2937     }
2938     // object (tos)
2939     __ mov(c_rarg3, esp);
2940     // c_rarg1: object pointer set up above (null if static)
2941     // c_rarg2: cache entry pointer
2942     // c_rarg3: jvalue object on the stack
2943     __ call_VM(noreg,
2944                CAST_FROM_FN_PTR(address,
2945                                 InterpreterRuntime::post_field_modification),
2946                c_rarg1, c_rarg2, c_rarg3);
2947     __ load_field_entry(cache, index);
2948     __ bind(L1);
2949   }
2950 }
2951 
2952 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2953   transition(vtos, vtos);
2954 
2955   const Register cache     = r2;
2956   const Register index     = r3;
2957   const Register tos_state = r3;
2958   const Register obj       = r2;
2959   const Register off       = r19;
2960   const Register flags     = r6;
2961   const Register bc        = r4;
2962   const Register inline_klass = r5;
2963 
2964   resolve_cache_and_index_for_field(byte_no, cache, index);
2965   jvmti_post_field_mod(cache, index, is_static);
2966   load_resolved_field_entry(obj, cache, tos_state, off, flags, is_static);
2967 
2968   Label Done;
2969   {
2970     Label notVolatile;
2971     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
2972     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2973     __ bind(notVolatile);
2974   }
2975 
2976   // field address
2977   const Address field(obj, off);
2978 
2979   Label notByte, notBool, notInt, notShort, notChar,
2980         notLong, notFloat, notObj, notDouble;
2981 
2982   assert(btos == 0, "change code, btos != 0");
2983   __ cbnz(tos_state, notByte);
2984 
2985   // Don't rewrite putstatic, only putfield
2986   if (is_static) rc = may_not_rewrite;
2987 
2988   // btos
2989   {
2990     __ pop(btos);
2991     if (!is_static) pop_and_check_object(obj);
2992     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
2993     if (rc == may_rewrite) {
2994       patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2995     }
2996     __ b(Done);
2997   }
2998 
2999   __ bind(notByte);
3000   __ cmp(tos_state, (u1)ztos);
3001   __ br(Assembler::NE, notBool);
3002 
3003   // ztos
3004   {
3005     __ pop(ztos);
3006     if (!is_static) pop_and_check_object(obj);
3007     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
3008     if (rc == may_rewrite) {
3009       patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
3010     }
3011     __ b(Done);
3012   }
3013 
3014   __ bind(notBool);
3015   __ cmp(tos_state, (u1)atos);
3016   __ br(Assembler::NE, notObj);
3017 
3018   // atos
3019   {
3020      if (!EnableValhalla) {
3021       __ pop(atos);
3022       if (!is_static) pop_and_check_object(obj);
3023       // Store into the field
3024       // Clobbers: r10, r11, r3
3025       do_oop_store(_masm, field, r0, IN_HEAP);
3026       if (rc == may_rewrite) {
3027         patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
3028       }
3029       __ b(Done);
3030      } else { // Valhalla
3031       __ pop(atos);
3032       if (is_static) {
3033         Label is_nullable;
3034          __ test_field_is_not_null_free_inline_type(flags, noreg /* temp */, is_nullable);
3035          __ null_check(r0);  // FIXME JDK-8341120
3036          __ bind(is_nullable);
3037          do_oop_store(_masm, field, r0, IN_HEAP);
3038          __ b(Done);
3039       } else {
3040         Label null_free_reference, is_flat, rewrite_inline;
3041         __ test_field_is_flat(flags, noreg /*temp*/, is_flat);
3042         __ test_field_is_null_free_inline_type(flags, noreg /*temp*/, null_free_reference);
3043         pop_and_check_object(obj);
3044         // Store into the field
3045         // Clobbers: r10, r11, r3
3046         do_oop_store(_masm, field, r0, IN_HEAP);
3047         if (rc == may_rewrite) {
3048           patch_bytecode(Bytecodes::_fast_aputfield, bc, r19, true, byte_no);
3049         }
3050         __ b(Done);
3051         // Implementation of the inline type semantic
3052         __ bind(null_free_reference);
3053         __ null_check(r0);  // FIXME JDK-8341120
3054         pop_and_check_object(obj);
3055         // Store into the field
3056         // Clobbers: r10, r11, r3
3057         do_oop_store(_masm, field, r0, IN_HEAP);
3058         __ b(rewrite_inline);
3059         __ bind(is_flat);
3060         pop_and_check_object(r7);
3061         __ write_flat_field(cache, off, r3, r6, r7);
3062         __ bind(rewrite_inline);
3063         if (rc == may_rewrite) {
3064           patch_bytecode(Bytecodes::_fast_vputfield, bc, r19, true, byte_no);
3065         }
3066         __ b(Done);
3067       }
3068      }  // Valhalla
3069   }
3070 
3071   __ bind(notObj);
3072   __ cmp(tos_state, (u1)itos);
3073   __ br(Assembler::NE, notInt);
3074 
3075   // itos
3076   {
3077     __ pop(itos);
3078     if (!is_static) pop_and_check_object(obj);
3079     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
3080     if (rc == may_rewrite) {
3081       patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
3082     }
3083     __ b(Done);
3084   }
3085 
3086   __ bind(notInt);
3087   __ cmp(tos_state, (u1)ctos);
3088   __ br(Assembler::NE, notChar);
3089 
3090   // ctos
3091   {
3092     __ pop(ctos);
3093     if (!is_static) pop_and_check_object(obj);
3094     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
3095     if (rc == may_rewrite) {
3096       patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
3097     }
3098     __ b(Done);
3099   }
3100 
3101   __ bind(notChar);
3102   __ cmp(tos_state, (u1)stos);
3103   __ br(Assembler::NE, notShort);
3104 
3105   // stos
3106   {
3107     __ pop(stos);
3108     if (!is_static) pop_and_check_object(obj);
3109     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
3110     if (rc == may_rewrite) {
3111       patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
3112     }
3113     __ b(Done);
3114   }
3115 
3116   __ bind(notShort);
3117   __ cmp(tos_state, (u1)ltos);
3118   __ br(Assembler::NE, notLong);
3119 
3120   // ltos
3121   {
