1 /*
   2  * Copyright (c) 1997, 2013, 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 "interpreter/interpreter.hpp"
  27 #include "interpreter/interpreterRuntime.hpp"
  28 #include "interpreter/templateTable.hpp"
  29 #include "memory/universe.inline.hpp"
  30 #include "oops/methodData.hpp"
  31 #include "oops/objArrayKlass.hpp"
  32 #include "oops/oop.inline.hpp"
  33 #include "prims/methodHandles.hpp"
  34 #include "runtime/sharedRuntime.hpp"
  35 #include "runtime/stubRoutines.hpp"
  36 #include "runtime/synchronizer.hpp"
  37 #include "utilities/macros.hpp"
  38 
  39 #ifndef CC_INTERP
  40 #define __ _masm->
  41 
  42 // Misc helpers
  43 
  44 // Do an oop store like *(base + index + offset) = val
  45 // index can be noreg,
  46 static void do_oop_store(InterpreterMacroAssembler* _masm,
  47                          Register base,
  48                          Register index,
  49                          int offset,
  50                          Register val,
  51                          Register tmp,
  52                          BarrierSet::Name barrier,
  53                          bool precise) {
  54   assert(tmp != val && tmp != base && tmp != index, "register collision");
  55   assert(index == noreg || offset == 0, "only one offset");
  56   switch (barrier) {
  57 #if INCLUDE_ALL_GCS
  58     case BarrierSet::G1SATBCT:
  59     case BarrierSet::G1SATBCTLogging:
  60       {
  61         // Load and record the previous value.
  62         __ g1_write_barrier_pre(base, index, offset,
  63                                 noreg /* pre_val */,
  64                                 tmp, true /*preserve_o_regs*/);
  65 
  66         // G1 barrier needs uncompressed oop for region cross check.
  67         Register new_val = val;
  68         if (UseCompressedOops && val != G0) {
  69           new_val = tmp;
  70           __ mov(val, new_val);
  71         }
  72 
  73         if (index == noreg ) {
  74           assert(Assembler::is_simm13(offset), "fix this code");
  75           __ store_heap_oop(val, base, offset);
  76         } else {
  77           __ store_heap_oop(val, base, index);
  78         }
  79 
  80         // No need for post barrier if storing NULL
  81         if (val != G0) {
  82           if (precise) {
  83             if (index == noreg) {
  84               __ add(base, offset, base);
  85             } else {
  86               __ add(base, index, base);
  87             }
  88           }
  89           __ g1_write_barrier_post(base, new_val, tmp);
  90         }
  91       }
  92       break;
  93 #endif // INCLUDE_ALL_GCS
  94     case BarrierSet::CardTableModRef:
  95     case BarrierSet::CardTableExtension:
  96       {
  97         if (index == noreg ) {
  98           assert(Assembler::is_simm13(offset), "fix this code");
  99           __ store_heap_oop(val, base, offset);
 100         } else {
 101           __ store_heap_oop(val, base, index);
 102         }
 103         // No need for post barrier if storing NULL
 104         if (val != G0) {
 105           if (precise) {
 106             if (index == noreg) {
 107               __ add(base, offset, base);
 108             } else {
 109               __ add(base, index, base);
 110             }
 111           }
 112           __ card_write_barrier_post(base, val, tmp);
 113         }
 114       }
 115       break;
 116     case BarrierSet::ModRef:
 117     case BarrierSet::Other:
 118       ShouldNotReachHere();
 119       break;
 120     default      :
 121       ShouldNotReachHere();
 122 
 123   }
 124 }
 125 
 126 
 127 //----------------------------------------------------------------------------------------------------
 128 // Platform-dependent initialization
 129 
 130 void TemplateTable::pd_initialize() {
 131   // (none)
 132 }
 133 
 134 
 135 //----------------------------------------------------------------------------------------------------
 136 // Condition conversion
 137 Assembler::Condition ccNot(TemplateTable::Condition cc) {
 138   switch (cc) {
 139     case TemplateTable::equal        : return Assembler::notEqual;
 140     case TemplateTable::not_equal    : return Assembler::equal;
 141     case TemplateTable::less         : return Assembler::greaterEqual;
 142     case TemplateTable::less_equal   : return Assembler::greater;
 143     case TemplateTable::greater      : return Assembler::lessEqual;
 144     case TemplateTable::greater_equal: return Assembler::less;
 145   }
 146   ShouldNotReachHere();
 147   return Assembler::zero;
 148 }
 149 
 150 //----------------------------------------------------------------------------------------------------
 151 // Miscelaneous helper routines
 152 
 153 
 154 Address TemplateTable::at_bcp(int offset) {
 155   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 156   return Address(Lbcp, offset);
 157 }
 158 
 159 
 160 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 161                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 162                                    int byte_no) {
 163   // With sharing on, may need to test Method* flag.
 164   if (!RewriteBytecodes)  return;
 165   Label L_patch_done;
 166 
 167   switch (bc) {
 168   case Bytecodes::_fast_aputfield:
 169   case Bytecodes::_fast_bputfield:
 170   case Bytecodes::_fast_cputfield:
 171   case Bytecodes::_fast_dputfield:
 172   case Bytecodes::_fast_fputfield:
 173   case Bytecodes::_fast_iputfield:
 174   case Bytecodes::_fast_lputfield:
 175   case Bytecodes::_fast_sputfield:
 176     {
 177       // We skip bytecode quickening for putfield instructions when
 178       // the put_code written to the constant pool cache is zero.
 179       // This is required so that every execution of this instruction
 180       // calls out to InterpreterRuntime::resolve_get_put to do
 181       // additional, required work.
 182       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 183       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 184       __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
 185       __ set(bc, bc_reg);
 186       __ cmp_and_br_short(temp_reg, 0, Assembler::equal, Assembler::pn, L_patch_done);  // don't patch
 187     }
 188     break;
 189   default:
 190     assert(byte_no == -1, "sanity");
 191     if (load_bc_into_bc_reg) {
 192       __ set(bc, bc_reg);
 193     }
 194   }
 195 
 196   if (JvmtiExport::can_post_breakpoint()) {
 197     Label L_fast_patch;
 198     __ ldub(at_bcp(0), temp_reg);
 199     __ cmp_and_br_short(temp_reg, Bytecodes::_breakpoint, Assembler::notEqual, Assembler::pt, L_fast_patch);
 200     // perform the quickening, slowly, in the bowels of the breakpoint table
 201     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, bc_reg);
 202     __ ba_short(L_patch_done);
 203     __ bind(L_fast_patch);
 204   }
 205 
 206 #ifdef ASSERT
 207   Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
 208   Label L_okay;
 209   __ ldub(at_bcp(0), temp_reg);
 210   __ cmp(temp_reg, orig_bytecode);
 211   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 212   __ delayed()->cmp(temp_reg, bc_reg);
 213   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 214   __ delayed()->nop();
 215   __ stop("patching the wrong bytecode");
 216   __ bind(L_okay);
 217 #endif
 218 
 219   // patch bytecode
 220   __ stb(bc_reg, at_bcp(0));
 221   __ bind(L_patch_done);
 222 }
 223 
 224 //----------------------------------------------------------------------------------------------------
 225 // Individual instructions
 226 
 227 void TemplateTable::nop() {
 228   transition(vtos, vtos);
 229   // nothing to do
 230 }
 231 
 232 void TemplateTable::shouldnotreachhere() {
 233   transition(vtos, vtos);
 234   __ stop("shouldnotreachhere bytecode");
 235 }
 236 
 237 void TemplateTable::aconst_null() {
 238   transition(vtos, atos);
 239   __ clr(Otos_i);
 240 }
 241 
 242 
 243 void TemplateTable::iconst(int value) {
 244   transition(vtos, itos);
 245   __ set(value, Otos_i);
 246 }
 247 
 248 
 249 void TemplateTable::lconst(int value) {
 250   transition(vtos, ltos);
 251   assert(value >= 0, "check this code");
 252 #ifdef _LP64
 253   __ set(value, Otos_l);
 254 #else
 255   __ set(value, Otos_l2);
 256   __ clr( Otos_l1);
 257 #endif
 258 }
 259 
 260 
 261 void TemplateTable::fconst(int value) {
 262   transition(vtos, ftos);
 263   static float zero = 0.0, one = 1.0, two = 2.0;
 264   float* p;
 265   switch( value ) {
 266    default: ShouldNotReachHere();
 267    case 0:  p = &zero;  break;
 268    case 1:  p = &one;   break;
 269    case 2:  p = &two;   break;
 270   }
 271   AddressLiteral a(p);
 272   __ sethi(a, G3_scratch);
 273   __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
 274 }
 275 
 276 
 277 void TemplateTable::dconst(int value) {
 278   transition(vtos, dtos);
 279   static double zero = 0.0, one = 1.0;
 280   double* p;
 281   switch( value ) {
 282    default: ShouldNotReachHere();
 283    case 0:  p = &zero;  break;
 284    case 1:  p = &one;   break;
 285   }
 286   AddressLiteral a(p);
 287   __ sethi(a, G3_scratch);
 288   __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
 289 }
 290 
 291 
 292 // %%%%% Should factore most snippet templates across platforms
 293 
 294 void TemplateTable::bipush() {
 295   transition(vtos, itos);
 296   __ ldsb( at_bcp(1), Otos_i );
 297 }
 298 
 299 void TemplateTable::sipush() {
 300   transition(vtos, itos);
 301   __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
 302 }
 303 
 304 void TemplateTable::ldc(bool wide) {
 305   transition(vtos, vtos);
 306   Label call_ldc, notInt, isString, notString, notClass, exit;
 307 
 308   if (wide) {
 309     __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 310   } else {
 311     __ ldub(Lbcp, 1, O1);
 312   }
 313   __ get_cpool_and_tags(O0, O2);
 314 
 315   const int base_offset = ConstantPool::header_size() * wordSize;
 316   const int tags_offset = Array<u1>::base_offset_in_bytes();
 317 
 318   // get type from tags
 319   __ add(O2, tags_offset, O2);
 320   __ ldub(O2, O1, O2);
 321 
 322   // unresolved class? If so, must resolve
 323   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClass, Assembler::equal, Assembler::pt, call_ldc);
 324 
 325   // unresolved class in error state
 326   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClassInError, Assembler::equal, Assembler::pn, call_ldc);
 327 
 328   __ cmp(O2, JVM_CONSTANT_Class);      // need to call vm to get java mirror of the class
 329   __ brx(Assembler::notEqual, true, Assembler::pt, notClass);
 330   __ delayed()->add(O0, base_offset, O0);
 331 
 332   __ bind(call_ldc);
 333   __ set(wide, O1);
 334   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), O1);
 335   __ push(atos);
 336   __ ba_short(exit);
 337 
 338   __ bind(notClass);
 339  // __ add(O0, base_offset, O0);
 340   __ sll(O1, LogBytesPerWord, O1);
 341   __ cmp(O2, JVM_CONSTANT_Integer);
 342   __ brx(Assembler::notEqual, true, Assembler::pt, notInt);
 343   __ delayed()->cmp(O2, JVM_CONSTANT_String);
 344   __ ld(O0, O1, Otos_i);
 345   __ push(itos);
 346   __ ba_short(exit);
 347 
 348   __ bind(notInt);
 349  // __ cmp(O2, JVM_CONSTANT_String);
 350   __ brx(Assembler::notEqual, true, Assembler::pt, notString);
 351   __ delayed()->ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
 352   __ bind(isString);
 353   __ stop("string should be rewritten to fast_aldc");
 354   __ ba_short(exit);
 355 
 356   __ bind(notString);
 357  // __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
 358   __ push(ftos);
 359 
 360   __ bind(exit);
 361 }
 362 
 363 // Fast path for caching oop constants.
 364 // %%% We should use this to handle Class and String constants also.
 365 // %%% It will simplify the ldc/primitive path considerably.
 366 void TemplateTable::fast_aldc(bool wide) {
 367   transition(vtos, atos);
 368 
 369   int index_size = wide ? sizeof(u2) : sizeof(u1);
 370   Label resolved;
 371 
 372   // We are resolved if the resolved reference cache entry contains a
 373   // non-null object (CallSite, etc.)
 374   assert_different_registers(Otos_i, G3_scratch);
 375   __ get_cache_index_at_bcp(Otos_i, G3_scratch, 1, index_size);  // load index => G3_scratch
 376   __ load_resolved_reference_at_index(Otos_i, G3_scratch);
 377   __ tst(Otos_i);
 378   __ br(Assembler::notEqual, false, Assembler::pt, resolved);
 379   __ delayed()->set((int)bytecode(), O1);
 380 
 381   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 382 
 383   // first time invocation - must resolve first
 384   __ call_VM(Otos_i, entry, O1);
 385   __ bind(resolved);
 386   __ verify_oop(Otos_i);
 387 }
 388 
 389 
 390 void TemplateTable::ldc2_w() {
 391   transition(vtos, vtos);
 392   Label Long, exit;
 393 
 394   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 395   __ get_cpool_and_tags(O0, O2);
 396 
 397   const int base_offset = ConstantPool::header_size() * wordSize;
 398   const int tags_offset = Array<u1>::base_offset_in_bytes();
 399   // get type from tags
 400   __ add(O2, tags_offset, O2);
 401   __ ldub(O2, O1, O2);
 402 
 403   __ sll(O1, LogBytesPerWord, O1);
 404   __ add(O0, O1, G3_scratch);
 405 
 406   __ cmp_and_brx_short(O2, JVM_CONSTANT_Double, Assembler::notEqual, Assembler::pt, Long);
 407   // A double can be placed at word-aligned locations in the constant pool.
 408   // Check out Conversions.java for an example.
 409   // Also ConstantPool::header_size() is 20, which makes it very difficult
 410   // to double-align double on the constant pool.  SG, 11/7/97
 411 #ifdef _LP64
 412   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
 413 #else
 414   FloatRegister f = Ftos_d;
 415   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset, f);
 416   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset + sizeof(jdouble)/2,
 417          f->successor());
 418 #endif
 419   __ push(dtos);
 420   __ ba_short(exit);
 421 
 422   __ bind(Long);
 423 #ifdef _LP64
 424   __ ldx(G3_scratch, base_offset, Otos_l);
 425 #else
 426   __ ld(G3_scratch, base_offset, Otos_l);
 427   __ ld(G3_scratch, base_offset + sizeof(jlong)/2, Otos_l->successor());
 428 #endif
 429   __ push(ltos);
 430 
 431   __ bind(exit);
 432 }
 433 
 434 
 435 void TemplateTable::locals_index(Register reg, int offset) {
 436   __ ldub( at_bcp(offset), reg );
 437 }
 438 
 439 
 440 void TemplateTable::locals_index_wide(Register reg) {
 441   // offset is 2, not 1, because Lbcp points to wide prefix code
 442   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
 443 }
 444 
 445 void TemplateTable::iload() {
 446   transition(vtos, itos);
 447   // Rewrite iload,iload  pair into fast_iload2
 448   //         iload,caload pair into fast_icaload
 449   if (RewriteFrequentPairs) {
 450     Label rewrite, done;
 451 
 452     // get next byte
 453     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
 454 
 455     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 456     // last two iloads in a pair.  Comparing against fast_iload means that
 457     // the next bytecode is neither an iload or a caload, and therefore
 458     // an iload pair.
