1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "gc/shared/barrierSetAssembler.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "interpreter/interp_masm.hpp"
  30 #include "interpreter/templateTable.hpp"
  31 #include "memory/universe.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/objArrayKlass.hpp"
  34 #include "oops/oop.inline.hpp"
  35 #include "prims/methodHandles.hpp"
  36 #include "runtime/frame.inline.hpp"
  37 #include "runtime/safepointMechanism.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 #include "runtime/synchronizer.hpp"
  41 #include "utilities/macros.hpp"
  42 
  43 #define __ _masm->
  44 
  45 // Misc helpers
  46 
  47 // Do an oop store like *(base + index + offset) = val
  48 // index can be noreg,
  49 static void do_oop_store(InterpreterMacroAssembler* _masm,
  50                          Register base,
  51                          Register index,
  52                          int offset,
  53                          Register val,
  54                          Register tmp,
  55                          DecoratorSet decorators = 0) {
  56   assert(tmp != val && tmp != base && tmp != index, "register collision");
  57   assert(index == noreg || offset == 0, "only one offset");
  58   if (index == noreg) {
  59     __ store_heap_oop(val, base, offset, tmp, decorators);
  60   } else {
  61     __ store_heap_oop(val, base, index, tmp, decorators);
  62   }
  63 }
  64 
  65 // Do an oop load like val = *(base + index + offset)
  66 // index can be noreg.
  67 static void do_oop_load(InterpreterMacroAssembler* _masm,
  68                         Register base,
  69                         Register index,
  70                         int offset,
  71                         Register dst,
  72                         Register tmp,
  73                         DecoratorSet decorators = 0) {
  74   assert(tmp != dst && tmp != base && tmp != index, "register collision");
  75   assert(index == noreg || offset == 0, "only one offset");
  76   if (index == noreg) {
  77     __ load_heap_oop(base, offset, dst, tmp, decorators);
  78   } else {
  79     __ load_heap_oop(base, index, dst, tmp, decorators);
  80   }
  81 }
  82 
  83 
  84 //----------------------------------------------------------------------------------------------------
  85 // Platform-dependent initialization
  86 
  87 void TemplateTable::pd_initialize() {
  88   // (none)
  89 }
  90 
  91 
  92 //----------------------------------------------------------------------------------------------------
  93 // Condition conversion
  94 Assembler::Condition ccNot(TemplateTable::Condition cc) {
  95   switch (cc) {
  96     case TemplateTable::equal        : return Assembler::notEqual;
  97     case TemplateTable::not_equal    : return Assembler::equal;
  98     case TemplateTable::less         : return Assembler::greaterEqual;
  99     case TemplateTable::less_equal   : return Assembler::greater;
 100     case TemplateTable::greater      : return Assembler::lessEqual;
 101     case TemplateTable::greater_equal: return Assembler::less;
 102   }
 103   ShouldNotReachHere();
 104   return Assembler::zero;
 105 }
 106 
 107 //----------------------------------------------------------------------------------------------------
 108 // Miscelaneous helper routines
 109 
 110 
 111 Address TemplateTable::at_bcp(int offset) {
 112   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 113   return Address(Lbcp, offset);
 114 }
 115 
 116 
 117 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 118                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 119                                    int byte_no) {
 120   // With sharing on, may need to test Method* flag.
 121   if (!RewriteBytecodes)  return;
 122   Label L_patch_done;
 123 
 124   switch (bc) {
 125   case Bytecodes::_fast_aputfield:
 126   case Bytecodes::_fast_bputfield:
 127   case Bytecodes::_fast_zputfield:
 128   case Bytecodes::_fast_cputfield:
 129   case Bytecodes::_fast_dputfield:
 130   case Bytecodes::_fast_fputfield:
 131   case Bytecodes::_fast_iputfield:
 132   case Bytecodes::_fast_lputfield:
 133   case Bytecodes::_fast_sputfield:
 134     {
 135       // We skip bytecode quickening for putfield instructions when
 136       // the put_code written to the constant pool cache is zero.
 137       // This is required so that every execution of this instruction
 138       // calls out to InterpreterRuntime::resolve_get_put to do
 139       // additional, required work.
 140       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 141       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 142       __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
 143       __ set(bc, bc_reg);
 144       __ cmp_and_br_short(temp_reg, 0, Assembler::equal, Assembler::pn, L_patch_done);  // don't patch
 145     }
 146     break;
 147   default:
 148     assert(byte_no == -1, "sanity");
 149     if (load_bc_into_bc_reg) {
 150       __ set(bc, bc_reg);
 151     }
 152   }
 153 
 154   if (JvmtiExport::can_post_breakpoint()) {
 155     Label L_fast_patch;
 156     __ ldub(at_bcp(0), temp_reg);
 157     __ cmp_and_br_short(temp_reg, Bytecodes::_breakpoint, Assembler::notEqual, Assembler::pt, L_fast_patch);
 158     // perform the quickening, slowly, in the bowels of the breakpoint table
 159     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, bc_reg);
 160     __ ba_short(L_patch_done);
 161     __ bind(L_fast_patch);
 162   }
 163 
 164 #ifdef ASSERT
 165   Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
 166   Label L_okay;
 167   __ ldub(at_bcp(0), temp_reg);
 168   __ cmp(temp_reg, orig_bytecode);
 169   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 170   __ delayed()->cmp(temp_reg, bc_reg);
 171   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 172   __ delayed()->nop();
 173   __ stop("patching the wrong bytecode");
 174   __ bind(L_okay);
 175 #endif
 176 
 177   // patch bytecode
 178   __ stb(bc_reg, at_bcp(0));
 179   __ bind(L_patch_done);
 180 }
 181 
 182 //----------------------------------------------------------------------------------------------------
 183 // Individual instructions
 184 
 185 void TemplateTable::nop() {
 186   transition(vtos, vtos);
 187   // nothing to do
 188 }
 189 
 190 void TemplateTable::shouldnotreachhere() {
 191   transition(vtos, vtos);
 192   __ stop("shouldnotreachhere bytecode");
 193 }
 194 
 195 void TemplateTable::aconst_null() {
 196   transition(vtos, atos);
 197   __ clr(Otos_i);
 198 }
 199 
 200 
 201 void TemplateTable::iconst(int value) {
 202   transition(vtos, itos);
 203   __ set(value, Otos_i);
 204 }
 205 
 206 
 207 void TemplateTable::lconst(int value) {
 208   transition(vtos, ltos);
 209   assert(value >= 0, "check this code");
 210   __ set(value, Otos_l);
 211 }
 212 
 213 
 214 void TemplateTable::fconst(int value) {
 215   transition(vtos, ftos);
 216   static float zero = 0.0, one = 1.0, two = 2.0;
 217   float* p;
 218   switch( value ) {
 219    default: ShouldNotReachHere();
 220    case 0:  p = &zero;  break;
 221    case 1:  p = &one;   break;
 222    case 2:  p = &two;   break;
 223   }
 224   AddressLiteral a(p);
 225   __ sethi(a, G3_scratch);
 226   __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
 227 }
 228 
 229 
 230 void TemplateTable::dconst(int value) {
 231   transition(vtos, dtos);
 232   static double zero = 0.0, one = 1.0;
 233   double* p;
 234   switch( value ) {
 235    default: ShouldNotReachHere();
 236    case 0:  p = &zero;  break;
 237    case 1:  p = &one;   break;
 238   }
 239   AddressLiteral a(p);
 240   __ sethi(a, G3_scratch);
 241   __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
 242 }
 243 
 244 
 245 // %%%%% Should factore most snippet templates across platforms
 246 
 247 void TemplateTable::bipush() {
 248   transition(vtos, itos);
 249   __ ldsb( at_bcp(1), Otos_i );
 250 }
 251 
 252 void TemplateTable::sipush() {
 253   transition(vtos, itos);
 254   __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
 255 }
 256 
 257 void TemplateTable::ldc(bool wide) {
 258   transition(vtos, vtos);
 259   Label call_ldc, notInt, isString, notString, notClass, notFloat, exit;
 260 
 261   if (wide) {
 262     __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 263   } else {
 264     __ ldub(Lbcp, 1, O1);
 265   }
 266   __ get_cpool_and_tags(O0, O2);
 267 
 268   const int base_offset = ConstantPool::header_size() * wordSize;
 269   const int tags_offset = Array<u1>::base_offset_in_bytes();
 270 
 271   // get type from tags
 272   __ add(O2, tags_offset, O2);
 273   __ ldub(O2, O1, O2);
 274 
 275   // unresolved class? If so, must resolve
 276   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClass, Assembler::equal, Assembler::pt, call_ldc);
 277 
 278   // unresolved class in error state
 279   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClassInError, Assembler::equal, Assembler::pn, call_ldc);
 280 
 281   __ cmp(O2, JVM_CONSTANT_Class);      // need to call vm to get java mirror of the class
 282   __ brx(Assembler::notEqual, true, Assembler::pt, notClass);
 283   __ delayed()->add(O0, base_offset, O0);
 284 
 285   __ bind(call_ldc);
 286   __ set(wide, O1);
 287   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), O1);
 288   __ push(atos);
 289   __ ba(exit);
 290   __ delayed()->nop();
 291 
 292   __ bind(notClass);
 293  // __ add(O0, base_offset, O0);
 294   __ sll(O1, LogBytesPerWord, O1);
 295   __ cmp(O2, JVM_CONSTANT_Integer);
 296   __ brx(Assembler::notEqual, true, Assembler::pt, notInt);
 297   __ delayed()->cmp(O2, JVM_CONSTANT_String);
 298   __ ld(O0, O1, Otos_i);
 299   __ push(itos);
 300   __ ba(exit);
 301   __ delayed()->nop();
 302 
 303   __ bind(notInt);
 304  // __ cmp(O2, JVM_CONSTANT_String);
 305   __ brx(Assembler::notEqual, true, Assembler::pt, notString);
 306   __ delayed()->cmp(O2, JVM_CONSTANT_Float);
 307   __ bind(isString);
 308   __ stop("string should be rewritten to fast_aldc");
 309   __ ba(exit);
 310   __ delayed()->nop();
 311 
 312   __ bind(notString);
 313  //__ cmp(O2, JVM_CONSTANT_Float);
 314   __ brx(Assembler::notEqual, true, Assembler::pt, notFloat);
 315   __ delayed()->nop();
 316   __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
 317   __ push(ftos);
 318   __ ba(exit);
 319   __ delayed()->nop();
 320 
 321   // assume the tag is for condy; if not, the VM runtime will tell us
 322   __ bind(notFloat);
 323   condy_helper(exit);
 324 
 325   __ bind(exit);
 326 }
 327 
 328 // Fast path for caching oop constants.
 329 // %%% We should use this to handle Class and String constants also.
 330 // %%% It will simplify the ldc/primitive path considerably.
 331 void TemplateTable::fast_aldc(bool wide) {
 332   transition(vtos, atos);
 333 
 334   int index_size = wide ? sizeof(u2) : sizeof(u1);
 335   Label resolved;
 336 
 337   // We are resolved if the resolved reference cache entry contains a
 338   // non-null object (CallSite, etc.)
 339   assert_different_registers(Otos_i, G3_scratch);
 340   __ get_cache_index_at_bcp(Otos_i, G3_scratch, 1, index_size);  // load index => G3_scratch
 341   __ load_resolved_reference_at_index(Otos_i, G3_scratch, Lscratch);
 342   __ tst(Otos_i);
 343   __ br(Assembler::notEqual, false, Assembler::pt, resolved);
 344   __ delayed()->set((int)bytecode(), O1);
 345 
 346   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 347 
 348   // first time invocation - must resolve first
 349   __ call_VM(Otos_i, entry, O1);
 350   __ bind(resolved);
 351 
 352   { // Check for the null sentinel.
 353     // If we just called the VM, it already did the mapping for us,
 354     // but it's harmless to retry.
 355     Label notNull;
 356     __ set(ExternalAddress((address)Universe::the_null_sentinel_addr()), G3_scratch);
 357     __ ld_ptr(G3_scratch, 0, G3_scratch);
 358     __ cmp(G3_scratch, Otos_i);
 359     __ br(Assembler::notEqual, true, Assembler::pt, notNull);
 360     __ delayed()->nop();
 361     __ clr(Otos_i);  // NULL object reference
 362     __ bind(notNull);
 363   }
 364 
 365   // Safe to call with 0 result
 366   __ verify_oop(Otos_i);
 367 }
 368 
 369 void TemplateTable::ldc2_w() {
 370   transition(vtos, vtos);
 371   Label notDouble, notLong, exit;
 372 
 373   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 374   __ get_cpool_and_tags(O0, O2);
 375 
 376   const int base_offset = ConstantPool::header_size() * wordSize;
 377   const int tags_offset = Array<u1>::base_offset_in_bytes();
 378   // get type from tags
 379   __ add(O2, tags_offset, O2);
 380   __ ldub(O2, O1, O2);
 381 
 382   __ sll(O1, LogBytesPerWord, O1);
 383   __ add(O0, O1, G3_scratch);
 384 
 385   __ cmp_and_brx_short(O2, JVM_CONSTANT_Double, Assembler::notEqual, Assembler::pt, notDouble);
 386   // A double can be placed at word-aligned locations in the constant pool.
 387   // Check out Conversions.java for an example.