3122     __ pop(ltos);
3123     if (!is_static) pop_and_check_object(obj);
3124     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
3125     if (rc == may_rewrite) {
3126       patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
3127     }
3128     __ b(Done);
3129   }
3130 
3131   __ bind(notLong);
3132   __ cmp(tos_state, (u1)ftos);
3133   __ br(Assembler::NE, notFloat);
3134 
3135   // ftos
3136   {
3137     __ pop(ftos);
3138     if (!is_static) pop_and_check_object(obj);
3139     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3140     if (rc == may_rewrite) {
3141       patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
3142     }
3143     __ b(Done);
3144   }
3145 
3146   __ bind(notFloat);
3147 #ifdef ASSERT
3148   __ cmp(tos_state, (u1)dtos);
3149   __ br(Assembler::NE, notDouble);
3150 #endif
3151 
3152   // dtos
3153   {
3154     __ pop(dtos);
3155     if (!is_static) pop_and_check_object(obj);
3156     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
3157     if (rc == may_rewrite) {
3158       patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
3159     }
3160   }
3161 
3162 #ifdef ASSERT
3163   __ b(Done);
3164 
3165   __ bind(notDouble);
3166   __ stop("Bad state");
3167 #endif
3168 
3169   __ bind(Done);
3170 
3171   {
3172     Label notVolatile;
3173     __ tbz(flags, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3174     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3175     __ bind(notVolatile);
3176   }
3177 }
3178 
3179 void TemplateTable::putfield(int byte_no)
3180 {
3181   putfield_or_static(byte_no, false);
3182 }
3183 
3184 void TemplateTable::nofast_putfield(int byte_no) {
3185   putfield_or_static(byte_no, false, may_not_rewrite);
3186 }
3187 
3188 void TemplateTable::putstatic(int byte_no) {
3189   putfield_or_static(byte_no, true);
3190 }
3191 
3192 void TemplateTable::jvmti_post_fast_field_mod() {
3193   if (JvmtiExport::can_post_field_modification()) {
3194     // Check to see if a field modification watch has been set before
3195     // we take the time to call into the VM.
3196     Label L2;
3197     __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3198     __ ldrw(c_rarg3, Address(rscratch1));
3199     __ cbzw(c_rarg3, L2);
3200     __ pop_ptr(r19);                  // copy the object pointer from tos
3201     __ verify_oop(r19);
3202     __ push_ptr(r19);                 // put the object pointer back on tos
3203     // Save tos values before call_VM() clobbers them. Since we have
3204     // to do it for every data type, we use the saved values as the
3205     // jvalue object.
3206     switch (bytecode()) {          // load values into the jvalue object
3207     case Bytecodes::_fast_vputfield: //fall through
3208     case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
3209     case Bytecodes::_fast_bputfield: // fall through
3210     case Bytecodes::_fast_zputfield: // fall through
3211     case Bytecodes::_fast_sputfield: // fall through
3212     case Bytecodes::_fast_cputfield: // fall through
3213     case Bytecodes::_fast_iputfield: __ push_i(r0); break;
3214     case Bytecodes::_fast_dputfield: __ push_d(); break;
3215     case Bytecodes::_fast_fputfield: __ push_f(); break;
3216     case Bytecodes::_fast_lputfield: __ push_l(r0); break;
3217 
3218     default:
3219       ShouldNotReachHere();
3220     }
3221     __ mov(c_rarg3, esp);             // points to jvalue on the stack
3222     // access constant pool cache entry
3223     __ load_field_entry(c_rarg2, r0);
3224     __ verify_oop(r19);
3225     // r19: object pointer copied above
3226     // c_rarg2: cache entry pointer
3227     // c_rarg3: jvalue object on the stack
3228     __ call_VM(noreg,
3229                CAST_FROM_FN_PTR(address,
3230                                 InterpreterRuntime::post_field_modification),
3231                r19, c_rarg2, c_rarg3);
3232 
3233     switch (bytecode()) {             // restore tos values
3234     case Bytecodes::_fast_vputfield: //fall through
3235     case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
3236     case Bytecodes::_fast_bputfield: // fall through
3237     case Bytecodes::_fast_zputfield: // fall through
3238     case Bytecodes::_fast_sputfield: // fall through
3239     case Bytecodes::_fast_cputfield: // fall through
3240     case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
3241     case Bytecodes::_fast_dputfield: __ pop_d(); break;
3242     case Bytecodes::_fast_fputfield: __ pop_f(); break;
3243     case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
3244     default: break;
3245     }
3246     __ bind(L2);
3247   }
3248 }
3249 
3250 void TemplateTable::fast_storefield(TosState state)
3251 {
3252   transition(state, vtos);
3253 
3254   ByteSize base = ConstantPoolCache::base_offset();
3255 
3256   jvmti_post_fast_field_mod();
3257 
3258   // access constant pool cache
3259   __ load_field_entry(r2, r1);
3260 
3261   // R1: field offset, R2: field holder, R5: flags
3262   load_resolved_field_entry(r2, r2, noreg, r1, r5);
3263 
3264   {
3265     Label notVolatile;
3266     __ tbz(r5, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3267     __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
3268     __ bind(notVolatile);
3269   }
3270 
3271   Label notVolatile;
3272 
3273   // Get object from stack
3274   pop_and_check_object(r2);
3275 
3276   // field address
3277   const Address field(r2, r1);
3278 
3279   // access field
3280   switch (bytecode()) {
3281   case Bytecodes::_fast_vputfield:
3282     {
3283       Label is_flat, has_null_marker, done;
3284       __ test_field_is_flat(r5, noreg /* temp */, is_flat);
3285       __ null_check(r0);
3286       do_oop_store(_masm, field, r0, IN_HEAP);
3287       __ b(done);
3288       __ bind(is_flat);
3289       __ load_field_entry(r4, r5);
3290       // Re-shuffle registers because of VM calls calling convention
3291       __ mov(r19, r1);
3292       __ mov(r7, r2);
3293       __ write_flat_field(r4, r19, r6, r8, r7);
3294       __ bind(done);