 459     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_iload, Assembler::equal, Assembler::pn, done);
 460 
 461     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
 462     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 463     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
 464 
 465     __ cmp(G3_scratch, (int)Bytecodes::_caload);
 466     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 467     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
 468 
 469     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
 470     // rewrite
 471     // G4_scratch: fast bytecode
 472     __ bind(rewrite);
 473     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
 474     __ bind(done);
 475   }
 476 
 477   // Get the local value into tos
 478   locals_index(G3_scratch);
 479   __ access_local_int( G3_scratch, Otos_i );
 480 }
 481 
 482 void TemplateTable::fast_iload2() {
 483   transition(vtos, itos);
 484   locals_index(G3_scratch);
 485   __ access_local_int( G3_scratch, Otos_i );
 486   __ push_i();
 487   locals_index(G3_scratch, 3);  // get next bytecode's local index.
 488   __ access_local_int( G3_scratch, Otos_i );
 489 }
 490 
 491 void TemplateTable::fast_iload() {
 492   transition(vtos, itos);
 493   locals_index(G3_scratch);
 494   __ access_local_int( G3_scratch, Otos_i );
 495 }
 496 
 497 void TemplateTable::lload() {
 498   transition(vtos, ltos);
 499   locals_index(G3_scratch);
 500   __ access_local_long( G3_scratch, Otos_l );
 501 }
 502 
 503 
 504 void TemplateTable::fload() {
 505   transition(vtos, ftos);
 506   locals_index(G3_scratch);
 507   __ access_local_float( G3_scratch, Ftos_f );
 508 }
 509 
 510 
 511 void TemplateTable::dload() {
 512   transition(vtos, dtos);
 513   locals_index(G3_scratch);
 514   __ access_local_double( G3_scratch, Ftos_d );
 515 }
 516 
 517 
 518 void TemplateTable::aload() {
 519   transition(vtos, atos);
 520   locals_index(G3_scratch);
 521   __ access_local_ptr( G3_scratch, Otos_i);
 522 }
 523 
 524 
 525 void TemplateTable::wide_iload() {
 526   transition(vtos, itos);
 527   locals_index_wide(G3_scratch);
 528   __ access_local_int( G3_scratch, Otos_i );
 529 }
 530 
 531 
 532 void TemplateTable::wide_lload() {
 533   transition(vtos, ltos);
 534   locals_index_wide(G3_scratch);
 535   __ access_local_long( G3_scratch, Otos_l );
 536 }
 537 
 538 
 539 void TemplateTable::wide_fload() {
 540   transition(vtos, ftos);
 541   locals_index_wide(G3_scratch);
 542   __ access_local_float( G3_scratch, Ftos_f );
 543 }
 544 
 545 
 546 void TemplateTable::wide_dload() {
 547   transition(vtos, dtos);
 548   locals_index_wide(G3_scratch);
 549   __ access_local_double( G3_scratch, Ftos_d );
 550 }
 551 
 552 
 553 void TemplateTable::wide_aload() {
 554   transition(vtos, atos);
 555   locals_index_wide(G3_scratch);
 556   __ access_local_ptr( G3_scratch, Otos_i );
 557   __ verify_oop(Otos_i);
 558 }
 559 
 560 
 561 void TemplateTable::iaload() {
 562   transition(itos, itos);
 563   // Otos_i: index
 564   // tos: array
 565   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 566   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
 567 }
 568 
 569 
 570 void TemplateTable::laload() {
 571   transition(itos, ltos);
 572   // Otos_i: index
 573   // O2: array
 574   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 575   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
 576 }
 577 
 578 
 579 void TemplateTable::faload() {
 580   transition(itos, ftos);
 581   // Otos_i: index
 582   // O2: array
 583   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 584   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
 585 }
 586 
 587 
 588 void TemplateTable::daload() {
 589   transition(itos, dtos);
 590   // Otos_i: index
 591   // O2: array
 592   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 593   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
 594 }
 595 
 596 
 597 void TemplateTable::aaload() {
 598   transition(itos, atos);
 599   // Otos_i: index
 600   // tos: array
 601   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
 602   __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
 603   __ verify_oop(Otos_i);
 604 }
 605 
 606 
 607 void TemplateTable::baload() {
 608   transition(itos, itos);
 609   // Otos_i: index
 610   // tos: array
 611   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
 612   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
 613 }
 614 
 615 
 616 void TemplateTable::caload() {
 617   transition(itos, itos);
 618   // Otos_i: index
 619   // tos: array
 620   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 621   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 622 }
 623 
 624 void TemplateTable::fast_icaload() {
 625   transition(vtos, itos);
 626   // Otos_i: index
 627   // tos: array
 628   locals_index(G3_scratch);
 629   __ access_local_int( G3_scratch, Otos_i );
 630   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 631   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 632 }
 633 
 634 
 635 void TemplateTable::saload() {
 636   transition(itos, itos);
 637   // Otos_i: index
 638   // tos: array
 639   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 640   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
 641 }
 642 
 643 
 644 void TemplateTable::iload(int n) {
 645   transition(vtos, itos);
 646   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 647 }
 648 
 649 
 650 void TemplateTable::lload(int n) {
 651   transition(vtos, ltos);
 652   assert(n+1 < Argument::n_register_parameters, "would need more code");
 653   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
 654 }
 655 
 656 
 657 void TemplateTable::fload(int n) {
 658   transition(vtos, ftos);
 659   assert(n < Argument::n_register_parameters, "would need more code");
 660   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
 661 }
 662 
 663 
 664 void TemplateTable::dload(int n) {
 665   transition(vtos, dtos);
 666   FloatRegister dst = Ftos_d;
 667   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
 668 }
 669 
 670 
 671 void TemplateTable::aload(int n) {
 672   transition(vtos, atos);
 673   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 674 }
 675 
 676 
 677 void TemplateTable::aload_0() {
 678   transition(vtos, atos);
 679 
 680   // According to bytecode histograms, the pairs:
 681   //
 682   // _aload_0, _fast_igetfield (itos)
 683   // _aload_0, _fast_agetfield (atos)
 684   // _aload_0, _fast_fgetfield (ftos)
 685   //
 686   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
 687   // bytecode checks the next bytecode and then rewrites the current
 688   // bytecode into a pair bytecode; otherwise it rewrites the current
 689   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
 690   //
 691   if (RewriteFrequentPairs) {
 692     Label rewrite, done;
 693 
 694     // get next byte
 695     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
 696 
 697     // do actual aload_0
 698     aload(0);
 699 
 700     // if _getfield then wait with rewrite
 701     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_getfield, Assembler::equal, Assembler::pn, done);
 702 
 703     // if _igetfield then rewrite to _fast_iaccess_0
 704     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 705     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
 706     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 707     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
 708 
 709     // if _agetfield then rewrite to _fast_aaccess_0
 710     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 711     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
 712     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 713     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
 714 
 715     // if _fgetfield then rewrite to _fast_faccess_0
 716     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 717     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
 718     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 719     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
 720 
 721     // else rewrite to _fast_aload0
 722     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 723     __ set(Bytecodes::_fast_aload_0, G4_scratch);
 724 
 725     // rewrite
 726     // G4_scratch: fast bytecode
 727     __ bind(rewrite);
 728     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
 729     __ bind(done);
 730   } else {
 731     aload(0);
 732   }
 733 }
 734 
 735 
 736 void TemplateTable::istore() {
 737   transition(itos, vtos);
 738   locals_index(G3_scratch);
 739   __ store_local_int( G3_scratch, Otos_i );
 740 }
 741 
 742 
 743 void TemplateTable::lstore() {
 744   transition(ltos, vtos);
 745   locals_index(G3_scratch);
 746   __ store_local_long( G3_scratch, Otos_l );
 747 }
 748 
 749 
 750 void TemplateTable::fstore() {
 751   transition(ftos, vtos);
 752   locals_index(G3_scratch);
 753   __ store_local_float( G3_scratch, Ftos_f );
 754 }
 755 
 756 
 757 void TemplateTable::dstore() {
 758   transition(dtos, vtos);
 759   locals_index(G3_scratch);
 760   __ store_local_double( G3_scratch, Ftos_d );
 761 }
 762 
 763 
 764 void TemplateTable::astore() {
 765   transition(vtos, vtos);
 766   __ load_ptr(0, Otos_i);
 767   __ inc(Lesp, Interpreter::stackElementSize);
 768   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 769   locals_index(G3_scratch);
 770   __ store_local_ptr(G3_scratch, Otos_i);
 771 }
 772 
 773 
 774 void TemplateTable::wide_istore() {
 775   transition(vtos, vtos);
 776   __ pop_i();
 777   locals_index_wide(G3_scratch);
 778   __ store_local_int( G3_scratch, Otos_i );
 779 }
 780 
 781 
 782 void TemplateTable::wide_lstore() {
 783   transition(vtos, vtos);
 784   __ pop_l();
 785   locals_index_wide(G3_scratch);
 786   __ store_local_long( G3_scratch, Otos_l );
 787 }
 788 
 789 
 790 void TemplateTable::wide_fstore() {
 791   transition(vtos, vtos);
 792   __ pop_f();
 793   locals_index_wide(G3_scratch);
 794   __ store_local_float( G3_scratch, Ftos_f );
 795 }
 796 
 797 
 798 void TemplateTable::wide_dstore() {
 799   transition(vtos, vtos);
 800   __ pop_d();
 801   locals_index_wide(G3_scratch);
 802   __ store_local_double( G3_scratch, Ftos_d );
 803 }
 804 
 805 
 806 void TemplateTable::wide_astore() {
 807   transition(vtos, vtos);
 808   __ load_ptr(0, Otos_i);
 809   __ inc(Lesp, Interpreter::stackElementSize);
 810   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 811   locals_index_wide(G3_scratch);
 812   __ store_local_ptr(G3_scratch, Otos_i);
 813 }
 814 
 815 
 816 void TemplateTable::iastore() {
 817   transition(itos, vtos);
 818   __ pop_i(O2); // index
 819   // Otos_i: val
 820   // O3: array
 821   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 822   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
 823 }
 824 
 825 
 826 void TemplateTable::lastore() {
 827   transition(ltos, vtos);
 828   __ pop_i(O2); // index
 829   // Otos_l: val
 830   // O3: array
 831   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 832   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
 833 }
 834 
 835 
 836 void TemplateTable::fastore() {
 837   transition(ftos, vtos);
 838   __ pop_i(O2); // index
 839   // Ftos_f: val
 840   // O3: array
 841   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 842   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
 843 }
 844 
 845 
 846 void TemplateTable::dastore() {
 847   transition(dtos, vtos);
 848   __ pop_i(O2); // index
 849   // Fos_d: val
 850   // O3: array
 851   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 852   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
 853 }
 854 
 855 
 856 void TemplateTable::aastore() {
 857   Label store_ok, is_null, done;
 858   transition(vtos, vtos);
 859   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
 860   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
 861   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
 862   // Otos_i: val
 863   // O2: index
 864   // O3: array
 865   __ verify_oop(Otos_i);
 866   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
 867 
 868   // do array store check - check for NULL value first
 869   __ br_null_short( Otos_i, Assembler::pn, is_null );
 870 
 871   __ load_klass(O3, O4); // get array klass
 872   __ load_klass(Otos_i, O5); // get value klass
 873 
 874   // do fast instanceof cache test
 875 
 876   __ ld_ptr(O4,     in_bytes(ObjArrayKlass::element_klass_offset()),  O4);
 877 
 878   assert(Otos_i == O0, "just checking");
 879 
 880   // Otos_i:    value
 881   // O1:        addr - offset
 882   // O2:        index
 883   // O3:        array
 884   // O4:        array element klass
 885   // O5:        value klass
 886 
 887   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
 888 
 889   // Generate a fast subtype check.  Branch to store_ok if no
 890   // failure.  Throw if failure.
 891   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
 892 
 893   // Not a subtype; so must throw exception
 894   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
 895 
 896   // Store is OK.
 897   __ bind(store_ok);
 898   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
 899 
 900   __ ba(done);
 901   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
 902 
 903   __ bind(is_null);
 904   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
 905 
 906   __ profile_null_seen(G3_scratch);
 907   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
 908   __ bind(done);
 909 }
 910 
 911 
 912 void TemplateTable::bastore() {
 913   transition(itos, vtos);
 914   __ pop_i(O2); // index
 915   // Otos_i: val
 916   // O3: array
 917   __ index_check(O3, O2, 0, G3_scratch, O2);
 918   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
 919 }
 920 
 921 
 922 void TemplateTable::castore() {
 923   transition(itos, vtos);
 924   __ pop_i(O2); // index
 925   // Otos_i: val
 926   // O3: array
 927   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
 928   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
 929 }
 930 
 931 
 932 void TemplateTable::sastore() {
 933   // %%%%% Factor across platform
 934   castore();
 935 }
 936 
 937 
 938 void TemplateTable::istore(int n) {
 939   transition(itos, vtos);
 940   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
 941 }
 942 
 943 
 944 void TemplateTable::lstore(int n) {
 945   transition(ltos, vtos);
 946   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
 947   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
 948 
 949 }
 950 
 951 
 952 void TemplateTable::fstore(int n) {
 953   transition(ftos, vtos);
 954   assert(n < Argument::n_register_parameters, "only handle register cases");
 955   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
 956 }
 957 
 958 
 959 void TemplateTable::dstore(int n) {
 960   transition(dtos, vtos);
 961   FloatRegister src = Ftos_d;
 962   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
 963 }
 964 
 965 
 966 void TemplateTable::astore(int n) {
 967   transition(vtos, vtos);
 968   __ load_ptr(0, Otos_i);
 969   __ inc(Lesp, Interpreter::stackElementSize);
 970   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 971   __ store_local_ptr(n, Otos_i);
 972 }
 973 
 974 
 975 void TemplateTable::pop() {
 976   transition(vtos, vtos);
 977   __ inc(Lesp, Interpreter::stackElementSize);
 978 }
 979 
 980 
 981 void TemplateTable::pop2() {
 982   transition(vtos, vtos);
 983   __ inc(Lesp, 2 * Interpreter::stackElementSize);
 984 }
 985 
 986 
 987 void TemplateTable::dup() {
 988   transition(vtos, vtos);
 989   // stack: ..., a
 990   // load a and tag
 991   __ load_ptr(0, Otos_i);
 992   __ push_ptr(Otos_i);
 993   // stack: ..., a, a
 994 }
 995 
 996 
 997 void TemplateTable::dup_x1() {
 998   transition(vtos, vtos);
 999   // stack: ..., a, b
1000   __ load_ptr( 1, G3_scratch);  // get a
1001   __ load_ptr( 0, Otos_l1);     // get b
1002   __ store_ptr(1, Otos_l1);     // put b
1003   __ store_ptr(0, G3_scratch);  // put a - like swap
1004   __ push_ptr(Otos_l1);         // push b
1005   // stack: ..., b, a, b
1006 }
1007 
1008 
1009 void TemplateTable::dup_x2() {
1010   transition(vtos, vtos);
1011   // stack: ..., a, b, c
1012   // get c and push on stack, reuse registers
1013   __ load_ptr( 0, G3_scratch);  // get c
1014   __ push_ptr(G3_scratch);      // push c with tag
1015   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
1016   // (stack offsets n+1 now)
1017   __ load_ptr( 3, Otos_l1);     // get a
1018   __ store_ptr(3, G3_scratch);  // put c at 3
1019   // stack: ..., c, b, c, c  (a in reg)
1020   __ load_ptr( 2, G3_scratch);  // get b
1021   __ store_ptr(2, Otos_l1);     // put a at 2
1022   // stack: ..., c, a, c, c  (b in reg)
1023   __ store_ptr(1, G3_scratch);  // put b at 1
1024   // stack: ..., c, a, b, c
1025 }
1026 
1027 
1028 void TemplateTable::dup2() {
1029   transition(vtos, vtos);
1030   __ load_ptr(1, G3_scratch);  // get a
1031   __ load_ptr(0, Otos_l1);     // get b
1032   __ push_ptr(G3_scratch);     // push a
1033   __ push_ptr(Otos_l1);        // push b
1034   // stack: ..., a, b, a, b
1035 }
1036 
1037 
1038 void TemplateTable::dup2_x1() {
1039   transition(vtos, vtos);
1040   // stack: ..., a, b, c
1041   __ load_ptr( 1, Lscratch);    // get b
1042   __ load_ptr( 2, Otos_l1);     // get a
1043   __ store_ptr(2, Lscratch);    // put b at a
1044   // stack: ..., b, b, c
1045   __ load_ptr( 0, G3_scratch);  // get c
1046   __ store_ptr(1, G3_scratch);  // put c at b
1047   // stack: ..., b, c, c
1048   __ store_ptr(0, Otos_l1);     // put a at c
1049   // stack: ..., b, c, a
1050   __ push_ptr(Lscratch);        // push b
1051   __ push_ptr(G3_scratch);      // push c
1052   // stack: ..., b, c, a, b, c
1053 }
1054 
1055 
1056 // The spec says that these types can be a mixture of category 1 (1 word)
1057 // types and/or category 2 types (long and doubles)
1058 void TemplateTable::dup2_x2() {
1059   transition(vtos, vtos);
1060   // stack: ..., a, b, c, d
1061   __ load_ptr( 1, Lscratch);    // get c
1062   __ load_ptr( 3, Otos_l1);     // get a
1063   __ store_ptr(3, Lscratch);    // put c at 3
1064   __ store_ptr(1, Otos_l1);     // put a at 1
1065   // stack: ..., c, b, a, d
1066   __ load_ptr( 2, G3_scratch);  // get b
1067   __ load_ptr( 0, Otos_l1);     // get d
1068   __ store_ptr(0, G3_scratch);  // put b at 0
1069   __ store_ptr(2, Otos_l1);     // put d at 2
1070   // stack: ..., c, d, a, b
1071   __ push_ptr(Lscratch);        // push c
1072   __ push_ptr(Otos_l1);         // push d
1073   // stack: ..., c, d, a, b, c, d
1074 }
1075 
1076 
1077 void TemplateTable::swap() {
1078   transition(vtos, vtos);
1079   // stack: ..., a, b
1080   __ load_ptr( 1, G3_scratch);  // get a
1081   __ load_ptr( 0, Otos_l1);     // get b
1082   __ store_ptr(0, G3_scratch);  // put b
1083   __ store_ptr(1, Otos_l1);     // put a
1084   // stack: ..., b, a
1085 }
1086 
1087 
1088 void TemplateTable::iop2(Operation op) {
1089   transition(itos, itos);
1090   __ pop_i(O1);
1091   switch (op) {
1092    case  add:  __  add(O1, Otos_i, Otos_i);  break;
1093    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
1094      // %%%%% Mul may not exist: better to call .mul?
