1 /*
   2  * Copyright (c) 1997, 2016, 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/interp_masm.hpp"
  29 #include "interpreter/templateTable.hpp"
  30 #include "memory/universe.inline.hpp"
  31 #include "oops/methodData.hpp"
  32 #include "oops/objArrayKlass.hpp"
  33 #include "oops/oop.inline.hpp"
  34 #include "prims/methodHandles.hpp"
  35 #include "runtime/sharedRuntime.hpp"
  36 #include "runtime/stubRoutines.hpp"
  37 #include "runtime/synchronizer.hpp"
  38 #include "utilities/macros.hpp"
  39 
  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::G1SATBCTLogging:
  59       {
  60         // Load and record the previous value.
  61         __ g1_write_barrier_pre(base, index, offset,
  62                                 noreg /* pre_val */,
  63                                 tmp, true /*preserve_o_regs*/);
  64 
  65         // G1 barrier needs uncompressed oop for region cross check.
  66         Register new_val = val;
  67         if (UseCompressedOops && val != G0) {
  68           new_val = tmp;
  69           __ mov(val, new_val);
  70         }
  71 
  72         if (index == noreg ) {
  73           assert(Assembler::is_simm13(offset), "fix this code");
  74           __ store_heap_oop(val, base, offset);
  75         } else {
  76           __ store_heap_oop(val, base, index);
  77         }
  78 
  79         // No need for post barrier if storing NULL
  80         if (val != G0) {
  81           if (precise) {
  82             if (index == noreg) {
  83               __ add(base, offset, base);
  84             } else {
  85               __ add(base, index, base);
  86             }
  87           }
  88           __ g1_write_barrier_post(base, new_val, tmp);
  89         }
  90       }
  91       break;
  92 #endif // INCLUDE_ALL_GCS
  93     case BarrierSet::CardTableForRS:
  94     case BarrierSet::CardTableExtension:
  95       {
  96         if (index == noreg ) {
  97           assert(Assembler::is_simm13(offset), "fix this code");
  98           __ store_heap_oop(val, base, offset);
  99         } else {
 100           __ store_heap_oop(val, base, index);
 101         }
 102         // No need for post barrier if storing NULL
 103         if (val != G0) {
 104           if (precise) {
 105             if (index == noreg) {
 106               __ add(base, offset, base);
 107             } else {
 108               __ add(base, index, base);
 109             }
 110           }
 111           __ card_write_barrier_post(base, val, tmp);
 112         }
 113       }
 114       break;
 115     case BarrierSet::ModRef:
 116       ShouldNotReachHere();
 117       break;
 118     default      :
 119       ShouldNotReachHere();
 120 
 121   }
 122 }
 123 
 124 
 125 //----------------------------------------------------------------------------------------------------
 126 // Platform-dependent initialization
 127 
 128 void TemplateTable::pd_initialize() {
 129   // (none)
 130 }
 131 
 132 
 133 //----------------------------------------------------------------------------------------------------
 134 // Condition conversion
 135 Assembler::Condition ccNot(TemplateTable::Condition cc) {
 136   switch (cc) {
 137     case TemplateTable::equal        : return Assembler::notEqual;
 138     case TemplateTable::not_equal    : return Assembler::equal;
 139     case TemplateTable::less         : return Assembler::greaterEqual;
 140     case TemplateTable::less_equal   : return Assembler::greater;
 141     case TemplateTable::greater      : return Assembler::lessEqual;
 142     case TemplateTable::greater_equal: return Assembler::less;
 143   }
 144   ShouldNotReachHere();
 145   return Assembler::zero;
 146 }
 147 
 148 //----------------------------------------------------------------------------------------------------
 149 // Miscelaneous helper routines
 150 
 151 
 152 Address TemplateTable::at_bcp(int offset) {
 153   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 154   return Address(Lbcp, offset);
 155 }
 156 
 157 
 158 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 159                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 160                                    int byte_no) {
 161   // With sharing on, may need to test Method* flag.
 162   if (!RewriteBytecodes)  return;
 163   Label L_patch_done;
 164 
 165   switch (bc) {
 166   case Bytecodes::_fast_aputfield:
 167   case Bytecodes::_fast_bputfield:
 168   case Bytecodes::_fast_zputfield:
 169   case Bytecodes::_fast_cputfield:
 170   case Bytecodes::_fast_dputfield:
 171   case Bytecodes::_fast_fputfield:
 172   case Bytecodes::_fast_iputfield:
 173   case Bytecodes::_fast_lputfield:
 174   case Bytecodes::_fast_sputfield:
 175     {
 176       // We skip bytecode quickening for putfield instructions when
 177       // the put_code written to the constant pool cache is zero.
 178       // This is required so that every execution of this instruction
 179       // calls out to InterpreterRuntime::resolve_get_put to do
 180       // additional, required work.
 181       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 182       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 183       __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
 184       __ set(bc, bc_reg);
 185       __ cmp_and_br_short(temp_reg, 0, Assembler::equal, Assembler::pn, L_patch_done);  // don't patch
 186     }
 187     break;
 188   default:
 189     assert(byte_no == -1, "sanity");
 190     if (load_bc_into_bc_reg) {
 191       __ set(bc, bc_reg);
 192     }
 193   }
 194 
 195   if (JvmtiExport::can_post_breakpoint()) {
 196     Label L_fast_patch;
 197     __ ldub(at_bcp(0), temp_reg);
 198     __ cmp_and_br_short(temp_reg, Bytecodes::_breakpoint, Assembler::notEqual, Assembler::pt, L_fast_patch);
 199     // perform the quickening, slowly, in the bowels of the breakpoint table
 200     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, bc_reg);
 201     __ ba_short(L_patch_done);
 202     __ bind(L_fast_patch);
 203   }
 204 
 205 #ifdef ASSERT
 206   Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
 207   Label L_okay;
 208   __ ldub(at_bcp(0), temp_reg);
 209   __ cmp(temp_reg, orig_bytecode);
 210   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 211   __ delayed()->cmp(temp_reg, bc_reg);
 212   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 213   __ delayed()->nop();
 214   __ stop("patching the wrong bytecode");
 215   __ bind(L_okay);
 216 #endif
 217 
 218   // patch bytecode
 219   __ stb(bc_reg, at_bcp(0));
 220   __ bind(L_patch_done);
 221 }
 222 
 223 //----------------------------------------------------------------------------------------------------
 224 // Individual instructions
 225 
 226 void TemplateTable::nop() {
 227   transition(vtos, vtos);
 228   // nothing to do
 229 }
 230 
 231 void TemplateTable::shouldnotreachhere() {
 232   transition(vtos, vtos);
 233   __ stop("shouldnotreachhere bytecode");
 234 }
 235 
 236 void TemplateTable::aconst_null() {
 237   transition(vtos, atos);
 238   __ clr(Otos_i);
 239 }
 240 
 241 
 242 void TemplateTable::iconst(int value) {
 243   transition(vtos, itos);
 244   __ set(value, Otos_i);
 245 }
 246 
 247 
 248 void TemplateTable::lconst(int value) {
 249   transition(vtos, ltos);
 250   assert(value >= 0, "check this code");
 251   __ set(value, Otos_l);
 252 }
 253 
 254 
 255 void TemplateTable::fconst(int value) {
 256   transition(vtos, ftos);
 257   static float zero = 0.0, one = 1.0, two = 2.0;
 258   float* p;
 259   switch( value ) {
 260    default: ShouldNotReachHere();
 261    case 0:  p = &zero;  break;
 262    case 1:  p = &one;   break;
 263    case 2:  p = &two;   break;
 264   }
 265   AddressLiteral a(p);
 266   __ sethi(a, G3_scratch);
 267   __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
 268 }
 269 
 270 
 271 void TemplateTable::dconst(int value) {
 272   transition(vtos, dtos);
 273   static double zero = 0.0, one = 1.0;
 274   double* p;
 275   switch( value ) {
 276    default: ShouldNotReachHere();
 277    case 0:  p = &zero;  break;
 278    case 1:  p = &one;   break;
 279   }
 280   AddressLiteral a(p);
 281   __ sethi(a, G3_scratch);
 282   __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
 283 }
 284 
 285 
 286 // %%%%% Should factore most snippet templates across platforms
 287 
 288 void TemplateTable::bipush() {
 289   transition(vtos, itos);
 290   __ ldsb( at_bcp(1), Otos_i );
 291 }
 292 
 293 void TemplateTable::sipush() {
 294   transition(vtos, itos);
 295   __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
 296 }
 297 
 298 void TemplateTable::ldc(bool wide) {
 299   transition(vtos, vtos);
 300   Label call_ldc, notInt, isString, notString, notClass, exit;
 301 
 302   if (wide) {
 303     __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 304   } else {
 305     __ ldub(Lbcp, 1, O1);
 306   }
 307   __ get_cpool_and_tags(O0, O2);
 308 
 309   const int base_offset = ConstantPool::header_size() * wordSize;
 310   const int tags_offset = Array<u1>::base_offset_in_bytes();
 311 
 312   // get type from tags
 313   __ add(O2, tags_offset, O2);
 314   __ ldub(O2, O1, O2);
 315 
 316   // unresolved class? If so, must resolve
 317   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClass, Assembler::equal, Assembler::pt, call_ldc);
 318 
 319   // unresolved class in error state
 320   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClassInError, Assembler::equal, Assembler::pn, call_ldc);
 321 
 322   __ cmp(O2, JVM_CONSTANT_Class);      // need to call vm to get java mirror of the class
 323   __ brx(Assembler::notEqual, true, Assembler::pt, notClass);
 324   __ delayed()->add(O0, base_offset, O0);
 325 
 326   __ bind(call_ldc);
 327   __ set(wide, O1);
 328   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), O1);
 329   __ push(atos);
 330   __ ba_short(exit);
 331 
 332   __ bind(notClass);
 333  // __ add(O0, base_offset, O0);
 334   __ sll(O1, LogBytesPerWord, O1);
 335   __ cmp(O2, JVM_CONSTANT_Integer);
 336   __ brx(Assembler::notEqual, true, Assembler::pt, notInt);
 337   __ delayed()->cmp(O2, JVM_CONSTANT_String);
 338   __ ld(O0, O1, Otos_i);
 339   __ push(itos);
 340   __ ba_short(exit);
 341 
 342   __ bind(notInt);
 343  // __ cmp(O2, JVM_CONSTANT_String);
 344   __ brx(Assembler::notEqual, true, Assembler::pt, notString);
 345   __ delayed()->ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
 346   __ bind(isString);
 347   __ stop("string should be rewritten to fast_aldc");
 348   __ ba_short(exit);
 349 
 350   __ bind(notString);
 351  // __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
 352   __ push(ftos);
 353 
 354   __ bind(exit);
 355 }
 356 
 357 // Fast path for caching oop constants.
 358 // %%% We should use this to handle Class and String constants also.
 359 // %%% It will simplify the ldc/primitive path considerably.
 360 void TemplateTable::fast_aldc(bool wide) {
 361   transition(vtos, atos);
 362 
 363   int index_size = wide ? sizeof(u2) : sizeof(u1);
 364   Label resolved;
 365 
 366   // We are resolved if the resolved reference cache entry contains a
 367   // non-null object (CallSite, etc.)
 368   assert_different_registers(Otos_i, G3_scratch);
 369   __ get_cache_index_at_bcp(Otos_i, G3_scratch, 1, index_size);  // load index => G3_scratch
 370   __ load_resolved_reference_at_index(Otos_i, G3_scratch);
 371   __ tst(Otos_i);
 372   __ br(Assembler::notEqual, false, Assembler::pt, resolved);
 373   __ delayed()->set((int)bytecode(), O1);
 374 
 375   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 376 
 377   // first time invocation - must resolve first
 378   __ call_VM(Otos_i, entry, O1);
 379   __ bind(resolved);
 380   __ verify_oop(Otos_i);
 381 }
 382 
 383 void TemplateTable::ldc2_w() {
 384   transition(vtos, vtos);
 385   Label Long, exit;
 386 
 387   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 388   __ get_cpool_and_tags(O0, O2);
 389 
 390   const int base_offset = ConstantPool::header_size() * wordSize;
 391   const int tags_offset = Array<u1>::base_offset_in_bytes();
 392   // get type from tags
 393   __ add(O2, tags_offset, O2);
 394   __ ldub(O2, O1, O2);
 395 
 396   __ sll(O1, LogBytesPerWord, O1);
 397   __ add(O0, O1, G3_scratch);
 398 
 399   __ cmp_and_brx_short(O2, JVM_CONSTANT_Double, Assembler::notEqual, Assembler::pt, Long);
 400   // A double can be placed at word-aligned locations in the constant pool.