 388   // Also ConstantPool::header_size() is 20, which makes it very difficult
 389   // to double-align double on the constant pool.  SG, 11/7/97
 390   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
 391   __ push(dtos);
 392   __ ba_short(exit);
 393 
 394   __ bind(notDouble);
 395   __ cmp_and_brx_short(O2, JVM_CONSTANT_Long, Assembler::notEqual, Assembler::pt, notLong);
 396   __ ldx(G3_scratch, base_offset, Otos_l);
 397   __ push(ltos);
 398   __ ba_short(exit);
 399 
 400   __ bind(notLong);
 401   condy_helper(exit);
 402 
 403   __ bind(exit);
 404 }
 405 
 406 void TemplateTable::condy_helper(Label& exit) {
 407   Register Robj = Otos_i;
 408   Register Roffset = G4_scratch;
 409   Register Rflags = G1_scratch;
 410 
 411   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 412 
 413   __ set((int)bytecode(), O1);
 414   __ call_VM(Robj, entry, O1);
 415 
 416   // Get vm_result_2 has flags = (tos, off) using format CPCE::_flags
 417   __ get_vm_result_2(G3_scratch);
 418 
 419   // Get offset
 420   __ set((int)ConstantPoolCacheEntry::field_index_mask, Roffset);
 421   __ and3(G3_scratch, Roffset, Roffset);
 422 
 423   // compute type
 424   __ srl(G3_scratch, ConstantPoolCacheEntry::tos_state_shift, Rflags);
 425   // Make sure we don't need to mask Rflags after the above shift
 426   ConstantPoolCacheEntry::verify_tos_state_shift();
 427 
 428   switch (bytecode()) {
 429   case Bytecodes::_ldc:
 430   case Bytecodes::_ldc_w:
 431     {
 432       // tos in (itos, ftos, stos, btos, ctos, ztos)
 433       Label notInt, notFloat, notShort, notByte, notChar, notBool;
 434       __ cmp(Rflags, itos);
 435       __ br(Assembler::notEqual, false, Assembler::pt, notInt);
 436       __ delayed()->cmp(Rflags, ftos);
 437       // itos
 438       __ ld(Robj, Roffset, Otos_i);
 439       __ push(itos);
 440       __ ba_short(exit);
 441 
 442       __ bind(notInt);
 443       __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
 444       __ delayed()->cmp(Rflags, stos);
 445       // ftos
 446       __ ldf(FloatRegisterImpl::S, Robj, Roffset, Ftos_f);
 447       __ push(ftos);
 448       __ ba_short(exit);
 449 
 450       __ bind(notFloat);
 451       __ br(Assembler::notEqual, false, Assembler::pt, notShort);
 452       __ delayed()->cmp(Rflags, btos);
 453       // stos
 454       __ ldsh(Robj, Roffset, Otos_i);
 455       __ push(itos);
 456       __ ba_short(exit);
 457 
 458       __ bind(notShort);
 459       __ br(Assembler::notEqual, false, Assembler::pt, notByte);
 460       __ delayed()->cmp(Rflags, ctos);
 461       // btos
 462       __ ldsb(Robj, Roffset, Otos_i);
 463       __ push(itos);
 464       __ ba_short(exit);
 465 
 466       __ bind(notByte);
 467       __ br(Assembler::notEqual, false, Assembler::pt, notChar);
 468       __ delayed()->cmp(Rflags, ztos);
 469       // ctos
 470       __ lduh(Robj, Roffset, Otos_i);
 471       __ push(itos);
 472       __ ba_short(exit);
 473 
 474       __ bind(notChar);
 475       __ br(Assembler::notEqual, false, Assembler::pt, notBool);
 476       __ delayed()->nop();
 477       // ztos
 478       __ ldsb(Robj, Roffset, Otos_i);
 479       __ push(itos);
 480       __ ba_short(exit);
 481 
 482       __ bind(notBool);
 483       break;
 484     }
 485 
 486   case Bytecodes::_ldc2_w:
 487     {
 488       Label notLong, notDouble;
 489       __ cmp(Rflags, ltos);
 490       __ br(Assembler::notEqual, false, Assembler::pt, notLong);
 491       __ delayed()->cmp(Rflags, dtos);
 492       // ltos
 493       // load must be atomic
 494       __ ld_long(Robj, Roffset, Otos_l);
 495       __ push(ltos);
 496       __ ba_short(exit);
 497 
 498       __ bind(notLong);
 499       __ br(Assembler::notEqual, false, Assembler::pt, notDouble);
 500       __ delayed()->nop();
 501       // dtos
 502       __ ldf(FloatRegisterImpl::D, Robj, Roffset, Ftos_d);
 503       __ push(dtos);
 504       __ ba_short(exit);
 505 
 506       __ bind(notDouble);
 507       break;
 508     }
 509 
 510   default:
 511     ShouldNotReachHere();
 512   }
 513 
 514   __ stop("bad ldc/condy");
 515 
 516   __ bind(exit);
 517 }
 518 
 519 void TemplateTable::locals_index(Register reg, int offset) {
 520   __ ldub( at_bcp(offset), reg );
 521 }
 522 
 523 void TemplateTable::locals_index_wide(Register reg) {
 524   // offset is 2, not 1, because Lbcp points to wide prefix code
 525   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
 526 }
 527 
 528 void TemplateTable::iload() {
 529   iload_internal();
 530 }
 531 
 532 void TemplateTable::nofast_iload() {
 533   iload_internal(may_not_rewrite);
 534 }
 535 
 536 void TemplateTable::iload_internal(RewriteControl rc) {
 537   transition(vtos, itos);
 538   // Rewrite iload,iload  pair into fast_iload2
 539   //         iload,caload pair into fast_icaload
 540   if (RewriteFrequentPairs && rc == may_rewrite) {
 541     Label rewrite, done;
 542 
 543     // get next byte
 544     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
 545 
 546     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 547     // last two iloads in a pair.  Comparing against fast_iload means that
 548     // the next bytecode is neither an iload or a caload, and therefore
 549     // an iload pair.
 550     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_iload, Assembler::equal, Assembler::pn, done);
 551 
 552     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
 553     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 554     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
 555 
 556     __ cmp(G3_scratch, (int)Bytecodes::_caload);
 557     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 558     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
 559 
 560     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
 561     // rewrite
 562     // G4_scratch: fast bytecode
 563     __ bind(rewrite);
 564     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
 565     __ bind(done);
 566   }
 567 
 568   // Get the local value into tos
 569   locals_index(G3_scratch);
 570   __ access_local_int( G3_scratch, Otos_i );
 571 }
 572 
 573 void TemplateTable::fast_iload2() {
 574   transition(vtos, itos);
 575   locals_index(G3_scratch);
 576   __ access_local_int( G3_scratch, Otos_i );
 577   __ push_i();
 578   locals_index(G3_scratch, 3);  // get next bytecode's local index.
 579   __ access_local_int( G3_scratch, Otos_i );
 580 }
 581 
 582 void TemplateTable::fast_iload() {
 583   transition(vtos, itos);
 584   locals_index(G3_scratch);
 585   __ access_local_int( G3_scratch, Otos_i );
 586 }
 587 
 588 void TemplateTable::lload() {
 589   transition(vtos, ltos);
 590   locals_index(G3_scratch);
 591   __ access_local_long( G3_scratch, Otos_l );
 592 }
 593 
 594 
 595 void TemplateTable::fload() {
 596   transition(vtos, ftos);
 597   locals_index(G3_scratch);
 598   __ access_local_float( G3_scratch, Ftos_f );
 599 }
 600 
 601 
 602 void TemplateTable::dload() {
 603   transition(vtos, dtos);
 604   locals_index(G3_scratch);
 605   __ access_local_double( G3_scratch, Ftos_d );
 606 }
 607 
 608 
 609 void TemplateTable::aload() {
 610   transition(vtos, atos);
 611   locals_index(G3_scratch);
 612   __ access_local_ptr( G3_scratch, Otos_i);
 613 }
 614 
 615 
 616 void TemplateTable::wide_iload() {
 617   transition(vtos, itos);
 618   locals_index_wide(G3_scratch);
 619   __ access_local_int( G3_scratch, Otos_i );
 620 }
 621 
 622 
 623 void TemplateTable::wide_lload() {
 624   transition(vtos, ltos);
 625   locals_index_wide(G3_scratch);
 626   __ access_local_long( G3_scratch, Otos_l );
 627 }
 628 
 629 
 630 void TemplateTable::wide_fload() {
 631   transition(vtos, ftos);
 632   locals_index_wide(G3_scratch);
 633   __ access_local_float( G3_scratch, Ftos_f );
 634 }
 635 
 636 
 637 void TemplateTable::wide_dload() {
 638   transition(vtos, dtos);
 639   locals_index_wide(G3_scratch);
 640   __ access_local_double( G3_scratch, Ftos_d );
 641 }
 642 
 643 
 644 void TemplateTable::wide_aload() {
 645   transition(vtos, atos);
 646   locals_index_wide(G3_scratch);
 647   __ access_local_ptr( G3_scratch, Otos_i );
 648   __ verify_oop(Otos_i);
 649 }
 650 
 651 
 652 void TemplateTable::iaload() {
 653   transition(itos, itos);
 654   // Otos_i: index
 655   // tos: array
 656   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 657   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
 658 }
 659 
 660 
 661 void TemplateTable::laload() {
 662   transition(itos, ltos);
 663   // Otos_i: index
 664   // O2: array
 665   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 666   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
 667 }
 668 
 669 
 670 void TemplateTable::faload() {
 671   transition(itos, ftos);
 672   // Otos_i: index
 673   // O2: array
 674   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 675   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
 676 }
 677 
 678 
 679 void TemplateTable::daload() {
 680   transition(itos, dtos);
 681   // Otos_i: index
 682   // O2: array
 683   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 684   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
 685 }
 686 
 687 
 688 void TemplateTable::aaload() {
 689   transition(itos, atos);
 690   // Otos_i: index
 691   // tos: array
 692   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
 693   do_oop_load(_masm,
 694               O3,
 695               noreg,
 696               arrayOopDesc::base_offset_in_bytes(T_OBJECT),
 697               Otos_i,
 698               G3_scratch,
 699               IN_HEAP_ARRAY);
 700   __ verify_oop(Otos_i);
 701 }
 702 
 703 
 704 void TemplateTable::baload() {
 705   transition(itos, itos);
 706   // Otos_i: index
 707   // tos: array
 708   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
 709   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
 710 }
 711 
 712 
 713 void TemplateTable::caload() {
 714   transition(itos, itos);
 715   // Otos_i: index
 716   // tos: array
 717   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 718   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 719 }
 720 
 721 void TemplateTable::fast_icaload() {
 722   transition(vtos, itos);
 723   // Otos_i: index
 724   // tos: array
 725   locals_index(G3_scratch);
 726   __ access_local_int( G3_scratch, Otos_i );
 727   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 728   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 729 }
 730 
 731 
 732 void TemplateTable::saload() {
 733   transition(itos, itos);
 734   // Otos_i: index
 735   // tos: array
 736   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 737   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
 738 }
 739 
 740 
 741 void TemplateTable::iload(int n) {
 742   transition(vtos, itos);
 743   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 744 }
 745 
 746 
 747 void TemplateTable::lload(int n) {
 748   transition(vtos, ltos);
 749   assert(n+1 < Argument::n_register_parameters, "would need more code");
 750   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
 751 }
 752 
 753 
 754 void TemplateTable::fload(int n) {
 755   transition(vtos, ftos);
 756   assert(n < Argument::n_register_parameters, "would need more code");
 757   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
 758 }
 759 
 760 
 761 void TemplateTable::dload(int n) {
 762   transition(vtos, dtos);
 763   FloatRegister dst = Ftos_d;
 764   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
 765 }
 766 
 767 
 768 void TemplateTable::aload(int n) {
 769   transition(vtos, atos);
 770   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 771 }
 772 
 773 void TemplateTable::aload_0() {
 774   aload_0_internal();
 775 }
 776 
 777 void TemplateTable::nofast_aload_0() {
 778   aload_0_internal(may_not_rewrite);
 779 }
 780 
 781 void TemplateTable::aload_0_internal(RewriteControl rc) {
 782   transition(vtos, atos);
 783 
 784   // According to bytecode histograms, the pairs:
 785   //
 786   // _aload_0, _fast_igetfield (itos)
 787   // _aload_0, _fast_agetfield (atos)
 788   // _aload_0, _fast_fgetfield (ftos)
 789   //
 790   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
 791   // bytecode checks the next bytecode and then rewrites the current
 792   // bytecode into a pair bytecode; otherwise it rewrites the current
 793   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
 794   //
 795   if (RewriteFrequentPairs && rc == may_rewrite) {
 796     Label rewrite, done;
 797 
 798     // get next byte
 799     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
 800 
 801     // if _getfield then wait with rewrite
 802     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_getfield, Assembler::equal, Assembler::pn, done);
 803 
 804     // if _igetfield then rewrite to _fast_iaccess_0
 805     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 806     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
 807     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 808     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
 809 
 810     // if _agetfield then rewrite to _fast_aaccess_0
 811     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 812     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
 813     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 814     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
 815 
 816     // if _fgetfield then rewrite to _fast_faccess_0
 817     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 818     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
 819     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 820     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
 821 
 822     // else rewrite to _fast_aload0
 823     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 824     __ set(Bytecodes::_fast_aload_0, G4_scratch);
 825 
 826     // rewrite
 827     // G4_scratch: fast bytecode
 828     __ bind(rewrite);
 829     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
 830     __ bind(done);
 831   }
 832 
 833   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 834   aload(0);
 835 }
 836 
 837 void TemplateTable::istore() {
 838   transition(itos, vtos);
 839   locals_index(G3_scratch);
 840   __ store_local_int( G3_scratch, Otos_i );
 841 }
 842 
 843 
 844 void TemplateTable::lstore() {
 845   transition(ltos, vtos);
 846   locals_index(G3_scratch);
 847   __ store_local_long( G3_scratch, Otos_l );
 848 }
 849 
 850 
 851 void TemplateTable::fstore() {
 852   transition(ftos, vtos);
 853   locals_index(G3_scratch);
 854   __ store_local_float( G3_scratch, Ftos_f );
 855 }
 856 
 857 
 858 void TemplateTable::dstore() {
 859   transition(dtos, vtos);
 860   locals_index(G3_scratch);
 861   __ store_local_double( G3_scratch, Ftos_d );
 862 }
 863 
 864 
 865 void TemplateTable::astore() {
 866   transition(vtos, vtos);
 867   __ load_ptr(0, Otos_i);
 868   __ inc(Lesp, Interpreter::stackElementSize);
 869   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 870   locals_index(G3_scratch);
 871   __ store_local_ptr(G3_scratch, Otos_i);
 872 }
 873 
 874 
 875 void TemplateTable::wide_istore() {
 876   transition(vtos, vtos);
 877   __ pop_i();
 878   locals_index_wide(G3_scratch);
 879   __ store_local_int( G3_scratch, Otos_i );
 880 }
 881 
 882 
 883 void TemplateTable::wide_lstore() {
 884   transition(vtos, vtos);
 885   __ pop_l();
 886   locals_index_wide(G3_scratch);
 887   __ store_local_long( G3_scratch, Otos_l );
 888 }
 889 
 890 
 891 void TemplateTable::wide_fstore() {
 892   transition(vtos, vtos);
 893   __ pop_f();
 894   locals_index_wide(G3_scratch);
 895   __ store_local_float( G3_scratch, Ftos_f );
 896 }
 897 
 898 
 899 void TemplateTable::wide_dstore() {
 900   transition(vtos, vtos);
 901   __ pop_d();
 902   locals_index_wide(G3_scratch);
 903   __ store_local_double( G3_scratch, Ftos_d );
 904 }
 905 
 906 
 907 void TemplateTable::wide_astore() {
 908   transition(vtos, vtos);
 909   __ load_ptr(0, Otos_i);
 910   __ inc(Lesp, Interpreter::stackElementSize);
 911   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 912   locals_index_wide(G3_scratch);
 913   __ store_local_ptr(G3_scratch, Otos_i);
 914 }
 915 
 916 
 917 void TemplateTable::iastore() {
 918   transition(itos, vtos);
 919   __ pop_i(O2); // index
 920   // Otos_i: val
 921   // O3: array
 922   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 923   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
 924 }
 925 
 926 
 927 void TemplateTable::lastore() {
 928   transition(ltos, vtos);
 929   __ pop_i(O2); // index
 930   // Otos_l: val
 931   // O3: array
 932   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 933   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
 934 }
 935 
 936 
 937 void TemplateTable::fastore() {
 938   transition(ftos, vtos);
 939   __ pop_i(O2); // index
 940   // Ftos_f: val
 941   // O3: array
 942   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 943   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
 944 }
 945 
 946 
 947 void TemplateTable::dastore() {
 948   transition(dtos, vtos);
 949   __ pop_i(O2); // index
 950   // Fos_d: val
 951   // O3: array
 952   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 953   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
 954 }
 955 
 956 
 957 void TemplateTable::aastore() {
 958   Label store_ok, is_null, done;
 959   transition(vtos, vtos);
 960   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
 961   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
 962   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
 963   // Otos_i: val
 964   // O2: index
 965   // O3: array
 966   __ verify_oop(Otos_i);
 967   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
 968 
 969   // do array store check - check for NULL value first
 970   __ br_null_short( Otos_i, Assembler::pn, is_null );
 971 
 972   __ load_klass(O3, O4); // get array klass
 973   __ load_klass(Otos_i, O5); // get value klass
 974 
 975   // do fast instanceof cache test
 976 
 977   __ ld_ptr(O4,     in_bytes(ObjArrayKlass::element_klass_offset()),  O4);
 978 
 979   assert(Otos_i == O0, "just checking");
 980 
 981   // Otos_i:    value
 982   // O1:        addr - offset
 983   // O2:        index
 984   // O3:        array
 985   // O4:        array element klass
 986   // O5:        value klass
 987 
 988   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
 989 
 990   // Generate a fast subtype check.  Branch to store_ok if no
 991   // failure.  Throw if failure.
 992   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
 993 
 994   // Not a subtype; so must throw exception
 995   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
 996 
 997   // Store is OK.
 998   __ bind(store_ok);
 999   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, IN_HEAP_ARRAY);
1000 
1001   __ ba(done);
1002   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
1003 
1004   __ bind(is_null);
1005   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, IN_HEAP_ARRAY);
1006 
1007   __ profile_null_seen(G3_scratch);
1008   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
1009   __ bind(done);
1010 }
1011 
1012 
1013 void TemplateTable::bastore() {
1014   transition(itos, vtos);
1015   __ pop_i(O2); // index
1016   // Otos_i: val
1017   // O2: index
1018   // O3: array
1019   __ index_check(O3, O2, 0, G3_scratch, O2);
1020   // Need to check whether array is boolean or byte
1021   // since both types share the bastore bytecode.