1095    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
1096    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
1097    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
1098    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
1099    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
1100    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
1101    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
1102    default: ShouldNotReachHere();
1103   }
1104 }
1105 
1106 
1107 void TemplateTable::lop2(Operation op) {
1108   transition(ltos, ltos);
1109   __ pop_l(O2);
1110   switch (op) {
1111 #ifdef _LP64
1112    case  add:  __  add(O2, Otos_l, Otos_l);  break;
1113    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
1114    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
1115    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
1116    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
1117 #else
1118    case  add:  __ addcc(O3, Otos_l2, Otos_l2);  __ addc(O2, Otos_l1, Otos_l1);  break;
1119    case  sub:  __ subcc(O3, Otos_l2, Otos_l2);  __ subc(O2, Otos_l1, Otos_l1);  break;
1120    case _and:  __  and3(O3, Otos_l2, Otos_l2);  __ and3(O2, Otos_l1, Otos_l1);  break;
1121    case  _or:  __   or3(O3, Otos_l2, Otos_l2);  __  or3(O2, Otos_l1, Otos_l1);  break;
1122    case _xor:  __  xor3(O3, Otos_l2, Otos_l2);  __ xor3(O2, Otos_l1, Otos_l1);  break;
1123 #endif
1124    default: ShouldNotReachHere();
1125   }
1126 }
1127 
1128 
1129 void TemplateTable::idiv() {
1130   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
1131   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
1132 
1133   transition(itos, itos);
1134   __ pop_i(O1); // get 1st op
1135 
1136   // Y contains upper 32 bits of result, set it to 0 or all ones
1137   __ wry(G0);
1138   __ mov(~0, G3_scratch);
1139 
1140   __ tst(O1);
1141      Label neg;
1142   __ br(Assembler::negative, true, Assembler::pn, neg);
1143   __ delayed()->wry(G3_scratch);
1144   __ bind(neg);
1145 
1146      Label ok;
1147   __ tst(Otos_i);
1148   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
1149 
1150   const int min_int = 0x80000000;
1151   Label regular;
1152   __ cmp(Otos_i, -1);
1153   __ br(Assembler::notEqual, false, Assembler::pt, regular);
1154 #ifdef _LP64
1155   // Don't put set in delay slot
1156   // Set will turn into multiple instructions in 64 bit mode
1157   __ delayed()->nop();
1158   __ set(min_int, G4_scratch);
1159 #else
1160   __ delayed()->set(min_int, G4_scratch);
1161 #endif
1162   Label done;
1163   __ cmp(O1, G4_scratch);
1164   __ br(Assembler::equal, true, Assembler::pt, done);
1165   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
1166 
1167   __ bind(regular);
1168   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
1169   __ bind(done);
1170 }
1171 
1172 
1173 void TemplateTable::irem() {
1174   transition(itos, itos);
1175   __ mov(Otos_i, O2); // save divisor
1176   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
1177   __ smul(Otos_i, O2, Otos_i);
1178   __ sub(O1, Otos_i, Otos_i);
1179 }
1180 
1181 
1182 void TemplateTable::lmul() {
1183   transition(ltos, ltos);
1184   __ pop_l(O2);
1185 #ifdef _LP64
1186   __ mulx(Otos_l, O2, Otos_l);
1187 #else
1188   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lmul));
1189 #endif
1190 
1191 }
1192 
1193 
1194 void TemplateTable::ldiv() {
1195   transition(ltos, ltos);
1196 
1197   // check for zero
1198   __ pop_l(O2);
1199 #ifdef _LP64
1200   __ tst(Otos_l);
1201   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1202   __ sdivx(O2, Otos_l, Otos_l);
1203 #else
1204   __ orcc(Otos_l1, Otos_l2, G0);
1205   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1206   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1207 #endif
1208 }
1209 
1210 
1211 void TemplateTable::lrem() {
1212   transition(ltos, ltos);
1213 
1214   // check for zero
1215   __ pop_l(O2);
1216 #ifdef _LP64
1217   __ tst(Otos_l);
1218   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1219   __ sdivx(O2, Otos_l, Otos_l2);
1220   __ mulx (Otos_l2, Otos_l, Otos_l2);
1221   __ sub  (O2, Otos_l2, Otos_l);
1222 #else
1223   __ orcc(Otos_l1, Otos_l2, G0);
1224   __ throw_if_not_icc(Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1225   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1226 #endif
1227 }
1228 
1229 
1230 void TemplateTable::lshl() {
1231   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
1232 
1233   __ pop_l(O2);                          // shift value in O2, O3
1234 #ifdef _LP64
1235   __ sllx(O2, Otos_i, Otos_l);
1236 #else
1237   __ lshl(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1238 #endif
1239 }
1240 
1241 
1242 void TemplateTable::lshr() {
1243   transition(itos, ltos); // %%%% see lshl comment
1244 
1245   __ pop_l(O2);                          // shift value in O2, O3
1246 #ifdef _LP64
1247   __ srax(O2, Otos_i, Otos_l);
1248 #else
1249   __ lshr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1250 #endif
1251 }
1252 
1253 
1254 
1255 void TemplateTable::lushr() {
1256   transition(itos, ltos); // %%%% see lshl comment
1257 
1258   __ pop_l(O2);                          // shift value in O2, O3
1259 #ifdef _LP64
1260   __ srlx(O2, Otos_i, Otos_l);
1261 #else
1262   __ lushr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1263 #endif
1264 }
1265 
1266 
1267 void TemplateTable::fop2(Operation op) {
1268   transition(ftos, ftos);
1269   switch (op) {
1270    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1271    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1272    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1273    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1274    case  rem:
1275      assert(Ftos_f == F0, "just checking");
1276 #ifdef _LP64
1277      // LP64 calling conventions use F1, F3 for passing 2 floats
1278      __ pop_f(F1);
1279      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
1280 #else
1281      __ pop_i(O0);
1282      __ stf(FloatRegisterImpl::S, Ftos_f, __ d_tmp);
1283      __ ld( __ d_tmp, O1 );
1284 #endif
1285      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1286      assert( Ftos_f == F0, "fix this code" );
1287      break;
1288 
1289    default: ShouldNotReachHere();
1290   }
1291 }
1292 
1293 
1294 void TemplateTable::dop2(Operation op) {
1295   transition(dtos, dtos);
1296   switch (op) {
1297    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1298    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1299    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1300    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1301    case  rem:
1302 #ifdef _LP64
1303      // Pass arguments in D0, D2
1304      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
1305      __ pop_d( F0 );
1306 #else
1307      // Pass arguments in O0O1, O2O3
1308      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
1309      __ ldd( __ d_tmp, O2 );
1310      __ pop_d(Ftos_f);
1311      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
1312      __ ldd( __ d_tmp, O0 );
1313 #endif
1314      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1315      assert( Ftos_d == F0, "fix this code" );
1316      break;
1317 
1318    default: ShouldNotReachHere();
1319   }
1320 }
1321 
1322 
1323 void TemplateTable::ineg() {
1324   transition(itos, itos);
1325   __ neg(Otos_i);
1326 }
1327 
1328 
1329 void TemplateTable::lneg() {
1330   transition(ltos, ltos);
1331 #ifdef _LP64
1332   __ sub(G0, Otos_l, Otos_l);
1333 #else
1334   __ lneg(Otos_l1, Otos_l2);
1335 #endif
1336 }
1337 
1338 
1339 void TemplateTable::fneg() {
1340   transition(ftos, ftos);
1341   __ fneg(FloatRegisterImpl::S, Ftos_f);
1342 }
1343 
1344 
1345 void TemplateTable::dneg() {
1346   transition(dtos, dtos);
1347   // v8 has fnegd if source and dest are the same
1348   __ fneg(FloatRegisterImpl::D, Ftos_f);
1349 }
1350 
1351 
1352 void TemplateTable::iinc() {
1353   transition(vtos, vtos);
1354   locals_index(G3_scratch);
1355   __ ldsb(Lbcp, 2, O2);  // load constant
1356   __ access_local_int(G3_scratch, Otos_i);
1357   __ add(Otos_i, O2, Otos_i);
1358   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1359 }
1360 
1361 
1362 void TemplateTable::wide_iinc() {
1363   transition(vtos, vtos);
1364   locals_index_wide(G3_scratch);
1365   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
1366   __ access_local_int(G3_scratch, Otos_i);
1367   __ add(Otos_i, O3, Otos_i);
1368   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1369 }
1370 
1371 
1372 void TemplateTable::convert() {
1373 // %%%%% Factor this first part accross platforms
1374   #ifdef ASSERT
1375     TosState tos_in  = ilgl;
1376     TosState tos_out = ilgl;
1377     switch (bytecode()) {
1378       case Bytecodes::_i2l: // fall through
1379       case Bytecodes::_i2f: // fall through
1380       case Bytecodes::_i2d: // fall through
1381       case Bytecodes::_i2b: // fall through
1382       case Bytecodes::_i2c: // fall through
1383       case Bytecodes::_i2s: tos_in = itos; break;
1384       case Bytecodes::_l2i: // fall through
1385       case Bytecodes::_l2f: // fall through
1386       case Bytecodes::_l2d: tos_in = ltos; break;
1387       case Bytecodes::_f2i: // fall through
1388       case Bytecodes::_f2l: // fall through
1389       case Bytecodes::_f2d: tos_in = ftos; break;
1390       case Bytecodes::_d2i: // fall through
1391       case Bytecodes::_d2l: // fall through
1392       case Bytecodes::_d2f: tos_in = dtos; break;
1393       default             : ShouldNotReachHere();
1394     }
1395     switch (bytecode()) {
1396       case Bytecodes::_l2i: // fall through
1397       case Bytecodes::_f2i: // fall through
1398       case Bytecodes::_d2i: // fall through
1399       case Bytecodes::_i2b: // fall through
1400       case Bytecodes::_i2c: // fall through
1401       case Bytecodes::_i2s: tos_out = itos; break;
1402       case Bytecodes::_i2l: // fall through
1403       case Bytecodes::_f2l: // fall through
1404       case Bytecodes::_d2l: tos_out = ltos; break;
1405       case Bytecodes::_i2f: // fall through
1406       case Bytecodes::_l2f: // fall through
1407       case Bytecodes::_d2f: tos_out = ftos; break;
1408       case Bytecodes::_i2d: // fall through
1409       case Bytecodes::_l2d: // fall through
1410       case Bytecodes::_f2d: tos_out = dtos; break;
1411       default             : ShouldNotReachHere();
1412     }
1413     transition(tos_in, tos_out);
1414   #endif
1415 
1416 
1417   // Conversion
1418   Label done;
1419   switch (bytecode()) {
1420    case Bytecodes::_i2l:
1421 #ifdef _LP64
1422     // Sign extend the 32 bits
1423     __ sra ( Otos_i, 0, Otos_l );
1424 #else
1425     __ addcc(Otos_i, 0, Otos_l2);
1426     __ br(Assembler::greaterEqual, true, Assembler::pt, done);
1427     __ delayed()->clr(Otos_l1);
1428     __ set(~0, Otos_l1);
1429 #endif
1430     break;
1431 
1432    case Bytecodes::_i2f:
1433     __ st(Otos_i, __ d_tmp );
1434     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1435     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
1436     break;
1437 
1438    case Bytecodes::_i2d:
1439     __ st(Otos_i, __ d_tmp);
1440     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1441     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
1442     break;
1443 
1444    case Bytecodes::_i2b:
1445     __ sll(Otos_i, 24, Otos_i);
1446     __ sra(Otos_i, 24, Otos_i);
1447     break;
1448 
1449    case Bytecodes::_i2c:
1450     __ sll(Otos_i, 16, Otos_i);
1451     __ srl(Otos_i, 16, Otos_i);
1452     break;
1453 
1454    case Bytecodes::_i2s:
1455     __ sll(Otos_i, 16, Otos_i);
1456     __ sra(Otos_i, 16, Otos_i);
1457     break;
1458 
1459    case Bytecodes::_l2i:
1460 #ifndef _LP64
1461     __ mov(Otos_l2, Otos_i);
1462 #else
1463     // Sign-extend into the high 32 bits
1464     __ sra(Otos_l, 0, Otos_i);
1465 #endif
1466     break;
1467 
1468    case Bytecodes::_l2f:
1469    case Bytecodes::_l2d:
1470     __ st_long(Otos_l, __ d_tmp);
1471     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
1472 
1473     if (VM_Version::v9_instructions_work()) {
1474       if (bytecode() == Bytecodes::_l2f) {
1475         __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
1476       } else {
1477         __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
1478       }
1479     } else {
1480       __ call_VM_leaf(
1481         Lscratch,
1482         bytecode() == Bytecodes::_l2f
1483           ? CAST_FROM_FN_PTR(address, SharedRuntime::l2f)
1484           : CAST_FROM_FN_PTR(address, SharedRuntime::l2d)
1485       );
1486     }
1487     break;
1488 
1489   case Bytecodes::_f2i:  {
1490       Label isNaN;
1491       // result must be 0 if value is NaN; test by comparing value to itself
1492       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
1493       // According to the v8 manual, you have to have a non-fp instruction
1494       // between fcmp and fb.