 401   // Check out Conversions.java for an example.
 402   // Also ConstantPool::header_size() is 20, which makes it very difficult
 403   // to double-align double on the constant pool.  SG, 11/7/97
 404   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
 405   __ push(dtos);
 406   __ ba_short(exit);
 407 
 408   __ bind(Long);
 409   __ ldx(G3_scratch, base_offset, Otos_l);
 410   __ push(ltos);
 411 
 412   __ bind(exit);
 413 }
 414 
 415 void TemplateTable::locals_index(Register reg, int offset) {
 416   __ ldub( at_bcp(offset), reg );
 417 }
 418 
 419 void TemplateTable::locals_index_wide(Register reg) {
 420   // offset is 2, not 1, because Lbcp points to wide prefix code
 421   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
 422 }
 423 
 424 void TemplateTable::iload() {
 425   iload_internal();
 426 }
 427 
 428 void TemplateTable::nofast_iload() {
 429   iload_internal(may_not_rewrite);
 430 }
 431 
 432 void TemplateTable::iload_internal(RewriteControl rc) {
 433   transition(vtos, itos);
 434   // Rewrite iload,iload  pair into fast_iload2
 435   //         iload,caload pair into fast_icaload
 436   if (RewriteFrequentPairs && rc == may_rewrite) {
 437     Label rewrite, done;
 438 
 439     // get next byte
 440     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
 441 
 442     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 443     // last two iloads in a pair.  Comparing against fast_iload means that
 444     // the next bytecode is neither an iload or a caload, and therefore
 445     // an iload pair.
 446     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_iload, Assembler::equal, Assembler::pn, done);
 447 
 448     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
 449     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 450     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
 451 
 452     __ cmp(G3_scratch, (int)Bytecodes::_caload);
 453     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 454     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
 455 
 456     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
 457     // rewrite
 458     // G4_scratch: fast bytecode
 459     __ bind(rewrite);
 460     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
 461     __ bind(done);
 462   }
 463 
 464   // Get the local value into tos
 465   locals_index(G3_scratch);
 466   __ access_local_int( G3_scratch, Otos_i );
 467 }
 468 
 469 void TemplateTable::fast_iload2() {
 470   transition(vtos, itos);
 471   locals_index(G3_scratch);
 472   __ access_local_int( G3_scratch, Otos_i );
 473   __ push_i();
 474   locals_index(G3_scratch, 3);  // get next bytecode's local index.
 475   __ access_local_int( G3_scratch, Otos_i );
 476 }
 477 
 478 void TemplateTable::fast_iload() {
 479   transition(vtos, itos);
 480   locals_index(G3_scratch);
 481   __ access_local_int( G3_scratch, Otos_i );
 482 }
 483 
 484 void TemplateTable::lload() {
 485   transition(vtos, ltos);
 486   locals_index(G3_scratch);
 487   __ access_local_long( G3_scratch, Otos_l );
 488 }
 489 
 490 
 491 void TemplateTable::fload() {
 492   transition(vtos, ftos);
 493   locals_index(G3_scratch);
 494   __ access_local_float( G3_scratch, Ftos_f );
 495 }
 496 
 497 
 498 void TemplateTable::dload() {
 499   transition(vtos, dtos);
 500   locals_index(G3_scratch);
 501   __ access_local_double( G3_scratch, Ftos_d );
 502 }
 503 
 504 
 505 void TemplateTable::aload() {
 506   transition(vtos, atos);
 507   locals_index(G3_scratch);
 508   __ access_local_ptr( G3_scratch, Otos_i);
 509 }
 510 
 511 
 512 void TemplateTable::wide_iload() {
 513   transition(vtos, itos);
 514   locals_index_wide(G3_scratch);
 515   __ access_local_int( G3_scratch, Otos_i );
 516 }
 517 
 518 
 519 void TemplateTable::wide_lload() {
 520   transition(vtos, ltos);
 521   locals_index_wide(G3_scratch);
 522   __ access_local_long( G3_scratch, Otos_l );
 523 }
 524 
 525 
 526 void TemplateTable::wide_fload() {
 527   transition(vtos, ftos);
 528   locals_index_wide(G3_scratch);
 529   __ access_local_float( G3_scratch, Ftos_f );
 530 }
 531 
 532 
 533 void TemplateTable::wide_dload() {
 534   transition(vtos, dtos);
 535   locals_index_wide(G3_scratch);
 536   __ access_local_double( G3_scratch, Ftos_d );
 537 }
 538 
 539 
 540 void TemplateTable::wide_aload() {
 541   transition(vtos, atos);
 542   locals_index_wide(G3_scratch);
 543   __ access_local_ptr( G3_scratch, Otos_i );
 544   __ verify_oop(Otos_i);
 545 }
 546 
 547 
 548 void TemplateTable::iaload() {
 549   transition(itos, itos);
 550   // Otos_i: index
 551   // tos: array
 552   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 553   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
 554 }
 555 
 556 
 557 void TemplateTable::laload() {
 558   transition(itos, ltos);
 559   // Otos_i: index
 560   // O2: array
 561   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 562   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
 563 }
 564 
 565 
 566 void TemplateTable::faload() {
 567   transition(itos, ftos);
 568   // Otos_i: index
 569   // O2: array
 570   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 571   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
 572 }
 573 
 574 
 575 void TemplateTable::daload() {
 576   transition(itos, dtos);
 577   // Otos_i: index
 578   // O2: array
 579   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 580   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
 581 }
 582 
 583 
 584 void TemplateTable::aaload() {
 585   transition(itos, atos);
 586   // Otos_i: index
 587   // tos: array
 588   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
 589   __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
 590   __ verify_oop(Otos_i);
 591 }
 592 
 593 
 594 void TemplateTable::baload() {
 595   transition(itos, itos);
 596   // Otos_i: index
 597   // tos: array
 598   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
 599   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
 600 }
 601 
 602 
 603 void TemplateTable::caload() {
 604   transition(itos, itos);
 605   // Otos_i: index
 606   // tos: array
 607   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 608   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 609 }
 610 
 611 void TemplateTable::fast_icaload() {
 612   transition(vtos, itos);
 613   // Otos_i: index
 614   // tos: array
 615   locals_index(G3_scratch);
 616   __ access_local_int( G3_scratch, Otos_i );
 617   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 618   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 619 }
 620 
 621 
 622 void TemplateTable::saload() {
 623   transition(itos, itos);
 624   // Otos_i: index
 625   // tos: array
 626   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 627   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
 628 }
 629 
 630 
 631 void TemplateTable::iload(int n) {
 632   transition(vtos, itos);
 633   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 634 }
 635 
 636 
 637 void TemplateTable::lload(int n) {
 638   transition(vtos, ltos);
 639   assert(n+1 < Argument::n_register_parameters, "would need more code");
 640   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
 641 }
 642 
 643 
 644 void TemplateTable::fload(int n) {
 645   transition(vtos, ftos);
 646   assert(n < Argument::n_register_parameters, "would need more code");
 647   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
 648 }
 649 
 650 
 651 void TemplateTable::dload(int n) {
 652   transition(vtos, dtos);
 653   FloatRegister dst = Ftos_d;
 654   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
 655 }
 656 
 657 
 658 void TemplateTable::aload(int n) {
 659   transition(vtos, atos);
 660   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 661 }
 662 
 663 void TemplateTable::aload_0() {
 664   aload_0_internal();
 665 }
 666 
 667 void TemplateTable::nofast_aload_0() {
 668   aload_0_internal(may_not_rewrite);
 669 }
 670 
 671 void TemplateTable::aload_0_internal(RewriteControl rc) {
 672   transition(vtos, atos);
 673 
 674   // According to bytecode histograms, the pairs:
 675   //
 676   // _aload_0, _fast_igetfield (itos)
 677   // _aload_0, _fast_agetfield (atos)
 678   // _aload_0, _fast_fgetfield (ftos)
 679   //
 680   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
 681   // bytecode checks the next bytecode and then rewrites the current
 682   // bytecode into a pair bytecode; otherwise it rewrites the current
 683   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
 684   //
 685   if (RewriteFrequentPairs && rc == may_rewrite) {
 686     Label rewrite, done;
 687 
 688     // get next byte
 689     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
 690 
 691     // if _getfield then wait with rewrite
 692     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_getfield, Assembler::equal, Assembler::pn, done);
 693 
 694     // if _igetfield then rewrite to _fast_iaccess_0
 695     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 696     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
 697     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 698     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
 699 
 700     // if _agetfield then rewrite to _fast_aaccess_0
 701     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 702     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
 703     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 704     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
 705 
 706     // if _fgetfield then rewrite to _fast_faccess_0
 707     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 708     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
 709     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 710     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
 711 
 712     // else rewrite to _fast_aload0
 713     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 714     __ set(Bytecodes::_fast_aload_0, G4_scratch);
 715 
 716     // rewrite
 717     // G4_scratch: fast bytecode
 718     __ bind(rewrite);
 719     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
 720     __ bind(done);
 721   }
 722 
 723   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 724   aload(0);
 725 }
 726 
 727 void TemplateTable::istore() {
 728   transition(itos, vtos);
 729   locals_index(G3_scratch);
 730   __ store_local_int( G3_scratch, Otos_i );
 731 }
 732 
 733 
 734 void TemplateTable::lstore() {
 735   transition(ltos, vtos);
 736   locals_index(G3_scratch);
 737   __ store_local_long( G3_scratch, Otos_l );
 738 }
 739 
 740 
 741 void TemplateTable::fstore() {
 742   transition(ftos, vtos);
 743   locals_index(G3_scratch);
 744   __ store_local_float( G3_scratch, Ftos_f );
 745 }
 746 
 747 
 748 void TemplateTable::dstore() {
 749   transition(dtos, vtos);
 750   locals_index(G3_scratch);
 751   __ store_local_double( G3_scratch, Ftos_d );
 752 }
 753 
 754 
 755 void TemplateTable::astore() {
 756   transition(vtos, vtos);
 757   __ load_ptr(0, Otos_i);
 758   __ inc(Lesp, Interpreter::stackElementSize);
 759   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 760   locals_index(G3_scratch);
 761   __ store_local_ptr(G3_scratch, Otos_i);
 762 }
 763 
 764 
 765 void TemplateTable::wide_istore() {
 766   transition(vtos, vtos);
 767   __ pop_i();
 768   locals_index_wide(G3_scratch);
 769   __ store_local_int( G3_scratch, Otos_i );
 770 }
 771 
 772 
 773 void TemplateTable::wide_lstore() {
 774   transition(vtos, vtos);
 775   __ pop_l();
 776   locals_index_wide(G3_scratch);
 777   __ store_local_long( G3_scratch, Otos_l );
 778 }
 779 
 780 
 781 void TemplateTable::wide_fstore() {
 782   transition(vtos, vtos);
 783   __ pop_f();
 784   locals_index_wide(G3_scratch);
 785   __ store_local_float( G3_scratch, Ftos_f );
 786 }
 787 
 788 
 789 void TemplateTable::wide_dstore() {
 790   transition(vtos, vtos);
 791   __ pop_d();
 792   locals_index_wide(G3_scratch);
 793   __ store_local_double( G3_scratch, Ftos_d );
 794 }
 795 
 796 
 797 void TemplateTable::wide_astore() {
 798   transition(vtos, vtos);
 799   __ load_ptr(0, Otos_i);
 800   __ inc(Lesp, Interpreter::stackElementSize);
 801   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 802   locals_index_wide(G3_scratch);
 803   __ store_local_ptr(G3_scratch, Otos_i);
 804 }
 805 
 806 
 807 void TemplateTable::iastore() {
 808   transition(itos, vtos);
 809   __ pop_i(O2); // index
 810   // Otos_i: val
 811   // O3: array
 812   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 813   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
 814 }
 815 
 816 
 817 void TemplateTable::lastore() {
 818   transition(ltos, vtos);
 819   __ pop_i(O2); // index
 820   // Otos_l: val
 821   // O3: array
 822   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 823   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
 824 }
 825 
 826 
 827 void TemplateTable::fastore() {
 828   transition(ftos, vtos);
 829   __ pop_i(O2); // index
 830   // Ftos_f: val
 831   // O3: array
 832   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 833   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
 834 }
 835 
 836 
 837 void TemplateTable::dastore() {
 838   transition(dtos, vtos);
 839   __ pop_i(O2); // index
 840   // Fos_d: val
 841   // O3: array
 842   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 843   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
 844 }
 845 
 846 
 847 void TemplateTable::aastore() {
 848   Label store_ok, is_null, done;
 849   transition(vtos, vtos);
 850   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
 851   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
 852   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
 853   // Otos_i: val
 854   // O2: index
 855   // O3: array
 856   __ verify_oop(Otos_i);
 857   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
 858 
 859   // do array store check - check for NULL value first
 860   __ br_null_short( Otos_i, Assembler::pn, is_null );
 861 
 862   __ load_klass(O3, O4); // get array klass
 863   __ load_klass(Otos_i, O5); // get value klass
 864 
 865   // do fast instanceof cache test
 866 
 867   __ ld_ptr(O4,     in_bytes(ObjArrayKlass::element_klass_offset()),  O4);
 868 
 869   assert(Otos_i == O0, "just checking");
 870 
 871   // Otos_i:    value
 872   // O1:        addr - offset
 873   // O2:        index
 874   // O3:        array
 875   // O4:        array element klass
 876   // O5:        value klass
 877 
 878   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
 879 
 880   // Generate a fast subtype check.  Branch to store_ok if no
 881   // failure.  Throw if failure.