1022   __ load_klass(O3, G4_scratch);
1023   __ ld(G4_scratch, in_bytes(Klass::layout_helper_offset()), G4_scratch);
1024   __ set(Klass::layout_helper_boolean_diffbit(), G3_scratch);
1025   __ andcc(G3_scratch, G4_scratch, G0);
1026   Label L_skip;
1027   __ br(Assembler::zero, false, Assembler::pn, L_skip);
1028   __ delayed()->nop();
1029   __ and3(Otos_i, 1, Otos_i);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
1030   __ bind(L_skip);
1031   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1032 }
1033 
1034 
1035 void TemplateTable::castore() {
1036   transition(itos, vtos);
1037   __ pop_i(O2); // index
1038   // Otos_i: val
1039   // O3: array
1040   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
1041   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
1042 }
1043 
1044 
1045 void TemplateTable::sastore() {
1046   // %%%%% Factor across platform
1047   castore();
1048 }
1049 
1050 
1051 void TemplateTable::istore(int n) {
1052   transition(itos, vtos);
1053   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
1054 }
1055 
1056 
1057 void TemplateTable::lstore(int n) {
1058   transition(ltos, vtos);
1059   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
1060   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
1061 
1062 }
1063 
1064 
1065 void TemplateTable::fstore(int n) {
1066   transition(ftos, vtos);
1067   assert(n < Argument::n_register_parameters, "only handle register cases");
1068   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
1069 }
1070 
1071 
1072 void TemplateTable::dstore(int n) {
1073   transition(dtos, vtos);
1074   FloatRegister src = Ftos_d;
1075   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
1076 }
1077 
1078 
1079 void TemplateTable::astore(int n) {
1080   transition(vtos, vtos);
1081   __ load_ptr(0, Otos_i);
1082   __ inc(Lesp, Interpreter::stackElementSize);
1083   __ verify_oop_or_return_address(Otos_i, G3_scratch);
1084   __ store_local_ptr(n, Otos_i);
1085 }
1086 
1087 
1088 void TemplateTable::pop() {
1089   transition(vtos, vtos);
1090   __ inc(Lesp, Interpreter::stackElementSize);
1091 }
1092 
1093 
1094 void TemplateTable::pop2() {
1095   transition(vtos, vtos);
1096   __ inc(Lesp, 2 * Interpreter::stackElementSize);
1097 }
1098 
1099 
1100 void TemplateTable::dup() {
1101   transition(vtos, vtos);
1102   // stack: ..., a
1103   // load a and tag
1104   __ load_ptr(0, Otos_i);
1105   __ push_ptr(Otos_i);
1106   // stack: ..., a, a
1107 }
1108 
1109 
1110 void TemplateTable::dup_x1() {
1111   transition(vtos, vtos);
1112   // stack: ..., a, b
1113   __ load_ptr( 1, G3_scratch);  // get a
1114   __ load_ptr( 0, Otos_l1);     // get b
1115   __ store_ptr(1, Otos_l1);     // put b
1116   __ store_ptr(0, G3_scratch);  // put a - like swap
1117   __ push_ptr(Otos_l1);         // push b
1118   // stack: ..., b, a, b
1119 }
1120 
1121 
1122 void TemplateTable::dup_x2() {
1123   transition(vtos, vtos);
1124   // stack: ..., a, b, c
1125   // get c and push on stack, reuse registers
1126   __ load_ptr( 0, G3_scratch);  // get c
1127   __ push_ptr(G3_scratch);      // push c with tag
1128   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
1129   // (stack offsets n+1 now)
1130   __ load_ptr( 3, Otos_l1);     // get a
1131   __ store_ptr(3, G3_scratch);  // put c at 3
1132   // stack: ..., c, b, c, c  (a in reg)
1133   __ load_ptr( 2, G3_scratch);  // get b
1134   __ store_ptr(2, Otos_l1);     // put a at 2
1135   // stack: ..., c, a, c, c  (b in reg)
1136   __ store_ptr(1, G3_scratch);  // put b at 1
1137   // stack: ..., c, a, b, c
1138 }
1139 
1140 
1141 void TemplateTable::dup2() {
1142   transition(vtos, vtos);
1143   __ load_ptr(1, G3_scratch);  // get a
1144   __ load_ptr(0, Otos_l1);     // get b
1145   __ push_ptr(G3_scratch);     // push a
1146   __ push_ptr(Otos_l1);        // push b
1147   // stack: ..., a, b, a, b
1148 }
1149 
1150 
1151 void TemplateTable::dup2_x1() {
1152   transition(vtos, vtos);
1153   // stack: ..., a, b, c
1154   __ load_ptr( 1, Lscratch);    // get b
1155   __ load_ptr( 2, Otos_l1);     // get a
1156   __ store_ptr(2, Lscratch);    // put b at a
1157   // stack: ..., b, b, c
1158   __ load_ptr( 0, G3_scratch);  // get c
1159   __ store_ptr(1, G3_scratch);  // put c at b
1160   // stack: ..., b, c, c
1161   __ store_ptr(0, Otos_l1);     // put a at c
1162   // stack: ..., b, c, a
1163   __ push_ptr(Lscratch);        // push b
1164   __ push_ptr(G3_scratch);      // push c
1165   // stack: ..., b, c, a, b, c
1166 }
1167 
1168 
1169 // The spec says that these types can be a mixture of category 1 (1 word)
1170 // types and/or category 2 types (long and doubles)
1171 void TemplateTable::dup2_x2() {
1172   transition(vtos, vtos);
1173   // stack: ..., a, b, c, d
1174   __ load_ptr( 1, Lscratch);    // get c
1175   __ load_ptr( 3, Otos_l1);     // get a
1176   __ store_ptr(3, Lscratch);    // put c at 3
1177   __ store_ptr(1, Otos_l1);     // put a at 1
1178   // stack: ..., c, b, a, d
1179   __ load_ptr( 2, G3_scratch);  // get b
1180   __ load_ptr( 0, Otos_l1);     // get d
1181   __ store_ptr(0, G3_scratch);  // put b at 0
1182   __ store_ptr(2, Otos_l1);     // put d at 2
1183   // stack: ..., c, d, a, b
1184   __ push_ptr(Lscratch);        // push c
1185   __ push_ptr(Otos_l1);         // push d
1186   // stack: ..., c, d, a, b, c, d
1187 }
1188 
1189 
1190 void TemplateTable::swap() {
1191   transition(vtos, vtos);
1192   // stack: ..., a, b
1193   __ load_ptr( 1, G3_scratch);  // get a
1194   __ load_ptr( 0, Otos_l1);     // get b
1195   __ store_ptr(0, G3_scratch);  // put b
1196   __ store_ptr(1, Otos_l1);     // put a
1197   // stack: ..., b, a
1198 }
1199 
1200 
1201 void TemplateTable::iop2(Operation op) {
1202   transition(itos, itos);
1203   __ pop_i(O1);
1204   switch (op) {
1205    case  add:  __  add(O1, Otos_i, Otos_i);  break;
1206    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
1207      // %%%%% Mul may not exist: better to call .mul?
1208    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
1209    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
1210    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
1211    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
1212    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
1213    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
1214    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
1215    default: ShouldNotReachHere();
1216   }
1217 }
1218 
1219 
1220 void TemplateTable::lop2(Operation op) {
1221   transition(ltos, ltos);
1222   __ pop_l(O2);
1223   switch (op) {
1224    case  add:  __  add(O2, Otos_l, Otos_l);  break;
1225    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
1226    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
1227    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
1228    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
1229    default: ShouldNotReachHere();
1230   }
1231 }
1232 
1233 
1234 void TemplateTable::idiv() {
1235   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
1236   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
1237 
1238   transition(itos, itos);
1239   __ pop_i(O1); // get 1st op
1240 
1241   // Y contains upper 32 bits of result, set it to 0 or all ones
1242   __ wry(G0);
1243   __ mov(~0, G3_scratch);
1244 
1245   __ tst(O1);
1246      Label neg;
1247   __ br(Assembler::negative, true, Assembler::pn, neg);
1248   __ delayed()->wry(G3_scratch);
1249   __ bind(neg);
1250 
1251      Label ok;
1252   __ tst(Otos_i);
1253   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
1254 
1255   const int min_int = 0x80000000;
1256   Label regular;
1257   __ cmp(Otos_i, -1);
1258   __ br(Assembler::notEqual, false, Assembler::pt, regular);
1259   // Don't put set in delay slot
1260   // Set will turn into multiple instructions in 64 bit mode
1261   __ delayed()->nop();
1262   __ set(min_int, G4_scratch);
1263   Label done;
1264   __ cmp(O1, G4_scratch);
1265   __ br(Assembler::equal, true, Assembler::pt, done);
1266   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
1267 
1268   __ bind(regular);
1269   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
1270   __ bind(done);
1271 }
1272 
1273 
1274 void TemplateTable::irem() {
1275   transition(itos, itos);
1276   __ mov(Otos_i, O2); // save divisor
1277   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
1278   __ smul(Otos_i, O2, Otos_i);
1279   __ sub(O1, Otos_i, Otos_i);
1280 }
1281 
1282 
1283 void TemplateTable::lmul() {
1284   transition(ltos, ltos);
1285   __ pop_l(O2);
1286   __ mulx(Otos_l, O2, Otos_l);
1287 
1288 }
1289 
1290 
1291 void TemplateTable::ldiv() {
1292   transition(ltos, ltos);
1293 
1294   // check for zero
1295   __ pop_l(O2);
1296   __ tst(Otos_l);
1297   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1298   __ sdivx(O2, Otos_l, Otos_l);
1299 }
1300 
1301 
1302 void TemplateTable::lrem() {
1303   transition(ltos, ltos);
1304 
1305   // check for zero
1306   __ pop_l(O2);
1307   __ tst(Otos_l);
1308   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1309   __ sdivx(O2, Otos_l, Otos_l2);
1310   __ mulx (Otos_l2, Otos_l, Otos_l2);
1311   __ sub  (O2, Otos_l2, Otos_l);
1312 }
1313 
1314 
1315 void TemplateTable::lshl() {
1316   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
1317 
1318   __ pop_l(O2);                          // shift value in O2, O3
1319   __ sllx(O2, Otos_i, Otos_l);
1320 }
1321 
1322 
1323 void TemplateTable::lshr() {
1324   transition(itos, ltos); // %%%% see lshl comment
1325 
1326   __ pop_l(O2);                          // shift value in O2, O3
1327   __ srax(O2, Otos_i, Otos_l);
1328 }
1329 
1330 
1331 
1332 void TemplateTable::lushr() {
1333   transition(itos, ltos); // %%%% see lshl comment
1334 
1335   __ pop_l(O2);                          // shift value in O2, O3
1336   __ srlx(O2, Otos_i, Otos_l);
1337 }
1338 
1339 
1340 void TemplateTable::fop2(Operation op) {
1341   transition(ftos, ftos);
1342   switch (op) {
1343    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1344    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1345    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1346    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1347    case  rem:
1348      assert(Ftos_f == F0, "just checking");
1349      // LP64 calling conventions use F1, F3 for passing 2 floats
1350      __ pop_f(F1);
1351      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
1352      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1353      assert( Ftos_f == F0, "fix this code" );
1354      break;
1355 
1356    default: ShouldNotReachHere();
1357   }
1358 }
1359 
1360 
1361 void TemplateTable::dop2(Operation op) {
1362   transition(dtos, dtos);
1363   switch (op) {
1364    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1365    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1366    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1367    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1368    case  rem:
1369      // Pass arguments in D0, D2
1370      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
1371      __ pop_d( F0 );
1372      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1373      assert( Ftos_d == F0, "fix this code" );
1374      break;
1375 
1376    default: ShouldNotReachHere();
1377   }
1378 }
1379 
1380 
1381 void TemplateTable::ineg() {
1382   transition(itos, itos);
1383   __ neg(Otos_i);
1384 }
1385 
1386 
1387 void TemplateTable::lneg() {
1388   transition(ltos, ltos);
1389   __ sub(G0, Otos_l, Otos_l);
1390 }
1391 
1392 
1393 void TemplateTable::fneg() {
1394   transition(ftos, ftos);
1395   __ fneg(FloatRegisterImpl::S, Ftos_f, Ftos_f);
1396 }
1397 
1398 
1399 void TemplateTable::dneg() {
1400   transition(dtos, dtos);
1401   __ fneg(FloatRegisterImpl::D, Ftos_f, Ftos_f);
1402 }
1403 
1404 
1405 void TemplateTable::iinc() {
1406   transition(vtos, vtos);
1407   locals_index(G3_scratch);
1408   __ ldsb(Lbcp, 2, O2);  // load constant
1409   __ access_local_int(G3_scratch, Otos_i);
1410   __ add(Otos_i, O2, Otos_i);
1411   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1412 }
1413 
1414 
1415 void TemplateTable::wide_iinc() {
1416   transition(vtos, vtos);
1417   locals_index_wide(G3_scratch);
1418   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
1419   __ access_local_int(G3_scratch, Otos_i);
1420   __ add(Otos_i, O3, Otos_i);
1421   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1422 }
1423 
1424 
1425 void TemplateTable::convert() {
1426 // %%%%% Factor this first part accross platforms
1427   #ifdef ASSERT
1428     TosState tos_in  = ilgl;
1429     TosState tos_out = ilgl;
1430     switch (bytecode()) {
1431       case Bytecodes::_i2l: // fall through
1432       case Bytecodes::_i2f: // fall through
1433       case Bytecodes::_i2d: // fall through
1434       case Bytecodes::_i2b: // fall through
1435       case Bytecodes::_i2c: // fall through
1436       case Bytecodes::_i2s: tos_in = itos; break;
1437       case Bytecodes::_l2i: // fall through
1438       case Bytecodes::_l2f: // fall through
1439       case Bytecodes::_l2d: tos_in = ltos; break;
1440       case Bytecodes::_f2i: // fall through
1441       case Bytecodes::_f2l: // fall through
1442       case Bytecodes::_f2d: tos_in = ftos; break;
1443       case Bytecodes::_d2i: // fall through
1444       case Bytecodes::_d2l: // fall through
1445       case Bytecodes::_d2f: tos_in = dtos; break;
1446       default             : ShouldNotReachHere();
1447     }
1448     switch (bytecode()) {
1449       case Bytecodes::_l2i: // fall through
1450       case Bytecodes::_f2i: // fall through
1451       case Bytecodes::_d2i: // fall through
1452       case Bytecodes::_i2b: // fall through
1453       case Bytecodes::_i2c: // fall through
1454       case Bytecodes::_i2s: tos_out = itos; break;
1455       case Bytecodes::_i2l: // fall through
1456       case Bytecodes::_f2l: // fall through
1457       case Bytecodes::_d2l: tos_out = ltos; break;
1458       case Bytecodes::_i2f: // fall through
1459       case Bytecodes::_l2f: // fall through
1460       case Bytecodes::_d2f: tos_out = ftos; break;
1461       case Bytecodes::_i2d: // fall through
1462       case Bytecodes::_l2d: // fall through
1463       case Bytecodes::_f2d: tos_out = dtos; break;
1464       default             : ShouldNotReachHere();
1465     }
1466     transition(tos_in, tos_out);
1467   #endif
1468 
1469 
1470   // Conversion
1471   Label done;
1472   switch (bytecode()) {
1473    case Bytecodes::_i2l:
1474     // Sign extend the 32 bits
1475     __ sra ( Otos_i, 0, Otos_l );
1476     break;
1477 
1478    case Bytecodes::_i2f:
1479     __ st(Otos_i, __ d_tmp );
1480     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1481     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
1482     break;
1483 
1484    case Bytecodes::_i2d:
1485     __ st(Otos_i, __ d_tmp);
1486     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1487     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
1488     break;
1489 
1490    case Bytecodes::_i2b:
1491     __ sll(Otos_i, 24, Otos_i);
1492     __ sra(Otos_i, 24, Otos_i);
1493     break;
1494 
1495    case Bytecodes::_i2c:
1496     __ sll(Otos_i, 16, Otos_i);
1497     __ srl(Otos_i, 16, Otos_i);
1498     break;
1499 
1500    case Bytecodes::_i2s:
1501     __ sll(Otos_i, 16, Otos_i);
1502     __ sra(Otos_i, 16, Otos_i);
1503     break;
1504 
1505    case Bytecodes::_l2i:
1506     // Sign-extend into the high 32 bits
1507     __ sra(Otos_l, 0, Otos_i);
1508     break;
1509 
1510    case Bytecodes::_l2f:
1511    case Bytecodes::_l2d:
1512     __ st_long(Otos_l, __ d_tmp);
1513     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
1514 
1515     if (bytecode() == Bytecodes::_l2f) {
1516       __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
1517     } else {
1518       __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
1519     }
1520     break;
1521 
1522   case Bytecodes::_f2i:  {
1523       Label isNaN;
1524       // result must be 0 if value is NaN; test by comparing value to itself
1525       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
1526       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
1527       __ delayed()->clr(Otos_i);                                     // NaN
1528       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
1529       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
1530       __ ld(__ d_tmp, Otos_i);
1531       __ bind(isNaN);
1532     }
1533     break;
1534 
1535    case Bytecodes::_f2l:
1536     // must uncache tos
1537     __ push_f();
1538     __ pop_f(F1);
1539     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1540     break;
1541 
1542    case Bytecodes::_f2d:
1543     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
1544     break;
1545 
1546    case Bytecodes::_d2i:
1547    case Bytecodes::_d2l:
1548     // must uncache tos
1549     __ push_d();
1550     // LP64 calling conventions pass first double arg in D0
1551     __ pop_d( Ftos_d );
1552     __ call_VM_leaf(Lscratch,
1553         bytecode() == Bytecodes::_d2i
1554           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
1555           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1556     break;
1557 
1558     case Bytecodes::_d2f:
1559       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
1560     break;
1561 
1562     default: ShouldNotReachHere();
1563   }
1564   __ bind(done);
1565 }
1566 
1567 
1568 void TemplateTable::lcmp() {
1569   transition(ltos, itos);
1570 
1571   __ pop_l(O1); // pop off value 1, value 2 is in O0
1572   __ lcmp( O1, Otos_l, Otos_i );
1573 }
1574 
1575 
1576 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1577 
1578   if (is_float) __ pop_f(F2);
1579   else          __ pop_d(F2);
1580 
1581   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
1582 
1583   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
1584 }
1585 
1586 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1587   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1588   __ verify_thread();
1589 
1590   const Register O2_bumped_count = O2;
1591   __ profile_taken_branch(G3_scratch, O2_bumped_count);
1592 
1593   // get (wide) offset to O1_disp
1594   const Register O1_disp = O1;
1595   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
1596   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
1597 
1598   // Handle all the JSR stuff here, then exit.