1495       if (!VM_Version::v9_instructions_work()) {
1496         __ nop();
1497       }
1498       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
1499       __ delayed()->clr(Otos_i);                                     // NaN
1500       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
1501       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
1502       __ ld(__ d_tmp, Otos_i);
1503       __ bind(isNaN);
1504     }
1505     break;
1506 
1507    case Bytecodes::_f2l:
1508     // must uncache tos
1509     __ push_f();
1510 #ifdef _LP64
1511     __ pop_f(F1);
1512 #else
1513     __ pop_i(O0);
1514 #endif
1515     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1516     break;
1517 
1518    case Bytecodes::_f2d:
1519     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
1520     break;
1521 
1522    case Bytecodes::_d2i:
1523    case Bytecodes::_d2l:
1524     // must uncache tos
1525     __ push_d();
1526 #ifdef _LP64
1527     // LP64 calling conventions pass first double arg in D0
1528     __ pop_d( Ftos_d );
1529 #else
1530     __ pop_i( O0 );
1531     __ pop_i( O1 );
1532 #endif
1533     __ call_VM_leaf(Lscratch,
1534         bytecode() == Bytecodes::_d2i
1535           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
1536           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1537     break;
1538 
1539     case Bytecodes::_d2f:
1540     if (VM_Version::v9_instructions_work()) {
1541       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
1542     }
1543     else {
1544       // must uncache tos
1545       __ push_d();
1546       __ pop_i(O0);
1547       __ pop_i(O1);
1548       __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::d2f));
1549     }
1550     break;
1551 
1552     default: ShouldNotReachHere();
1553   }
1554   __ bind(done);
1555 }
1556 
1557 
1558 void TemplateTable::lcmp() {
1559   transition(ltos, itos);
1560 
1561 #ifdef _LP64
1562   __ pop_l(O1); // pop off value 1, value 2 is in O0
1563   __ lcmp( O1, Otos_l, Otos_i );
1564 #else
1565   __ pop_l(O2); // cmp O2,3 to O0,1
1566   __ lcmp( O2, O3, Otos_l1, Otos_l2, Otos_i );
1567 #endif
1568 }
1569 
1570 
1571 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1572 
1573   if (is_float) __ pop_f(F2);
1574   else          __ pop_d(F2);
1575 
1576   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
1577 
1578   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
1579 }
1580 
1581 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1582   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1583   __ verify_thread();
1584 
1585   const Register O2_bumped_count = O2;
1586   __ profile_taken_branch(G3_scratch, O2_bumped_count);
1587 
1588   // get (wide) offset to O1_disp
1589   const Register O1_disp = O1;
1590   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
1591   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
1592 
1593   // Handle all the JSR stuff here, then exit.
1594   // It's much shorter and cleaner than intermingling with the
1595   // non-JSR normal-branch stuff occurring below.
1596   if( is_jsr ) {
1597     // compute return address as bci in Otos_i
1598     __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1599     __ sub(Lbcp, G3_scratch, G3_scratch);
1600     __ sub(G3_scratch, in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
1601 
1602     // Bump Lbcp to target of JSR
1603     __ add(Lbcp, O1_disp, Lbcp);
1604     // Push returnAddress for "ret" on stack
1605     __ push_ptr(Otos_i);
1606     // And away we go!
1607     __ dispatch_next(vtos);
1608     return;
1609   }
1610 
1611   // Normal (non-jsr) branch handling
1612 
1613   // Save the current Lbcp
1614   const Register l_cur_bcp = Lscratch;
1615   __ mov( Lbcp, l_cur_bcp );
1616 
1617   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1618   if ( increment_invocation_counter_for_backward_branches ) {
1619     Label Lforward;
1620     // check branch direction
1621     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
1622     // Bump bytecode pointer by displacement (take the branch)
1623     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
1624 
1625     const Register Rcounters = G3_scratch;
1626     __ get_method_counters(Lmethod, Rcounters, Lforward);
1627 
1628     if (TieredCompilation) {
1629       Label Lno_mdo, Loverflow;
1630       int increment = InvocationCounter::count_increment;
1631       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1632       if (ProfileInterpreter) {
1633         // If no method data exists, go to profile_continue.
1634         __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch);
1635         __ br_null_short(G4_scratch, Assembler::pn, Lno_mdo);
1636 
1637         // Increment backedge counter in the MDO
1638         Address mdo_backedge_counter(G4_scratch, in_bytes(MethodData::backedge_counter_offset()) +
1639                                                  in_bytes(InvocationCounter::counter_offset()));
1640         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, O0,
1641                                    Assembler::notZero, &Lforward);
1642         __ ba_short(Loverflow);
1643       }
1644 
1645       // If there's no MDO, increment counter in MethodCounters*
1646       __ bind(Lno_mdo);
1647       Address backedge_counter(Rcounters,
1648               in_bytes(MethodCounters::backedge_counter_offset()) +
1649               in_bytes(InvocationCounter::counter_offset()));
1650       __ increment_mask_and_jump(backedge_counter, increment, mask, G4_scratch, O0,
1651                                  Assembler::notZero, &Lforward);
1652       __ bind(Loverflow);
1653 
1654       // notify point for loop, pass branch bytecode
1655       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), l_cur_bcp);
1656 
1657       // Was an OSR adapter generated?
1658       // O0 = osr nmethod
1659       __ br_null_short(O0, Assembler::pn, Lforward);
1660 
1661       // Has the nmethod been invalidated already?
1662       __ ld(O0, nmethod::entry_bci_offset(), O2);
1663       __ cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, Lforward);
1664 
1665       // migrate the interpreter frame off of the stack
1666 
1667       __ mov(G2_thread, L7);
1668       // save nmethod
1669       __ mov(O0, L6);
1670       __ set_last_Java_frame(SP, noreg);
1671       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
1672       __ reset_last_Java_frame();
1673       __ mov(L7, G2_thread);
1674 
1675       // move OSR nmethod to I1
1676       __ mov(L6, I1);
1677 
1678       // OSR buffer to I0
1679       __ mov(O0, I0);
1680 
1681       // remove the interpreter frame
1682       __ restore(I5_savedSP, 0, SP);
1683 
1684       // Jump to the osr code.
1685       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
1686       __ jmp(O2, G0);
1687       __ delayed()->nop();
1688 
1689     } else {
1690       // Update Backedge branch separately from invocations
1691       const Register G4_invoke_ctr = G4;
1692       __ increment_backedge_counter(Rcounters, G4_invoke_ctr, G1_scratch);
1693       if (ProfileInterpreter) {
1694         __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_scratch, Lforward);
1695         if (UseOnStackReplacement) {
1696           __ test_backedge_count_for_osr(O2_bumped_count, l_cur_bcp, G3_scratch);
1697         }
1698       } else {
1699         if (UseOnStackReplacement) {
1700           __ test_backedge_count_for_osr(G4_invoke_ctr, l_cur_bcp, G3_scratch);
1701         }
1702       }
1703     }
1704 
1705     __ bind(Lforward);
1706   } else
1707     // Bump bytecode pointer by displacement (take the branch)
1708     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
1709 
1710   // continue with bytecode @ target
1711   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1712   // %%%%% and changing dispatch_next to dispatch_only
1713   __ dispatch_next(vtos);
1714 }
1715 
1716 
1717 // Note Condition in argument is TemplateTable::Condition
1718 // arg scope is within class scope
1719 
1720 void TemplateTable::if_0cmp(Condition cc) {
1721   // no pointers, integer only!
1722   transition(itos, vtos);
1723   // assume branch is more often taken than not (loops use backward branches)
1724   __ cmp( Otos_i, 0);
1725   __ if_cmp(ccNot(cc), false);
1726 }
1727 
1728 
1729 void TemplateTable::if_icmp(Condition cc) {
1730   transition(itos, vtos);
1731   __ pop_i(O1);
1732   __ cmp(O1, Otos_i);
1733   __ if_cmp(ccNot(cc), false);
1734 }
1735 
1736 
1737 void TemplateTable::if_nullcmp(Condition cc) {
1738   transition(atos, vtos);
1739   __ tst(Otos_i);
1740   __ if_cmp(ccNot(cc), true);
1741 }
1742 
1743 
1744 void TemplateTable::if_acmp(Condition cc) {
1745   transition(atos, vtos);
1746   __ pop_ptr(O1);
1747   __ verify_oop(O1);
1748   __ verify_oop(Otos_i);
1749   __ cmp(O1, Otos_i);
1750   __ if_cmp(ccNot(cc), true);
1751 }
1752 
1753 
1754 
1755 void TemplateTable::ret() {
1756   transition(vtos, vtos);
1757   locals_index(G3_scratch);
1758   __ access_local_returnAddress(G3_scratch, Otos_i);
1759   // Otos_i contains the bci, compute the bcp from that
1760 
1761 #ifdef _LP64
1762 #ifdef ASSERT
1763   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
1764   // the result.  The return address (really a BCI) was stored with an
1765   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
1766   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
1767   // loaded value.
1768   { Label zzz ;
1769      __ set (65536, G3_scratch) ;
1770      __ cmp (Otos_i, G3_scratch) ;
1771      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
1772      __ delayed()->nop();
1773      __ stop("BCI is in the wrong register half?");
1774      __ bind (zzz) ;
1775   }
1776 #endif
1777 #endif
1778 
1779   __ profile_ret(vtos, Otos_i, G4_scratch);
1780 
1781   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1782   __ add(G3_scratch, Otos_i, G3_scratch);
1783   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1784   __ dispatch_next(vtos);
1785 }
1786 
1787 
1788 void TemplateTable::wide_ret() {
1789   transition(vtos, vtos);
1790   locals_index_wide(G3_scratch);
1791   __ access_local_returnAddress(G3_scratch, Otos_i);
1792   // Otos_i contains the bci, compute the bcp from that
1793 
1794   __ profile_ret(vtos, Otos_i, G4_scratch);
1795 
1796   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1797   __ add(G3_scratch, Otos_i, G3_scratch);
1798   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1799   __ dispatch_next(vtos);
1800 }
1801 
1802 
1803 void TemplateTable::tableswitch() {
1804   transition(itos, vtos);
1805   Label default_case, continue_execution;
1806 
1807   // align bcp
1808   __ add(Lbcp, BytesPerInt, O1);
1809   __ and3(O1, -BytesPerInt, O1);
1810   // load lo, hi
1811   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
1812   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
1813 #ifdef _LP64
1814   // Sign extend the 32 bits
1815   __ sra ( Otos_i, 0, Otos_i );
1816 #endif /* _LP64 */
1817 
1818   // check against lo & hi
1819   __ cmp( Otos_i, O2);
1820   __ br( Assembler::less, false, Assembler::pn, default_case);
1821   __ delayed()->cmp( Otos_i, O3 );
1822   __ br( Assembler::greater, false, Assembler::pn, default_case);
1823   // lookup dispatch offset
1824   __ delayed()->sub(Otos_i, O2, O2);
1825   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
1826   __ sll(O2, LogBytesPerInt, O2);
1827   __ add(O2, 3 * BytesPerInt, O2);
1828   __ ba(continue_execution);
1829   __ delayed()->ld(O1, O2, O2);
1830   // handle default
1831   __ bind(default_case);
1832   __ profile_switch_default(O3);
1833   __ ld(O1, 0, O2); // get default offset
1834   // continue execution
1835   __ bind(continue_execution);
1836   __ add(Lbcp, O2, Lbcp);
1837   __ dispatch_next(vtos);
1838 }
1839 
1840 
1841 void TemplateTable::lookupswitch() {
1842   transition(itos, itos);
1843   __ stop("lookupswitch bytecode should have been rewritten");
1844 }
1845 
1846 void TemplateTable::fast_linearswitch() {
1847   transition(itos, vtos);
1848     Label loop_entry, loop, found, continue_execution;
1849   // align bcp
1850   __ add(Lbcp, BytesPerInt, O1);
1851   __ and3(O1, -BytesPerInt, O1);
1852  // set counter
1853   __ ld(O1, BytesPerInt, O2);
1854   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
1855   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
1856   __ ba(loop_entry);
1857   __ delayed()->add(O3, O2, O2); // counter now points past last pair
1858 
1859   // table search
1860   __ bind(loop);
1861   __ cmp(O4, Otos_i);
1862   __ br(Assembler::equal, true, Assembler::pn, found);
1863   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
1864   __ inc(O3, 2 * BytesPerInt);
1865 
1866   __ bind(loop_entry);
1867   __ cmp(O2, O3);
1868   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
1869   __ delayed()->ld(O3, 0, O4);
1870 
1871   // default case
1872   __ ld(O1, 0, O4); // get default offset
1873   if (ProfileInterpreter) {
1874     __ profile_switch_default(O3);
1875     __ ba_short(continue_execution);
1876   }
1877 
1878   // entry found -> get offset
1879   __ bind(found);
1880   if (ProfileInterpreter) {
1881     __ sub(O3, O1, O3);
1882     __ sub(O3, 2*BytesPerInt, O3);
1883     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
1884     __ profile_switch_case(O3, O1, O2, G3_scratch);
1885 
1886     __ bind(continue_execution);
1887   }
1888   __ add(Lbcp, O4, Lbcp);
1889   __ dispatch_next(vtos);
1890 }
1891 
1892 
1893 void TemplateTable::fast_binaryswitch() {
1894   transition(itos, vtos);
1895   // Implementation using the following core algorithm: (copied from Intel)
1896   //
1897   // int binary_search(int key, LookupswitchPair* array, int n) {
1898   //   // Binary search according to "Methodik des Programmierens" by
1899   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1900   //   int i = 0;
1901   //   int j = n;
1902   //   while (i+1 < j) {
1903   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1904   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1905   //     // where a stands for the array and assuming that the (inexisting)
1906   //     // element a[n] is infinitely big.
1907   //     int h = (i + j) >> 1;
1908   //     // i < h < j
1909   //     if (key < array[h].fast_match()) {
1910   //       j = h;
1911   //     } else {
1912   //       i = h;
1913   //     }
1914   //   }
1915   //   // R: a[i] <= key < a[i+1] or Q
1916   //   // (i.e., if key is within array, i is the correct index)
1917   //   return i;
1918   // }
1919 
1920   // register allocation
1921   assert(Otos_i == O0, "alias checking");
1922   const Register Rkey     = Otos_i;                    // already set (tosca)
1923   const Register Rarray   = O1;
1924   const Register Ri       = O2;
1925   const Register Rj       = O3;
1926   const Register Rh       = O4;
1927   const Register Rscratch = O5;
1928 
1929   const int log_entry_size = 3;
1930   const int entry_size = 1 << log_entry_size;
1931 
1932   Label found;
1933   // Find Array start
1934   __ add(Lbcp, 3 * BytesPerInt, Rarray);
1935   __ and3(Rarray, -BytesPerInt, Rarray);
1936   // initialize i & j (in delay slot)
1937   __ clr( Ri );
1938 
1939   // and start
1940   Label entry;
1941   __ ba(entry);
1942   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
1943   // (Rj is already in the native byte-ordering.)