 882   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
 883 
 884   // Not a subtype; so must throw exception
 885   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
 886 
 887   // Store is OK.
 888   __ bind(store_ok);
 889   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
 890 
 891   __ ba(done);
 892   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
 893 
 894   __ bind(is_null);
 895   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
 896 
 897   __ profile_null_seen(G3_scratch);
 898   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
 899   __ bind(done);
 900 }
 901 
 902 
 903 void TemplateTable::bastore() {
 904   transition(itos, vtos);
 905   __ pop_i(O2); // index
 906   // Otos_i: val
 907   // O2: index
 908   // O3: array
 909   __ index_check(O3, O2, 0, G3_scratch, O2);
 910   // Need to check whether array is boolean or byte
 911   // since both types share the bastore bytecode.
 912   __ load_klass(O3, G4_scratch);
 913   __ ld(G4_scratch, in_bytes(Klass::layout_helper_offset()), G4_scratch);
 914   __ set(Klass::layout_helper_boolean_diffbit(), G3_scratch);
 915   __ andcc(G3_scratch, G4_scratch, G0);
 916   Label L_skip;
 917   __ br(Assembler::zero, false, Assembler::pn, L_skip);
 918   __ delayed()->nop();
 919   __ and3(Otos_i, 1, Otos_i);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
 920   __ bind(L_skip);
 921   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
 922 }
 923 
 924 
 925 void TemplateTable::castore() {
 926   transition(itos, vtos);
 927   __ pop_i(O2); // index
 928   // Otos_i: val
 929   // O3: array
 930   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
 931   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
 932 }
 933 
 934 
 935 void TemplateTable::sastore() {
 936   // %%%%% Factor across platform
 937   castore();
 938 }
 939 
 940 
 941 void TemplateTable::istore(int n) {
 942   transition(itos, vtos);
 943   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
 944 }
 945 
 946 
 947 void TemplateTable::lstore(int n) {
 948   transition(ltos, vtos);
 949   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
 950   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
 951 
 952 }
 953 
 954 
 955 void TemplateTable::fstore(int n) {
 956   transition(ftos, vtos);
 957   assert(n < Argument::n_register_parameters, "only handle register cases");
 958   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
 959 }
 960 
 961 
 962 void TemplateTable::dstore(int n) {
 963   transition(dtos, vtos);
 964   FloatRegister src = Ftos_d;
 965   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
 966 }
 967 
 968 
 969 void TemplateTable::astore(int n) {
 970   transition(vtos, vtos);
 971   __ load_ptr(0, Otos_i);
 972   __ inc(Lesp, Interpreter::stackElementSize);
 973   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 974   __ store_local_ptr(n, Otos_i);
 975 }
 976 
 977 
 978 void TemplateTable::pop() {
 979   transition(vtos, vtos);
 980   __ inc(Lesp, Interpreter::stackElementSize);
 981 }
 982 
 983 
 984 void TemplateTable::pop2() {
 985   transition(vtos, vtos);
 986   __ inc(Lesp, 2 * Interpreter::stackElementSize);
 987 }
 988 
 989 
 990 void TemplateTable::dup() {
 991   transition(vtos, vtos);
 992   // stack: ..., a
 993   // load a and tag
 994   __ load_ptr(0, Otos_i);
 995   __ push_ptr(Otos_i);
 996   // stack: ..., a, a
 997 }
 998 
 999 
1000 void TemplateTable::dup_x1() {
1001   transition(vtos, vtos);
1002   // stack: ..., a, b
1003   __ load_ptr( 1, G3_scratch);  // get a
1004   __ load_ptr( 0, Otos_l1);     // get b
1005   __ store_ptr(1, Otos_l1);     // put b
1006   __ store_ptr(0, G3_scratch);  // put a - like swap
1007   __ push_ptr(Otos_l1);         // push b
1008   // stack: ..., b, a, b
1009 }
1010 
1011 
1012 void TemplateTable::dup_x2() {
1013   transition(vtos, vtos);
1014   // stack: ..., a, b, c
1015   // get c and push on stack, reuse registers
1016   __ load_ptr( 0, G3_scratch);  // get c
1017   __ push_ptr(G3_scratch);      // push c with tag
1018   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
1019   // (stack offsets n+1 now)
1020   __ load_ptr( 3, Otos_l1);     // get a
1021   __ store_ptr(3, G3_scratch);  // put c at 3
1022   // stack: ..., c, b, c, c  (a in reg)
1023   __ load_ptr( 2, G3_scratch);  // get b
1024   __ store_ptr(2, Otos_l1);     // put a at 2
1025   // stack: ..., c, a, c, c  (b in reg)
1026   __ store_ptr(1, G3_scratch);  // put b at 1
1027   // stack: ..., c, a, b, c
1028 }
1029 
1030 
1031 void TemplateTable::dup2() {
1032   transition(vtos, vtos);
1033   __ load_ptr(1, G3_scratch);  // get a
1034   __ load_ptr(0, Otos_l1);     // get b
1035   __ push_ptr(G3_scratch);     // push a
1036   __ push_ptr(Otos_l1);        // push b
1037   // stack: ..., a, b, a, b
1038 }
1039 
1040 
1041 void TemplateTable::dup2_x1() {
1042   transition(vtos, vtos);
1043   // stack: ..., a, b, c
1044   __ load_ptr( 1, Lscratch);    // get b
1045   __ load_ptr( 2, Otos_l1);     // get a
1046   __ store_ptr(2, Lscratch);    // put b at a
1047   // stack: ..., b, b, c
1048   __ load_ptr( 0, G3_scratch);  // get c
1049   __ store_ptr(1, G3_scratch);  // put c at b
1050   // stack: ..., b, c, c
1051   __ store_ptr(0, Otos_l1);     // put a at c
1052   // stack: ..., b, c, a
1053   __ push_ptr(Lscratch);        // push b
1054   __ push_ptr(G3_scratch);      // push c
1055   // stack: ..., b, c, a, b, c
1056 }
1057 
1058 
1059 // The spec says that these types can be a mixture of category 1 (1 word)
1060 // types and/or category 2 types (long and doubles)
1061 void TemplateTable::dup2_x2() {
1062   transition(vtos, vtos);
1063   // stack: ..., a, b, c, d
1064   __ load_ptr( 1, Lscratch);    // get c
1065   __ load_ptr( 3, Otos_l1);     // get a
1066   __ store_ptr(3, Lscratch);    // put c at 3
1067   __ store_ptr(1, Otos_l1);     // put a at 1
1068   // stack: ..., c, b, a, d
1069   __ load_ptr( 2, G3_scratch);  // get b
1070   __ load_ptr( 0, Otos_l1);     // get d
1071   __ store_ptr(0, G3_scratch);  // put b at 0
1072   __ store_ptr(2, Otos_l1);     // put d at 2
1073   // stack: ..., c, d, a, b
1074   __ push_ptr(Lscratch);        // push c
1075   __ push_ptr(Otos_l1);         // push d
1076   // stack: ..., c, d, a, b, c, d
1077 }
1078 
1079 
1080 void TemplateTable::swap() {
1081   transition(vtos, vtos);
1082   // stack: ..., a, b
1083   __ load_ptr( 1, G3_scratch);  // get a
1084   __ load_ptr( 0, Otos_l1);     // get b
1085   __ store_ptr(0, G3_scratch);  // put b
1086   __ store_ptr(1, Otos_l1);     // put a
1087   // stack: ..., b, a
1088 }
1089 
1090 
1091 void TemplateTable::iop2(Operation op) {
1092   transition(itos, itos);
1093   __ pop_i(O1);
1094   switch (op) {
1095    case  add:  __  add(O1, Otos_i, Otos_i);  break;
1096    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
1097      // %%%%% Mul may not exist: better to call .mul?