1599   // It's much shorter and cleaner than intermingling with the
1600   // non-JSR normal-branch stuff occurring below.
1601   if( is_jsr ) {
1602     // compute return address as bci in Otos_i
1603     __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1604     __ sub(Lbcp, G3_scratch, G3_scratch);
1605     __ sub(G3_scratch, in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
1606 
1607     // Bump Lbcp to target of JSR
1608     __ add(Lbcp, O1_disp, Lbcp);
1609     // Push returnAddress for "ret" on stack
1610     __ push_ptr(Otos_i);
1611     // And away we go!
1612     __ dispatch_next(vtos, 0, true);
1613     return;
1614   }
1615 
1616   // Normal (non-jsr) branch handling
1617 
1618   // Save the current Lbcp
1619   const Register l_cur_bcp = Lscratch;
1620   __ mov( Lbcp, l_cur_bcp );
1621 
1622   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1623   if ( increment_invocation_counter_for_backward_branches ) {
1624     Label Lforward;
1625     // check branch direction
1626     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
1627     // Bump bytecode pointer by displacement (take the branch)
1628     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
1629 
1630     const Register G3_method_counters = G3_scratch;
1631     __ get_method_counters(Lmethod, G3_method_counters, Lforward);
1632 
1633     if (TieredCompilation) {
1634       Label Lno_mdo, Loverflow;
1635       int increment = InvocationCounter::count_increment;
1636       if (ProfileInterpreter) {
1637         // If no method data exists, go to profile_continue.
1638         __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch);
1639         __ br_null_short(G4_scratch, Assembler::pn, Lno_mdo);
1640 
1641         // Increment backedge counter in the MDO
1642         Address mdo_backedge_counter(G4_scratch, in_bytes(MethodData::backedge_counter_offset()) +
1643                                                  in_bytes(InvocationCounter::counter_offset()));
1644         Address mask(G4_scratch, in_bytes(MethodData::backedge_mask_offset()));
1645         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, O0,
1646                                    (UseOnStackReplacement ? Assembler::notZero : Assembler::always), &Lforward);
1647         __ ba_short(Loverflow);
1648       }
1649 
1650       // If there's no MDO, increment counter in MethodCounters*
1651       __ bind(Lno_mdo);
1652       Address backedge_counter(G3_method_counters,
1653               in_bytes(MethodCounters::backedge_counter_offset()) +
1654               in_bytes(InvocationCounter::counter_offset()));
1655       Address mask(G3_method_counters, in_bytes(MethodCounters::backedge_mask_offset()));
1656       __ increment_mask_and_jump(backedge_counter, increment, mask, G4_scratch, O0,
1657                                  (UseOnStackReplacement ? Assembler::notZero : Assembler::always), &Lforward);
1658       __ bind(Loverflow);
1659 
1660       // notify point for loop, pass branch bytecode
1661       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), l_cur_bcp);
1662 
1663       // Was an OSR adapter generated?
1664       // O0 = osr nmethod
1665       __ br_null_short(O0, Assembler::pn, Lforward);
1666 
1667       // Has the nmethod been invalidated already?
1668       __ ldub(O0, nmethod::state_offset(), O2);
1669       __ cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, Lforward);
1670 
1671       // migrate the interpreter frame off of the stack
1672 
1673       __ mov(G2_thread, L7);
1674       // save nmethod
1675       __ mov(O0, L6);
1676       __ set_last_Java_frame(SP, noreg);
1677       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
1678       __ reset_last_Java_frame();
1679       __ mov(L7, G2_thread);
1680 
1681       // move OSR nmethod to I1
1682       __ mov(L6, I1);
1683 
1684       // OSR buffer to I0
1685       __ mov(O0, I0);
1686 
1687       // remove the interpreter frame
1688       __ restore(I5_savedSP, 0, SP);
1689 
1690       // Jump to the osr code.
1691       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
1692       __ jmp(O2, G0);
1693       __ delayed()->nop();
1694 
1695     } else { // not TieredCompilation
1696       // Update Backedge branch separately from invocations
1697       const Register G4_invoke_ctr = G4;
1698       __ increment_backedge_counter(G3_method_counters, G4_invoke_ctr, G1_scratch);
1699       if (ProfileInterpreter) {
1700         __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_method_counters, G1_scratch, Lforward);
1701         if (UseOnStackReplacement) {
1702 
1703           __ test_backedge_count_for_osr(O2_bumped_count, G3_method_counters, l_cur_bcp, G1_scratch);
1704         }
1705       } else {
1706         if (UseOnStackReplacement) {
1707           __ test_backedge_count_for_osr(G4_invoke_ctr, G3_method_counters, l_cur_bcp, G1_scratch);
1708         }
1709       }
1710     }
1711 
1712     __ bind(Lforward);
1713   } else
1714     // Bump bytecode pointer by displacement (take the branch)
1715     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
1716 
1717   // continue with bytecode @ target
1718   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1719   // %%%%% and changing dispatch_next to dispatch_only
1720   __ dispatch_next(vtos, 0, true);
1721 }
1722 
1723 
1724 // Note Condition in argument is TemplateTable::Condition
1725 // arg scope is within class scope
1726 
1727 void TemplateTable::if_0cmp(Condition cc) {
1728   // no pointers, integer only!
1729   transition(itos, vtos);
1730   // assume branch is more often taken than not (loops use backward branches)
1731   __ cmp( Otos_i, 0);
1732   __ if_cmp(ccNot(cc), false);
1733 }
1734 
1735 
1736 void TemplateTable::if_icmp(Condition cc) {
1737   transition(itos, vtos);
1738   __ pop_i(O1);
1739   __ cmp(O1, Otos_i);
1740   __ if_cmp(ccNot(cc), false);
1741 }
1742 
1743 
1744 void TemplateTable::if_nullcmp(Condition cc) {
1745   transition(atos, vtos);
1746   __ tst(Otos_i);
1747   __ if_cmp(ccNot(cc), true);
1748 }
1749 
1750 
1751 void TemplateTable::if_acmp(Condition cc) {
1752   transition(atos, vtos);
1753   __ pop_ptr(O1);
1754   __ verify_oop(O1);
1755   __ verify_oop(Otos_i);
1756   __ cmp(O1, Otos_i);
1757   __ if_cmp(ccNot(cc), true);
1758 }
1759 
1760 
1761 
1762 void TemplateTable::ret() {
1763   transition(vtos, vtos);
1764   locals_index(G3_scratch);
1765   __ access_local_returnAddress(G3_scratch, Otos_i);
1766   // Otos_i contains the bci, compute the bcp from that
1767 
1768 #ifdef ASSERT
1769   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
1770   // the result.  The return address (really a BCI) was stored with an
1771   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
1772   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
1773   // loaded value.
1774   { Label zzz ;
1775      __ set (65536, G3_scratch) ;
1776      __ cmp (Otos_i, G3_scratch) ;
1777      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
1778      __ delayed()->nop();
1779      __ stop("BCI is in the wrong register half?");
1780      __ bind (zzz) ;
1781   }
1782 #endif
1783 
1784   __ profile_ret(vtos, Otos_i, G4_scratch);
1785 
1786   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1787   __ add(G3_scratch, Otos_i, G3_scratch);
1788   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1789   __ dispatch_next(vtos, 0, true);
1790 }
1791 
1792 
1793 void TemplateTable::wide_ret() {
1794   transition(vtos, vtos);
1795   locals_index_wide(G3_scratch);
1796   __ access_local_returnAddress(G3_scratch, Otos_i);
1797   // Otos_i contains the bci, compute the bcp from that
1798 
1799   __ profile_ret(vtos, Otos_i, G4_scratch);
1800 
1801   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1802   __ add(G3_scratch, Otos_i, G3_scratch);
1803   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1804   __ dispatch_next(vtos, 0, true);
1805 }
1806 
1807 
1808 void TemplateTable::tableswitch() {
1809   transition(itos, vtos);
1810   Label default_case, continue_execution;
1811 
1812   // align bcp
1813   __ add(Lbcp, BytesPerInt, O1);
1814   __ and3(O1, -BytesPerInt, O1);
1815   // load lo, hi
1816   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
1817   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
1818   // Sign extend the 32 bits
1819   __ sra ( Otos_i, 0, Otos_i );
1820 
1821   // check against lo & hi
1822   __ cmp( Otos_i, O2);
1823   __ br( Assembler::less, false, Assembler::pn, default_case);
1824   __ delayed()->cmp( Otos_i, O3 );
1825   __ br( Assembler::greater, false, Assembler::pn, default_case);
1826   // lookup dispatch offset
1827   __ delayed()->sub(Otos_i, O2, O2);
1828   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
1829   __ sll(O2, LogBytesPerInt, O2);
1830   __ add(O2, 3 * BytesPerInt, O2);
1831   __ ba(continue_execution);
1832   __ delayed()->ld(O1, O2, O2);
1833   // handle default
1834   __ bind(default_case);
1835   __ profile_switch_default(O3);
1836   __ ld(O1, 0, O2); // get default offset
1837   // continue execution
1838   __ bind(continue_execution);
1839   __ add(Lbcp, O2, Lbcp);
1840   __ dispatch_next(vtos, 0, true);
1841 }
1842 
1843 
1844 void TemplateTable::lookupswitch() {
1845   transition(itos, itos);
1846   __ stop("lookupswitch bytecode should have been rewritten");
1847 }
1848 
1849 void TemplateTable::fast_linearswitch() {
1850   transition(itos, vtos);
1851     Label loop_entry, loop, found, continue_execution;
1852   // align bcp
1853   __ add(Lbcp, BytesPerInt, O1);
1854   __ and3(O1, -BytesPerInt, O1);
1855  // set counter
1856   __ ld(O1, BytesPerInt, O2);
1857   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
1858   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
1859   __ ba(loop_entry);
1860   __ delayed()->add(O3, O2, O2); // counter now points past last pair
1861 
1862   // table search
1863   __ bind(loop);
1864   __ cmp(O4, Otos_i);
1865   __ br(Assembler::equal, true, Assembler::pn, found);
1866   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
1867   __ inc(O3, 2 * BytesPerInt);
1868 
1869   __ bind(loop_entry);
1870   __ cmp(O2, O3);
1871   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
1872   __ delayed()->ld(O3, 0, O4);
1873 
1874   // default case
1875   __ ld(O1, 0, O4); // get default offset
1876   if (ProfileInterpreter) {
1877     __ profile_switch_default(O3);
1878     __ ba_short(continue_execution);
1879   }
1880 
1881   // entry found -> get offset
1882   __ bind(found);
1883   if (ProfileInterpreter) {
1884     __ sub(O3, O1, O3);
1885     __ sub(O3, 2*BytesPerInt, O3);
1886     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
1887     __ profile_switch_case(O3, O1, O2, G3_scratch);
1888 
1889     __ bind(continue_execution);
1890   }
1891   __ add(Lbcp, O4, Lbcp);
1892   __ dispatch_next(vtos, 0, true);
1893 }
1894 
1895 
1896 void TemplateTable::fast_binaryswitch() {
1897   transition(itos, vtos);
1898   // Implementation using the following core algorithm: (copied from Intel)
1899   //
1900   // int binary_search(int key, LookupswitchPair* array, int n) {
1901   //   // Binary search according to "Methodik des Programmierens" by
1902   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1903   //   int i = 0;
1904   //   int j = n;
1905   //   while (i+1 < j) {
1906   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1907   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1908   //     // where a stands for the array and assuming that the (inexisting)
1909   //     // element a[n] is infinitely big.
1910   //     int h = (i + j) >> 1;
1911   //     // i < h < j
1912   //     if (key < array[h].fast_match()) {
1913   //       j = h;
1914   //     } else {
1915   //       i = h;
1916   //     }
1917   //   }
1918   //   // R: a[i] <= key < a[i+1] or Q
1919   //   // (i.e., if key is within array, i is the correct index)
1920   //   return i;
1921   // }
1922 
1923   // register allocation
1924   assert(Otos_i == O0, "alias checking");
1925   const Register Rkey     = Otos_i;                    // already set (tosca)
1926   const Register Rarray   = O1;
1927   const Register Ri       = O2;
1928   const Register Rj       = O3;
1929   const Register Rh       = O4;
1930   const Register Rscratch = O5;
1931 
1932   const int log_entry_size = 3;
1933   const int entry_size = 1 << log_entry_size;
1934 
1935   Label found;
1936   // Find Array start
1937   __ add(Lbcp, 3 * BytesPerInt, Rarray);
1938   __ and3(Rarray, -BytesPerInt, Rarray);
1939   // initialize i & j (in delay slot)
1940   __ clr( Ri );
1941 
1942   // and start
1943   Label entry;
1944   __ ba(entry);
1945   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
1946   // (Rj is already in the native byte-ordering.)