1944 
1945   // binary search loop
1946   { Label loop;
1947     __ bind( loop );
1948     // int h = (i + j) >> 1;
1949     __ sra( Rh, 1, Rh );
1950     // if (key < array[h].fast_match()) {
1951     //   j = h;
1952     // } else {
1953     //   i = h;
1954     // }
1955     __ sll( Rh, log_entry_size, Rscratch );
1956     __ ld( Rarray, Rscratch, Rscratch );
1957     // (Rscratch is already in the native byte-ordering.)
1958     __ cmp( Rkey, Rscratch );
1959     if ( VM_Version::v9_instructions_work() ) {
1960       __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
1961       __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
1962     }
1963     else {
1964       Label end_of_if;
1965       __ br( Assembler::less, true, Assembler::pt, end_of_if );
1966       __ delayed()->mov( Rh, Rj ); // if (<) Rj = Rh
1967       __ mov( Rh, Ri );            // else i = h
1968       __ bind(end_of_if);          // }
1969     }
1970 
1971     // while (i+1 < j)
1972     __ bind( entry );
1973     __ add( Ri, 1, Rscratch );
1974     __ cmp(Rscratch, Rj);
1975     __ br( Assembler::less, true, Assembler::pt, loop );
1976     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
1977   }
1978 
1979   // end of binary search, result index is i (must check again!)
1980   Label default_case;
1981   Label continue_execution;
1982   if (ProfileInterpreter) {
1983     __ mov( Ri, Rh );              // Save index in i for profiling
1984   }
1985   __ sll( Ri, log_entry_size, Ri );
1986   __ ld( Rarray, Ri, Rscratch );
1987   // (Rscratch is already in the native byte-ordering.)
1988   __ cmp( Rkey, Rscratch );
1989   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
1990   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
1991 
1992   // entry found -> j = offset
1993   __ inc( Ri, BytesPerInt );
1994   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
1995   __ ld( Rarray, Ri, Rj );
1996   // (Rj is already in the native byte-ordering.)
1997 
1998   if (ProfileInterpreter) {
1999     __ ba_short(continue_execution);
2000   }
2001 
2002   __ bind(default_case); // fall through (if not profiling)
2003   __ profile_switch_default(Ri);
2004 
2005   __ bind(continue_execution);
2006   __ add( Lbcp, Rj, Lbcp );
2007   __ dispatch_next( vtos );
2008 }
2009 
2010 
2011 void TemplateTable::_return(TosState state) {
2012   transition(state, state);
2013   assert(_desc->calls_vm(), "inconsistent calls_vm information");
2014 
2015   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2016     assert(state == vtos, "only valid state");
2017     __ mov(G0, G3_scratch);
2018     __ access_local_ptr(G3_scratch, Otos_i);
2019     __ load_klass(Otos_i, O2);
2020     __ set(JVM_ACC_HAS_FINALIZER, G3);
2021     __ ld(O2, in_bytes(Klass::access_flags_offset()), O2);
2022     __ andcc(G3, O2, G0);
2023     Label skip_register_finalizer;
2024     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
2025     __ delayed()->nop();
2026 
2027     // Call out to do finalizer registration
2028     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
2029 
2030     __ bind(skip_register_finalizer);
2031   }
2032 
2033   __ remove_activation(state, /* throw_monitor_exception */ true);
2034 
2035   // The caller's SP was adjusted upon method entry to accomodate
2036   // the callee's non-argument locals. Undo that adjustment.
2037   __ ret();                             // return to caller
2038   __ delayed()->restore(I5_savedSP, G0, SP);
2039 }
2040 
2041 
2042 // ----------------------------------------------------------------------------
2043 // Volatile variables demand their effects be made known to all CPU's in
2044 // order.  Store buffers on most chips allow reads & writes to reorder; the
2045 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2046 // memory barrier (i.e., it's not sufficient that the interpreter does not
2047 // reorder volatile references, the hardware also must not reorder them).
2048 //
2049 // According to the new Java Memory Model (JMM):
2050 // (1) All volatiles are serialized wrt to each other.
2051 // ALSO reads & writes act as aquire & release, so:
2052 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2053 // the read float up to before the read.  It's OK for non-volatile memory refs
2054 // that happen before the volatile read to float down below it.
2055 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2056 // that happen BEFORE the write float down to after the write.  It's OK for
2057 // non-volatile memory refs that happen after the volatile write to float up
2058 // before it.
2059 //
2060 // We only put in barriers around volatile refs (they are expensive), not
2061 // _between_ memory refs (that would require us to track the flavor of the
2062 // previous memory refs).  Requirements (2) and (3) require some barriers
2063 // before volatile stores and after volatile loads.  These nearly cover
2064 // requirement (1) but miss the volatile-store-volatile-load case.  This final
2065 // case is placed after volatile-stores although it could just as well go
2066 // before volatile-loads.
2067 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
2068   // Helper function to insert a is-volatile test and memory barrier
2069   // All current sparc implementations run in TSO, needing only StoreLoad
2070   if ((order_constraint & Assembler::StoreLoad) == 0) return;
2071   __ membar( order_constraint );
2072 }
2073 
2074 // ----------------------------------------------------------------------------
2075 void TemplateTable::resolve_cache_and_index(int byte_no,
2076                                             Register Rcache,
2077                                             Register index,
2078                                             size_t index_size) {
2079   // Depends on cpCacheOop layout!
2080   Label resolved;
2081 
2082     assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2083     __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, Lbyte_code, byte_no, 1, index_size);
2084     __ cmp(Lbyte_code, (int) bytecode());  // have we resolved this bytecode?
2085     __ br(Assembler::equal, false, Assembler::pt, resolved);
2086     __ delayed()->set((int)bytecode(), O1);
2087 
2088   address entry;
2089   switch (bytecode()) {
2090     case Bytecodes::_getstatic      : // fall through
2091     case Bytecodes::_putstatic      : // fall through
2092     case Bytecodes::_getfield       : // fall through
2093     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2094     case Bytecodes::_invokevirtual  : // fall through
2095     case Bytecodes::_invokespecial  : // fall through
2096     case Bytecodes::_invokestatic   : // fall through
2097     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
2098     case Bytecodes::_invokehandle   : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);  break;
2099     case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
2100     default:
2101       fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2102       break;
2103   }
2104   // first time invocation - must resolve first
2105   __ call_VM(noreg, entry, O1);
2106   // Update registers with resolved info
2107   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2108   __ bind(resolved);
2109 }
2110 
2111 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2112                                                Register method,
2113                                                Register itable_index,
2114                                                Register flags,
2115                                                bool is_invokevirtual,
2116                                                bool is_invokevfinal,
2117                                                bool is_invokedynamic) {
2118   // Uses both G3_scratch and G4_scratch
2119   Register cache = G3_scratch;
2120   Register index = G4_scratch;
2121   assert_different_registers(cache, method, itable_index);
2122 
2123   // determine constant pool cache field offsets
2124   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2125   const int method_offset = in_bytes(
2126       ConstantPoolCache::base_offset() +
2127       ((byte_no == f2_byte)
2128        ? ConstantPoolCacheEntry::f2_offset()
2129        : ConstantPoolCacheEntry::f1_offset()
2130       )
2131     );
2132   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2133                                     ConstantPoolCacheEntry::flags_offset());
2134   // access constant pool cache fields
2135   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2136                                     ConstantPoolCacheEntry::f2_offset());
2137 
2138   if (is_invokevfinal) {
2139     __ get_cache_and_index_at_bcp(cache, index, 1);
2140     __ ld_ptr(Address(cache, method_offset), method);
2141   } else {
2142     size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2143     resolve_cache_and_index(byte_no, cache, index, index_size);
2144     __ ld_ptr(Address(cache, method_offset), method);
2145   }
2146 
2147   if (itable_index != noreg) {
2148     // pick up itable or appendix index from f2 also:
2149     __ ld_ptr(Address(cache, index_offset), itable_index);
2150   }
2151   __ ld_ptr(Address(cache, flags_offset), flags);
2152 }
2153 
2154 // The Rcache register must be set before call
2155 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2156                                               Register Rcache,
2157                                               Register index,
2158                                               Register Roffset,
2159                                               Register Rflags,
2160                                               bool is_static) {
2161   assert_different_registers(Rcache, Rflags, Roffset);
2162 
2163   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2164 
2165   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2166   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2167   if (is_static) {
2168     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
2169     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2170     __ ld_ptr( Robj, mirror_offset, Robj);
2171   }
2172 }
2173 
2174 // The registers Rcache and index expected to be set before call.
2175 // Correct values of the Rcache and index registers are preserved.
2176 void TemplateTable::jvmti_post_field_access(Register Rcache,
2177                                             Register index,
2178                                             bool is_static,
2179                                             bool has_tos) {
2180   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2181 
2182   if (JvmtiExport::can_post_field_access()) {
2183     // Check to see if a field access watch has been set before we take
2184     // the time to call into the VM.
2185     Label Label1;
2186     assert_different_registers(Rcache, index, G1_scratch);
2187     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
2188     __ load_contents(get_field_access_count_addr, G1_scratch);
2189     __ cmp_and_br_short(G1_scratch, 0, Assembler::equal, Assembler::pt, Label1);
2190 
2191     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
2192 
2193     if (is_static) {
2194       __ clr(Otos_i);
2195     } else {
2196       if (has_tos) {
2197       // save object pointer before call_VM() clobbers it
2198         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
2199       } else {
2200         // Load top of stack (do not pop the value off the stack);
2201         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
2202       }
2203       __ verify_oop(Otos_i);
2204     }
2205     // Otos_i: object pointer or NULL if static
2206     // Rcache: cache entry pointer
2207     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2208                Otos_i, Rcache);
2209     if (!is_static && has_tos) {
2210       __ pop_ptr(Otos_i);  // restore object pointer
2211       __ verify_oop(Otos_i);
2212     }
2213     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2214     __ bind(Label1);
2215   }
2216 }
2217 
2218 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2219   transition(vtos, vtos);
2220 
2221   Register Rcache = G3_scratch;
2222   Register index  = G4_scratch;
2223   Register Rclass = Rcache;
2224   Register Roffset= G4_scratch;
2225   Register Rflags = G1_scratch;
2226   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2227 
2228   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2229   jvmti_post_field_access(Rcache, index, is_static, false);
2230   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2231 
2232   if (!is_static) {
2233     pop_and_check_object(Rclass);
2234   } else {
2235     __ verify_oop(Rclass);
2236   }
2237 
2238   Label exit;
2239 
2240   Assembler::Membar_mask_bits membar_bits =
2241     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2242 
2243   if (__ membar_has_effect(membar_bits)) {
2244     // Get volatile flag
2245     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2246     __ and3(Rflags, Lscratch, Lscratch);
2247   }
2248 
2249   Label checkVolatile;
2250 
2251   // compute field type
2252   Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
2253   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2254   // Make sure we don't need to mask Rflags after the above shift
2255   ConstantPoolCacheEntry::verify_tos_state_shift();
2256 
2257   // Check atos before itos for getstatic, more likely (in Queens at least)
2258   __ cmp(Rflags, atos);
2259   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2260   __ delayed() ->cmp(Rflags, itos);
2261 
2262   // atos
2263   __ load_heap_oop(Rclass, Roffset, Otos_i);
2264   __ verify_oop(Otos_i);
2265   __ push(atos);
2266   if (!is_static) {
2267     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
2268   }
2269   __ ba(checkVolatile);
2270   __ delayed()->tst(Lscratch);
2271 
2272   __ bind(notObj);
2273 
2274   // cmp(Rflags, itos);
2275   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2276   __ delayed() ->cmp(Rflags, ltos);
2277 
2278   // itos
2279   __ ld(Rclass, Roffset, Otos_i);
2280   __ push(itos);
2281   if (!is_static) {
2282     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
2283   }
2284   __ ba(checkVolatile);
2285   __ delayed()->tst(Lscratch);
2286 
2287   __ bind(notInt);
2288 
2289   // cmp(Rflags, ltos);
2290   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2291   __ delayed() ->cmp(Rflags, btos);
2292 
2293   // ltos
2294   // load must be atomic
2295   __ ld_long(Rclass, Roffset, Otos_l);
2296   __ push(ltos);
2297   if (!is_static) {
2298     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
2299   }
2300   __ ba(checkVolatile);
2301   __ delayed()->tst(Lscratch);
2302 
2303   __ bind(notLong);
2304 
2305   // cmp(Rflags, btos);
2306   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2307   __ delayed() ->cmp(Rflags, ctos);
2308 
2309   // btos
2310   __ ldsb(Rclass, Roffset, Otos_i);
2311   __ push(itos);
2312   if (!is_static) {
2313     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
2314   }
2315   __ ba(checkVolatile);
2316   __ delayed()->tst(Lscratch);
2317 
2318   __ bind(notByte);
2319 
2320   // cmp(Rflags, ctos);
2321   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2322   __ delayed() ->cmp(Rflags, stos);
2323 
2324   // ctos
2325   __ lduh(Rclass, Roffset, Otos_i);
2326   __ push(itos);
2327   if (!is_static) {
2328     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
2329   }
2330   __ ba(checkVolatile);
2331   __ delayed()->tst(Lscratch);
2332 
2333   __ bind(notChar);
2334 
2335   // cmp(Rflags, stos);
2336   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2337   __ delayed() ->cmp(Rflags, ftos);
2338 
2339   // stos
2340   __ ldsh(Rclass, Roffset, Otos_i);
2341   __ push(itos);
2342   if (!is_static) {
2343     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
2344   }
2345   __ ba(checkVolatile);
2346   __ delayed()->tst(Lscratch);
2347 
2348   __ bind(notShort);
2349 
2350 
2351   // cmp(Rflags, ftos);
2352   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
2353   __ delayed() ->tst(Lscratch);
2354 
2355   // ftos
2356   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
2357   __ push(ftos);
2358   if (!is_static) {
2359     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
2360   }
2361   __ ba(checkVolatile);
2362   __ delayed()->tst(Lscratch);
2363 
2364   __ bind(notFloat);
2365 
2366 
2367   // dtos
2368   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
2369   __ push(dtos);
2370   if (!is_static) {
2371     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
2372   }
2373 
2374   __ bind(checkVolatile);
2375   if (__ membar_has_effect(membar_bits)) {
2376     // __ tst(Lscratch); executed in delay slot
2377     __ br(Assembler::zero, false, Assembler::pt, exit);
2378     __ delayed()->nop();
2379     volatile_barrier(membar_bits);
2380   }
2381 
2382   __ bind(exit);
2383 }
2384 
2385 
2386 void TemplateTable::getfield(int byte_no) {
2387   getfield_or_static(byte_no, false);
2388 }
2389 
2390 void TemplateTable::getstatic(int byte_no) {
2391   getfield_or_static(byte_no, true);
2392 }
2393 
2394 
2395 void TemplateTable::fast_accessfield(TosState state) {
2396   transition(atos, state);
2397   Register Rcache  = G3_scratch;
2398   Register index   = G4_scratch;
2399   Register Roffset = G4_scratch;
2400   Register Rflags  = Rcache;
2401   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2402 
2403   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2404   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
2405 
2406   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2407 
2408   __ null_check(Otos_i);
2409   __ verify_oop(Otos_i);
2410 
2411   Label exit;
2412 
2413   Assembler::Membar_mask_bits membar_bits =
2414     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2415   if (__ membar_has_effect(membar_bits)) {
2416     // Get volatile flag
2417     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
2418     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2419   }
2420 
2421   switch (bytecode()) {
2422     case Bytecodes::_fast_bgetfield:
2423       __ ldsb(Otos_i, Roffset, Otos_i);
2424       break;
2425     case Bytecodes::_fast_cgetfield:
2426       __ lduh(Otos_i, Roffset, Otos_i);
2427       break;
2428     case Bytecodes::_fast_sgetfield:
2429       __ ldsh(Otos_i, Roffset, Otos_i);
2430       break;
2431     case Bytecodes::_fast_igetfield:
2432       __ ld(Otos_i, Roffset, Otos_i);
2433       break;
2434     case Bytecodes::_fast_lgetfield:
2435       __ ld_long(Otos_i, Roffset, Otos_l);
2436       break;
2437     case Bytecodes::_fast_fgetfield:
2438       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
2439       break;
2440     case Bytecodes::_fast_dgetfield:
2441       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
2442       break;
2443     case Bytecodes::_fast_agetfield:
2444       __ load_heap_oop(Otos_i, Roffset, Otos_i);
2445       break;
2446     default:
2447       ShouldNotReachHere();
2448   }
2449 
2450   if (__ membar_has_effect(membar_bits)) {
2451     __ btst(Lscratch, Rflags);
2452     __ br(Assembler::zero, false, Assembler::pt, exit);
2453     __ delayed()->nop();
2454     volatile_barrier(membar_bits);
2455     __ bind(exit);
2456   }
2457 
2458   if (state == atos) {
2459     __ verify_oop(Otos_i);    // does not blow flags!