3295     }
3296     break;
3297   case Bytecodes::_fast_aputfield:
3298     // Clobbers: r10, r11, r3
3299     do_oop_store(_masm, field, r0, IN_HEAP);
3300     break;
3301   case Bytecodes::_fast_lputfield:
3302     __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg, noreg);
3303     break;
3304   case Bytecodes::_fast_iputfield:
3305     __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg, noreg);
3306     break;
3307   case Bytecodes::_fast_zputfield:
3308     __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg, noreg);
3309     break;
3310   case Bytecodes::_fast_bputfield:
3311     __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg, noreg);
3312     break;
3313   case Bytecodes::_fast_sputfield:
3314     __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg, noreg);
3315     break;
3316   case Bytecodes::_fast_cputfield:
3317     __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg, noreg);
3318     break;
3319   case Bytecodes::_fast_fputfield:
3320     __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg, noreg);
3321     break;
3322   case Bytecodes::_fast_dputfield:
3323     __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg, noreg);
3324     break;
3325   default:
3326     ShouldNotReachHere();
3327   }
3328 
3329   {
3330     Label notVolatile;
3331     __ tbz(r5, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3332     __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3333     __ bind(notVolatile);
3334   }
3335 }
3336 
3337 
3338 void TemplateTable::fast_accessfield(TosState state)
3339 {
3340   transition(atos, state);
3341   // Do the JVMTI work here to avoid disturbing the register state below
3342   if (JvmtiExport::can_post_field_access()) {
3343     // Check to see if a field access watch has been set before we
3344     // take the time to call into the VM.
3345     Label L1;
3346     __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3347     __ ldrw(r2, Address(rscratch1));
3348     __ cbzw(r2, L1);
3349     // access constant pool cache entry
3350     __ load_field_entry(c_rarg2, rscratch2);
3351     __ verify_oop(r0);
3352     __ push_ptr(r0);  // save object pointer before call_VM() clobbers it
3353     __ mov(c_rarg1, r0);
3354     // c_rarg1: object pointer copied above
3355     // c_rarg2: cache entry pointer
3356     __ call_VM(noreg,
3357                CAST_FROM_FN_PTR(address,
3358                                 InterpreterRuntime::post_field_access),
3359                c_rarg1, c_rarg2);
3360     __ pop_ptr(r0); // restore object pointer
3361     __ bind(L1);
3362   }
3363 
3364   // access constant pool cache
3365   __ load_field_entry(r2, r1);
3366 
3367   __ load_sized_value(r1, Address(r2, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3368   __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3369 
3370   // r0: object
3371   __ verify_oop(r0);
3372   __ null_check(r0);
3373   const Address field(r0, r1);
3374 
3375   // 8179954: We need to make sure that the code generated for
3376   // volatile accesses forms a sequentially-consistent set of
3377   // operations when combined with STLR and LDAR.  Without a leading
3378   // membar it's possible for a simple Dekker test to fail if loads
3379   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3380   // the stores in one method and we interpret the loads in another.
3381   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3382     Label notVolatile;
3383     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3384     __ membar(MacroAssembler::AnyAny);
3385     __ bind(notVolatile);
3386   }
3387 
3388   // access field
3389   switch (bytecode()) {
3390   case Bytecodes::_fast_vgetfield:
3391     {
3392       Register index = r4, tmp = r7;
3393       // field is flat
3394       __ load_unsigned_short(index, Address(r2, in_bytes(ResolvedFieldEntry::field_index_offset())));
3395       __ read_flat_field(r2, index, r1, tmp /* temp */, r0);
3396       __ verify_oop(r0);
3397     }
3398     break;
3399   case Bytecodes::_fast_agetfield:
3400     do_oop_load(_masm, field, r0, IN_HEAP);
3401     __ verify_oop(r0);
3402     break;
3403   case Bytecodes::_fast_lgetfield:
3404     __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
3405     break;
3406   case Bytecodes::_fast_igetfield:
3407     __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
3408     break;
3409   case Bytecodes::_fast_bgetfield:
3410     __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
3411     break;
3412   case Bytecodes::_fast_sgetfield:
3413     __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
3414     break;
3415   case Bytecodes::_fast_cgetfield:
3416     __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
3417     break;
3418   case Bytecodes::_fast_fgetfield:
3419     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3420     break;
3421   case Bytecodes::_fast_dgetfield:
3422     __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3423     break;
3424   default:
3425     ShouldNotReachHere();
3426   }
3427   {
3428     Label notVolatile;
3429     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3430     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3431     __ bind(notVolatile);
3432   }
3433 }
3434 
3435 void TemplateTable::fast_xaccess(TosState state)
3436 {
3437   transition(vtos, state);
3438 
3439   // get receiver
3440   __ ldr(r0, aaddress(0));
3441   // access constant pool cache
3442   __ load_field_entry(r2, r3, 2);
3443   __ load_sized_value(r1, Address(r2, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
3444 
3445   // 8179954: We need to make sure that the code generated for
3446   // volatile accesses forms a sequentially-consistent set of
3447   // operations when combined with STLR and LDAR.  Without a leading
3448   // membar it's possible for a simple Dekker test to fail if loads
3449   // use LDR;DMB but stores use STLR.  This can happen if C2 compiles
3450   // the stores in one method and we interpret the loads in another.