1947 
1948   // binary search loop
1949   { Label loop;
1950     __ bind( loop );
1951     // int h = (i + j) >> 1;
1952     __ sra( Rh, 1, Rh );
1953     // if (key < array[h].fast_match()) {
1954     //   j = h;
1955     // } else {
1956     //   i = h;
1957     // }
1958     __ sll( Rh, log_entry_size, Rscratch );
1959     __ ld( Rarray, Rscratch, Rscratch );
1960     // (Rscratch is already in the native byte-ordering.)
1961     __ cmp( Rkey, Rscratch );
1962     __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
1963     __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
1964 
1965     // while (i+1 < j)
1966     __ bind( entry );
1967     __ add( Ri, 1, Rscratch );
1968     __ cmp(Rscratch, Rj);
1969     __ br( Assembler::less, true, Assembler::pt, loop );
1970     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
1971   }
1972 
1973   // end of binary search, result index is i (must check again!)
1974   Label default_case;
1975   Label continue_execution;
1976   if (ProfileInterpreter) {
1977     __ mov( Ri, Rh );              // Save index in i for profiling
1978   }
1979   __ sll( Ri, log_entry_size, Ri );
1980   __ ld( Rarray, Ri, Rscratch );
1981   // (Rscratch is already in the native byte-ordering.)
1982   __ cmp( Rkey, Rscratch );
1983   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
1984   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
1985 
1986   // entry found -> j = offset
1987   __ inc( Ri, BytesPerInt );
1988   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
1989   __ ld( Rarray, Ri, Rj );
1990   // (Rj is already in the native byte-ordering.)
1991 
1992   if (ProfileInterpreter) {
1993     __ ba_short(continue_execution);
1994   }
1995 
1996   __ bind(default_case); // fall through (if not profiling)
1997   __ profile_switch_default(Ri);
1998 
1999   __ bind(continue_execution);
2000   __ add( Lbcp, Rj, Lbcp );
2001   __ dispatch_next(vtos, 0, true);
2002 }
2003 
2004 
2005 void TemplateTable::_return(TosState state) {
2006   transition(state, state);
2007   assert(_desc->calls_vm(), "inconsistent calls_vm information");
2008 
2009   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2010     assert(state == vtos, "only valid state");
2011     __ mov(G0, G3_scratch);
2012     __ access_local_ptr(G3_scratch, Otos_i);
2013     __ load_klass(Otos_i, O2);
2014     __ set(JVM_ACC_HAS_FINALIZER, G3);
2015     __ ld(O2, in_bytes(Klass::access_flags_offset()), O2);
2016     __ andcc(G3, O2, G0);
2017     Label skip_register_finalizer;
2018     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
2019     __ delayed()->nop();
2020 
2021     // Call out to do finalizer registration
2022     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
2023 
2024     __ bind(skip_register_finalizer);
2025   }
2026 
2027   if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2028     Label no_safepoint;
2029     __ ldx(Address(G2_thread, Thread::polling_page_offset()), G3_scratch, 0);
2030     __ btst(SafepointMechanism::poll_bit(), G3_scratch);
2031     __ br(Assembler::zero, false, Assembler::pt, no_safepoint);
2032     __ delayed()->nop();
2033     __ push(state);
2034     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint));
2035     __ pop(state);
2036     __ bind(no_safepoint);
2037   }
2038 
2039   // Narrow result if state is itos but result type is smaller.
2040   // Need to narrow in the return bytecode rather than in generate_return_entry
2041   // since compiled code callers expect the result to already be narrowed.
2042   if (state == itos) {
2043     __ narrow(Otos_i);
2044   }
2045   __ remove_activation(state, /* throw_monitor_exception */ true);
2046 
2047   // The caller's SP was adjusted upon method entry to accomodate
2048   // the callee's non-argument locals. Undo that adjustment.
2049   __ ret();                             // return to caller
2050   __ delayed()->restore(I5_savedSP, G0, SP);
2051 }
2052 
2053 
2054 // ----------------------------------------------------------------------------
2055 // Volatile variables demand their effects be made known to all CPU's in
2056 // order.  Store buffers on most chips allow reads & writes to reorder; the
2057 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2058 // memory barrier (i.e., it's not sufficient that the interpreter does not
2059 // reorder volatile references, the hardware also must not reorder them).
2060 //
2061 // According to the new Java Memory Model (JMM):
2062 // (1) All volatiles are serialized wrt to each other.
2063 // ALSO reads & writes act as aquire & release, so:
2064 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2065 // the read float up to before the read.  It's OK for non-volatile memory refs
2066 // that happen before the volatile read to float down below it.
2067 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2068 // that happen BEFORE the write float down to after the write.  It's OK for
2069 // non-volatile memory refs that happen after the volatile write to float up
2070 // before it.
2071 //
2072 // We only put in barriers around volatile refs (they are expensive), not
2073 // _between_ memory refs (that would require us to track the flavor of the
2074 // previous memory refs).  Requirements (2) and (3) require some barriers
2075 // before volatile stores and after volatile loads.  These nearly cover
2076 // requirement (1) but miss the volatile-store-volatile-load case.  This final
2077 // case is placed after volatile-stores although it could just as well go
2078 // before volatile-loads.
2079 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
2080   // Helper function to insert a is-volatile test and memory barrier
2081   // All current sparc implementations run in TSO, needing only StoreLoad
2082   if ((order_constraint & Assembler::StoreLoad) == 0) return;
2083   __ membar( order_constraint );
2084 }
2085 
2086 // ----------------------------------------------------------------------------
2087 void TemplateTable::resolve_cache_and_index(int byte_no,
2088                                             Register Rcache,
2089                                             Register index,
2090                                             size_t index_size) {
2091   // Depends on cpCacheOop layout!
2092 
2093   Label resolved;
2094   Bytecodes::Code code = bytecode();
2095   switch (code) {
2096   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2097   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2098   }
2099 
2100   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2101   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, Lbyte_code, byte_no, 1, index_size);
2102   __ cmp(Lbyte_code, code);  // have we resolved this bytecode?
2103   __ br(Assembler::equal, false, Assembler::pt, resolved);
2104   __ delayed()->set(code, O1);
2105 
2106   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2107   // first time invocation - must resolve first
2108   __ call_VM(noreg, entry, O1);
2109   // Update registers with resolved info
2110   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2111   __ bind(resolved);
2112 }
2113 
2114 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2115                                                Register method,
2116                                                Register itable_index,
2117                                                Register flags,
2118                                                bool is_invokevirtual,
2119                                                bool is_invokevfinal,
2120                                                bool is_invokedynamic) {
2121   // Uses both G3_scratch and G4_scratch
2122   Register cache = G3_scratch;
2123   Register index = G4_scratch;
2124   assert_different_registers(cache, method, itable_index);
2125 
2126   // determine constant pool cache field offsets
2127   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2128   const int method_offset = in_bytes(
2129       ConstantPoolCache::base_offset() +
2130       ((byte_no == f2_byte)
2131        ? ConstantPoolCacheEntry::f2_offset()
2132        : ConstantPoolCacheEntry::f1_offset()
2133       )
2134     );
2135   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2136                                     ConstantPoolCacheEntry::flags_offset());
2137   // access constant pool cache fields
2138   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2139                                     ConstantPoolCacheEntry::f2_offset());
2140 
2141   if (is_invokevfinal) {
2142     __ get_cache_and_index_at_bcp(cache, index, 1);
2143     __ ld_ptr(Address(cache, method_offset), method);
2144   } else {
2145     size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2146     resolve_cache_and_index(byte_no, cache, index, index_size);
2147     __ ld_ptr(Address(cache, method_offset), method);
2148   }
2149 
2150   if (itable_index != noreg) {
2151     // pick up itable or appendix index from f2 also:
2152     __ ld_ptr(Address(cache, index_offset), itable_index);
2153   }
2154   __ ld_ptr(Address(cache, flags_offset), flags);
2155 }
2156 
2157 // The Rcache register must be set before call
2158 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2159                                               Register Rcache,
2160                                               Register index,
2161                                               Register Roffset,
2162                                               Register Rflags,
2163                                               bool is_static) {
2164   assert_different_registers(Rcache, Rflags, Roffset, Lscratch);
2165 
2166   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2167 
2168   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2169   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2170   if (is_static) {
2171     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
2172     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2173     __ ld_ptr( Robj, mirror_offset, Robj);
2174     __ resolve_oop_handle(Robj, Lscratch);
2175   }
2176 }
2177 
2178 // The registers Rcache and index expected to be set before call.
2179 // Correct values of the Rcache and index registers are preserved.
2180 void TemplateTable::jvmti_post_field_access(Register Rcache,
2181                                             Register index,
2182                                             bool is_static,
2183                                             bool has_tos) {
2184   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2185 
2186   if (JvmtiExport::can_post_field_access()) {
2187     // Check to see if a field access watch has been set before we take
2188     // the time to call into the VM.
2189     Label Label1;
2190     assert_different_registers(Rcache, index, G1_scratch);
2191     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
2192     __ load_contents(get_field_access_count_addr, G1_scratch);
2193     __ cmp_and_br_short(G1_scratch, 0, Assembler::equal, Assembler::pt, Label1);
2194 
2195     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
2196 
2197     if (is_static) {
2198       __ clr(Otos_i);
2199     } else {
2200       if (has_tos) {
2201       // save object pointer before call_VM() clobbers it
2202         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
2203       } else {
2204         // Load top of stack (do not pop the value off the stack);
2205         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
2206       }
2207       __ verify_oop(Otos_i);
2208     }
2209     // Otos_i: object pointer or NULL if static
2210     // Rcache: cache entry pointer
2211     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2212                Otos_i, Rcache);
2213     if (!is_static && has_tos) {
2214       __ pop_ptr(Otos_i);  // restore object pointer
2215       __ verify_oop(Otos_i);
2216     }
2217     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2218     __ bind(Label1);
2219   }
2220 }
2221 
2222 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2223   transition(vtos, vtos);
2224 
2225   Register Rcache = G3_scratch;
2226   Register index  = G4_scratch;
2227   Register Rclass = Rcache;
2228   Register Roffset= G4_scratch;
2229   Register Rflags = G1_scratch;
2230   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2231 
2232   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2233   jvmti_post_field_access(Rcache, index, is_static, false);
2234   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2235 
2236   if (!is_static) {
2237     pop_and_check_object(Rclass);
2238   } else {
2239     __ verify_oop(Rclass);
2240   }
2241 
2242   Label exit;
2243 
2244   Assembler::Membar_mask_bits membar_bits =
2245     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2246 
2247   if (__ membar_has_effect(membar_bits)) {
2248     // Get volatile flag
2249     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2250     __ and3(Rflags, Lscratch, Lscratch);
2251   }
2252 
2253   Label checkVolatile;
2254 
2255   // compute field type
2256   Label notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
2257   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2258   // Make sure we don't need to mask Rflags after the above shift
2259   ConstantPoolCacheEntry::verify_tos_state_shift();
2260 
2261   // Check atos before itos for getstatic, more likely (in Queens at least)
2262   __ cmp(Rflags, atos);
2263   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2264   __ delayed() ->cmp(Rflags, itos);
2265 
2266   // atos
2267   do_oop_load(_masm, Rclass, Roffset, 0, Otos_i, noreg);
2268   __ verify_oop(Otos_i);
2269   __ push(atos);
2270   if (!is_static && rc == may_rewrite) {
2271     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
2272   }
2273   __ ba(checkVolatile);
2274   __ delayed()->tst(Lscratch);
2275 
2276   __ bind(notObj);
2277 
2278   // cmp(Rflags, itos);
2279   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2280   __ delayed() ->cmp(Rflags, ltos);
2281 
2282   // itos
2283   __ ld(Rclass, Roffset, Otos_i);
2284   __ push(itos);
2285   if (!is_static && rc == may_rewrite) {
2286     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
2287   }
2288   __ ba(checkVolatile);
2289   __ delayed()->tst(Lscratch);
2290 
2291   __ bind(notInt);
2292 
2293   // cmp(Rflags, ltos);
2294   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2295   __ delayed() ->cmp(Rflags, btos);
2296 
2297   // ltos
2298   // load must be atomic
2299   __ ld_long(Rclass, Roffset, Otos_l);
2300   __ push(ltos);
2301   if (!is_static && rc == may_rewrite) {
2302     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
2303   }
2304   __ ba(checkVolatile);
2305   __ delayed()->tst(Lscratch);
2306 
2307   __ bind(notLong);
2308 
2309   // cmp(Rflags, btos);
2310   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2311   __ delayed() ->cmp(Rflags, ztos);
2312 
2313   // btos
2314   __ ldsb(Rclass, Roffset, Otos_i);
2315   __ push(itos);
2316   if (!is_static && rc == may_rewrite) {
2317     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
2318   }
2319   __ ba(checkVolatile);
2320   __ delayed()->tst(Lscratch);
2321 
2322   __ bind(notByte);
2323 
2324   // cmp(Rflags, ztos);
2325   __ br(Assembler::notEqual, false, Assembler::pt, notBool);
2326   __ delayed() ->cmp(Rflags, ctos);
2327 
2328   // ztos
2329   __ ldsb(Rclass, Roffset, Otos_i);
2330   __ push(itos);
2331   if (!is_static && rc == may_rewrite) {
2332     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2333     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
2334   }
2335   __ ba(checkVolatile);
2336   __ delayed()->tst(Lscratch);
2337 
2338   __ bind(notBool);
2339 
2340   // cmp(Rflags, ctos);
2341   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2342   __ delayed() ->cmp(Rflags, stos);
2343 
2344   // ctos
2345   __ lduh(Rclass, Roffset, Otos_i);
2346   __ push(itos);
2347   if (!is_static && rc == may_rewrite) {
2348     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
2349   }
2350   __ ba(checkVolatile);
2351   __ delayed()->tst(Lscratch);
2352 
2353   __ bind(notChar);
2354 
2355   // cmp(Rflags, stos);
2356   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2357   __ delayed() ->cmp(Rflags, ftos);
2358 
2359   // stos
2360   __ ldsh(Rclass, Roffset, Otos_i);
2361   __ push(itos);
2362   if (!is_static && rc == may_rewrite) {
2363     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
2364   }
2365   __ ba(checkVolatile);
2366   __ delayed()->tst(Lscratch);
2367 
2368   __ bind(notShort);
2369 
2370 
2371   // cmp(Rflags, ftos);
2372   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
2373   __ delayed() ->tst(Lscratch);
2374 
2375   // ftos
2376   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
2377   __ push(ftos);
2378   if (!is_static && rc == may_rewrite) {
2379     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
2380   }
2381   __ ba(checkVolatile);
2382   __ delayed()->tst(Lscratch);
2383 
2384   __ bind(notFloat);
2385 
2386 
2387   // dtos
2388   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
2389   __ push(dtos);
2390   if (!is_static && rc == may_rewrite) {
2391     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
2392   }
2393 
2394   __ bind(checkVolatile);
2395   if (__ membar_has_effect(membar_bits)) {
2396     // __ tst(Lscratch); executed in delay slot
2397     __ br(Assembler::zero, false, Assembler::pt, exit);
2398     __ delayed()->nop();
2399     volatile_barrier(membar_bits);
2400   }
2401 
2402   __ bind(exit);
2403 }
2404 
2405 void TemplateTable::getfield(int byte_no) {
2406   getfield_or_static(byte_no, false);
2407 }
2408 
2409 void TemplateTable::nofast_getfield(int byte_no) {
2410   getfield_or_static(byte_no, false, may_not_rewrite);
2411 }
2412 
2413 void TemplateTable::getstatic(int byte_no) {
2414   getfield_or_static(byte_no, true);
2415 }
2416 
2417 void TemplateTable::fast_accessfield(TosState state) {
2418   transition(atos, state);
2419   Register Rcache  = G3_scratch;
2420   Register index   = G4_scratch;
2421   Register Roffset = G4_scratch;
2422   Register Rflags  = Rcache;
2423   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2424 
2425   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2426   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
2427 
2428   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2429 
2430   __ null_check(Otos_i);
2431   __ verify_oop(Otos_i);
2432 
2433   Label exit;
2434 
2435   Assembler::Membar_mask_bits membar_bits =
2436     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2437   if (__ membar_has_effect(membar_bits)) {
2438     // Get volatile flag
2439     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
2440     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2441   }
2442 
2443   switch (bytecode()) {
2444     case Bytecodes::_fast_bgetfield:
2445       __ ldsb(Otos_i, Roffset, Otos_i);
2446       break;
2447     case Bytecodes::_fast_cgetfield:
2448       __ lduh(Otos_i, Roffset, Otos_i);
2449       break;
2450     case Bytecodes::_fast_sgetfield:
2451       __ ldsh(Otos_i, Roffset, Otos_i);
2452       break;
2453     case Bytecodes::_fast_igetfield:
2454       __ ld(Otos_i, Roffset, Otos_i);
2455       break;
2456     case Bytecodes::_fast_lgetfield:
2457       __ ld_long(Otos_i, Roffset, Otos_l);
2458       break;
2459     case Bytecodes::_fast_fgetfield:
2460       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
2461       break;
2462     case Bytecodes::_fast_dgetfield:
2463       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
2464       break;
2465     case Bytecodes::_fast_agetfield:
2466       do_oop_load(_masm, Otos_i, Roffset, 0, Otos_i, noreg);
2467       break;
2468     default:
2469       ShouldNotReachHere();
2470   }
2471 
2472   if (__ membar_has_effect(membar_bits)) {
2473     __ btst(Lscratch, Rflags);
2474     __ br(Assembler::zero, false, Assembler::pt, exit);
2475     __ delayed()->nop();
2476     volatile_barrier(membar_bits);
2477     __ bind(exit);
2478   }
2479 
2480   if (state == atos) {
2481     __ verify_oop(Otos_i);    // does not blow flags!