2460   }
2461 }
2462 
2463 void TemplateTable::jvmti_post_fast_field_mod() {
2464   if (JvmtiExport::can_post_field_modification()) {
2465     // Check to see if a field modification watch has been set before we take
2466     // the time to call into the VM.
2467     Label done;
2468     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2469     __ load_contents(get_field_modification_count_addr, G4_scratch);
2470     __ cmp_and_br_short(G4_scratch, 0, Assembler::equal, Assembler::pt, done);
2471     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
2472     __ verify_oop(G4_scratch);
2473     __ push_ptr(G4_scratch);    // put the object pointer back on tos
2474     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
2475     // Save tos values before call_VM() clobbers them. Since we have
2476     // to do it for every data type, we use the saved values as the
2477     // jvalue object.
2478     switch (bytecode()) {  // save tos values before call_VM() clobbers them
2479     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
2480     case Bytecodes::_fast_bputfield: // fall through
2481     case Bytecodes::_fast_sputfield: // fall through
2482     case Bytecodes::_fast_cputfield: // fall through
2483     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
2484     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
2485     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
2486     // get words in right order for use as jvalue object
2487     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
2488     }
2489     // setup pointer to jvalue object
2490     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
2491     // G4_scratch:  object pointer
2492     // G1_scratch: cache entry pointer
2493     // G3_scratch: jvalue object on the stack
2494     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
2495     switch (bytecode()) {             // restore tos values
2496     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
2497     case Bytecodes::_fast_bputfield: // fall through
2498     case Bytecodes::_fast_sputfield: // fall through
2499     case Bytecodes::_fast_cputfield: // fall through
2500     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
2501     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
2502     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
2503     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
2504     }
2505     __ bind(done);
2506   }
2507 }
2508 
2509 // The registers Rcache and index expected to be set before call.
2510 // The function may destroy various registers, just not the Rcache and index registers.
2511 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
2512   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2513 
2514   if (JvmtiExport::can_post_field_modification()) {
2515     // Check to see if a field modification watch has been set before we take
2516     // the time to call into the VM.
2517     Label Label1;
2518     assert_different_registers(Rcache, index, G1_scratch);
2519     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2520     __ load_contents(get_field_modification_count_addr, G1_scratch);
2521     __ cmp_and_br_short(G1_scratch, 0, Assembler::zero, Assembler::pt, Label1);
2522 
2523     // The Rcache and index registers have been already set.
2524     // This allows to eliminate this call but the Rcache and index
2525     // registers must be correspondingly used after this line.
2526     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
2527 
2528     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
2529     if (is_static) {
2530       // Life is simple.  Null out the object pointer.
2531       __ clr(G4_scratch);
2532     } else {
2533       Register Rflags = G1_scratch;
2534       // Life is harder. The stack holds the value on top, followed by the
2535       // object.  We don't know the size of the value, though; it could be
2536       // one or two words depending on its type. As a result, we must find
2537       // the type to determine where the object is.
2538 
2539       Label two_word, valsizeknown;
2540       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2541       __ mov(Lesp, G4_scratch);
2542       __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2543       // Make sure we don't need to mask Rflags after the above shift
2544       ConstantPoolCacheEntry::verify_tos_state_shift();
2545       __ cmp(Rflags, ltos);
2546       __ br(Assembler::equal, false, Assembler::pt, two_word);
2547       __ delayed()->cmp(Rflags, dtos);
2548       __ br(Assembler::equal, false, Assembler::pt, two_word);
2549       __ delayed()->nop();
2550       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
2551       __ ba_short(valsizeknown);
2552       __ bind(two_word);
2553 
2554       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
2555 
2556       __ bind(valsizeknown);
2557       // setup object pointer
2558       __ ld_ptr(G4_scratch, 0, G4_scratch);
2559       __ verify_oop(G4_scratch);
2560     }
2561     // setup pointer to jvalue object
2562     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
2563     // G4_scratch:  object pointer or NULL if static
2564     // G3_scratch: cache entry pointer
2565     // G1_scratch: jvalue object on the stack
2566     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2567                G4_scratch, G3_scratch, G1_scratch);
2568     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2569     __ bind(Label1);
2570   }
2571 }
2572 
2573 void TemplateTable::pop_and_check_object(Register r) {
2574   __ pop_ptr(r);
2575   __ null_check(r);  // for field access must check obj.
2576   __ verify_oop(r);
2577 }
2578 
2579 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2580   transition(vtos, vtos);
2581   Register Rcache = G3_scratch;
2582   Register index  = G4_scratch;
2583   Register Rclass = Rcache;
2584   Register Roffset= G4_scratch;
2585   Register Rflags = G1_scratch;
2586   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2587 
2588   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2589   jvmti_post_field_mod(Rcache, index, is_static);
2590   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2591 
2592   Assembler::Membar_mask_bits read_bits =
2593     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2594   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2595 
2596   Label notVolatile, checkVolatile, exit;
2597   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2598     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2599     __ and3(Rflags, Lscratch, Lscratch);
2600 
2601     if (__ membar_has_effect(read_bits)) {
2602       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2603       volatile_barrier(read_bits);
2604       __ bind(notVolatile);
2605     }
2606   }
2607 
2608   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2609   // Make sure we don't need to mask Rflags after the above shift
2610   ConstantPoolCacheEntry::verify_tos_state_shift();
2611 
2612   // compute field type
2613   Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
2614 
2615   if (is_static) {
2616     // putstatic with object type most likely, check that first
2617     __ cmp(Rflags, atos);
2618     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2619     __ delayed()->cmp(Rflags, itos);
2620 
2621     // atos
2622     {
2623       __ pop_ptr();
2624       __ verify_oop(Otos_i);
2625       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2626       __ ba(checkVolatile);
2627       __ delayed()->tst(Lscratch);
2628     }
2629 
2630     __ bind(notObj);
2631     // cmp(Rflags, itos);
2632     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2633     __ delayed()->cmp(Rflags, btos);
2634 
2635     // itos
2636     {
2637       __ pop_i();
2638       __ st(Otos_i, Rclass, Roffset);
2639       __ ba(checkVolatile);
2640       __ delayed()->tst(Lscratch);
2641     }
2642 
2643     __ bind(notInt);
2644   } else {
2645     // putfield with int type most likely, check that first
2646     __ cmp(Rflags, itos);
2647     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2648     __ delayed()->cmp(Rflags, atos);
2649 
2650     // itos
2651     {
2652       __ pop_i();
2653       pop_and_check_object(Rclass);
2654       __ st(Otos_i, Rclass, Roffset);
2655       patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch, true, byte_no);
2656       __ ba(checkVolatile);
2657       __ delayed()->tst(Lscratch);
2658     }
2659 
2660     __ bind(notInt);
2661     // cmp(Rflags, atos);
2662     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2663     __ delayed()->cmp(Rflags, btos);
2664 
2665     // atos
2666     {
2667       __ pop_ptr();
2668       pop_and_check_object(Rclass);
2669       __ verify_oop(Otos_i);
2670       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2671       patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch, true, byte_no);
2672       __ ba(checkVolatile);
2673       __ delayed()->tst(Lscratch);
2674     }
2675 
2676     __ bind(notObj);
2677   }
2678 
2679   // cmp(Rflags, btos);
2680   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2681   __ delayed()->cmp(Rflags, ltos);
2682 
2683   // btos
2684   {
2685     __ pop_i();
2686     if (!is_static) pop_and_check_object(Rclass);
2687     __ stb(Otos_i, Rclass, Roffset);
2688     if (!is_static) {
2689       patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch, true, byte_no);
2690     }
2691     __ ba(checkVolatile);
2692     __ delayed()->tst(Lscratch);
2693   }
2694 
2695   __ bind(notByte);
2696   // cmp(Rflags, ltos);
2697   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2698   __ delayed()->cmp(Rflags, ctos);
2699 
2700   // ltos
2701   {
2702     __ pop_l();
2703     if (!is_static) pop_and_check_object(Rclass);
2704     __ st_long(Otos_l, Rclass, Roffset);
2705     if (!is_static) {
2706       patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch, true, byte_no);
2707     }
2708     __ ba(checkVolatile);
2709     __ delayed()->tst(Lscratch);
2710   }
2711 
2712   __ bind(notLong);
2713   // cmp(Rflags, ctos);
2714   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2715   __ delayed()->cmp(Rflags, stos);
2716 
2717   // ctos (char)
2718   {
2719     __ pop_i();
2720     if (!is_static) pop_and_check_object(Rclass);
2721     __ sth(Otos_i, Rclass, Roffset);
2722     if (!is_static) {
2723       patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch, true, byte_no);
2724     }
2725     __ ba(checkVolatile);
2726     __ delayed()->tst(Lscratch);
2727   }
2728 
2729   __ bind(notChar);
2730   // cmp(Rflags, stos);
2731   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2732   __ delayed()->cmp(Rflags, ftos);
2733 
2734   // stos (short)
2735   {
2736     __ pop_i();
2737     if (!is_static) pop_and_check_object(Rclass);
2738     __ sth(Otos_i, Rclass, Roffset);
2739     if (!is_static) {
2740       patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch, true, byte_no);
2741     }
2742     __ ba(checkVolatile);
2743     __ delayed()->tst(Lscratch);
2744   }
2745 
2746   __ bind(notShort);
2747   // cmp(Rflags, ftos);
2748   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
2749   __ delayed()->nop();
2750 
2751   // ftos
2752   {
2753     __ pop_f();
2754     if (!is_static) pop_and_check_object(Rclass);
2755     __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2756     if (!is_static) {
2757       patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch, true, byte_no);
2758     }
2759     __ ba(checkVolatile);
2760     __ delayed()->tst(Lscratch);
2761   }
2762 
2763   __ bind(notFloat);
2764 
2765   // dtos
2766   {
2767     __ pop_d();
2768     if (!is_static) pop_and_check_object(Rclass);
2769     __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2770     if (!is_static) {
2771       patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch, true, byte_no);
2772     }
2773   }
2774 
2775   __ bind(checkVolatile);
2776   __ tst(Lscratch);
2777 
2778   if (__ membar_has_effect(write_bits)) {
2779     // __ tst(Lscratch); in delay slot
2780     __ br(Assembler::zero, false, Assembler::pt, exit);
2781     __ delayed()->nop();
2782     volatile_barrier(Assembler::StoreLoad);
2783     __ bind(exit);
2784   }
2785 }
2786 
2787 void TemplateTable::fast_storefield(TosState state) {
2788   transition(state, vtos);
2789   Register Rcache = G3_scratch;
2790   Register Rclass = Rcache;
2791   Register Roffset= G4_scratch;
2792   Register Rflags = G1_scratch;
2793   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2794 
2795   jvmti_post_fast_field_mod();
2796 
2797   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
2798 
2799   Assembler::Membar_mask_bits read_bits =
2800     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2801   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2802 
2803   Label notVolatile, checkVolatile, exit;
2804   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2805     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2806     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2807     __ and3(Rflags, Lscratch, Lscratch);
2808     if (__ membar_has_effect(read_bits)) {
2809       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2810       volatile_barrier(read_bits);
2811       __ bind(notVolatile);
2812     }
2813   }
2814 
2815   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2816   pop_and_check_object(Rclass);
2817 
2818   switch (bytecode()) {
2819     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
2820     case Bytecodes::_fast_cputfield: /* fall through */
2821     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
2822     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
2823     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
2824     case Bytecodes::_fast_fputfield:
2825       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2826       break;
2827     case Bytecodes::_fast_dputfield:
2828       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2829       break;
2830     case Bytecodes::_fast_aputfield:
2831       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2832       break;
2833     default:
2834       ShouldNotReachHere();
2835   }
2836 
2837   if (__ membar_has_effect(write_bits)) {
2838     __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, exit);
2839     volatile_barrier(Assembler::StoreLoad);
2840     __ bind(exit);
2841   }
2842 }
2843 
2844 
2845 void TemplateTable::putfield(int byte_no) {
2846   putfield_or_static(byte_no, false);
2847 }
2848 
2849 void TemplateTable::putstatic(int byte_no) {
2850   putfield_or_static(byte_no, true);
2851 }
2852 
2853 
2854 void TemplateTable::fast_xaccess(TosState state) {
2855   transition(vtos, state);
2856   Register Rcache = G3_scratch;
2857   Register Roffset = G4_scratch;
2858   Register Rflags  = G4_scratch;
2859   Register Rreceiver = Lscratch;
2860 
2861   __ ld_ptr(Llocals, 0, Rreceiver);
2862 
2863   // access constant pool cache  (is resolved)
2864   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
2865   __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
2866   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
2867 
2868   __ verify_oop(Rreceiver);
2869   __ null_check(Rreceiver);
2870   if (state == atos) {
2871     __ load_heap_oop(Rreceiver, Roffset, Otos_i);
2872   } else if (state == itos) {
2873     __ ld (Rreceiver, Roffset, Otos_i) ;
2874   } else if (state == ftos) {
2875     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
2876   } else {
2877     ShouldNotReachHere();
2878   }
2879 
2880   Assembler::Membar_mask_bits membar_bits =
2881     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2882   if (__ membar_has_effect(membar_bits)) {
2883 
2884     // Get is_volatile value in Rflags and check if membar is needed
2885     __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
2886 
2887     // Test volatile
2888     Label notVolatile;
2889     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2890     __ btst(Rflags, Lscratch);
2891     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
2892     __ delayed()->nop();
2893     volatile_barrier(membar_bits);
2894     __ bind(notVolatile);
2895   }
2896 
2897   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
2898   __ sub(Lbcp, 1, Lbcp);
2899 }
2900 
2901 //----------------------------------------------------------------------------------------------------
2902 // Calls
2903 
2904 void TemplateTable::count_calls(Register method, Register temp) {
2905   // implemented elsewhere
2906   ShouldNotReachHere();
2907 }
2908 
2909 void TemplateTable::prepare_invoke(int byte_no,
2910                                    Register method,  // linked method (or i-klass)
2911                                    Register ra,      // return address
2912                                    Register index,   // itable index, MethodType, etc.
2913                                    Register recv,    // if caller wants to see it
2914                                    Register flags    // if caller wants to test it
2915                                    ) {
2916   // determine flags
2917   const Bytecodes::Code code = bytecode();
2918   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
2919   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
2920   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
2921   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
2922   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
2923   const bool load_receiver       = (recv != noreg);
2924   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2925   assert(recv  == noreg || recv  == O0, "");
2926   assert(flags == noreg || flags == O1, "");
2927 
2928   // setup registers & access constant pool cache
2929   if (recv  == noreg)  recv  = O0;
2930   if (flags == noreg)  flags = O1;
2931   const Register temp = O2;
2932   assert_different_registers(method, ra, index, recv, flags, temp);
2933 
2934   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2935 
2936   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
2937 
2938   // maybe push appendix to arguments
2939   if (is_invokedynamic || is_invokehandle) {
2940     Label L_no_push;
2941     __ set((1 << ConstantPoolCacheEntry::has_appendix_shift), temp);
2942     __ btst(flags, temp);
2943     __ br(Assembler::zero, false, Assembler::pt, L_no_push);
2944     __ delayed()->nop();
2945     // Push the appendix as a trailing parameter.