3451   if (!CompilerConfig::is_c1_or_interpreter_only_no_jvmci()) {
3452     Label notVolatile;
3453     __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3454     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3455     __ membar(MacroAssembler::AnyAny);
3456     __ bind(notVolatile);
3457   }
3458 
3459   // make sure exception is reported in correct bcp range (getfield is
3460   // next instruction)
3461   __ increment(rbcp);
3462   __ null_check(r0);
3463   switch (state) {
3464   case itos:
3465     __ access_load_at(T_INT, IN_HEAP, r0, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3466     break;
3467   case atos:
3468     do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3469     __ verify_oop(r0);
3470     break;
3471   case ftos:
3472     __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3473     break;
3474   default:
3475     ShouldNotReachHere();
3476   }
3477 
3478   {
3479     Label notVolatile;
3480     __ load_unsigned_byte(r3, Address(r2, in_bytes(ResolvedFieldEntry::flags_offset())));
3481     __ tbz(r3, ResolvedFieldEntry::is_volatile_shift, notVolatile);
3482     __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3483     __ bind(notVolatile);
3484   }
3485 
3486   __ decrement(rbcp);
3487 }
3488 
3489 
3490 
3491 //-----------------------------------------------------------------------------
3492 // Calls
3493 
3494 void TemplateTable::prepare_invoke(Register cache, Register recv) {
3495 
3496   Bytecodes::Code code = bytecode();
3497   const bool load_receiver       = (code != Bytecodes::_invokestatic) && (code != Bytecodes::_invokedynamic);
3498 
3499   // save 'interpreter return address'
3500   __ save_bcp();
3501 
3502   // Load TOS state for later
3503   __ load_unsigned_byte(rscratch2, Address(cache, in_bytes(ResolvedMethodEntry::type_offset())));
3504 
3505   // load receiver if needed (note: no return address pushed yet)
3506   if (load_receiver) {
3507     __ load_unsigned_short(recv, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
3508     __ add(rscratch1, esp, recv, ext::uxtx, 3);
3509     __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3510     __ verify_oop(recv);
3511   }
3512 
3513   // load return address
3514   {
3515     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3516     __ mov(rscratch1, table_addr);
3517     __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3518   }
3519 }
3520 
3521 
3522 void TemplateTable::invokevirtual_helper(Register index,
3523                                          Register recv,
3524                                          Register flags)
3525 {
3526   // Uses temporary registers r0, r3
3527   assert_different_registers(index, recv, r0, r3);
3528   // Test for an invoke of a final method
3529   Label notFinal;
3530   __ tbz(flags, ResolvedMethodEntry::is_vfinal_shift, notFinal);
3531 
3532   const Register method = index;  // method must be rmethod
3533   assert(method == rmethod,
3534          "Method must be rmethod for interpreter calling convention");
3535 
3536   // do the call - the index is actually the method to call
3537   // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3538 
3539   // It's final, need a null check here!
3540   __ null_check(recv);
3541 
3542   // profile this call
3543   __ profile_final_call(r0);
3544   __ profile_arguments_type(r0, method, r4, true);
3545 
3546   __ jump_from_interpreted(method, r0);
3547 
3548   __ bind(notFinal);
3549 
3550   // get receiver klass
3551   __ load_klass(r0, recv);
3552 
3553   // profile this call
3554   __ profile_virtual_call(r0, rlocals, r3);
3555 
3556   // get target Method & entry point
3557   __ lookup_virtual_method(r0, index, method);
3558   __ profile_arguments_type(r3, method, r4, true);
3559   // FIXME -- this looks completely redundant. is it?