2482   }
2483 }
2484 
2485 void TemplateTable::jvmti_post_fast_field_mod() {
2486   if (JvmtiExport::can_post_field_modification()) {
2487     // Check to see if a field modification watch has been set before we take
2488     // the time to call into the VM.
2489     Label done;
2490     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2491     __ load_contents(get_field_modification_count_addr, G4_scratch);
2492     __ cmp_and_br_short(G4_scratch, 0, Assembler::equal, Assembler::pt, done);
2493     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
2494     __ verify_oop(G4_scratch);
2495     __ push_ptr(G4_scratch);    // put the object pointer back on tos
2496     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
2497     // Save tos values before call_VM() clobbers them. Since we have
2498     // to do it for every data type, we use the saved values as the
2499     // jvalue object.
2500     switch (bytecode()) {  // save tos values before call_VM() clobbers them
2501     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
2502     case Bytecodes::_fast_bputfield: // fall through
2503     case Bytecodes::_fast_zputfield: // fall through
2504     case Bytecodes::_fast_sputfield: // fall through
2505     case Bytecodes::_fast_cputfield: // fall through
2506     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
2507     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
2508     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
2509     // get words in right order for use as jvalue object
2510     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
2511     }
2512     // setup pointer to jvalue object
2513     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
2514     // G4_scratch:  object pointer
2515     // G1_scratch: cache entry pointer
2516     // G3_scratch: jvalue object on the stack
2517     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
2518     switch (bytecode()) {             // restore tos values
2519     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
2520     case Bytecodes::_fast_bputfield: // fall through
2521     case Bytecodes::_fast_zputfield: // fall through
2522     case Bytecodes::_fast_sputfield: // fall through
2523     case Bytecodes::_fast_cputfield: // fall through
2524     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
2525     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
2526     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
2527     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
2528     }
2529     __ bind(done);
2530   }
2531 }
2532 
2533 // The registers Rcache and index expected to be set before call.
2534 // The function may destroy various registers, just not the Rcache and index registers.
2535 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
2536   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2537 
2538   if (JvmtiExport::can_post_field_modification()) {
2539     // Check to see if a field modification watch has been set before we take
2540     // the time to call into the VM.
2541     Label Label1;
2542     assert_different_registers(Rcache, index, G1_scratch);
2543     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2544     __ load_contents(get_field_modification_count_addr, G1_scratch);
2545     __ cmp_and_br_short(G1_scratch, 0, Assembler::zero, Assembler::pt, Label1);
2546 
2547     // The Rcache and index registers have been already set.
2548     // This allows to eliminate this call but the Rcache and index
2549     // registers must be correspondingly used after this line.
2550     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
2551 
2552     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
2553     if (is_static) {
2554       // Life is simple.  Null out the object pointer.
2555       __ clr(G4_scratch);
2556     } else {
2557       Register Rflags = G1_scratch;
2558       // Life is harder. The stack holds the value on top, followed by the
2559       // object.  We don't know the size of the value, though; it could be
2560       // one or two words depending on its type. As a result, we must find
2561       // the type to determine where the object is.
2562 
2563       Label two_word, valsizeknown;
2564       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2565       __ mov(Lesp, G4_scratch);
2566       __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2567       // Make sure we don't need to mask Rflags after the above shift
2568       ConstantPoolCacheEntry::verify_tos_state_shift();
2569       __ cmp(Rflags, ltos);
2570       __ br(Assembler::equal, false, Assembler::pt, two_word);
2571       __ delayed()->cmp(Rflags, dtos);
2572       __ br(Assembler::equal, false, Assembler::pt, two_word);
2573       __ delayed()->nop();
2574       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
2575       __ ba_short(valsizeknown);
2576       __ bind(two_word);
2577 
2578       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
2579 
2580       __ bind(valsizeknown);
2581       // setup object pointer
2582       __ ld_ptr(G4_scratch, 0, G4_scratch);
2583       __ verify_oop(G4_scratch);
2584     }
2585     // setup pointer to jvalue object
2586     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
2587     // G4_scratch:  object pointer or NULL if static
2588     // G3_scratch: cache entry pointer
2589     // G1_scratch: jvalue object on the stack
2590     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2591                G4_scratch, G3_scratch, G1_scratch);
2592     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2593     __ bind(Label1);
2594   }
2595 }
2596 
2597 void TemplateTable::pop_and_check_object(Register r) {
2598   __ pop_ptr(r);
2599   __ null_check(r);  // for field access must check obj.
2600   __ verify_oop(r);
2601 }
2602 
2603 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2604   transition(vtos, vtos);
2605   Register Rcache = G3_scratch;
2606   Register index  = G4_scratch;
2607   Register Rclass = Rcache;
2608   Register Roffset= G4_scratch;
2609   Register Rflags = G1_scratch;
2610   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2611 
2612   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2613   jvmti_post_field_mod(Rcache, index, is_static);
2614   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2615 
2616   Assembler::Membar_mask_bits read_bits =
2617     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2618   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2619 
2620   Label notVolatile, checkVolatile, exit;
2621   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2622     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2623     __ and3(Rflags, Lscratch, Lscratch);
2624 
2625     if (__ membar_has_effect(read_bits)) {
2626       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2627       volatile_barrier(read_bits);
2628       __ bind(notVolatile);
2629     }
2630   }
2631 
2632   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2633   // Make sure we don't need to mask Rflags after the above shift
2634   ConstantPoolCacheEntry::verify_tos_state_shift();
2635 
2636   // compute field type
2637   Label notInt, notShort, notChar, notObj, notByte, notBool, notLong, notFloat;
2638 
2639   if (is_static) {
2640     // putstatic with object type most likely, check that first
2641     __ cmp(Rflags, atos);
2642     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2643     __ delayed()->cmp(Rflags, itos);
2644 
2645     // atos
2646     {
2647       __ pop_ptr();
2648       __ verify_oop(Otos_i);
2649       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch);
2650       __ ba(checkVolatile);
2651       __ delayed()->tst(Lscratch);
2652     }
2653 
2654     __ bind(notObj);
2655     // cmp(Rflags, itos);
2656     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2657     __ delayed()->cmp(Rflags, btos);
2658 
2659     // itos
2660     {
2661       __ pop_i();
2662       __ st(Otos_i, Rclass, Roffset);
2663       __ ba(checkVolatile);
2664       __ delayed()->tst(Lscratch);
2665     }
2666 
2667     __ bind(notInt);
2668   } else {
2669     // putfield with int type most likely, check that first
2670     __ cmp(Rflags, itos);
2671     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2672     __ delayed()->cmp(Rflags, atos);
2673 
2674     // itos
2675     {
2676       __ pop_i();
2677       pop_and_check_object(Rclass);
2678       __ st(Otos_i, Rclass, Roffset);
2679       if (rc == may_rewrite) patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch, true, byte_no);
2680       __ ba(checkVolatile);
2681       __ delayed()->tst(Lscratch);
2682     }
2683 
2684     __ bind(notInt);
2685     // cmp(Rflags, atos);
2686     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2687     __ delayed()->cmp(Rflags, btos);
2688 
2689     // atos
2690     {
2691       __ pop_ptr();
2692       pop_and_check_object(Rclass);
2693       __ verify_oop(Otos_i);
2694       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch);
2695       if (rc == may_rewrite) patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch, true, byte_no);
2696       __ ba(checkVolatile);
2697       __ delayed()->tst(Lscratch);
2698     }
2699 
2700     __ bind(notObj);
2701   }
2702 
2703   // cmp(Rflags, btos);
2704   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2705   __ delayed()->cmp(Rflags, ztos);
2706 
2707   // btos
2708   {
2709     __ pop_i();
2710     if (!is_static) pop_and_check_object(Rclass);
2711     __ stb(Otos_i, Rclass, Roffset);
2712     if (!is_static && rc == may_rewrite) {
2713       patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch, true, byte_no);
2714     }
2715     __ ba(checkVolatile);
2716     __ delayed()->tst(Lscratch);
2717   }
2718 
2719   __ bind(notByte);
2720 
2721   // cmp(Rflags, btos);
2722   __ br(Assembler::notEqual, false, Assembler::pt, notBool);
2723   __ delayed()->cmp(Rflags, ltos);
2724 
2725   // ztos
2726   {
2727     __ pop_i();
2728     if (!is_static) pop_and_check_object(Rclass);
2729     __ and3(Otos_i, 1, Otos_i);
2730     __ stb(Otos_i, Rclass, Roffset);
2731     if (!is_static && rc == may_rewrite) {
2732       patch_bytecode(Bytecodes::_fast_zputfield, G3_scratch, G4_scratch, true, byte_no);
2733     }
2734     __ ba(checkVolatile);
2735     __ delayed()->tst(Lscratch);
2736   }
2737 
2738   __ bind(notBool);
2739   // cmp(Rflags, ltos);
2740   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2741   __ delayed()->cmp(Rflags, ctos);
2742 
2743   // ltos
2744   {
2745     __ pop_l();
2746     if (!is_static) pop_and_check_object(Rclass);
2747     __ st_long(Otos_l, Rclass, Roffset);
2748     if (!is_static && rc == may_rewrite) {
2749       patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch, true, byte_no);
2750     }
2751     __ ba(checkVolatile);
2752     __ delayed()->tst(Lscratch);
2753   }
2754 
2755   __ bind(notLong);
2756   // cmp(Rflags, ctos);
2757   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2758   __ delayed()->cmp(Rflags, stos);
2759 
2760   // ctos (char)
2761   {
2762     __ pop_i();
2763     if (!is_static) pop_and_check_object(Rclass);
2764     __ sth(Otos_i, Rclass, Roffset);
2765     if (!is_static && rc == may_rewrite) {
2766       patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch, true, byte_no);
2767     }
2768     __ ba(checkVolatile);
2769     __ delayed()->tst(Lscratch);
2770   }
2771 
2772   __ bind(notChar);
2773   // cmp(Rflags, stos);
2774   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2775   __ delayed()->cmp(Rflags, ftos);
2776 
2777   // stos (short)
2778   {
2779     __ pop_i();
2780     if (!is_static) pop_and_check_object(Rclass);
2781     __ sth(Otos_i, Rclass, Roffset);
2782     if (!is_static && rc == may_rewrite) {
2783       patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch, true, byte_no);
2784     }
2785     __ ba(checkVolatile);
2786     __ delayed()->tst(Lscratch);
2787   }
2788 
2789   __ bind(notShort);
2790   // cmp(Rflags, ftos);
2791   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
2792   __ delayed()->nop();
2793 
2794   // ftos
2795   {
2796     __ pop_f();
2797     if (!is_static) pop_and_check_object(Rclass);
2798     __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2799     if (!is_static && rc == may_rewrite) {
2800       patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch, true, byte_no);
2801     }
2802     __ ba(checkVolatile);
2803     __ delayed()->tst(Lscratch);
2804   }
2805 
2806   __ bind(notFloat);
2807 
2808   // dtos
2809   {
2810     __ pop_d();
2811     if (!is_static) pop_and_check_object(Rclass);
2812     __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2813     if (!is_static && rc == may_rewrite) {
2814       patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch, true, byte_no);
2815     }
2816   }
2817 
2818   __ bind(checkVolatile);
2819   __ tst(Lscratch);
2820 
2821   if (__ membar_has_effect(write_bits)) {
2822     // __ tst(Lscratch); in delay slot
2823     __ br(Assembler::zero, false, Assembler::pt, exit);
2824     __ delayed()->nop();
2825     volatile_barrier(Assembler::StoreLoad);
2826     __ bind(exit);
2827   }
2828 }
2829 
2830 void TemplateTable::fast_storefield(TosState state) {
2831   transition(state, vtos);
2832   Register Rcache = G3_scratch;
2833   Register Rclass = Rcache;
2834   Register Roffset= G4_scratch;
2835   Register Rflags = G1_scratch;
2836   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2837 
2838   jvmti_post_fast_field_mod();
2839 
2840   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
2841 
2842   Assembler::Membar_mask_bits read_bits =
2843     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2844   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2845 
2846   Label notVolatile, checkVolatile, exit;
2847   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2848     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2849     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2850     __ and3(Rflags, Lscratch, Lscratch);
2851     if (__ membar_has_effect(read_bits)) {
2852       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2853       volatile_barrier(read_bits);
2854       __ bind(notVolatile);
2855     }
2856   }
2857 
2858   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2859   pop_and_check_object(Rclass);
2860 
2861   switch (bytecode()) {
2862     case Bytecodes::_fast_zputfield: __ and3(Otos_i, 1, Otos_i);  // fall through to bputfield
2863     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
2864     case Bytecodes::_fast_cputfield: /* fall through */
2865     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
2866     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
2867     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
2868     case Bytecodes::_fast_fputfield:
2869       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2870       break;
2871     case Bytecodes::_fast_dputfield:
2872       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2873       break;
2874     case Bytecodes::_fast_aputfield:
2875       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch);
2876       break;
2877     default:
2878       ShouldNotReachHere();
2879   }
2880 
2881   if (__ membar_has_effect(write_bits)) {
2882     __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, exit);
2883     volatile_barrier(Assembler::StoreLoad);
2884     __ bind(exit);
2885   }
2886 }
2887 
2888 void TemplateTable::putfield(int byte_no) {
2889   putfield_or_static(byte_no, false);
2890 }
2891 
2892 void TemplateTable::nofast_putfield(int byte_no) {
2893   putfield_or_static(byte_no, false, may_not_rewrite);
2894 }
2895 
2896 void TemplateTable::putstatic(int byte_no) {
2897   putfield_or_static(byte_no, true);
2898 }
2899 
2900 void TemplateTable::fast_xaccess(TosState state) {
2901   transition(vtos, state);
2902   Register Rcache = G3_scratch;
2903   Register Roffset = G4_scratch;
2904   Register Rflags  = G4_scratch;
2905   Register Rreceiver = Lscratch;
2906 
2907   __ ld_ptr(Llocals, 0, Rreceiver);
2908 
2909   // access constant pool cache  (is resolved)
2910   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
2911   __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
2912   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
2913 
2914   __ verify_oop(Rreceiver);
2915   __ null_check(Rreceiver);
2916   if (state == atos) {
2917     do_oop_load(_masm, Rreceiver, Roffset, 0, Otos_i, noreg);
2918   } else if (state == itos) {
2919     __ ld (Rreceiver, Roffset, Otos_i) ;
2920   } else if (state == ftos) {
2921     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
2922   } else {
2923     ShouldNotReachHere();
2924   }
2925 
2926   Assembler::Membar_mask_bits membar_bits =
2927     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2928   if (__ membar_has_effect(membar_bits)) {
2929 
2930     // Get is_volatile value in Rflags and check if membar is needed
2931     __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
2932 
2933     // Test volatile
2934     Label notVolatile;
2935     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2936     __ btst(Rflags, Lscratch);
2937     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
2938     __ delayed()->nop();
2939     volatile_barrier(membar_bits);
2940     __ bind(notVolatile);
2941   }
2942 
2943   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
2944   __ sub(Lbcp, 1, Lbcp);
2945 }
2946 
2947 //----------------------------------------------------------------------------------------------------