1098    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
1099    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
1100    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
1101    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
1102    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
1103    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
1104    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
1105    default: ShouldNotReachHere();
1106   }
1107 }
1108 
1109 
1110 void TemplateTable::lop2(Operation op) {
1111   transition(ltos, ltos);
1112   __ pop_l(O2);
1113   switch (op) {
1114    case  add:  __  add(O2, Otos_l, Otos_l);  break;
1115    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
1116    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
1117    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
1118    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
1119    default: ShouldNotReachHere();
1120   }
1121 }
1122 
1123 
1124 void TemplateTable::idiv() {
1125   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
1126   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
1127 
1128   transition(itos, itos);
1129   __ pop_i(O1); // get 1st op
1130 
1131   // Y contains upper 32 bits of result, set it to 0 or all ones
1132   __ wry(G0);
1133   __ mov(~0, G3_scratch);
1134 
1135   __ tst(O1);
1136      Label neg;
1137   __ br(Assembler::negative, true, Assembler::pn, neg);
1138   __ delayed()->wry(G3_scratch);
1139   __ bind(neg);
1140 
1141      Label ok;
1142   __ tst(Otos_i);
1143   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
1144 
1145   const int min_int = 0x80000000;
1146   Label regular;
1147   __ cmp(Otos_i, -1);
1148   __ br(Assembler::notEqual, false, Assembler::pt, regular);
1149   // Don't put set in delay slot
1150   // Set will turn into multiple instructions in 64 bit mode
1151   __ delayed()->nop();
1152   __ set(min_int, G4_scratch);
1153   Label done;
1154   __ cmp(O1, G4_scratch);
1155   __ br(Assembler::equal, true, Assembler::pt, done);
1156   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
1157 
1158   __ bind(regular);
1159   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
1160   __ bind(done);
1161 }
1162 
1163 
1164 void TemplateTable::irem() {
1165   transition(itos, itos);
1166   __ mov(Otos_i, O2); // save divisor
1167   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
1168   __ smul(Otos_i, O2, Otos_i);
1169   __ sub(O1, Otos_i, Otos_i);
1170 }
1171 
1172 
1173 void TemplateTable::lmul() {
1174   transition(ltos, ltos);
1175   __ pop_l(O2);
1176   __ mulx(Otos_l, O2, Otos_l);
1177 
1178 }
1179 
1180 
1181 void TemplateTable::ldiv() {
1182   transition(ltos, ltos);
1183 
1184   // check for zero
1185   __ pop_l(O2);
1186   __ tst(Otos_l);
1187   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1188   __ sdivx(O2, Otos_l, Otos_l);
1189 }
1190 
1191 
1192 void TemplateTable::lrem() {
1193   transition(ltos, ltos);
1194 
1195   // check for zero
1196   __ pop_l(O2);
1197   __ tst(Otos_l);
1198   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1199   __ sdivx(O2, Otos_l, Otos_l2);
1200   __ mulx (Otos_l2, Otos_l, Otos_l2);
1201   __ sub  (O2, Otos_l2, Otos_l);
1202 }
1203 
1204 
1205 void TemplateTable::lshl() {
1206   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
1207 
1208   __ pop_l(O2);                          // shift value in O2, O3
1209   __ sllx(O2, Otos_i, Otos_l);
1210 }
1211 
1212 
1213 void TemplateTable::lshr() {
1214   transition(itos, ltos); // %%%% see lshl comment
1215 
1216   __ pop_l(O2);                          // shift value in O2, O3
1217   __ srax(O2, Otos_i, Otos_l);
1218 }
1219 
1220 
1221 
1222 void TemplateTable::lushr() {
1223   transition(itos, ltos); // %%%% see lshl comment
1224 
1225   __ pop_l(O2);                          // shift value in O2, O3
1226   __ srlx(O2, Otos_i, Otos_l);
1227 }
1228 
1229 
1230 void TemplateTable::fop2(Operation op) {
1231   transition(ftos, ftos);
1232   switch (op) {
1233    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1234    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1235    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1236    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1237    case  rem:
1238      assert(Ftos_f == F0, "just checking");
1239      // LP64 calling conventions use F1, F3 for passing 2 floats
1240      __ pop_f(F1);
1241      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
1242      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1243      assert( Ftos_f == F0, "fix this code" );
1244      break;
1245 
1246    default: ShouldNotReachHere();
1247   }
1248 }
1249 
1250 
1251 void TemplateTable::dop2(Operation op) {
1252   transition(dtos, dtos);
1253   switch (op) {
1254    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1255    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1256    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1257    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1258    case  rem:
1259      // Pass arguments in D0, D2
1260      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
1261      __ pop_d( F0 );
1262      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1263      assert( Ftos_d == F0, "fix this code" );
1264      break;
1265 
1266    default: ShouldNotReachHere();
1267   }
1268 }
1269 
1270 
1271 void TemplateTable::ineg() {
1272   transition(itos, itos);
1273   __ neg(Otos_i);
1274 }
1275 
1276 
1277 void TemplateTable::lneg() {
1278   transition(ltos, ltos);
1279   __ sub(G0, Otos_l, Otos_l);
1280 }
1281 
1282 
1283 void TemplateTable::fneg() {
1284   transition(ftos, ftos);
1285   __ fneg(FloatRegisterImpl::S, Ftos_f, Ftos_f);
1286 }
1287 
1288 
1289 void TemplateTable::dneg() {
1290   transition(dtos, dtos);
1291   __ fneg(FloatRegisterImpl::D, Ftos_f, Ftos_f);
1292 }
1293 
1294 
1295 void TemplateTable::iinc() {
1296   transition(vtos, vtos);
1297   locals_index(G3_scratch);
1298   __ ldsb(Lbcp, 2, O2);  // load constant
1299   __ access_local_int(G3_scratch, Otos_i);
1300   __ add(Otos_i, O2, Otos_i);
1301   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1302 }
1303 
1304 
1305 void TemplateTable::wide_iinc() {
1306   transition(vtos, vtos);
1307   locals_index_wide(G3_scratch);
1308   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
1309   __ access_local_int(G3_scratch, Otos_i);
1310   __ add(Otos_i, O3, Otos_i);
1311   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1312 }
1313 
1314 
1315 void TemplateTable::convert() {
1316 // %%%%% Factor this first part accross platforms
1317   #ifdef ASSERT
1318     TosState tos_in  = ilgl;
1319     TosState tos_out = ilgl;
1320     switch (bytecode()) {
1321       case Bytecodes::_i2l: // fall through
1322       case Bytecodes::_i2f: // fall through
1323       case Bytecodes::_i2d: // fall through
1324       case Bytecodes::_i2b: // fall through
1325       case Bytecodes::_i2c: // fall through
1326       case Bytecodes::_i2s: tos_in = itos; break;
1327       case Bytecodes::_l2i: // fall through
1328       case Bytecodes::_l2f: // fall through
1329       case Bytecodes::_l2d: tos_in = ltos; break;
1330       case Bytecodes::_f2i: // fall through
1331       case Bytecodes::_f2l: // fall through
1332       case Bytecodes::_f2d: tos_in = ftos; break;
1333       case Bytecodes::_d2i: // fall through
1334       case Bytecodes::_d2l: // fall through
1335       case Bytecodes::_d2f: tos_in = dtos; break;
1336       default             : ShouldNotReachHere();
1337     }
1338     switch (bytecode()) {
1339       case Bytecodes::_l2i: // fall through
1340       case Bytecodes::_f2i: // fall through
1341       case Bytecodes::_d2i: // fall through
1342       case Bytecodes::_i2b: // fall through
1343       case Bytecodes::_i2c: // fall through
1344       case Bytecodes::_i2s: tos_out = itos; break;
1345       case Bytecodes::_i2l: // fall through
1346       case Bytecodes::_f2l: // fall through
1347       case Bytecodes::_d2l: tos_out = ltos; break;
1348       case Bytecodes::_i2f: // fall through
1349       case Bytecodes::_l2f: // fall through
1350       case Bytecodes::_d2f: tos_out = ftos; break;
1351       case Bytecodes::_i2d: // fall through
1352       case Bytecodes::_l2d: // fall through
1353       case Bytecodes::_f2d: tos_out = dtos; break;
1354       default             : ShouldNotReachHere();
1355     }
1356     transition(tos_in, tos_out);
1357   #endif
1358 
1359 
1360   // Conversion
1361   Label done;
1362   switch (bytecode()) {
1363    case Bytecodes::_i2l:
1364     // Sign extend the 32 bits
1365     __ sra ( Otos_i, 0, Otos_l );
1366     break;
1367 
1368    case Bytecodes::_i2f:
1369     __ st(Otos_i, __ d_tmp );
1370     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1371     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
1372     break;
1373 
1374    case Bytecodes::_i2d:
1375     __ st(Otos_i, __ d_tmp);
1376     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1377     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
1378     break;
1379 
1380    case Bytecodes::_i2b:
1381     __ sll(Otos_i, 24, Otos_i);
1382     __ sra(Otos_i, 24, Otos_i);
1383     break;
1384 
1385    case Bytecodes::_i2c:
1386     __ sll(Otos_i, 16, Otos_i);
1387     __ srl(Otos_i, 16, Otos_i);
1388     break;
1389 
1390    case Bytecodes::_i2s:
1391     __ sll(Otos_i, 16, Otos_i);
1392     __ sra(Otos_i, 16, Otos_i);
1393     break;
1394 
1395    case Bytecodes::_l2i:
1396     // Sign-extend into the high 32 bits
1397     __ sra(Otos_l, 0, Otos_i);
1398     break;
1399 
1400    case Bytecodes::_l2f:
1401    case Bytecodes::_l2d:
1402     __ st_long(Otos_l, __ d_tmp);
1403     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
1404 
1405     if (bytecode() == Bytecodes::_l2f) {
1406       __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
1407     } else {
1408       __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
1409     }
1410     break;
1411 
1412   case Bytecodes::_f2i:  {
1413       Label isNaN;
1414       // result must be 0 if value is NaN; test by comparing value to itself
1415       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
1416       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
1417       __ delayed()->clr(Otos_i);                                     // NaN
1418       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
1419       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
1420       __ ld(__ d_tmp, Otos_i);
1421       __ bind(isNaN);
1422     }
1423     break;
1424 
1425    case Bytecodes::_f2l:
1426     // must uncache tos
1427     __ push_f();
1428     __ pop_f(F1);
1429     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1430     break;
1431 
1432    case Bytecodes::_f2d:
1433     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
1434     break;
1435 
1436    case Bytecodes::_d2i:
1437    case Bytecodes::_d2l:
1438     // must uncache tos
1439     __ push_d();
1440     // LP64 calling conventions pass first double arg in D0
1441     __ pop_d( Ftos_d );
1442     __ call_VM_leaf(Lscratch,
1443         bytecode() == Bytecodes::_d2i
1444           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
1445           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1446     break;
1447 
1448     case Bytecodes::_d2f:
1449       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
1450     break;
1451 
1452     default: ShouldNotReachHere();
1453   }
1454   __ bind(done);
1455 }
1456 
1457 
1458 void TemplateTable::lcmp() {
1459   transition(ltos, itos);
1460 
1461   __ pop_l(O1); // pop off value 1, value 2 is in O0
1462   __ lcmp( O1, Otos_l, Otos_i );
1463 }
1464 
1465 
1466 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1467 
1468   if (is_float) __ pop_f(F2);
1469   else          __ pop_d(F2);
1470 
1471   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
1472 
1473   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
1474 }
1475 
1476 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1477   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1478   __ verify_thread();
1479 
1480   const Register O2_bumped_count = O2;
1481   __ profile_taken_branch(G3_scratch, O2_bumped_count);
1482 
1483   // get (wide) offset to O1_disp
1484   const Register O1_disp = O1;
1485   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
1486   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
1487 
1488   // Handle all the JSR stuff here, then exit.
1489   // It's much shorter and cleaner than intermingling with the
1490   // non-JSR normal-branch stuff occurring below.
1491   if( is_jsr ) {
1492     // compute return address as bci in Otos_i
1493     __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1494     __ sub(Lbcp, G3_scratch, G3_scratch);
1495     __ sub(G3_scratch, in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
1496 
1497     // Bump Lbcp to target of JSR
1498     __ add(Lbcp, O1_disp, Lbcp);
1499     // Push returnAddress for "ret" on stack
1500     __ push_ptr(Otos_i);
1501     // And away we go!
1502     __ dispatch_next(vtos);
1503     return;
1504   }
1505 
1506   // Normal (non-jsr) branch handling
1507 
1508   // Save the current Lbcp
1509   const Register l_cur_bcp = Lscratch;
1510   __ mov( Lbcp, l_cur_bcp );
1511 
1512   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1513   if ( increment_invocation_counter_for_backward_branches ) {
1514     Label Lforward;
1515     // check branch direction
1516     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
1517     // Bump bytecode pointer by displacement (take the branch)
1518     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
1519 
1520     const Register G3_method_counters = G3_scratch;
1521     __ get_method_counters(Lmethod, G3_method_counters, Lforward);
1522 
1523     if (TieredCompilation) {
1524       Label Lno_mdo, Loverflow;
1525       int increment = InvocationCounter::count_increment;
1526       if (ProfileInterpreter) {
1527         // If no method data exists, go to profile_continue.