3560   // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3561   __ jump_from_interpreted(method, r3);
3562 }
3563 
3564 void TemplateTable::invokevirtual(int byte_no)
3565 {
3566   transition(vtos, vtos);
3567   assert(byte_no == f2_byte, "use this argument");
3568 
3569   load_resolved_method_entry_virtual(r2,      // ResolvedMethodEntry*
3570                                      rmethod, // Method* or itable index
3571                                      r3);     // flags
3572   prepare_invoke(r2, r2); // recv
3573 
3574   // rmethod: index (actually a Method*)
3575   // r2: receiver
3576   // r3: flags
3577 
3578   invokevirtual_helper(rmethod, r2, r3);
3579 }
3580 
3581 void TemplateTable::invokespecial(int byte_no)
3582 {
3583   transition(vtos, vtos);
3584   assert(byte_no == f1_byte, "use this argument");
3585 
3586   load_resolved_method_entry_special_or_static(r2,      // ResolvedMethodEntry*
3587                                                rmethod, // Method*
3588                                                r3);     // flags
3589   prepare_invoke(r2, r2);  // get receiver also for null check
3590   __ verify_oop(r2);
3591   __ null_check(r2);
3592   // do the call
3593   __ profile_call(r0);
3594   __ profile_arguments_type(r0, rmethod, rbcp, false);
3595   __ jump_from_interpreted(rmethod, r0);
3596 }
3597 
3598 void TemplateTable::invokestatic(int byte_no)
3599 {
3600   transition(vtos, vtos);
3601   assert(byte_no == f1_byte, "use this argument");
3602 
3603   load_resolved_method_entry_special_or_static(r2,      // ResolvedMethodEntry*
3604                                                rmethod, // Method*
3605                                                r3);     // flags
3606   prepare_invoke(r2, r2);  // get receiver also for null check
3607 
3608   // do the call
3609   __ profile_call(r0);
3610   __ profile_arguments_type(r0, rmethod, r4, false);
3611   __ jump_from_interpreted(rmethod, r0);
3612 }
3613 
3614 void TemplateTable::fast_invokevfinal(int byte_no)
3615 {
3616   __ call_Unimplemented();
3617 }
3618 
3619 void TemplateTable::invokeinterface(int byte_no) {
3620   transition(vtos, vtos);
3621   assert(byte_no == f1_byte, "use this argument");
3622 
3623   load_resolved_method_entry_interface(r2,      // ResolvedMethodEntry*
3624                                        r0,      // Klass*
3625                                        rmethod, // Method* or itable/vtable index
3626                                        r3);     // flags
3627   prepare_invoke(r2, r2); // receiver
3628 
3629   // r0: interface klass (from f1)
3630   // rmethod: method (from f2)
3631   // r2: receiver
3632   // r3: flags
3633 
3634   // First check for Object case, then private interface method,
3635   // then regular interface method.
3636 
3637   // Special case of invokeinterface called for virtual method of
3638   // java.lang.Object.  See cpCache.cpp for details.
3639   Label notObjectMethod;
3640   __ tbz(r3, ResolvedMethodEntry::is_forced_virtual_shift, notObjectMethod);
3641 
3642   invokevirtual_helper(rmethod, r2, r3);
3643   __ bind(notObjectMethod);
3644 
3645   Label no_such_interface;
3646 
3647   // Check for private method invocation - indicated by vfinal
3648   Label notVFinal;
3649   __ tbz(r3, ResolvedMethodEntry::is_vfinal_shift, notVFinal);
3650 
3651   // Get receiver klass into r3
3652   __ load_klass(r3, r2);
3653 
3654   Label subtype;
3655   __ check_klass_subtype(r3, r0, r4, subtype);
3656   // If we get here the typecheck failed
3657   __ b(no_such_interface);
3658   __ bind(subtype);
3659 
3660   __ profile_final_call(r0);
3661   __ profile_arguments_type(r0, rmethod, r4, true);
3662   __ jump_from_interpreted(rmethod, r0);
3663 
3664   __ bind(notVFinal);
3665 
3666   // Get receiver klass into r3
3667   __ restore_locals();
3668   __ load_klass(r3, r2);
3669 
3670   Label no_such_method;
3671 
3672   // Preserve method for throw_AbstractMethodErrorVerbose.
3673   __ mov(r16, rmethod);
3674   // Receiver subtype check against REFC.
3675   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3676   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3677                              r3, r0, noreg,
3678                              // outputs: scan temp. reg, scan temp. reg
3679                              rscratch2, r13,
3680                              no_such_interface,
3681                              /*return_method=*/false);
3682 
3683   // profile this call
3684   __ profile_virtual_call(r3, r13, r19);
3685 
3686   // Get declaring interface class from method, and itable index
3687 
3688   __ load_method_holder(r0, rmethod);
3689   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3690   __ subw(rmethod, rmethod, Method::itable_index_max);
3691   __ negw(rmethod, rmethod);
3692 
3693   // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3694   __ mov(rlocals, r3);
3695   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3696                              rlocals, r0, rmethod,
3697                              // outputs: method, scan temp. reg
3698                              rmethod, r13,
3699                              no_such_interface);
3700 
3701   // rmethod,: Method to call
3702   // r2: receiver
3703   // Check for abstract method error
3704   // Note: This should be done more efficiently via a throw_abstract_method_error
3705   //       interpreter entry point and a conditional jump to it in case of a null
3706   //       method.
3707   __ cbz(rmethod, no_such_method);
3708 
3709   __ profile_arguments_type(r3, rmethod, r13, true);
3710 
3711   // do the call
3712   // r2: receiver
3713   // rmethod,: Method
3714   __ jump_from_interpreted(rmethod, r3);
3715   __ should_not_reach_here();
3716 
3717   // exception handling code follows...
3718   // note: must restore interpreter registers to canonical
3719   //       state for exception handling to work correctly!
3720 
3721   __ bind(no_such_method);
3722   // throw exception
3723   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3724   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3725   // Pass arguments for generating a verbose error message.
3726   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3727   // the call_VM checks for exception, so we should never return here.
3728   __ should_not_reach_here();
3729 
3730   __ bind(no_such_interface);
3731   // throw exception
3732   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
3733   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
3734   // Pass arguments for generating a verbose error message.
3735   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3736                    InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3737   // the call_VM checks for exception, so we should never return here.