2946     // This must be done before we get the receiver,
2947     // since the parameter_size includes it.
2948     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2949     __ load_resolved_reference_at_index(temp, index);
2950     __ verify_oop(temp);
2951     __ push_ptr(temp);  // push appendix (MethodType, CallSite, etc.)
2952     __ bind(L_no_push);
2953   }
2954 
2955   // load receiver if needed (after appendix is pushed so parameter size is correct)
2956   if (load_receiver) {
2957     __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, temp);  // get parameter size
2958     __ load_receiver(temp, recv);  //  __ argument_address uses Gargs but we need Lesp
2959     __ verify_oop(recv);
2960   }
2961 
2962   // compute return type
2963   __ srl(flags, ConstantPoolCacheEntry::tos_state_shift, ra);
2964   // Make sure we don't need to mask flags after the above shift
2965   ConstantPoolCacheEntry::verify_tos_state_shift();
2966   // load return address
2967   {
2968     const address table_addr = (is_invokeinterface || is_invokedynamic) ?
2969         (address)Interpreter::return_5_addrs_by_index_table() :
2970         (address)Interpreter::return_3_addrs_by_index_table();
2971     AddressLiteral table(table_addr);
2972     __ set(table, temp);
2973     __ sll(ra, LogBytesPerWord, ra);
2974     __ ld_ptr(Address(temp, ra), ra);
2975   }
2976 }
2977 
2978 
2979 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
2980   Register Rtemp = G4_scratch;
2981   Register Rcall = Rindex;
2982   assert_different_registers(Rcall, G5_method, Gargs, Rret);
2983 
2984   // get target Method* & entry point
2985   __ lookup_virtual_method(Rrecv, Rindex, G5_method);
2986   __ call_from_interpreter(Rcall, Gargs, Rret);
2987 }
2988 
2989 void TemplateTable::invokevirtual(int byte_no) {
2990   transition(vtos, vtos);
2991   assert(byte_no == f2_byte, "use this argument");
2992 
2993   Register Rscratch = G3_scratch;
2994   Register Rtemp    = G4_scratch;
2995   Register Rret     = Lscratch;
2996   Register O0_recv  = O0;
2997   Label notFinal;
2998 
2999   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
3000   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3001 
3002   // Check for vfinal
3003   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), G4_scratch);
3004   __ btst(Rret, G4_scratch);
3005   __ br(Assembler::zero, false, Assembler::pt, notFinal);
3006   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
3007 
3008   patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
3009 
3010   invokevfinal_helper(Rscratch, Rret);
3011 
3012   __ bind(notFinal);
3013 
3014   __ mov(G5_method, Rscratch);  // better scratch register
3015   __ load_receiver(G4_scratch, O0_recv);  // gets receiverOop
3016   // receiver is in O0_recv
3017   __ verify_oop(O0_recv);
3018 
3019   // get return address
3020   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
3021   __ set(table, Rtemp);
3022   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
3023   // Make sure we don't need to mask Rret after the above shift
3024   ConstantPoolCacheEntry::verify_tos_state_shift();
3025   __ sll(Rret,  LogBytesPerWord, Rret);
3026   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3027 
3028   // get receiver klass
3029   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
3030   __ load_klass(O0_recv, O0_recv);
3031   __ verify_klass_ptr(O0_recv);
3032 
3033   __ profile_virtual_call(O0_recv, O4);
3034 
3035   generate_vtable_call(O0_recv, Rscratch, Rret);
3036 }
3037 
3038 void TemplateTable::fast_invokevfinal(int byte_no) {
3039   transition(vtos, vtos);
3040   assert(byte_no == f2_byte, "use this argument");
3041 
3042   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
3043                              /*is_invokevfinal*/true, false);
3044   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3045   invokevfinal_helper(G3_scratch, Lscratch);
3046 }
3047 
3048 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
3049   Register Rtemp = G4_scratch;
3050 
3051   // Load receiver from stack slot
3052   __ ld_ptr(G5_method, in_bytes(Method::const_offset()), G4_scratch);
3053   __ lduh(G4_scratch, in_bytes(ConstMethod::size_of_parameters_offset()), G4_scratch);
3054   __ load_receiver(G4_scratch, O0);
3055 
3056   // receiver NULL check
3057   __ null_check(O0);
3058 
3059   __ profile_final_call(O4);
3060 
3061   // get return address
3062   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
3063   __ set(table, Rtemp);
3064   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
3065   // Make sure we don't need to mask Rret after the above shift
3066   ConstantPoolCacheEntry::verify_tos_state_shift();
3067   __ sll(Rret,  LogBytesPerWord, Rret);
3068   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3069 
3070 
3071   // do the call
3072   __ call_from_interpreter(Rscratch, Gargs, Rret);
3073 }
3074 
3075 
3076 void TemplateTable::invokespecial(int byte_no) {
3077   transition(vtos, vtos);
3078   assert(byte_no == f1_byte, "use this argument");
3079 
3080   const Register Rret     = Lscratch;
3081   const Register O0_recv  = O0;
3082   const Register Rscratch = G3_scratch;
3083 
3084   prepare_invoke(byte_no, G5_method, Rret, noreg, O0_recv);  // get receiver also for null check
3085   __ null_check(O0_recv);
3086 
3087   // do the call
3088   __ profile_call(O4);
3089   __ call_from_interpreter(Rscratch, Gargs, Rret);
3090 }
3091 
3092 
3093 void TemplateTable::invokestatic(int byte_no) {
3094   transition(vtos, vtos);
3095   assert(byte_no == f1_byte, "use this argument");
3096 
3097   const Register Rret     = Lscratch;
3098   const Register Rscratch = G3_scratch;
3099 
3100   prepare_invoke(byte_no, G5_method, Rret);  // get f1 Method*
3101 
3102   // do the call
3103   __ profile_call(O4);
3104   __ call_from_interpreter(Rscratch, Gargs, Rret);
3105 }
3106 
3107 void TemplateTable::invokeinterface_object_method(Register RKlass,
3108                                                   Register Rcall,
3109                                                   Register Rret,
3110                                                   Register Rflags) {
3111   Register Rscratch = G4_scratch;
3112   Register Rindex = Lscratch;
3113 
3114   assert_different_registers(Rscratch, Rindex, Rret);
3115 
3116   Label notFinal;
3117 
3118   // Check for vfinal
3119   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), Rscratch);
3120   __ btst(Rflags, Rscratch);
3121   __ br(Assembler::zero, false, Assembler::pt, notFinal);
3122   __ delayed()->nop();
3123 
3124   __ profile_final_call(O4);
3125 
3126   // do the call - the index (f2) contains the Method*
3127   assert_different_registers(G5_method, Gargs, Rcall);
3128   __ mov(Rindex, G5_method);
3129   __ call_from_interpreter(Rcall, Gargs, Rret);
3130   __ bind(notFinal);
3131 
3132   __ profile_virtual_call(RKlass, O4);
3133   generate_vtable_call(RKlass, Rindex, Rret);
3134 }
3135 
3136 
3137 void TemplateTable::invokeinterface(int byte_no) {
3138   transition(vtos, vtos);
3139   assert(byte_no == f1_byte, "use this argument");
3140 
3141   const Register Rinterface  = G1_scratch;
3142   const Register Rret        = G3_scratch;
3143   const Register Rindex      = Lscratch;
3144   const Register O0_recv     = O0;
3145   const Register O1_flags    = O1;
3146   const Register O2_Klass    = O2;
3147   const Register Rscratch    = G4_scratch;
3148   assert_different_registers(Rscratch, G5_method);
3149 
3150   prepare_invoke(byte_no, Rinterface, Rret, Rindex, O0_recv, O1_flags);
3151 
3152   // get receiver klass
3153   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
3154   __ load_klass(O0_recv, O2_Klass);
3155 
3156   // Special case of invokeinterface called for virtual method of
3157   // java.lang.Object.  See cpCacheOop.cpp for details.
3158   // This code isn't produced by javac, but could be produced by
3159   // another compliant java compiler.
3160   Label notMethod;
3161   __ set((1 << ConstantPoolCacheEntry::is_forced_virtual_shift), Rscratch);
3162   __ btst(O1_flags, Rscratch);
3163   __ br(Assembler::zero, false, Assembler::pt, notMethod);
3164   __ delayed()->nop();
3165 
3166   invokeinterface_object_method(O2_Klass, Rinterface, Rret, O1_flags);
3167 
3168   __ bind(notMethod);
3169 
3170   __ profile_virtual_call(O2_Klass, O4);
3171 
3172   //
3173   // find entry point to call
3174   //
3175 
3176   // compute start of first itableOffsetEntry (which is at end of vtable)
3177   const int base = InstanceKlass::vtable_start_offset() * wordSize;
3178   Label search;
3179   Register Rtemp = O1_flags;
3180 
3181   __ ld(O2_Klass, InstanceKlass::vtable_length_offset() * wordSize, Rtemp);
3182   if (align_object_offset(1) > 1) {
3183     __ round_to(Rtemp, align_object_offset(1));
3184   }
3185   __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
3186   if (Assembler::is_simm13(base)) {
3187     __ add(Rtemp, base, Rtemp);
3188   } else {
3189     __ set(base, Rscratch);
3190     __ add(Rscratch, Rtemp, Rtemp);
3191   }
3192   __ add(O2_Klass, Rtemp, Rscratch);
3193 
3194   __ bind(search);
3195 
3196   __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
3197   {
3198     Label ok;
3199 
3200     // Check that entry is non-null.  Null entries are probably a bytecode
3201     // problem.  If the interface isn't implemented by the receiver class,
3202     // the VM should throw IncompatibleClassChangeError.  linkResolver checks
3203     // this too but that's only if the entry isn't already resolved, so we
3204     // need to check again.
3205     __ br_notnull_short( Rtemp, Assembler::pt, ok);
3206     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3207     __ should_not_reach_here();
3208     __ bind(ok);
3209   }
3210 
3211   __ cmp(Rinterface, Rtemp);
3212   __ brx(Assembler::notEqual, true, Assembler::pn, search);
3213   __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
3214 
3215   // entry found and Rscratch points to it
3216   __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
3217 
3218   assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
3219   __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
3220   __ add(Rscratch, Rindex, Rscratch);
3221   __ ld_ptr(O2_Klass, Rscratch, G5_method);
3222 
3223   // Check for abstract method error.
3224   {
3225     Label ok;
3226     __ br_notnull_short(G5_method, Assembler::pt, ok);
3227     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3228     __ should_not_reach_here();
3229     __ bind(ok);
3230   }
3231 
3232   Register Rcall = Rinterface;
3233   assert_different_registers(Rcall, G5_method, Gargs, Rret);
3234 
3235   __ call_from_interpreter(Rcall, Gargs, Rret);
3236 }
3237 
3238 void TemplateTable::invokehandle(int byte_no) {
3239   transition(vtos, vtos);
3240   assert(byte_no == f1_byte, "use this argument");
3241 
3242   if (!EnableInvokeDynamic) {
3243     // rewriter does not generate this bytecode
3244     __ should_not_reach_here();
3245     return;
3246   }
3247 
3248   const Register Rret       = Lscratch;
3249   const Register G4_mtype   = G4_scratch;
3250   const Register O0_recv    = O0;
3251   const Register Rscratch   = G3_scratch;
3252 
3253   prepare_invoke(byte_no, G5_method, Rret, G4_mtype, O0_recv);
3254   __ null_check(O0_recv);
3255 
3256   // G4: MethodType object (from cpool->resolved_references[f1], if necessary)
3257   // G5: MH.invokeExact_MT method (from f2)
3258 
3259   // Note:  G4_mtype is already pushed (if necessary) by prepare_invoke
3260 
3261   // do the call
3262   __ verify_oop(G4_mtype);
3263   __ profile_final_call(O4);  // FIXME: profile the LambdaForm also
3264   __ call_from_interpreter(Rscratch, Gargs, Rret);
3265 }
3266 
3267 
3268 void TemplateTable::invokedynamic(int byte_no) {
3269   transition(vtos, vtos);
3270   assert(byte_no == f1_byte, "use this argument");
3271 
3272   if (!EnableInvokeDynamic) {
3273     // We should not encounter this bytecode if !EnableInvokeDynamic.
3274     // The verifier will stop it.  However, if we get past the verifier,
3275     // this will stop the thread in a reasonable way, without crashing the JVM.
3276     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3277                      InterpreterRuntime::throw_IncompatibleClassChangeError));
3278     // the call_VM checks for exception, so we should never return here.
3279     __ should_not_reach_here();
3280     return;
3281   }
3282 
3283   const Register Rret        = Lscratch;
3284   const Register G4_callsite = G4_scratch;
3285   const Register Rscratch    = G3_scratch;
3286 
3287   prepare_invoke(byte_no, G5_method, Rret, G4_callsite);
3288 
3289   // G4: CallSite object (from cpool->resolved_references[f1])
3290   // G5: MH.linkToCallSite method (from f2)
3291 
3292   // Note:  G4_callsite is already pushed by prepare_invoke
3293 
3294   // %%% should make a type profile for any invokedynamic that takes a ref argument
3295   // profile this call
3296   __ profile_call(O4);
3297 
3298   // do the call
3299   __ verify_oop(G4_callsite);
3300   __ call_from_interpreter(Rscratch, Gargs, Rret);
3301 }
3302 
3303 
3304 //----------------------------------------------------------------------------------------------------
3305 // Allocation
3306 
3307 void TemplateTable::_new() {
3308   transition(vtos, atos);
3309 
3310   Label slow_case;
3311   Label done;
3312   Label initialize_header;
3313   Label initialize_object;  // including clearing the fields
3314 
3315   Register RallocatedObject = Otos_i;
3316   Register RinstanceKlass = O1;
3317   Register Roffset = O3;
3318   Register Rscratch = O4;
3319 
3320   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3321   __ get_cpool_and_tags(Rscratch, G3_scratch);
3322   // make sure the class we're about to instantiate has been resolved
3323   // This is done before loading InstanceKlass to be consistent with the order
3324   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3325   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3326   __ ldub(G3_scratch, Roffset, G3_scratch);
3327   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3328   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3329   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3330   // get InstanceKlass
3331   //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
3332   __ add(Roffset, sizeof(ConstantPool), Roffset);
3333   __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
3334 
3335   // make sure klass is fully initialized:
3336   __ ldub(RinstanceKlass, in_bytes(InstanceKlass::init_state_offset()), G3_scratch);
3337   __ cmp(G3_scratch, InstanceKlass::fully_initialized);
3338   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3339   __ delayed()->ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3340 
3341   // get instance_size in InstanceKlass (already aligned)
3342   //__ ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3343 
3344   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
3345   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
3346   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
3347   __ delayed()->nop();
3348 
3349   // allocate the instance
3350   // 1) Try to allocate in the TLAB
3351   // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
3352   // 3) if the above fails (or is not applicable), go to a slow case
3353   // (creates a new TLAB, etc.)