2948 // Calls
2949 
2950 void TemplateTable::count_calls(Register method, Register temp) {
2951   // implemented elsewhere
2952   ShouldNotReachHere();
2953 }
2954 
2955 void TemplateTable::prepare_invoke(int byte_no,
2956                                    Register method,  // linked method (or i-klass)
2957                                    Register ra,      // return address
2958                                    Register index,   // itable index, MethodType, etc.
2959                                    Register recv,    // if caller wants to see it
2960                                    Register flags    // if caller wants to test it
2961                                    ) {
2962   // determine flags
2963   const Bytecodes::Code code = bytecode();
2964   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
2965   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
2966   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
2967   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
2968   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
2969   const bool load_receiver       = (recv != noreg);
2970   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2971   assert(recv  == noreg || recv  == O0, "");
2972   assert(flags == noreg || flags == O1, "");
2973 
2974   // setup registers & access constant pool cache
2975   if (recv  == noreg)  recv  = O0;
2976   if (flags == noreg)  flags = O1;
2977   const Register temp = O2;
2978   assert_different_registers(method, ra, index, recv, flags, temp);
2979 
2980   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2981 
2982   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
2983 
2984   // maybe push appendix to arguments
2985   if (is_invokedynamic || is_invokehandle) {
2986     Label L_no_push;
2987     __ set((1 << ConstantPoolCacheEntry::has_appendix_shift), temp);
2988     __ btst(flags, temp);
2989     __ br(Assembler::zero, false, Assembler::pt, L_no_push);
2990     __ delayed()->nop();
2991     // Push the appendix as a trailing parameter.
2992     // This must be done before we get the receiver,
2993     // since the parameter_size includes it.
2994     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2995     __ load_resolved_reference_at_index(temp, index, /*tmp*/recv);
2996     __ verify_oop(temp);
2997     __ push_ptr(temp);  // push appendix (MethodType, CallSite, etc.)
2998     __ bind(L_no_push);
2999   }
3000 
3001   // load receiver if needed (after appendix is pushed so parameter size is correct)
3002   if (load_receiver) {
3003     __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, temp);  // get parameter size
3004     __ load_receiver(temp, recv);  //  __ argument_address uses Gargs but we need Lesp
3005     __ verify_oop(recv);
3006   }
3007 
3008   // compute return type
3009   __ srl(flags, ConstantPoolCacheEntry::tos_state_shift, ra);
3010   // Make sure we don't need to mask flags after the above shift
3011   ConstantPoolCacheEntry::verify_tos_state_shift();
3012   // load return address
3013   {
3014     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3015     AddressLiteral table(table_addr);
3016     __ set(table, temp);
3017     __ sll(ra, LogBytesPerWord, ra);
3018     __ ld_ptr(Address(temp, ra), ra);
3019   }
3020 }
3021 
3022 
3023 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
3024   Register Rtemp = G4_scratch;
3025   Register Rcall = Rindex;
3026   assert_different_registers(Rcall, G5_method, Gargs, Rret);
3027 
3028   // get target Method* & entry point
3029   __ lookup_virtual_method(Rrecv, Rindex, G5_method);
3030   __ profile_arguments_type(G5_method, Rcall, Gargs, true);
3031   __ profile_called_method(G5_method, Rtemp);
3032   __ call_from_interpreter(Rcall, Gargs, Rret);
3033 }
3034 
3035 void TemplateTable::invokevirtual(int byte_no) {
3036   transition(vtos, vtos);
3037   assert(byte_no == f2_byte, "use this argument");
3038 
3039   Register Rscratch = G3_scratch;
3040   Register Rtemp    = G4_scratch;
3041   Register Rret     = Lscratch;
3042   Register O0_recv  = O0;
3043   Label notFinal;
3044 
3045   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
3046   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3047 
3048   // Check for vfinal
3049   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), G4_scratch);
3050   __ btst(Rret, G4_scratch);
3051   __ br(Assembler::zero, false, Assembler::pt, notFinal);
3052   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
3053 
3054   if (RewriteBytecodes && !UseSharedSpaces && !DumpSharedSpaces) {
3055     patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
3056   }
3057 
3058   invokevfinal_helper(Rscratch, Rret);
3059 
3060   __ bind(notFinal);
3061 
3062   __ mov(G5_method, Rscratch);  // better scratch register
3063   __ load_receiver(G4_scratch, O0_recv);  // gets receiverOop
3064   // receiver is in O0_recv
3065   __ verify_oop(O0_recv);
3066 
3067   // get return address
3068   AddressLiteral table(Interpreter::invoke_return_entry_table());
3069   __ set(table, Rtemp);
3070   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
3071   // Make sure we don't need to mask Rret after the above shift
3072   ConstantPoolCacheEntry::verify_tos_state_shift();
3073   __ sll(Rret,  LogBytesPerWord, Rret);
3074   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3075 
3076   // get receiver klass
3077   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
3078   __ load_klass(O0_recv, O0_recv);
3079   __ verify_klass_ptr(O0_recv);
3080 
3081   __ profile_virtual_call(O0_recv, O4);
3082 
3083   generate_vtable_call(O0_recv, Rscratch, Rret);
3084 }
3085 
3086 void TemplateTable::fast_invokevfinal(int byte_no) {
3087   transition(vtos, vtos);
3088   assert(byte_no == f2_byte, "use this argument");
3089 
3090   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
3091                              /*is_invokevfinal*/true, false);
3092   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3093   invokevfinal_helper(G3_scratch, Lscratch);
3094 }
3095 
3096 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
3097   Register Rtemp = G4_scratch;
3098 
3099   // Load receiver from stack slot
3100   __ ld_ptr(G5_method, in_bytes(Method::const_offset()), G4_scratch);
3101   __ lduh(G4_scratch, in_bytes(ConstMethod::size_of_parameters_offset()), G4_scratch);
3102   __ load_receiver(G4_scratch, O0);
3103 
3104   // receiver NULL check
3105   __ null_check(O0);
3106 
3107   __ profile_final_call(O4);
3108   __ profile_arguments_type(G5_method, Rscratch, Gargs, true);
3109 
3110   // get return address
3111   AddressLiteral table(Interpreter::invoke_return_entry_table());
3112   __ set(table, Rtemp);
3113   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
3114   // Make sure we don't need to mask Rret after the above shift
3115   ConstantPoolCacheEntry::verify_tos_state_shift();
3116   __ sll(Rret,  LogBytesPerWord, Rret);
3117   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3118 
3119 
3120   // do the call
3121   __ call_from_interpreter(Rscratch, Gargs, Rret);
3122 }
3123 
3124 
3125 void TemplateTable::invokespecial(int byte_no) {
3126   transition(vtos, vtos);
3127   assert(byte_no == f1_byte, "use this argument");
3128 
3129   const Register Rret     = Lscratch;
3130   const Register O0_recv  = O0;
3131   const Register Rscratch = G3_scratch;
3132 
3133   prepare_invoke(byte_no, G5_method, Rret, noreg, O0_recv);  // get receiver also for null check
3134   __ null_check(O0_recv);
3135 
3136   // do the call
3137   __ profile_call(O4);
3138   __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3139   __ call_from_interpreter(Rscratch, Gargs, Rret);
3140 }
3141 
3142 
3143 void TemplateTable::invokestatic(int byte_no) {
3144   transition(vtos, vtos);
3145   assert(byte_no == f1_byte, "use this argument");
3146 
3147   const Register Rret     = Lscratch;
3148   const Register Rscratch = G3_scratch;
3149 
3150   prepare_invoke(byte_no, G5_method, Rret);  // get f1 Method*
3151 
3152   // do the call
3153   __ profile_call(O4);
3154   __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3155   __ call_from_interpreter(Rscratch, Gargs, Rret);
3156 }
3157 
3158 void TemplateTable::invokeinterface_object_method(Register RKlass,
3159                                                   Register Rcall,
3160                                                   Register Rret,
3161                                                   Register Rflags) {
3162   Register Rscratch = G4_scratch;
3163   Register Rindex = Lscratch;
3164 
3165   assert_different_registers(Rscratch, Rindex, Rret);
3166 
3167   Label notFinal;
3168 
3169   // Check for vfinal
3170   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), Rscratch);
3171   __ btst(Rflags, Rscratch);
3172   __ br(Assembler::zero, false, Assembler::pt, notFinal);
3173   __ delayed()->nop();
3174 
3175   __ profile_final_call(O4);
3176 
3177   // do the call - the index (f2) contains the Method*
3178   assert_different_registers(G5_method, Gargs, Rcall);
3179   __ mov(Rindex, G5_method);
3180   __ profile_arguments_type(G5_method, Rcall, Gargs, true);
3181   __ call_from_interpreter(Rcall, Gargs, Rret);
3182   __ bind(notFinal);
3183 
3184   __ profile_virtual_call(RKlass, O4);
3185   generate_vtable_call(RKlass, Rindex, Rret);
3186 }
3187 
3188 
3189 void TemplateTable::invokeinterface(int byte_no) {
3190   transition(vtos, vtos);
3191   assert(byte_no == f1_byte, "use this argument");
3192 
3193   const Register Rinterface  = G1_scratch;
3194   const Register Rmethod     = Lscratch;
3195   const Register Rret        = G3_scratch;
3196   const Register O0_recv     = O0;
3197   const Register O1_flags    = O1;
3198   const Register O2_Klass    = O2;
3199   const Register Rscratch    = G4_scratch;
3200   assert_different_registers(Rscratch, G5_method);
3201 
3202   prepare_invoke(byte_no, Rinterface, Rret, Rmethod, O0_recv, O1_flags);
3203 
3204   // get receiver klass
3205   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
3206   __ load_klass(O0_recv, O2_Klass);
3207 
3208   // Special case of invokeinterface called for virtual method of
3209   // java.lang.Object.  See cpCacheOop.cpp for details.
3210   // This code isn't produced by javac, but could be produced by
3211   // another compliant java compiler.
3212   Label notMethod;
3213   __ set((1 << ConstantPoolCacheEntry::is_forced_virtual_shift), Rscratch);
3214   __ btst(O1_flags, Rscratch);
3215   __ br(Assembler::zero, false, Assembler::pt, notMethod);
3216   __ delayed()->nop();
3217 
3218   invokeinterface_object_method(O2_Klass, Rinterface, Rret, O1_flags);
3219 
3220   __ bind(notMethod);
3221 
3222   Register Rtemp = O1_flags;
3223 
3224   Label L_no_such_interface;
3225 
3226   // Receiver subtype check against REFC.
3227   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3228                              O2_Klass, Rinterface, noreg,
3229                              // outputs: temp reg1, temp reg2, temp reg3
3230                              G5_method, Rscratch, Rtemp,
3231                              L_no_such_interface,
3232                              /*return_method=*/false);
3233 
3234   __ profile_virtual_call(O2_Klass, O4);
3235 
3236   //
3237   // find entry point to call
3238   //
3239 
3240   // Get declaring interface class from method
3241   __ ld_ptr(Rmethod, Method::const_offset(), Rinterface);
3242   __ ld_ptr(Rinterface, ConstMethod::constants_offset(), Rinterface);
3243   __ ld_ptr(Rinterface, ConstantPool::pool_holder_offset_in_bytes(), Rinterface);
3244 
3245   // Get itable index from method
3246   const Register Rindex = G5_method;
3247   __ ld(Rmethod, Method::itable_index_offset(), Rindex);
3248   __ sub(Rindex, Method::itable_index_max, Rindex);
3249   __ neg(Rindex);
3250 
3251   // Preserve O2_Klass for throw_AbstractMethodErrorVerbose
3252   __ mov(O2_Klass, O4);
3253   __ lookup_interface_method(// inputs: rec. class, interface, itable index
3254                              O4, Rinterface, Rindex,
3255                              // outputs: method, scan temp reg, temp reg
3256                              G5_method, Rscratch, Rtemp,
3257                              L_no_such_interface);
3258 
3259   // Check for abstract method error.
3260   {
3261     Label ok;
3262     __ br_notnull_short(G5_method, Assembler::pt, ok);
3263     // Pass arguments for generating a verbose error message.
3264     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3265             O2_Klass, Rmethod);
3266     __ should_not_reach_here();
3267     __ bind(ok);
3268   }
3269 
3270   Register Rcall = Rinterface;
3271   assert_different_registers(Rcall, G5_method, Gargs, Rret);
3272 
3273   __ profile_arguments_type(G5_method, Rcall, Gargs, true);
3274   __ profile_called_method(G5_method, Rscratch);
3275   __ call_from_interpreter(Rcall, Gargs, Rret);
3276 
3277   __ bind(L_no_such_interface);
3278   // Pass arguments for generating a verbose error message.
3279   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3280           O2_Klass, Rinterface);
3281   __ should_not_reach_here();
3282 }
3283 
3284 void TemplateTable::invokehandle(int byte_no) {
3285   transition(vtos, vtos);
3286   assert(byte_no == f1_byte, "use this argument");
3287 
3288   const Register Rret       = Lscratch;
3289   const Register G4_mtype   = G4_scratch;
3290   const Register O0_recv    = O0;
3291   const Register Rscratch   = G3_scratch;
3292 
3293   prepare_invoke(byte_no, G5_method, Rret, G4_mtype, O0_recv);
3294   __ null_check(O0_recv);
3295 
3296   // G4: MethodType object (from cpool->resolved_references[f1], if necessary)
3297   // G5: MH.invokeExact_MT method (from f2)
3298 
3299   // Note:  G4_mtype is already pushed (if necessary) by prepare_invoke
3300 
3301   // do the call
3302   __ verify_oop(G4_mtype);
3303   __ profile_final_call(O4);  // FIXME: profile the LambdaForm also
3304   __ profile_arguments_type(G5_method, Rscratch, Gargs, true);
3305   __ call_from_interpreter(Rscratch, Gargs, Rret);
3306 }
3307 
3308 
3309 void TemplateTable::invokedynamic(int byte_no) {
3310   transition(vtos, vtos);
3311   assert(byte_no == f1_byte, "use this argument");
3312 
3313   const Register Rret        = Lscratch;
3314   const Register G4_callsite = G4_scratch;
3315   const Register Rscratch    = G3_scratch;
3316 
3317   prepare_invoke(byte_no, G5_method, Rret, G4_callsite);
3318 
3319   // G4: CallSite object (from cpool->resolved_references[f1])
3320   // G5: MH.linkToCallSite method (from f2)
3321 
3322   // Note:  G4_callsite is already pushed by prepare_invoke
3323 
3324   // %%% should make a type profile for any invokedynamic that takes a ref argument
3325   // profile this call
3326   __ profile_call(O4);
3327 
3328   // do the call
3329   __ verify_oop(G4_callsite);
3330   __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3331   __ call_from_interpreter(Rscratch, Gargs, Rret);
3332 }
3333 
3334 
3335 //----------------------------------------------------------------------------------------------------
3336 // Allocation
3337 
3338 void TemplateTable::_new() {
3339   transition(vtos, atos);
3340 
3341   Label slow_case;
3342   Label done;
3343   Label initialize_header;
3344   Label initialize_object;  // including clearing the fields
3345 
3346   Register RallocatedObject = Otos_i;
3347   Register RinstanceKlass = O1;
3348   Register Roffset = O3;
3349   Register Rscratch = O4;
3350 
3351   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3352   __ get_cpool_and_tags(Rscratch, G3_scratch);
3353   // make sure the class we're about to instantiate has been resolved
3354   // This is done before loading InstanceKlass to be consistent with the order
3355   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3356   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3357   __ ldub(G3_scratch, Roffset, G3_scratch);
3358   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3359   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3360   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3361   // get InstanceKlass
3362   __ load_resolved_klass_at_offset(Rscratch, Roffset, RinstanceKlass);
3363 
3364   // make sure klass is fully initialized:
3365   __ ldub(RinstanceKlass, in_bytes(InstanceKlass::init_state_offset()), G3_scratch);
3366   __ cmp(G3_scratch, InstanceKlass::fully_initialized);
3367   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3368   __ delayed()->ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3369 
3370   // get instance_size in InstanceKlass (already aligned)
3371   //__ ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3372 
3373   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
3374   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
3375   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
3376   __ delayed()->nop();
3377 
3378   // Allocate the instance:
3379   //  If TLAB is enabled:
3380   //    Try to allocate in the TLAB.