3738   __ should_not_reach_here();
3739   return;
3740 }
3741 
3742 void TemplateTable::invokehandle(int byte_no) {
3743   transition(vtos, vtos);
3744   assert(byte_no == f1_byte, "use this argument");
3745 
3746   load_resolved_method_entry_handle(r2,      // ResolvedMethodEntry*
3747                                     rmethod, // Method*
3748                                     r0,      // Resolved reference
3749                                     r3);     // flags
3750   prepare_invoke(r2, r2);
3751 
3752   __ verify_method_ptr(r2);
3753   __ verify_oop(r2);
3754   __ null_check(r2);
3755 
3756   // FIXME: profile the LambdaForm also
3757 
3758   // r13 is safe to use here as a scratch reg because it is about to
3759   // be clobbered by jump_from_interpreted().
3760   __ profile_final_call(r13);
3761   __ profile_arguments_type(r13, rmethod, r4, true);
3762 
3763   __ jump_from_interpreted(rmethod, r0);
3764 }
3765 
3766 void TemplateTable::invokedynamic(int byte_no) {
3767   transition(vtos, vtos);
3768   assert(byte_no == f1_byte, "use this argument");
3769 
3770   load_invokedynamic_entry(rmethod);
3771 
3772   // r0: CallSite object (from cpool->resolved_references[])
3773   // rmethod: MH.linkToCallSite method
3774 
3775   // Note:  r0_callsite is already pushed
3776 
3777   // %%% should make a type profile for any invokedynamic that takes a ref argument
3778   // profile this call
3779   __ profile_call(rbcp);
3780   __ profile_arguments_type(r3, rmethod, r13, false);
3781 
3782   __ verify_oop(r0);
3783 
3784   __ jump_from_interpreted(rmethod, r0);
3785 }
3786 
3787 
3788 //-----------------------------------------------------------------------------
3789 // Allocation
3790 
3791 void TemplateTable::_new() {
3792   transition(vtos, atos);
3793 
3794   __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3795   Label slow_case;
3796   Label done;
3797   Label initialize_header;
3798 
3799   __ get_cpool_and_tags(r4, r0);
3800   // Make sure the class we're about to instantiate has been resolved.
3801   // This is done before loading InstanceKlass to be consistent with the order
3802   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3803   const int tags_offset = Array<u1>::base_offset_in_bytes();
3804   __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3805   __ lea(rscratch1, Address(rscratch1, tags_offset));
3806   __ ldarb(rscratch1, rscratch1);
3807   __ cmp(rscratch1, (u1)JVM_CONSTANT_Class);
3808   __ br(Assembler::NE, slow_case);
3809 
3810   // get InstanceKlass
3811   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3812 
3813   // make sure klass is initialized
3814   assert(VM_Version::supports_fast_class_init_checks(), "Optimization requires support for fast class initialization checks");
3815   __ clinit_barrier(r4, rscratch1, nullptr /*L_fast_path*/, &slow_case);
3816 
3817   __ allocate_instance(r4, r0, r3, r1, true, slow_case);
3818   __ b(done);
3819 
3820   // slow case
3821   __ bind(slow_case);
3822   __ get_constant_pool(c_rarg1);
3823   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3824   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3825   __ verify_oop(r0);
3826 
3827   // continue
3828   __ bind(done);
3829   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3830   __ membar(Assembler::StoreStore);
3831 }
3832 
3833 void TemplateTable::newarray() {
3834   transition(itos, atos);
3835   __ load_unsigned_byte(c_rarg1, at_bcp(1));
3836   __ mov(c_rarg2, r0);
3837   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3838           c_rarg1, c_rarg2);
3839   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3840   __ membar(Assembler::StoreStore);
3841 }
3842 
3843 void TemplateTable::anewarray() {
3844   transition(itos, atos);
3845   __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3846   __ get_constant_pool(c_rarg1);
3847   __ mov(c_rarg3, r0);
3848   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3849           c_rarg1, c_rarg2, c_rarg3);
3850   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3851   __ membar(Assembler::StoreStore);
3852 }
3853 
3854 void TemplateTable::arraylength() {
3855   transition(atos, itos);
3856   __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3857 }
3858 
3859 void TemplateTable::checkcast()
3860 {
3861   transition(atos, atos);
3862   Label done, is_null, ok_is_subtype, quicked, resolved;
3863   __ cbz(r0, is_null);
3864 
3865   // Get cpool & tags index
3866   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3867   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3868   // See if bytecode has already been quicked
3869   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3870   __ lea(r1, Address(rscratch1, r19));
3871   __ ldarb(r1, r1);
3872   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3873   __ br(Assembler::EQ, quicked);
3874 
3875   __ push(atos); // save receiver for result, and for GC
3876   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3877   __ get_vm_result_metadata(r0, rthread);
3878   __ pop(r3); // restore receiver
3879   __ b(resolved);
3880 
3881   // Get superklass in r0 and subklass in r3
3882   __ bind(quicked);
3883   __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3884   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3885 
3886   __ bind(resolved);
3887   __ load_klass(r19, r3);
3888 
3889   // Generate subtype check.  Blows r2, r5.  Object in r3.
3890   // Superklass in r0.  Subklass in r19.
3891   __ gen_subtype_check(r19, ok_is_subtype);
3892 
3893   // Come here on failure
3894   __ push(r3);
3895   // object is at TOS
3896   __ b(Interpreter::_throw_ClassCastException_entry);
3897 
3898   // Come here on success
3899   __ bind(ok_is_subtype);
3900   __ mov(r0, r3); // Restore object in r3
3901 
3902   __ b(done);
3903   __ bind(is_null);
3904 
3905   // Collect counts on whether this test sees nulls a lot or not.