3354 
3355   const bool allow_shared_alloc =
3356     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3357 
3358   if(UseTLAB) {
3359     Register RoldTopValue = RallocatedObject;
3360     Register RtlabWasteLimitValue = G3_scratch;
3361     Register RnewTopValue = G1_scratch;
3362     Register RendValue = Rscratch;
3363     Register RfreeValue = RnewTopValue;
3364 
3365     // check if we can allocate in the TLAB
3366     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
3367     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
3368     __ add(RoldTopValue, Roffset, RnewTopValue);
3369 
3370     // if there is enough space, we do not CAS and do not clear
3371     __ cmp(RnewTopValue, RendValue);
3372     if(ZeroTLAB) {
3373       // the fields have already been cleared
3374       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
3375     } else {
3376       // initialize both the header and fields
3377       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
3378     }
3379     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
3380 
3381     if (allow_shared_alloc) {
3382       // Check if tlab should be discarded (refill_waste_limit >= free)
3383       __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
3384       __ sub(RendValue, RoldTopValue, RfreeValue);
3385 #ifdef _LP64
3386       __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
3387 #else
3388       __ srl(RfreeValue, LogHeapWordSize, RfreeValue);
3389 #endif
3390       __ cmp_and_brx_short(RtlabWasteLimitValue, RfreeValue, Assembler::greaterEqualUnsigned, Assembler::pt, slow_case); // tlab waste is small
3391 
3392       // increment waste limit to prevent getting stuck on this slow path
3393       __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
3394       __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
3395     } else {
3396       // No allocation in the shared eden.
3397       __ ba_short(slow_case);
3398     }
3399   }
3400 
3401   // Allocation in the shared Eden
3402   if (allow_shared_alloc) {
3403     Register RoldTopValue = G1_scratch;
3404     Register RtopAddr = G3_scratch;
3405     Register RnewTopValue = RallocatedObject;
3406     Register RendValue = Rscratch;
3407 
3408     __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
3409 
3410     Label retry;
3411     __ bind(retry);
3412     __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
3413     __ ld_ptr(RendValue, 0, RendValue);
3414     __ ld_ptr(RtopAddr, 0, RoldTopValue);
3415     __ add(RoldTopValue, Roffset, RnewTopValue);
3416 
3417     // RnewTopValue contains the top address after the new object
3418     // has been allocated.
3419     __ cmp_and_brx_short(RnewTopValue, RendValue, Assembler::greaterUnsigned, Assembler::pn, slow_case);
3420 
3421     __ casx_under_lock(RtopAddr, RoldTopValue, RnewTopValue,
3422       VM_Version::v9_instructions_work() ? NULL :
3423       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
3424 
3425     // if someone beat us on the allocation, try again, otherwise continue
3426     __ cmp_and_brx_short(RoldTopValue, RnewTopValue, Assembler::notEqual, Assembler::pn, retry);
3427 
3428     // bump total bytes allocated by this thread
3429     // RoldTopValue and RtopAddr are dead, so can use G1 and G3
3430     __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
3431   }
3432 
3433   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3434     // clear object fields
3435     __ bind(initialize_object);
3436     __ deccc(Roffset, sizeof(oopDesc));
3437     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
3438     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
3439 
3440     // initialize remaining object fields
3441     if (UseBlockZeroing) {
3442       // Use BIS for zeroing
3443       __ bis_zeroing(G3_scratch, Roffset, G1_scratch, initialize_header);
3444     } else {
3445       Label loop;
3446       __ subcc(Roffset, wordSize, Roffset);
3447       __ bind(loop);
3448       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
3449       __ st_ptr(G0, G3_scratch, Roffset);
3450       __ br(Assembler::notEqual, false, Assembler::pt, loop);
3451       __ delayed()->subcc(Roffset, wordSize, Roffset);
3452     }
3453     __ ba_short(initialize_header);
3454   }
3455 
3456   // slow case
3457   __ bind(slow_case);
3458   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
3459   __ get_constant_pool(O1);
3460 
3461   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
3462 
3463   __ ba_short(done);
3464 
3465   // Initialize the header: mark, klass
3466   __ bind(initialize_header);
3467 
3468   if (UseBiasedLocking) {
3469     __ ld_ptr(RinstanceKlass, in_bytes(Klass::prototype_header_offset()), G4_scratch);
3470   } else {
3471     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
3472   }
3473   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
3474   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
3475   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
3476 
3477   {
3478     SkipIfEqual skip_if(
3479       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
3480     // Trigger dtrace event
3481     __ push(atos);
3482     __ call_VM_leaf(noreg,
3483        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
3484     __ pop(atos);
3485   }
3486 
3487   // continue
3488   __ bind(done);
3489 }
3490 
3491 
3492 
3493 void TemplateTable::newarray() {
3494   transition(itos, atos);
3495   __ ldub(Lbcp, 1, O1);
3496      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
3497 }
3498 
3499 
3500 void TemplateTable::anewarray() {
3501   transition(itos, atos);
3502   __ get_constant_pool(O1);
3503   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
3504      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
3505 }
3506 
3507 
3508 void TemplateTable::arraylength() {
3509   transition(atos, itos);
3510   Label ok;
3511   __ verify_oop(Otos_i);
3512   __ tst(Otos_i);
3513   __ throw_if_not_1_x( Assembler::notZero, ok );
3514   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
3515   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3516 }
3517 
3518 
3519 void TemplateTable::checkcast() {
3520   transition(atos, atos);
3521   Label done, is_null, quicked, cast_ok, resolved;
3522   Register Roffset = G1_scratch;
3523   Register RobjKlass = O5;
3524   Register RspecifiedKlass = O4;
3525 
3526   // Check for casting a NULL
3527   __ br_null_short(Otos_i, Assembler::pn, is_null);
3528 
3529   // Get value klass in RobjKlass
3530   __ load_klass(Otos_i, RobjKlass); // get value klass
3531 
3532   // Get constant pool tag
3533   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3534 
3535   // See if the checkcast has been quickened
3536   __ get_cpool_and_tags(Lscratch, G3_scratch);
3537   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3538   __ ldub(G3_scratch, Roffset, G3_scratch);
3539   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3540   __ br(Assembler::equal, true, Assembler::pt, quicked);
3541   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3542 
3543   __ push_ptr(); // save receiver for result, and for GC
3544   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3545   __ get_vm_result_2(RspecifiedKlass);
3546   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3547 
3548   __ ba_short(resolved);
3549 
3550   // Extract target class from constant pool
3551   __ bind(quicked);
3552   __ add(Roffset, sizeof(ConstantPool), Roffset);
3553   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3554   __ bind(resolved);
3555   __ load_klass(Otos_i, RobjKlass); // get value klass
3556 
3557   // Generate a fast subtype check.  Branch to cast_ok if no
3558   // failure.  Throw exception if failure.
3559   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
3560 
3561   // Not a subtype; so must throw exception
3562   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
3563 
3564   __ bind(cast_ok);
3565 
3566   if (ProfileInterpreter) {
3567     __ ba_short(done);
3568   }
3569   __ bind(is_null);
3570   __ profile_null_seen(G3_scratch);
3571   __ bind(done);
3572 }
3573 
3574 
3575 void TemplateTable::instanceof() {
3576   Label done, is_null, quicked, resolved;
3577   transition(atos, itos);
3578   Register Roffset = G1_scratch;
3579   Register RobjKlass = O5;
3580   Register RspecifiedKlass = O4;
3581 
3582   // Check for casting a NULL
3583   __ br_null_short(Otos_i, Assembler::pt, is_null);
3584 
3585   // Get value klass in RobjKlass
3586   __ load_klass(Otos_i, RobjKlass); // get value klass
3587 
3588   // Get constant pool tag
3589   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3590 
3591   // See if the checkcast has been quickened
3592   __ get_cpool_and_tags(Lscratch, G3_scratch);
3593   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3594   __ ldub(G3_scratch, Roffset, G3_scratch);
3595   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3596   __ br(Assembler::equal, true, Assembler::pt, quicked);
3597   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3598 
3599   __ push_ptr(); // save receiver for result, and for GC
3600   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3601   __ get_vm_result_2(RspecifiedKlass);
3602   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3603 
3604   __ ba_short(resolved);
3605 
3606   // Extract target class from constant pool
3607   __ bind(quicked);
3608   __ add(Roffset, sizeof(ConstantPool), Roffset);
3609   __ get_constant_pool(Lscratch);
3610   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3611   __ bind(resolved);
3612   __ load_klass(Otos_i, RobjKlass); // get value klass
3613 
3614   // Generate a fast subtype check.  Branch to cast_ok if no
3615   // failure.  Return 0 if failure.
3616   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
3617   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
3618   // Not a subtype; return 0;
3619   __ clr( Otos_i );
3620 
3621   if (ProfileInterpreter) {
3622     __ ba_short(done);
3623   }
3624   __ bind(is_null);
3625   __ profile_null_seen(G3_scratch);
3626   __ bind(done);
3627 }
3628 
3629 void TemplateTable::_breakpoint() {
3630 
3631    // Note: We get here even if we are single stepping..
3632    // jbug inists on setting breakpoints at every bytecode
3633    // even if we are in single step mode.
3634 
3635    transition(vtos, vtos);
3636    // get the unpatched byte code
3637    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
3638    __ mov(O0, Lbyte_code);
3639 
3640    // post the breakpoint event
3641    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
3642 
3643    // complete the execution of original bytecode
3644    __ dispatch_normal(vtos);
3645 }
3646 
3647 
3648 //----------------------------------------------------------------------------------------------------
3649 // Exceptions
3650 
3651 void TemplateTable::athrow() {
3652   transition(atos, vtos);
3653 
3654   // This works because exception is cached in Otos_i which is same as O0,
3655   // which is same as what throw_exception_entry_expects
3656   assert(Otos_i == Oexception, "see explanation above");
3657 
3658   __ verify_oop(Otos_i);
3659   __ null_check(Otos_i);
3660   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
3661 }
3662 
3663 
3664 //----------------------------------------------------------------------------------------------------
3665 // Synchronization
3666 
3667 
3668 // See frame_sparc.hpp for monitor block layout.
3669 // Monitor elements are dynamically allocated by growing stack as needed.
3670 
3671 void TemplateTable::monitorenter() {
3672   transition(atos, vtos);
3673   __ verify_oop(Otos_i);
3674   // Try to acquire a lock on the object
3675   // Repeat until succeeded (i.e., until
3676   // monitorenter returns true).
3677 
3678   {   Label ok;
3679     __ tst(Otos_i);
3680     __ throw_if_not_1_x( Assembler::notZero,  ok);
3681     __ delayed()->mov(Otos_i, Lscratch); // save obj
3682     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3683   }
3684 
3685   assert(O0 == Otos_i, "Be sure where the object to lock is");
3686 
3687   // find a free slot in the monitor block
3688 
3689 
3690   // initialize entry pointer
3691   __ clr(O1); // points to free slot or NULL
3692 
3693   {
3694     Label entry, loop, exit;
3695     __ add( __ top_most_monitor(), O2 ); // last one to check
3696     __ ba( entry );
3697     __ delayed()->mov( Lmonitors, O3 ); // first one to check
3698 
3699 
3700     __ bind( loop );
3701 
3702     __ verify_oop(O4);          // verify each monitor's oop
3703     __ tst(O4); // is this entry unused?
3704     if (VM_Version::v9_instructions_work())
3705       __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
3706     else {
3707       Label L;
3708       __ br( Assembler::zero, true, Assembler::pn, L );
3709       __ delayed()->mov(O3, O1); // rememeber this one if match
3710       __ bind(L);
3711     }
3712 
3713     __ cmp(O4, O0); // check if current entry is for same object
3714     __ brx( Assembler::equal, false, Assembler::pn, exit );
3715     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
3716 
3717     __ bind( entry );
3718 
3719     __ cmp( O3, O2 );
3720     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3721     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
3722 
3723     __ bind( exit );
3724   }
3725 
3726   { Label allocated;
3727 
3728     // found free slot?
3729     __ br_notnull_short(O1, Assembler::pn, allocated);
3730 
3731     __ add_monitor_to_stack( false, O2, O3 );
3732     __ mov(Lmonitors, O1);
3733 
3734     __ bind(allocated);
3735   }
3736 
3737   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3738   // The object has already been poped from the stack, so the expression stack looks correct.
3739   __ inc(Lbcp);
3740 
3741   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
3742   __ lock_object(O1, O0);
3743 
3744   // check if there's enough space on the stack for the monitors after locking
3745   __ generate_stack_overflow_check(0);
3746 
3747   // The bcp has already been incremented. Just need to dispatch to next instruction.
3748   __ dispatch_next(vtos);
3749 }
3750 
3751 
3752 void TemplateTable::monitorexit() {
3753   transition(atos, vtos);
3754   __ verify_oop(Otos_i);
3755   __ tst(Otos_i);
3756   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
3757 
3758   assert(O0 == Otos_i, "just checking");
3759 
3760   { Label entry, loop, found;
3761     __ add( __ top_most_monitor(), O2 ); // last one to check
3762     __ ba(entry);
3763     // use Lscratch to hold monitor elem to check, start with most recent monitor,
3764     // By using a local it survives the call to the C routine.
3765     __ delayed()->mov( Lmonitors, Lscratch );
3766 
3767     __ bind( loop );
3768 
3769     __ verify_oop(O4);          // verify each monitor's oop
3770     __ cmp(O4, O0); // check if current entry is for desired object
3771     __ brx( Assembler::equal, true, Assembler::pt, found );
3772     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
3773 
3774     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
3775 
3776     __ bind( entry );
3777 
3778     __ cmp( Lscratch, O2 );
3779     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3780     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
3781 
3782     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3783     __ should_not_reach_here();
3784 
3785     __ bind(found);
3786   }
3787   __ unlock_object(O1);
3788 }
3789 
3790 
3791 //----------------------------------------------------------------------------------------------------
3792 // Wide instructions
3793 
3794 void TemplateTable::wide() {
3795   transition(vtos, vtos);
3796   __ ldub(Lbcp, 1, G3_scratch);// get next bc
3797   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
3798   AddressLiteral ep(Interpreter::_wentry_point);
3799   __ set(ep, G4_scratch);
3800   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
3801   __ jmp(G3_scratch, G0);
3802   __ delayed()->nop();
3803   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
3804 }
3805 
3806 
3807 //----------------------------------------------------------------------------------------------------
3808 // Multi arrays
3809 
3810 void TemplateTable::multianewarray() {
3811   transition(vtos, atos);
3812      // put ndims * wordSize into Lscratch
3813   __ ldub( Lbcp,     3,               Lscratch);
3814   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
3815      // Lesp points past last_dim, so set to O1 to first_dim address
3816   __ add(  Lesp,     Lscratch,        O1);
3817      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
3818   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
3819 }
3820 #endif /* !CC_INTERP */