1528         __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch);
1529         __ br_null_short(G4_scratch, Assembler::pn, Lno_mdo);
1530 
1531         // Increment backedge counter in the MDO
1532         Address mdo_backedge_counter(G4_scratch, in_bytes(MethodData::backedge_counter_offset()) +
1533                                                  in_bytes(InvocationCounter::counter_offset()));
1534         Address mask(G4_scratch, in_bytes(MethodData::backedge_mask_offset()));
1535         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, O0,
1536                                    (UseOnStackReplacement ? Assembler::notZero : Assembler::always), &Lforward);
1537         __ ba_short(Loverflow);
1538       }
1539 
1540       // If there's no MDO, increment counter in MethodCounters*
1541       __ bind(Lno_mdo);
1542       Address backedge_counter(G3_method_counters,
1543               in_bytes(MethodCounters::backedge_counter_offset()) +
1544               in_bytes(InvocationCounter::counter_offset()));
1545       Address mask(G3_method_counters, in_bytes(MethodCounters::backedge_mask_offset()));
1546       __ increment_mask_and_jump(backedge_counter, increment, mask, G4_scratch, O0,
1547                                  (UseOnStackReplacement ? Assembler::notZero : Assembler::always), &Lforward);
1548       __ bind(Loverflow);
1549 
1550       // notify point for loop, pass branch bytecode
1551       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), l_cur_bcp);
1552 
1553       // Was an OSR adapter generated?
1554       // O0 = osr nmethod
1555       __ br_null_short(O0, Assembler::pn, Lforward);
1556 
1557       // Has the nmethod been invalidated already?
1558       __ ldub(O0, nmethod::state_offset(), O2);
1559       __ cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, Lforward);
1560 
1561       // migrate the interpreter frame off of the stack
1562 
1563       __ mov(G2_thread, L7);
1564       // save nmethod
1565       __ mov(O0, L6);
1566       __ set_last_Java_frame(SP, noreg);
1567       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
1568       __ reset_last_Java_frame();
1569       __ mov(L7, G2_thread);
1570 
1571       // move OSR nmethod to I1
1572       __ mov(L6, I1);
1573 
1574       // OSR buffer to I0
1575       __ mov(O0, I0);
1576 
1577       // remove the interpreter frame
1578       __ restore(I5_savedSP, 0, SP);
1579 
1580       // Jump to the osr code.
1581       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
1582       __ jmp(O2, G0);
1583       __ delayed()->nop();
1584 
1585     } else { // not TieredCompilation
1586       // Update Backedge branch separately from invocations
1587       const Register G4_invoke_ctr = G4;
1588       __ increment_backedge_counter(G3_method_counters, G4_invoke_ctr, G1_scratch);
1589       if (ProfileInterpreter) {
1590         __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_method_counters, G1_scratch, Lforward);
1591         if (UseOnStackReplacement) {
1592 
1593           __ test_backedge_count_for_osr(O2_bumped_count, G3_method_counters, l_cur_bcp, G1_scratch);
1594         }
1595       } else {
1596         if (UseOnStackReplacement) {
1597           __ test_backedge_count_for_osr(G4_invoke_ctr, G3_method_counters, l_cur_bcp, G1_scratch);
1598         }
1599       }
1600     }
1601 
1602     __ bind(Lforward);
1603   } else
1604     // Bump bytecode pointer by displacement (take the branch)
1605     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
1606 
1607   // continue with bytecode @ target
1608   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1609   // %%%%% and changing dispatch_next to dispatch_only
1610   __ dispatch_next(vtos);
1611 }
1612 
1613 
1614 // Note Condition in argument is TemplateTable::Condition
1615 // arg scope is within class scope
1616 
1617 void TemplateTable::if_0cmp(Condition cc) {
1618   // no pointers, integer only!
1619   transition(itos, vtos);
1620   // assume branch is more often taken than not (loops use backward branches)
1621   __ cmp( Otos_i, 0);
1622   __ if_cmp(ccNot(cc), false);
1623 }
1624 
1625 
1626 void TemplateTable::if_icmp(Condition cc) {
1627   transition(itos, vtos);
1628   __ pop_i(O1);
1629   __ cmp(O1, Otos_i);
1630   __ if_cmp(ccNot(cc), false);
1631 }
1632 
1633 
1634 void TemplateTable::if_nullcmp(Condition cc) {
1635   transition(atos, vtos);
1636   __ tst(Otos_i);
1637   __ if_cmp(ccNot(cc), true);
1638 }
1639 
1640 
1641 void TemplateTable::if_acmp(Condition cc) {
1642   transition(atos, vtos);
1643   __ pop_ptr(O1);
1644   __ verify_oop(O1);
1645   __ verify_oop(Otos_i);
1646   __ cmp(O1, Otos_i);
1647   __ if_cmp(ccNot(cc), true);
1648 }
1649 
1650 
1651 
1652 void TemplateTable::ret() {
1653   transition(vtos, vtos);
1654   locals_index(G3_scratch);
1655   __ access_local_returnAddress(G3_scratch, Otos_i);
1656   // Otos_i contains the bci, compute the bcp from that
1657 
1658 #ifdef ASSERT
1659   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
1660   // the result.  The return address (really a BCI) was stored with an
1661   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
1662   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
1663   // loaded value.
1664   { Label zzz ;
1665      __ set (65536, G3_scratch) ;
1666      __ cmp (Otos_i, G3_scratch) ;
1667      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
1668      __ delayed()->nop();
1669      __ stop("BCI is in the wrong register half?");
1670      __ bind (zzz) ;
1671   }
1672 #endif
1673 
1674   __ profile_ret(vtos, Otos_i, G4_scratch);
1675 
1676   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1677   __ add(G3_scratch, Otos_i, G3_scratch);
1678   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1679   __ dispatch_next(vtos);
1680 }
1681 
1682 
1683 void TemplateTable::wide_ret() {
1684   transition(vtos, vtos);
1685   locals_index_wide(G3_scratch);
1686   __ access_local_returnAddress(G3_scratch, Otos_i);
1687   // Otos_i contains the bci, compute the bcp from that
1688 
1689   __ profile_ret(vtos, Otos_i, G4_scratch);
1690 
1691   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1692   __ add(G3_scratch, Otos_i, G3_scratch);
1693   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1694   __ dispatch_next(vtos);
1695 }
1696 
1697 
1698 void TemplateTable::tableswitch() {
1699   transition(itos, vtos);
1700   Label default_case, continue_execution;
1701 
1702   // align bcp
1703   __ add(Lbcp, BytesPerInt, O1);
1704   __ and3(O1, -BytesPerInt, O1);
1705   // load lo, hi
1706   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
1707   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
1708   // Sign extend the 32 bits
1709   __ sra ( Otos_i, 0, Otos_i );
1710 
1711   // check against lo & hi
1712   __ cmp( Otos_i, O2);
1713   __ br( Assembler::less, false, Assembler::pn, default_case);
1714   __ delayed()->cmp( Otos_i, O3 );
1715   __ br( Assembler::greater, false, Assembler::pn, default_case);
1716   // lookup dispatch offset
1717   __ delayed()->sub(Otos_i, O2, O2);
1718   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
1719   __ sll(O2, LogBytesPerInt, O2);
1720   __ add(O2, 3 * BytesPerInt, O2);
1721   __ ba(continue_execution);
1722   __ delayed()->ld(O1, O2, O2);
1723   // handle default
1724   __ bind(default_case);
1725   __ profile_switch_default(O3);
1726   __ ld(O1, 0, O2); // get default offset
1727   // continue execution
1728   __ bind(continue_execution);
1729   __ add(Lbcp, O2, Lbcp);
1730   __ dispatch_next(vtos);
1731 }
1732 
1733 
1734 void TemplateTable::lookupswitch() {
1735   transition(itos, itos);
1736   __ stop("lookupswitch bytecode should have been rewritten");
1737 }
1738 
1739 void TemplateTable::fast_linearswitch() {
1740   transition(itos, vtos);
1741     Label loop_entry, loop, found, continue_execution;
1742   // align bcp
1743   __ add(Lbcp, BytesPerInt, O1);
1744   __ and3(O1, -BytesPerInt, O1);
1745  // set counter
1746   __ ld(O1, BytesPerInt, O2);
1747   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
1748   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
1749   __ ba(loop_entry);
1750   __ delayed()->add(O3, O2, O2); // counter now points past last pair
1751 
1752   // table search
1753   __ bind(loop);
1754   __ cmp(O4, Otos_i);
1755   __ br(Assembler::equal, true, Assembler::pn, found);
1756   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
1757   __ inc(O3, 2 * BytesPerInt);
1758 
1759   __ bind(loop_entry);
1760   __ cmp(O2, O3);
1761   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
1762   __ delayed()->ld(O3, 0, O4);
1763 
1764   // default case
1765   __ ld(O1, 0, O4); // get default offset
1766   if (ProfileInterpreter) {
1767     __ profile_switch_default(O3);
1768     __ ba_short(continue_execution);
1769   }
1770 
1771   // entry found -> get offset
1772   __ bind(found);
1773   if (ProfileInterpreter) {
1774     __ sub(O3, O1, O3);
1775     __ sub(O3, 2*BytesPerInt, O3);
1776     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
1777     __ profile_switch_case(O3, O1, O2, G3_scratch);
1778 
1779     __ bind(continue_execution);
1780   }
1781   __ add(Lbcp, O4, Lbcp);
1782   __ dispatch_next(vtos);
1783 }
1784 
1785 
1786 void TemplateTable::fast_binaryswitch() {
1787   transition(itos, vtos);
1788   // Implementation using the following core algorithm: (copied from Intel)
1789   //
1790   // int binary_search(int key, LookupswitchPair* array, int n) {
1791   //   // Binary search according to "Methodik des Programmierens" by
1792   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1793   //   int i = 0;
1794   //   int j = n;
1795   //   while (i+1 < j) {
1796   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1797   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1798   //     // where a stands for the array and assuming that the (inexisting)
1799   //     // element a[n] is infinitely big.
1800   //     int h = (i + j) >> 1;
1801   //     // i < h < j
1802   //     if (key < array[h].fast_match()) {
1803   //       j = h;
1804   //     } else {
1805   //       i = h;
1806   //     }
1807   //   }
1808   //   // R: a[i] <= key < a[i+1] or Q
1809   //   // (i.e., if key is within array, i is the correct index)
1810   //   return i;
1811   // }
1812 
1813   // register allocation
1814   assert(Otos_i == O0, "alias checking");
1815   const Register Rkey     = Otos_i;                    // already set (tosca)
1816   const Register Rarray   = O1;
1817   const Register Ri       = O2;
1818   const Register Rj       = O3;
1819   const Register Rh       = O4;
1820   const Register Rscratch = O5;
1821 
1822   const int log_entry_size = 3;
1823   const int entry_size = 1 << log_entry_size;
1824 
1825   Label found;
1826   // Find Array start
1827   __ add(Lbcp, 3 * BytesPerInt, Rarray);
1828   __ and3(Rarray, -BytesPerInt, Rarray);
1829   // initialize i & j (in delay slot)
1830   __ clr( Ri );
1831 
1832   // and start
1833   Label entry;
1834   __ ba(entry);
1835   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
1836   // (Rj is already in the native byte-ordering.)
1837 
1838   // binary search loop
1839   { Label loop;
1840     __ bind( loop );
1841     // int h = (i + j) >> 1;
1842     __ sra( Rh, 1, Rh );
1843     // if (key < array[h].fast_match()) {
1844     //   j = h;
1845     // } else {
1846     //   i = h;
1847     // }
1848     __ sll( Rh, log_entry_size, Rscratch );
1849     __ ld( Rarray, Rscratch, Rscratch );
1850     // (Rscratch is already in the native byte-ordering.)
1851     __ cmp( Rkey, Rscratch );
1852     __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
1853     __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
1854 
1855     // while (i+1 < j)
1856     __ bind( entry );
1857     __ add( Ri, 1, Rscratch );
1858     __ cmp(Rscratch, Rj);
1859     __ br( Assembler::less, true, Assembler::pt, loop );
1860     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
1861   }
1862 
1863   // end of binary search, result index is i (must check again!)
1864   Label default_case;
1865   Label continue_execution;
1866   if (ProfileInterpreter) {
1867     __ mov( Ri, Rh );              // Save index in i for profiling
1868   }
1869   __ sll( Ri, log_entry_size, Ri );
1870   __ ld( Rarray, Ri, Rscratch );
1871   // (Rscratch is already in the native byte-ordering.)
1872   __ cmp( Rkey, Rscratch );
1873   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
1874   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
1875 
1876   // entry found -> j = offset
1877   __ inc( Ri, BytesPerInt );
1878   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
1879   __ ld( Rarray, Ri, Rj );
1880   // (Rj is already in the native byte-ordering.)