3906   if (ProfileInterpreter) {
3907     __ profile_null_seen(r2);
3908   }
3909 
3910   __ bind(done);
3911 }
3912 
3913 void TemplateTable::instanceof() {
3914   transition(atos, itos);
3915   Label done, is_null, ok_is_subtype, quicked, resolved;
3916   __ cbz(r0, is_null);
3917 
3918   // Get cpool & tags index
3919   __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3920   __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3921   // See if bytecode has already been quicked
3922   __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3923   __ lea(r1, Address(rscratch1, r19));
3924   __ ldarb(r1, r1);
3925   __ cmp(r1, (u1)JVM_CONSTANT_Class);
3926   __ br(Assembler::EQ, quicked);
3927 
3928   __ push(atos); // save receiver for result, and for GC
3929   call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3930   __ get_vm_result_metadata(r0, rthread);
3931   __ pop(r3); // restore receiver
3932   __ verify_oop(r3);
3933   __ load_klass(r3, r3);
3934   __ b(resolved);
3935 
3936   // Get superklass in r0 and subklass in r3
3937   __ bind(quicked);
3938   __ load_klass(r3, r0);
3939   __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3940 
3941   __ bind(resolved);
3942 
3943   // Generate subtype check.  Blows r2, r5
3944   // Superklass in r0.  Subklass in r3.
3945   __ gen_subtype_check(r3, ok_is_subtype);
3946 
3947   // Come here on failure
3948   __ mov(r0, 0);
3949   __ b(done);
3950   // Come here on success
3951   __ bind(ok_is_subtype);
3952   __ mov(r0, 1);
3953 
3954   // Collect counts on whether this test sees nulls a lot or not.
3955   if (ProfileInterpreter) {
3956     __ b(done);
3957     __ bind(is_null);
3958     __ profile_null_seen(r2);
3959   } else {
3960     __ bind(is_null);   // same as 'done'
3961   }
3962   __ bind(done);
3963   // r0 = 0: obj == nullptr or  obj is not an instanceof the specified klass
3964   // r0 = 1: obj != nullptr and obj is     an instanceof the specified klass
3965 }
3966 
3967 //-----------------------------------------------------------------------------
3968 // Breakpoints
3969 void TemplateTable::_breakpoint() {
3970   // Note: We get here even if we are single stepping..
3971   // jbug inists on setting breakpoints at every bytecode
3972   // even if we are in single step mode.
3973 
3974   transition(vtos, vtos);
3975 
3976   // get the unpatched byte code
3977   __ get_method(c_rarg1);
3978   __ call_VM(noreg,
3979              CAST_FROM_FN_PTR(address,
3980                               InterpreterRuntime::get_original_bytecode_at),
3981              c_rarg1, rbcp);
3982   __ mov(r19, r0);
3983 
3984   // post the breakpoint event
3985   __ call_VM(noreg,
3986              CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3987              rmethod, rbcp);
3988 
3989   // complete the execution of original bytecode
3990   __ mov(rscratch1, r19);
3991   __ dispatch_only_normal(vtos);
3992 }
3993 
3994 //-----------------------------------------------------------------------------
3995 // Exceptions
3996 
3997 void TemplateTable::athrow() {
3998   transition(atos, vtos);
3999   __ null_check(r0);
4000   __ b(Interpreter::throw_exception_entry());
4001 }
4002 
4003 //-----------------------------------------------------------------------------
4004 // Synchronization
4005 //
4006 // Note: monitorenter & exit are symmetric routines; which is reflected
4007 //       in the assembly code structure as well
4008 //
4009 // Stack layout:
4010 //
4011 // [expressions  ] <--- esp               = expression stack top
4012 // ..
4013 // [expressions  ]
4014 // [monitor entry] <--- monitor block top = expression stack bot
4015 // ..
4016 // [monitor entry]
4017 // [frame data   ] <--- monitor block bot
4018 // ...
4019 // [saved rfp    ] <--- rfp
4020 void TemplateTable::monitorenter()
4021 {
4022   transition(atos, vtos);
4023 
4024   // check for null object
4025   __ null_check(r0);
4026 
4027   Label is_inline_type;
4028   __ ldr(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
4029   __ test_markword_is_inline_type(rscratch1, is_inline_type);
4030 
4031   const Address monitor_block_top(
4032         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4033   const Address monitor_block_bot(
4034         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4035   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4036 
4037   Label allocated;
4038 
4039   // initialize entry pointer
4040   __ mov(c_rarg1, zr); // points to free slot or null
4041 
4042   // find a free slot in the monitor block (result in c_rarg1)
4043   {
4044     Label entry, loop, exit;
4045     __ ldr(c_rarg3, monitor_block_top); // derelativize pointer
4046     __ lea(c_rarg3, Address(rfp, c_rarg3, Address::lsl(Interpreter::logStackElementSize)));
4047     // c_rarg3 points to current entry, starting with top-most entry
4048 
4049     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4050 
4051     __ b(entry);
4052 
4053     __ bind(loop);
4054     // check if current entry is used
4055     // if not used then remember entry in c_rarg1
4056     __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset()));
4057     __ cmp(zr, rscratch1);
4058     __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
4059     // check if current entry is for same object
4060     __ cmp(r0, rscratch1);
4061     // if same object then stop searching
4062     __ br(Assembler::EQ, exit);
4063     // otherwise advance to next entry
4064     __ add(c_rarg3, c_rarg3, entry_size);
4065     __ bind(entry);
4066     // check if bottom reached
4067     __ cmp(c_rarg3, c_rarg2);
4068     // if not at bottom then check this entry
4069     __ br(Assembler::NE, loop);
4070     __ bind(exit);
4071   }
4072 
4073   __ cbnz(c_rarg1, allocated); // check if a slot has been found and
4074                             // if found, continue with that on
4075 
4076   // allocate one if there's no free slot
4077   {
4078     Label entry, loop;
4079     // 1. compute new pointers            // rsp: old expression stack top