3381   //    If fails, go to the slow path.
3382   //  Else If inline contiguous allocations are enabled:
3383   //    Try to allocate in eden.
3384   //    If fails due to heap end, go to slow path.
3385   //
3386   //  If TLAB is enabled OR inline contiguous is enabled:
3387   //    Initialize the allocation.
3388   //    Exit.
3389   //
3390   //  Go to slow path.
3391 
3392   const bool allow_shared_alloc =
3393     Universe::heap()->supports_inline_contig_alloc();
3394 
3395   if(UseTLAB) {
3396     Register RoldTopValue = RallocatedObject;
3397     Register RtlabWasteLimitValue = G3_scratch;
3398     Register RnewTopValue = G1_scratch;
3399     Register RendValue = Rscratch;
3400     Register RfreeValue = RnewTopValue;
3401 
3402     // check if we can allocate in the TLAB
3403     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
3404     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
3405     __ add(RoldTopValue, Roffset, RnewTopValue);
3406 
3407     // if there is enough space, we do not CAS and do not clear
3408     __ cmp(RnewTopValue, RendValue);
3409     if(ZeroTLAB) {
3410       // the fields have already been cleared
3411       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
3412     } else {
3413       // initialize both the header and fields
3414       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
3415     }
3416     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
3417 
3418     // Allocation does not fit in the TLAB.
3419     __ ba_short(slow_case);
3420   } else {
3421     // Allocation in the shared Eden
3422     if (allow_shared_alloc) {
3423       Register RoldTopValue = G1_scratch;
3424       Register RtopAddr = G3_scratch;
3425       Register RnewTopValue = RallocatedObject;
3426       Register RendValue = Rscratch;
3427 
3428       __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
3429 
3430       Label retry;
3431       __ bind(retry);
3432       __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
3433       __ ld_ptr(RendValue, 0, RendValue);
3434       __ ld_ptr(RtopAddr, 0, RoldTopValue);
3435       __ add(RoldTopValue, Roffset, RnewTopValue);
3436 
3437       // RnewTopValue contains the top address after the new object
3438       // has been allocated.
3439       __ cmp_and_brx_short(RnewTopValue, RendValue, Assembler::greaterUnsigned, Assembler::pn, slow_case);
3440 
3441       __ cas_ptr(RtopAddr, RoldTopValue, RnewTopValue);
3442 
3443       // if someone beat us on the allocation, try again, otherwise continue
3444       __ cmp_and_brx_short(RoldTopValue, RnewTopValue, Assembler::notEqual, Assembler::pn, retry);
3445 
3446       // bump total bytes allocated by this thread
3447       // RoldTopValue and RtopAddr are dead, so can use G1 and G3
3448       __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
3449     }
3450   }
3451 
3452   // If UseTLAB or allow_shared_alloc are true, the object is created above and
3453   // there is an initialize need. Otherwise, skip and go to the slow path.
3454   if (UseTLAB || allow_shared_alloc) {
3455     // clear object fields
3456     __ bind(initialize_object);
3457     __ deccc(Roffset, sizeof(oopDesc));
3458     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
3459     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
3460 
3461     // initialize remaining object fields
3462     if (UseBlockZeroing) {
3463       // Use BIS for zeroing
3464       __ bis_zeroing(G3_scratch, Roffset, G1_scratch, initialize_header);
3465     } else {
3466       Label loop;
3467       __ subcc(Roffset, wordSize, Roffset);
3468       __ bind(loop);
3469       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
3470       __ st_ptr(G0, G3_scratch, Roffset);
3471       __ br(Assembler::notEqual, false, Assembler::pt, loop);
3472       __ delayed()->subcc(Roffset, wordSize, Roffset);
3473     }
3474     __ ba_short(initialize_header);
3475   }
3476 
3477   // slow case
3478   __ bind(slow_case);
3479   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
3480   __ get_constant_pool(O1);
3481 
3482   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
3483 
3484   __ ba_short(done);
3485 
3486   // Initialize the header: mark, klass
3487   __ bind(initialize_header);
3488 
3489   if (UseBiasedLocking) {
3490     __ ld_ptr(RinstanceKlass, in_bytes(Klass::prototype_header_offset()), G4_scratch);
3491   } else {
3492     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
3493   }
3494   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
3495   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
3496   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
3497 
3498   {
3499     SkipIfEqual skip_if(
3500       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
3501     // Trigger dtrace event
3502     __ push(atos);
3503     __ call_VM_leaf(noreg,
3504        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
3505     __ pop(atos);
3506   }
3507 
3508   // continue
3509   __ bind(done);
3510 }
3511 
3512 
3513 
3514 void TemplateTable::newarray() {
3515   transition(itos, atos);
3516   __ ldub(Lbcp, 1, O1);
3517      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
3518 }
3519 
3520 
3521 void TemplateTable::anewarray() {
3522   transition(itos, atos);
3523   __ get_constant_pool(O1);
3524   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
3525      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
3526 }
3527 
3528 
3529 void TemplateTable::arraylength() {
3530   transition(atos, itos);
3531   Label ok;
3532   __ verify_oop(Otos_i);
3533   __ tst(Otos_i);
3534   __ throw_if_not_1_x( Assembler::notZero, ok );
3535   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
3536   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3537 }
3538 
3539 
3540 void TemplateTable::checkcast() {
3541   transition(atos, atos);
3542   Label done, is_null, quicked, cast_ok, resolved;
3543   Register Roffset = G1_scratch;
3544   Register RobjKlass = O5;
3545   Register RspecifiedKlass = O4;
3546 
3547   // Check for casting a NULL
3548   __ br_null(Otos_i, false, Assembler::pn, is_null);
3549   __ delayed()->nop();
3550 
3551   // Get value klass in RobjKlass
3552   __ load_klass(Otos_i, RobjKlass); // get value klass
3553 
3554   // Get constant pool tag
3555   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3556 
3557   // See if the checkcast has been quickened
3558   __ get_cpool_and_tags(Lscratch, G3_scratch);
3559   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3560   __ ldub(G3_scratch, Roffset, G3_scratch);
3561   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3562   __ br(Assembler::equal, true, Assembler::pt, quicked);
3563   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3564 
3565   __ push_ptr(); // save receiver for result, and for GC
3566   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3567   __ get_vm_result_2(RspecifiedKlass);
3568   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3569 
3570   __ ba_short(resolved);
3571 
3572   // Extract target class from constant pool
3573   __ bind(quicked);
3574   __ load_resolved_klass_at_offset(Lscratch, Roffset, RspecifiedKlass);
3575 
3576 
3577   __ bind(resolved);
3578   __ load_klass(Otos_i, RobjKlass); // get value klass
3579 
3580   // Generate a fast subtype check.  Branch to cast_ok if no
3581   // failure.  Throw exception if failure.
3582   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
3583 
3584   // Not a subtype; so must throw exception
3585   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
3586 
3587   __ bind(cast_ok);
3588 
3589   if (ProfileInterpreter) {
3590     __ ba_short(done);
3591   }
3592   __ bind(is_null);
3593   __ profile_null_seen(G3_scratch);
3594   __ bind(done);
3595 }
3596 
3597 
3598 void TemplateTable::instanceof() {
3599   Label done, is_null, quicked, resolved;
3600   transition(atos, itos);
3601   Register Roffset = G1_scratch;
3602   Register RobjKlass = O5;
3603   Register RspecifiedKlass = O4;
3604 
3605   // Check for casting a NULL
3606   __ br_null(Otos_i, false, Assembler::pt, is_null);
3607   __ delayed()->nop();
3608 
3609   // Get value klass in RobjKlass
3610   __ load_klass(Otos_i, RobjKlass); // get value klass
3611 
3612   // Get constant pool tag
3613   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3614 
3615   // See if the checkcast has been quickened
3616   __ get_cpool_and_tags(Lscratch, G3_scratch);
3617   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3618   __ ldub(G3_scratch, Roffset, G3_scratch);
3619   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3620   __ br(Assembler::equal, true, Assembler::pt, quicked);
3621   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3622 
3623   __ push_ptr(); // save receiver for result, and for GC
3624   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3625   __ get_vm_result_2(RspecifiedKlass);
3626   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3627 
3628   __ ba_short(resolved);
3629 
3630   // Extract target class from constant pool
3631   __ bind(quicked);
3632   __ get_constant_pool(Lscratch);
3633   __ load_resolved_klass_at_offset(Lscratch, Roffset, RspecifiedKlass);
3634 
3635   __ bind(resolved);
3636   __ load_klass(Otos_i, RobjKlass); // get value klass
3637 
3638   // Generate a fast subtype check.  Branch to cast_ok if no
3639   // failure.  Return 0 if failure.
3640   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
3641   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
3642   // Not a subtype; return 0;
3643   __ clr( Otos_i );
3644 
3645   if (ProfileInterpreter) {
3646     __ ba_short(done);
3647   }
3648   __ bind(is_null);
3649   __ profile_null_seen(G3_scratch);
3650   __ bind(done);
3651 }
3652 
3653 void TemplateTable::_breakpoint() {
3654 
3655    // Note: We get here even if we are single stepping..
3656    // jbug insists on setting breakpoints at every bytecode
3657    // even if we are in single step mode.
3658 
3659    transition(vtos, vtos);
3660    // get the unpatched byte code
3661    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
3662    __ mov(O0, Lbyte_code);
3663 
3664    // post the breakpoint event
3665    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
3666 
3667    // complete the execution of original bytecode
3668    __ dispatch_normal(vtos);
3669 }
3670 
3671 
3672 //----------------------------------------------------------------------------------------------------
3673 // Exceptions
3674 
3675 void TemplateTable::athrow() {
3676   transition(atos, vtos);
3677 
3678   // This works because exception is cached in Otos_i which is same as O0,
3679   // which is same as what throw_exception_entry_expects
3680   assert(Otos_i == Oexception, "see explanation above");
3681 
3682   __ verify_oop(Otos_i);
3683   __ null_check(Otos_i);
3684   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
3685 }
3686 
3687 
3688 //----------------------------------------------------------------------------------------------------
3689 // Synchronization
3690 
3691 
3692 // See frame_sparc.hpp for monitor block layout.
3693 // Monitor elements are dynamically allocated by growing stack as needed.
3694 
3695 void TemplateTable::monitorenter() {
3696   transition(atos, vtos);
3697   __ verify_oop(Otos_i);
3698   // Try to acquire a lock on the object
3699   // Repeat until succeeded (i.e., until
3700   // monitorenter returns true).
3701 
3702   {   Label ok;
3703     __ tst(Otos_i);
3704     __ throw_if_not_1_x( Assembler::notZero,  ok);
3705     __ delayed()->mov(Otos_i, Lscratch); // save obj
3706     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3707   }
3708 
3709   assert(O0 == Otos_i, "Be sure where the object to lock is");
3710 
3711   // find a free slot in the monitor block
3712 
3713 
3714   // initialize entry pointer
3715   __ clr(O1); // points to free slot or NULL
3716 
3717   {
3718     Label entry, loop, exit;
3719     __ add( __ top_most_monitor(), O2 ); // last one to check
3720     __ ba( entry );
3721     __ delayed()->mov( Lmonitors, O3 ); // first one to check
3722 
3723 
3724     __ bind( loop );
3725 
3726     __ verify_oop(O4);          // verify each monitor's oop
3727     __ tst(O4); // is this entry unused?
3728     __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
3729 
3730     __ cmp(O4, O0); // check if current entry is for same object
3731     __ brx( Assembler::equal, false, Assembler::pn, exit );
3732     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
3733 
3734     __ bind( entry );
3735 
3736     __ cmp( O3, O2 );
3737     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3738     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
3739 
3740     __ bind( exit );
3741   }
3742 
3743   { Label allocated;
3744 
3745     // found free slot?
3746     __ br_notnull_short(O1, Assembler::pn, allocated);
3747 
3748     __ add_monitor_to_stack( false, O2, O3 );
3749     __ mov(Lmonitors, O1);
3750 
3751     __ bind(allocated);
3752   }
3753 
3754   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3755   // The object has already been poped from the stack, so the expression stack looks correct.
3756   __ inc(Lbcp);
3757 
3758   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
3759   __ lock_object(O1, O0);
3760 
3761   // check if there's enough space on the stack for the monitors after locking
3762   __ generate_stack_overflow_check(0);
3763 
3764   // The bcp has already been incremented. Just need to dispatch to next instruction.
3765   __ dispatch_next(vtos);
3766 }
3767 
3768 
3769 void TemplateTable::monitorexit() {
3770   transition(atos, vtos);
3771   __ verify_oop(Otos_i);
3772   __ tst(Otos_i);
3773   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
3774 
3775   assert(O0 == Otos_i, "just checking");
3776 
3777   { Label entry, loop, found;
3778     __ add( __ top_most_monitor(), O2 ); // last one to check
3779     __ ba(entry);
3780     // use Lscratch to hold monitor elem to check, start with most recent monitor,
3781     // By using a local it survives the call to the C routine.
3782     __ delayed()->mov( Lmonitors, Lscratch );
3783 
3784     __ bind( loop );
3785 
3786     __ verify_oop(O4);          // verify each monitor's oop
3787     __ cmp(O4, O0); // check if current entry is for desired object
3788     __ brx( Assembler::equal, true, Assembler::pt, found );
3789     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
3790 
3791     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
3792 
3793     __ bind( entry );
3794 
3795     __ cmp( Lscratch, O2 );
3796     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3797     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
3798 
3799     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3800     __ should_not_reach_here();
3801 
3802     __ bind(found);
3803   }
3804   __ unlock_object(O1);
3805 }
3806 
3807 
3808 //----------------------------------------------------------------------------------------------------
3809 // Wide instructions
3810 
3811 void TemplateTable::wide() {
3812   transition(vtos, vtos);
3813   __ ldub(Lbcp, 1, G3_scratch);// get next bc
3814   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
3815   AddressLiteral ep(Interpreter::_wentry_point);
3816   __ set(ep, G4_scratch);
3817   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
3818   __ jmp(G3_scratch, G0);
3819   __ delayed()->nop();
3820   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
3821 }
3822 
3823 
3824 //----------------------------------------------------------------------------------------------------
3825 // Multi arrays
3826 
3827 void TemplateTable::multianewarray() {
3828   transition(vtos, atos);
3829      // put ndims * wordSize into Lscratch
3830   __ ldub( Lbcp,     3,               Lscratch);
3831   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
3832      // Lesp points past last_dim, so set to O1 to first_dim address
3833   __ add(  Lesp,     Lscratch,        O1);
3834      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
3835   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
3836 }