1881 
1882   if (ProfileInterpreter) {
1883     __ ba_short(continue_execution);
1884   }
1885 
1886   __ bind(default_case); // fall through (if not profiling)
1887   __ profile_switch_default(Ri);
1888 
1889   __ bind(continue_execution);
1890   __ add( Lbcp, Rj, Lbcp );
1891   __ dispatch_next( vtos );
1892 }
1893 
1894 
1895 void TemplateTable::_return(TosState state) {
1896   transition(state, state);
1897   assert(_desc->calls_vm(), "inconsistent calls_vm information");
1898 
1899   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
1900     assert(state == vtos, "only valid state");
1901     __ mov(G0, G3_scratch);
1902     __ access_local_ptr(G3_scratch, Otos_i);
1903     __ load_klass(Otos_i, O2);
1904     __ set(JVM_ACC_HAS_FINALIZER, G3);
1905     __ ld(O2, in_bytes(Klass::access_flags_offset()), O2);
1906     __ andcc(G3, O2, G0);
1907     Label skip_register_finalizer;
1908     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
1909     __ delayed()->nop();
1910 
1911     // Call out to do finalizer registration
1912     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
1913 
1914     __ bind(skip_register_finalizer);
1915   }
1916 
1917   // Narrow result if state is itos but result type is smaller.
1918   // Need to narrow in the return bytecode rather than in generate_return_entry
1919   // since compiled code callers expect the result to already be narrowed.
1920   if (state == itos) {
1921     __ narrow(Otos_i);
1922   }
1923   __ remove_activation(state, /* throw_monitor_exception */ true);
1924 
1925   // The caller's SP was adjusted upon method entry to accomodate
1926   // the callee's non-argument locals. Undo that adjustment.
1927   __ ret();                             // return to caller
1928   __ delayed()->restore(I5_savedSP, G0, SP);
1929 }
1930 
1931 
1932 // ----------------------------------------------------------------------------
1933 // Volatile variables demand their effects be made known to all CPU's in
1934 // order.  Store buffers on most chips allow reads & writes to reorder; the
1935 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
1936 // memory barrier (i.e., it's not sufficient that the interpreter does not
1937 // reorder volatile references, the hardware also must not reorder them).
1938 //
1939 // According to the new Java Memory Model (JMM):
1940 // (1) All volatiles are serialized wrt to each other.
1941 // ALSO reads & writes act as aquire & release, so:
1942 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
1943 // the read float up to before the read.  It's OK for non-volatile memory refs
1944 // that happen before the volatile read to float down below it.
1945 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
1946 // that happen BEFORE the write float down to after the write.  It's OK for
1947 // non-volatile memory refs that happen after the volatile write to float up
1948 // before it.
1949 //
1950 // We only put in barriers around volatile refs (they are expensive), not
1951 // _between_ memory refs (that would require us to track the flavor of the
1952 // previous memory refs).  Requirements (2) and (3) require some barriers
1953 // before volatile stores and after volatile loads.  These nearly cover
1954 // requirement (1) but miss the volatile-store-volatile-load case.  This final
1955 // case is placed after volatile-stores although it could just as well go
1956 // before volatile-loads.
1957 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
1958   // Helper function to insert a is-volatile test and memory barrier
1959   // All current sparc implementations run in TSO, needing only StoreLoad
1960   if ((order_constraint & Assembler::StoreLoad) == 0) return;
1961   __ membar( order_constraint );
1962 }
1963 
1964 // ----------------------------------------------------------------------------
1965 void TemplateTable::resolve_cache_and_index(int byte_no,
1966                                             Register Rcache,
1967                                             Register index,
1968                                             size_t index_size) {
1969   // Depends on cpCacheOop layout!
1970 
1971   Label resolved;
1972   Bytecodes::Code code = bytecode();
1973   switch (code) {
1974   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
1975   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
1976   }
1977 
1978   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
1979   __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, Lbyte_code, byte_no, 1, index_size);
1980   __ cmp(Lbyte_code, code);  // have we resolved this bytecode?
1981   __ br(Assembler::equal, false, Assembler::pt, resolved);
1982   __ delayed()->set(code, O1);
1983 
1984   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
1985   // first time invocation - must resolve first
1986   __ call_VM(noreg, entry, O1);
1987   // Update registers with resolved info
1988   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
1989   __ bind(resolved);
1990 }
1991 
1992 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
1993                                                Register method,
1994                                                Register itable_index,
1995                                                Register flags,
1996                                                bool is_invokevirtual,
1997                                                bool is_invokevfinal,
1998                                                bool is_invokedynamic) {
1999   // Uses both G3_scratch and G4_scratch
2000   Register cache = G3_scratch;
2001   Register index = G4_scratch;
2002   assert_different_registers(cache, method, itable_index);
2003 
2004   // determine constant pool cache field offsets
2005   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2006   const int method_offset = in_bytes(
2007       ConstantPoolCache::base_offset() +
2008       ((byte_no == f2_byte)
2009        ? ConstantPoolCacheEntry::f2_offset()
2010        : ConstantPoolCacheEntry::f1_offset()
2011       )
2012     );
2013   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2014                                     ConstantPoolCacheEntry::flags_offset());
2015   // access constant pool cache fields
2016   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2017                                     ConstantPoolCacheEntry::f2_offset());
2018 
2019   if (is_invokevfinal) {
2020     __ get_cache_and_index_at_bcp(cache, index, 1);
2021     __ ld_ptr(Address(cache, method_offset), method);
2022   } else {
2023     size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2024     resolve_cache_and_index(byte_no, cache, index, index_size);
2025     __ ld_ptr(Address(cache, method_offset), method);
2026   }
2027 
2028   if (itable_index != noreg) {
2029     // pick up itable or appendix index from f2 also:
2030     __ ld_ptr(Address(cache, index_offset), itable_index);
2031   }
2032   __ ld_ptr(Address(cache, flags_offset), flags);
2033 }
2034 
2035 // The Rcache register must be set before call
2036 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2037                                               Register Rcache,
2038                                               Register index,
2039                                               Register Roffset,
2040                                               Register Rflags,
2041                                               bool is_static) {
2042   assert_different_registers(Rcache, Rflags, Roffset);
2043 
2044   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2045 
2046   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2047   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2048   if (is_static) {
2049     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
2050     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2051     __ ld_ptr( Robj, mirror_offset, Robj);
2052   }
2053 }
2054 
2055 // The registers Rcache and index expected to be set before call.
2056 // Correct values of the Rcache and index registers are preserved.
2057 void TemplateTable::jvmti_post_field_access(Register Rcache,
2058                                             Register index,
2059                                             bool is_static,
2060                                             bool has_tos) {
2061   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2062 
2063   if (JvmtiExport::can_post_field_access()) {
2064     // Check to see if a field access watch has been set before we take
2065     // the time to call into the VM.
2066     Label Label1;
2067     assert_different_registers(Rcache, index, G1_scratch);
2068     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
2069     __ load_contents(get_field_access_count_addr, G1_scratch);
2070     __ cmp_and_br_short(G1_scratch, 0, Assembler::equal, Assembler::pt, Label1);
2071 
2072     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
2073 
2074     if (is_static) {
2075       __ clr(Otos_i);
2076     } else {
2077       if (has_tos) {
2078       // save object pointer before call_VM() clobbers it
2079         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
2080       } else {
2081         // Load top of stack (do not pop the value off the stack);
2082         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
2083       }
2084       __ verify_oop(Otos_i);
2085     }
2086     // Otos_i: object pointer or NULL if static
2087     // Rcache: cache entry pointer
2088     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2089                Otos_i, Rcache);
2090     if (!is_static && has_tos) {
2091       __ pop_ptr(Otos_i);  // restore object pointer
2092       __ verify_oop(Otos_i);
2093     }
2094     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2095     __ bind(Label1);
2096   }
2097 }
2098 
2099 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2100   transition(vtos, vtos);
2101 
2102   Register Rcache = G3_scratch;
2103   Register index  = G4_scratch;
2104   Register Rclass = Rcache;
2105   Register Roffset= G4_scratch;
2106   Register Rflags = G1_scratch;
2107   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2108 
2109   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2110   jvmti_post_field_access(Rcache, index, is_static, false);
2111   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2112 
2113   if (!is_static) {
2114     pop_and_check_object(Rclass);
2115   } else {
2116     __ verify_oop(Rclass);
2117   }
2118 
2119   Label exit;
2120 
2121   Assembler::Membar_mask_bits membar_bits =
2122     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2123 
2124   if (__ membar_has_effect(membar_bits)) {
2125     // Get volatile flag
2126     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2127     __ and3(Rflags, Lscratch, Lscratch);
2128   }
2129 
2130   Label checkVolatile;
2131 
2132   // compute field type
2133   Label notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
2134   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2135   // Make sure we don't need to mask Rflags after the above shift
2136   ConstantPoolCacheEntry::verify_tos_state_shift();
2137 
2138   // Check atos before itos for getstatic, more likely (in Queens at least)
2139   __ cmp(Rflags, atos);
2140   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2141   __ delayed() ->cmp(Rflags, itos);
2142 
2143   // atos
2144   __ load_heap_oop(Rclass, Roffset, Otos_i);
2145   __ verify_oop(Otos_i);
2146   __ push(atos);
2147   if (!is_static && rc == may_rewrite) {
2148     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
2149   }
2150   __ ba(checkVolatile);
2151   __ delayed()->tst(Lscratch);
2152 
2153   __ bind(notObj);
2154 
2155   // cmp(Rflags, itos);
2156   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2157   __ delayed() ->cmp(Rflags, ltos);
2158 
2159   // itos
2160   __ ld(Rclass, Roffset, Otos_i);
2161   __ push(itos);
2162   if (!is_static && rc == may_rewrite) {
2163     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
2164   }
2165   __ ba(checkVolatile);
2166   __ delayed()->tst(Lscratch);
2167 
2168   __ bind(notInt);
2169 
2170   // cmp(Rflags, ltos);
2171   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2172   __ delayed() ->cmp(Rflags, btos);
2173 
2174   // ltos
2175   // load must be atomic
2176   __ ld_long(Rclass, Roffset, Otos_l);
2177   __ push(ltos);
2178   if (!is_static && rc == may_rewrite) {
2179     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
2180   }
2181   __ ba(checkVolatile);
2182   __ delayed()->tst(Lscratch);
2183 
2184   __ bind(notLong);
2185 
2186   // cmp(Rflags, btos);
2187   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2188   __ delayed() ->cmp(Rflags, ztos);
2189 
2190   // btos
2191   __ ldsb(Rclass, Roffset, Otos_i);
2192   __ push(itos);
2193   if (!is_static && rc == may_rewrite) {
2194     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
2195   }
2196   __ ba(checkVolatile);
2197   __ delayed()->tst(Lscratch);
2198 
2199   __ bind(notByte);
2200 
2201   // cmp(Rflags, ztos);
2202   __ br(Assembler::notEqual, false, Assembler::pt, notBool);
2203   __ delayed() ->cmp(Rflags, ctos);
2204 
2205   // ztos
2206   __ ldsb(Rclass, Roffset, Otos_i);
2207   __ push(itos);
2208   if (!is_static && rc == may_rewrite) {
2209     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2210     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
2211   }
2212   __ ba(checkVolatile);
2213   __ delayed()->tst(Lscratch);
2214 
2215   __ bind(notBool);
2216 
2217   // cmp(Rflags, ctos);
2218   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2219   __ delayed() ->cmp(Rflags, stos);
2220 
2221   // ctos