4080 
4081     __ check_extended_sp();
4082     __ sub(sp, sp, entry_size);           // make room for the monitor
4083     __ sub(rscratch1, sp, rfp);
4084     __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
4085     __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
4086 
4087     __ ldr(c_rarg1, monitor_block_bot);   // derelativize pointer
4088     __ lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
4089     // c_rarg1 points to the old expression stack bottom
4090 
4091     __ sub(esp, esp, entry_size);         // move expression stack top
4092     __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
4093     __ mov(c_rarg3, esp);                 // set start value for copy loop
4094     __ sub(rscratch1, c_rarg1, rfp);      // relativize pointer
4095     __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
4096     __ str(rscratch1, monitor_block_bot);  // set new monitor block bottom
4097 
4098     __ b(entry);
4099     // 2. move expression stack contents
4100     __ bind(loop);
4101     __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
4102                                                    // word from old location
4103     __ str(c_rarg2, Address(c_rarg3, 0));          // and store it at new location
4104     __ add(c_rarg3, c_rarg3, wordSize);            // advance to next word
4105     __ bind(entry);
4106     __ cmp(c_rarg3, c_rarg1);        // check if bottom reached
4107     __ br(Assembler::NE, loop);      // if not at bottom then
4108                                      // copy next word
4109   }
4110 
4111   // call run-time routine
4112   // c_rarg1: points to monitor entry
4113   __ bind(allocated);
4114 
4115   // Increment bcp to point to the next bytecode, so exception
4116   // handling for async. exceptions work correctly.
4117   // The object has already been popped from the stack, so the
4118   // expression stack looks correct.
4119   __ increment(rbcp);
4120 
4121   // store object
4122   __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
4123   __ lock_object(c_rarg1);
4124 
4125   // check to make sure this monitor doesn't cause stack overflow after locking
4126   __ save_bcp();  // in case of exception
4127   __ generate_stack_overflow_check(0);
4128 
4129   // The bcp has already been incremented. Just need to dispatch to
4130   // next instruction.
4131   __ dispatch_next(vtos);
4132 
4133   __ bind(is_inline_type);
4134   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4135                     InterpreterRuntime::throw_identity_exception), r0);
4136   __ should_not_reach_here();
4137 }
4138 
4139 
4140 void TemplateTable::monitorexit()
4141 {
4142   transition(atos, vtos);
4143 
4144   // check for null object
4145   __ null_check(r0);
4146 
4147   const int is_inline_type_mask = markWord::inline_type_pattern;
4148   Label has_identity;
4149   __ ldr(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
4150   __ mov(rscratch2, is_inline_type_mask);
4151   __ andr(rscratch1, rscratch1, rscratch2);
4152   __ cmp(rscratch1, rscratch2);
4153   __ br(Assembler::NE, has_identity);
4154   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4155                      InterpreterRuntime::throw_illegal_monitor_state_exception));
4156   __ should_not_reach_here();
4157   __ bind(has_identity);
4158 
4159   const Address monitor_block_top(
4160         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4161   const Address monitor_block_bot(
4162         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4163   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
4164 
4165   Label found;
4166 
4167   // find matching slot
4168   {
4169     Label entry, loop;
4170     __ ldr(c_rarg1, monitor_block_top); // derelativize pointer
4171     __ lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
4172     // c_rarg1 points to current entry, starting with top-most entry
4173 
4174     __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4175                                         // of monitor block
4176     __ b(entry);
4177 
4178     __ bind(loop);
4179     // check if current entry is for same object
4180     __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
4181     __ cmp(r0, rscratch1);
4182     // if same object then stop searching
4183     __ br(Assembler::EQ, found);
4184     // otherwise advance to next entry
4185     __ add(c_rarg1, c_rarg1, entry_size);
4186     __ bind(entry);
4187     // check if bottom reached
4188     __ cmp(c_rarg1, c_rarg2);
4189     // if not at bottom then check this entry
4190     __ br(Assembler::NE, loop);
4191   }
4192 
4193   // error handling. Unlocking was not block-structured
4194   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4195                    InterpreterRuntime::throw_illegal_monitor_state_exception));
4196   __ should_not_reach_here();
4197 
4198   // call run-time routine
4199   __ bind(found);
4200   __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
4201   __ unlock_object(c_rarg1);
4202   __ pop_ptr(r0); // discard object
4203 }
4204 
4205 
4206 // Wide instructions
4207 void TemplateTable::wide()
4208 {
4209   __ load_unsigned_byte(r19, at_bcp(1));
4210   __ mov(rscratch1, (address)Interpreter::_wentry_point);
4211   __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
4212   __ br(rscratch1);
4213 }
4214 
4215 
4216 // Multi arrays
4217 void TemplateTable::multianewarray() {
4218   transition(vtos, atos);
4219   __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
4220   // last dim is on top of stack; we want address of first one:
4221   // first_addr = last_addr + (ndims - 1) * wordSize
4222   __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
4223   __ sub(c_rarg1, c_rarg1, wordSize);
4224   call_VM(r0,
4225           CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
4226           c_rarg1);
4227   __ load_unsigned_byte(r1, at_bcp(3));
4228   __ lea(esp, Address(esp, r1, Address::uxtw(3)));
4229 }