2222   __ lduh(Rclass, Roffset, Otos_i);
2223   __ push(itos);
2224   if (!is_static && rc == may_rewrite) {
2225     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
2226   }
2227   __ ba(checkVolatile);
2228   __ delayed()->tst(Lscratch);
2229 
2230   __ bind(notChar);
2231 
2232   // cmp(Rflags, stos);
2233   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2234   __ delayed() ->cmp(Rflags, ftos);
2235 
2236   // stos
2237   __ ldsh(Rclass, Roffset, Otos_i);
2238   __ push(itos);
2239   if (!is_static && rc == may_rewrite) {
2240     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
2241   }
2242   __ ba(checkVolatile);
2243   __ delayed()->tst(Lscratch);
2244 
2245   __ bind(notShort);
2246 
2247 
2248   // cmp(Rflags, ftos);
2249   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
2250   __ delayed() ->tst(Lscratch);
2251 
2252   // ftos
2253   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
2254   __ push(ftos);
2255   if (!is_static && rc == may_rewrite) {
2256     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
2257   }
2258   __ ba(checkVolatile);
2259   __ delayed()->tst(Lscratch);
2260 
2261   __ bind(notFloat);
2262 
2263 
2264   // dtos
2265   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
2266   __ push(dtos);
2267   if (!is_static && rc == may_rewrite) {
2268     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
2269   }
2270 
2271   __ bind(checkVolatile);
2272   if (__ membar_has_effect(membar_bits)) {
2273     // __ tst(Lscratch); executed in delay slot
2274     __ br(Assembler::zero, false, Assembler::pt, exit);
2275     __ delayed()->nop();
2276     volatile_barrier(membar_bits);
2277   }
2278 
2279   __ bind(exit);
2280 }
2281 
2282 void TemplateTable::getfield(int byte_no) {
2283   getfield_or_static(byte_no, false);
2284 }
2285 
2286 void TemplateTable::nofast_getfield(int byte_no) {
2287   getfield_or_static(byte_no, false, may_not_rewrite);
2288 }
2289 
2290 void TemplateTable::getstatic(int byte_no) {
2291   getfield_or_static(byte_no, true);
2292 }
2293 
2294 void TemplateTable::fast_accessfield(TosState state) {
2295   transition(atos, state);
2296   Register Rcache  = G3_scratch;
2297   Register index   = G4_scratch;
2298   Register Roffset = G4_scratch;
2299   Register Rflags  = Rcache;
2300   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2301 
2302   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2303   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
2304 
2305   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2306 
2307   __ null_check(Otos_i);
2308   __ verify_oop(Otos_i);
2309 
2310   Label exit;
2311 
2312   Assembler::Membar_mask_bits membar_bits =
2313     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2314   if (__ membar_has_effect(membar_bits)) {
2315     // Get volatile flag
2316     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
2317     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2318   }
2319 
2320   switch (bytecode()) {
2321     case Bytecodes::_fast_bgetfield:
2322       __ ldsb(Otos_i, Roffset, Otos_i);
2323       break;
2324     case Bytecodes::_fast_cgetfield:
2325       __ lduh(Otos_i, Roffset, Otos_i);
2326       break;
2327     case Bytecodes::_fast_sgetfield:
2328       __ ldsh(Otos_i, Roffset, Otos_i);
2329       break;
2330     case Bytecodes::_fast_igetfield:
2331       __ ld(Otos_i, Roffset, Otos_i);
2332       break;
2333     case Bytecodes::_fast_lgetfield:
2334       __ ld_long(Otos_i, Roffset, Otos_l);
2335       break;
2336     case Bytecodes::_fast_fgetfield:
2337       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
2338       break;
2339     case Bytecodes::_fast_dgetfield:
2340       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
2341       break;
2342     case Bytecodes::_fast_agetfield:
2343       __ load_heap_oop(Otos_i, Roffset, Otos_i);
2344       break;
2345     default:
2346       ShouldNotReachHere();
2347   }
2348 
2349   if (__ membar_has_effect(membar_bits)) {
2350     __ btst(Lscratch, Rflags);
2351     __ br(Assembler::zero, false, Assembler::pt, exit);
2352     __ delayed()->nop();
2353     volatile_barrier(membar_bits);
2354     __ bind(exit);
2355   }
2356 
2357   if (state == atos) {
2358     __ verify_oop(Otos_i);    // does not blow flags!
2359   }
2360 }
2361 
2362 void TemplateTable::jvmti_post_fast_field_mod() {
2363   if (JvmtiExport::can_post_field_modification()) {
2364     // Check to see if a field modification watch has been set before we take
2365     // the time to call into the VM.
2366     Label done;
2367     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2368     __ load_contents(get_field_modification_count_addr, G4_scratch);
2369     __ cmp_and_br_short(G4_scratch, 0, Assembler::equal, Assembler::pt, done);
2370     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
2371     __ verify_oop(G4_scratch);
2372     __ push_ptr(G4_scratch);    // put the object pointer back on tos
2373     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
2374     // Save tos values before call_VM() clobbers them. Since we have
2375     // to do it for every data type, we use the saved values as the
2376     // jvalue object.
2377     switch (bytecode()) {  // save tos values before call_VM() clobbers them
2378     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
2379     case Bytecodes::_fast_bputfield: // fall through
2380     case Bytecodes::_fast_zputfield: // fall through
2381     case Bytecodes::_fast_sputfield: // fall through
2382     case Bytecodes::_fast_cputfield: // fall through
2383     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
2384     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
2385     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
2386     // get words in right order for use as jvalue object
2387     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
2388     }
2389     // setup pointer to jvalue object
2390     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
2391     // G4_scratch:  object pointer
2392     // G1_scratch: cache entry pointer
2393     // G3_scratch: jvalue object on the stack
2394     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
2395     switch (bytecode()) {             // restore tos values
2396     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
2397     case Bytecodes::_fast_bputfield: // fall through
2398     case Bytecodes::_fast_zputfield: // fall through
2399     case Bytecodes::_fast_sputfield: // fall through
2400     case Bytecodes::_fast_cputfield: // fall through
2401     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
2402     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
2403     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
2404     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
2405     }
2406     __ bind(done);
2407   }
2408 }
2409 
2410 // The registers Rcache and index expected to be set before call.
2411 // The function may destroy various registers, just not the Rcache and index registers.
2412 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
2413   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2414 
2415   if (JvmtiExport::can_post_field_modification()) {
2416     // Check to see if a field modification watch has been set before we take
2417     // the time to call into the VM.
2418     Label Label1;
2419     assert_different_registers(Rcache, index, G1_scratch);
2420     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2421     __ load_contents(get_field_modification_count_addr, G1_scratch);
2422     __ cmp_and_br_short(G1_scratch, 0, Assembler::zero, Assembler::pt, Label1);
2423 
2424     // The Rcache and index registers have been already set.
2425     // This allows to eliminate this call but the Rcache and index
2426     // registers must be correspondingly used after this line.
2427     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
2428 
2429     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
2430     if (is_static) {
2431       // Life is simple.  Null out the object pointer.
2432       __ clr(G4_scratch);
2433     } else {
2434       Register Rflags = G1_scratch;
2435       // Life is harder. The stack holds the value on top, followed by the
2436       // object.  We don't know the size of the value, though; it could be
2437       // one or two words depending on its type. As a result, we must find
2438       // the type to determine where the object is.
2439 
2440       Label two_word, valsizeknown;
2441       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2442       __ mov(Lesp, G4_scratch);
2443       __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2444       // Make sure we don't need to mask Rflags after the above shift
2445       ConstantPoolCacheEntry::verify_tos_state_shift();
2446       __ cmp(Rflags, ltos);
2447       __ br(Assembler::equal, false, Assembler::pt, two_word);
2448       __ delayed()->cmp(Rflags, dtos);
2449       __ br(Assembler::equal, false, Assembler::pt, two_word);
2450       __ delayed()->nop();
2451       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
2452       __ ba_short(valsizeknown);
2453       __ bind(two_word);
2454 
2455       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
2456 
2457       __ bind(valsizeknown);
2458       // setup object pointer
2459       __ ld_ptr(G4_scratch, 0, G4_scratch);
2460       __ verify_oop(G4_scratch);
2461     }
2462     // setup pointer to jvalue object
2463     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
2464     // G4_scratch:  object pointer or NULL if static
2465     // G3_scratch: cache entry pointer
2466     // G1_scratch: jvalue object on the stack
2467     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2468                G4_scratch, G3_scratch, G1_scratch);
2469     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2470     __ bind(Label1);
2471   }
2472 }
2473 
2474 void TemplateTable::pop_and_check_object(Register r) {
2475   __ pop_ptr(r);
2476   __ null_check(r);  // for field access must check obj.
2477   __ verify_oop(r);
2478 }
2479 
2480 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2481   transition(vtos, vtos);
2482   Register Rcache = G3_scratch;
2483   Register index  = G4_scratch;
2484   Register Rclass = Rcache;
2485   Register Roffset= G4_scratch;
2486   Register Rflags = G1_scratch;
2487   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2488 
2489   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2490   jvmti_post_field_mod(Rcache, index, is_static);
2491   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2492 
2493   Assembler::Membar_mask_bits read_bits =
2494     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2495   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2496 
2497   Label notVolatile, checkVolatile, exit;
2498   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2499     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2500     __ and3(Rflags, Lscratch, Lscratch);
2501 
2502     if (__ membar_has_effect(read_bits)) {
2503       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2504       volatile_barrier(read_bits);
2505       __ bind(notVolatile);
2506     }
2507   }
2508 
2509   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2510   // Make sure we don't need to mask Rflags after the above shift
2511   ConstantPoolCacheEntry::verify_tos_state_shift();
2512 
2513   // compute field type
2514   Label notInt, notShort, notChar, notObj, notByte, notBool, notLong, notFloat;
2515 
2516   if (is_static) {
2517     // putstatic with object type most likely, check that first
2518     __ cmp(Rflags, atos);
2519     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2520     __ delayed()->cmp(Rflags, itos);
2521 
2522     // atos
2523     {
2524       __ pop_ptr();
2525       __ verify_oop(Otos_i);
2526       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2527       __ ba(checkVolatile);
2528       __ delayed()->tst(Lscratch);
2529     }
2530 
2531     __ bind(notObj);
2532     // cmp(Rflags, itos);
2533     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2534     __ delayed()->cmp(Rflags, btos);
2535 
2536     // itos
2537     {
2538       __ pop_i();
2539       __ st(Otos_i, Rclass, Roffset);
2540       __ ba(checkVolatile);
2541       __ delayed()->tst(Lscratch);
2542     }
2543 
2544     __ bind(notInt);
2545   } else {
2546     // putfield with int type most likely, check that first
2547     __ cmp(Rflags, itos);
2548     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2549     __ delayed()->cmp(Rflags, atos);
2550 
2551     // itos
2552     {
2553       __ pop_i();
2554       pop_and_check_object(Rclass);
2555       __ st(Otos_i, Rclass, Roffset);
2556       if (rc == may_rewrite) patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch, true, byte_no);
2557       __ ba(checkVolatile);
2558       __ delayed()->tst(Lscratch);
2559     }
2560 
2561     __ bind(notInt);
2562     // cmp(Rflags, atos);
2563     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2564     __ delayed()->cmp(Rflags, btos);
2565 
2566     // atos
2567     {
2568       __ pop_ptr();
2569       pop_and_check_object(Rclass);
2570       __ verify_oop(Otos_i);
2571       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2572       if (rc == may_rewrite) patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch, true, byte_no);
2573       __ ba(checkVolatile);
2574       __ delayed()->tst(Lscratch);
2575     }
2576 
2577     __ bind(notObj);
2578   }
2579 
2580   // cmp(Rflags, btos);
2581   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2582   __ delayed()->cmp(Rflags, ztos);
2583 
2584   // btos
2585   {
2586     __ pop_i();
2587     if (!is_static) pop_and_check_object(Rclass);
2588     __ stb(Otos_i, Rclass, Roffset);
2589     if (!is_static && rc == may_rewrite) {
2590       patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch, true, byte_no);
2591     }
2592     __ ba(checkVolatile);
2593     __ delayed()->tst(Lscratch);
2594   }
2595 
2596   __ bind(notByte);
2597 
2598   // cmp(Rflags, btos);
2599   __ br(Assembler::notEqual, false, Assembler::pt, notBool);
2600   __ delayed()->cmp(Rflags, ltos);
2601 
2602   // ztos
2603   {
2604     __ pop_i();
2605     if (!is_static) pop_and_check_object(Rclass);
2606     __ and3(Otos_i, 1, Otos_i);
2607     __ stb(Otos_i, Rclass, Roffset);
2608     if (!is_static && rc == may_rewrite) {
2609       patch_bytecode(Bytecodes::_fast_zputfield, G3_scratch, G4_scratch, true, byte_no);
2610     }
2611     __ ba(checkVolatile);
2612     __ delayed()->tst(Lscratch);
2613   }
2614 
2615   __ bind(notBool);
2616   // cmp(Rflags, ltos);
2617   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2618   __ delayed()->cmp(Rflags, ctos);
2619 
2620   // ltos
2621   {
2622     __ pop_l();
2623     if (!is_static) pop_and_check_object(Rclass);
2624     __ st_long(Otos_l, Rclass, Roffset);
2625     if (!is_static && rc == may_rewrite) {
2626       patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch, true, byte_no);
2627     }
2628     __ ba(checkVolatile);
2629     __ delayed()->tst(Lscratch);
2630   }
2631 
2632   __ bind(notLong);
2633   // cmp(Rflags, ctos);
2634   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2635   __ delayed()->cmp(Rflags, stos);
2636 
2637   // ctos (char)
2638   {
2639     __ pop_i();
2640     if (!is_static) pop_and_check_object(Rclass);
2641     __ sth(Otos_i, Rclass, Roffset);
2642     if (!is_static && rc == may_rewrite) {
2643       patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch, true, byte_no);
2644     }
2645     __ ba(checkVolatile);
2646     __ delayed()->tst(Lscratch);
2647   }
2648 
2649   __ bind(notChar);
2650   // cmp(Rflags, stos);
2651   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2652   __ delayed()->cmp(Rflags, ftos);
2653 
2654   // stos (short)
2655   {
2656     __ pop_i();
2657     if (!is_static) pop_and_check_object(Rclass);
2658     __ sth(Otos_i, Rclass, Roffset);
2659     if (!is_static && rc == may_rewrite) {
2660       patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch, true, byte_no);
2661     }
2662     __ ba(checkVolatile);
2663     __ delayed()->tst(Lscratch);
2664   }
2665 
2666   __ bind(notShort);
2667   // cmp(Rflags, ftos);
2668   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
2669   __ delayed()->nop();
2670 
2671   // ftos
2672   {
2673     __ pop_f();
2674     if (!is_static) pop_and_check_object(Rclass);
2675     __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2676     if (!is_static && rc == may_rewrite) {
2677       patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch, true, byte_no);
2678     }
2679     __ ba(checkVolatile);
2680     __ delayed()->tst(Lscratch);
2681   }
2682 
2683   __ bind(notFloat);
2684 
2685   // dtos
2686   {
2687     __ pop_d();
2688     if (!is_static) pop_and_check_object(Rclass);
2689     __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2690     if (!is_static && rc == may_rewrite) {
2691       patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch, true, byte_no);
2692     }
2693   }
2694 
2695   __ bind(checkVolatile);
2696   __ tst(Lscratch);
2697 
2698   if (__ membar_has_effect(write_bits)) {
2699     // __ tst(Lscratch); in delay slot
2700     __ br(Assembler::zero, false, Assembler::pt, exit);
2701     __ delayed()->nop();
2702     volatile_barrier(Assembler::StoreLoad);
2703     __ bind(exit);
2704   }
2705 }
2706 
2707 void TemplateTable::fast_storefield(TosState state) {
2708   transition(state, vtos);
2709   Register Rcache = G3_scratch;
2710   Register Rclass = Rcache;
2711   Register Roffset= G4_scratch;
2712   Register Rflags = G1_scratch;
2713   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2714 
2715   jvmti_post_fast_field_mod();
2716 
2717   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
2718 
2719   Assembler::Membar_mask_bits read_bits =
2720     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2721   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2722 
2723   Label notVolatile, checkVolatile, exit;
2724   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2725     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2726     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2727     __ and3(Rflags, Lscratch, Lscratch);
2728     if (__ membar_has_effect(read_bits)) {
2729       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2730       volatile_barrier(read_bits);
2731       __ bind(notVolatile);
2732     }
2733   }
2734 
2735   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2736   pop_and_check_object(Rclass);
2737 
2738   switch (bytecode()) {
2739     case Bytecodes::_fast_zputfield: __ and3(Otos_i, 1, Otos_i);  // fall through to bputfield
2740     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
2741     case Bytecodes::_fast_cputfield: /* fall through */
2742     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
2743     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
2744     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
2745     case Bytecodes::_fast_fputfield:
2746       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2747       break;
2748     case Bytecodes::_fast_dputfield:
2749       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2750       break;
2751     case Bytecodes::_fast_aputfield:
2752       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2753       break;
2754     default:
2755       ShouldNotReachHere();
2756   }
2757 
2758   if (__ membar_has_effect(write_bits)) {
2759     __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, exit);
2760     volatile_barrier(Assembler::StoreLoad);
2761     __ bind(exit);
2762   }
2763 }
2764 
2765 void TemplateTable::putfield(int byte_no) {
2766   putfield_or_static(byte_no, false);
2767 }
2768 
2769 void TemplateTable::nofast_putfield(int byte_no) {
2770   putfield_or_static(byte_no, false, may_not_rewrite);
2771 }
2772 
2773 void TemplateTable::putstatic(int byte_no) {
2774   putfield_or_static(byte_no, true);
2775 }
2776 
2777 void TemplateTable::fast_xaccess(TosState state) {
2778   transition(vtos, state);
2779   Register Rcache = G3_scratch;
2780   Register Roffset = G4_scratch;
2781   Register Rflags  = G4_scratch;
2782   Register Rreceiver = Lscratch;
2783 
2784   __ ld_ptr(Llocals, 0, Rreceiver);
2785 
2786   // access constant pool cache  (is resolved)
2787   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
2788   __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
2789   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
2790 
2791   __ verify_oop(Rreceiver);
2792   __ null_check(Rreceiver);
2793   if (state == atos) {
2794     __ load_heap_oop(Rreceiver, Roffset, Otos_i);
2795   } else if (state == itos) {
2796     __ ld (Rreceiver, Roffset, Otos_i) ;
2797   } else if (state == ftos) {
2798     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
2799   } else {
2800     ShouldNotReachHere();
2801   }
2802 
2803   Assembler::Membar_mask_bits membar_bits =
2804     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2805   if (__ membar_has_effect(membar_bits)) {
2806 
2807     // Get is_volatile value in Rflags and check if membar is needed
2808     __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
2809 
2810     // Test volatile
2811     Label notVolatile;
2812     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2813     __ btst(Rflags, Lscratch);
2814     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
2815     __ delayed()->nop();
2816     volatile_barrier(membar_bits);
2817     __ bind(notVolatile);
2818   }
2819 
2820   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
2821   __ sub(Lbcp, 1, Lbcp);
2822 }
2823 
2824 //----------------------------------------------------------------------------------------------------
2825 // Calls
2826 
2827 void TemplateTable::count_calls(Register method, Register temp) {
2828   // implemented elsewhere
2829   ShouldNotReachHere();
2830 }
2831 
2832 void TemplateTable::prepare_invoke(int byte_no,
2833                                    Register method,  // linked method (or i-klass)
2834                                    Register ra,      // return address
2835                                    Register index,   // itable index, MethodType, etc.
2836                                    Register recv,    // if caller wants to see it
2837                                    Register flags    // if caller wants to test it
2838                                    ) {
2839   // determine flags
2840   const Bytecodes::Code code = bytecode();
2841   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
2842   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
2843   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
2844   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
2845   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
2846   const bool load_receiver       = (recv != noreg);
2847   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2848   assert(recv  == noreg || recv  == O0, "");
2849   assert(flags == noreg || flags == O1, "");
2850 
2851   // setup registers & access constant pool cache
2852   if (recv  == noreg)  recv  = O0;
2853   if (flags == noreg)  flags = O1;
2854   const Register temp = O2;
2855   assert_different_registers(method, ra, index, recv, flags, temp);
2856 
2857   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2858 
2859   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
2860 
2861   // maybe push appendix to arguments
2862   if (is_invokedynamic || is_invokehandle) {
2863     Label L_no_push;
2864     __ set((1 << ConstantPoolCacheEntry::has_appendix_shift), temp);
2865     __ btst(flags, temp);
2866     __ br(Assembler::zero, false, Assembler::pt, L_no_push);
2867     __ delayed()->nop();
2868     // Push the appendix as a trailing parameter.
2869     // This must be done before we get the receiver,
2870     // since the parameter_size includes it.
2871     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2872     __ load_resolved_reference_at_index(temp, index);
2873     __ verify_oop(temp);
2874     __ push_ptr(temp);  // push appendix (MethodType, CallSite, etc.)
2875     __ bind(L_no_push);
2876   }
2877 
2878   // load receiver if needed (after appendix is pushed so parameter size is correct)
2879   if (load_receiver) {
2880     __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, temp);  // get parameter size
2881     __ load_receiver(temp, recv);  //  __ argument_address uses Gargs but we need Lesp
2882     __ verify_oop(recv);
2883   }
2884 
2885   // compute return type
2886   __ srl(flags, ConstantPoolCacheEntry::tos_state_shift, ra);
2887   // Make sure we don't need to mask flags after the above shift
2888   ConstantPoolCacheEntry::verify_tos_state_shift();
2889   // load return address
2890   {
2891     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2892     AddressLiteral table(table_addr);
2893     __ set(table, temp);
2894     __ sll(ra, LogBytesPerWord, ra);
2895     __ ld_ptr(Address(temp, ra), ra);
2896   }
2897 }
2898 
2899 
2900 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
2901   Register Rtemp = G4_scratch;
2902   Register Rcall = Rindex;
2903   assert_different_registers(Rcall, G5_method, Gargs, Rret);
2904 
2905   // get target Method* & entry point
2906   __ lookup_virtual_method(Rrecv, Rindex, G5_method);
2907   __ profile_arguments_type(G5_method, Rcall, Gargs, true);
2908   __ profile_called_method(G5_method, Rtemp);
2909   __ call_from_interpreter(Rcall, Gargs, Rret);
2910 }
2911 
2912 void TemplateTable::invokevirtual(int byte_no) {
2913   transition(vtos, vtos);
2914   assert(byte_no == f2_byte, "use this argument");
2915 
2916   Register Rscratch = G3_scratch;
2917   Register Rtemp    = G4_scratch;
2918   Register Rret     = Lscratch;
2919   Register O0_recv  = O0;
2920   Label notFinal;
2921 
2922   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
2923   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
2924 
2925   // Check for vfinal
2926   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), G4_scratch);
2927   __ btst(Rret, G4_scratch);
2928   __ br(Assembler::zero, false, Assembler::pt, notFinal);
2929   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
2930 
2931   if (RewriteBytecodes && !UseSharedSpaces) {
2932     patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
2933   }
2934 
2935   invokevfinal_helper(Rscratch, Rret);
2936 
2937   __ bind(notFinal);
2938 
2939   __ mov(G5_method, Rscratch);  // better scratch register
2940   __ load_receiver(G4_scratch, O0_recv);  // gets receiverOop
2941   // receiver is in O0_recv
2942   __ verify_oop(O0_recv);
2943 
2944   // get return address
2945   AddressLiteral table(Interpreter::invoke_return_entry_table());
2946   __ set(table, Rtemp);
2947   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
2948   // Make sure we don't need to mask Rret after the above shift
2949   ConstantPoolCacheEntry::verify_tos_state_shift();
2950   __ sll(Rret,  LogBytesPerWord, Rret);
2951   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
2952 
2953   // get receiver klass
2954   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
2955   __ load_klass(O0_recv, O0_recv);
2956   __ verify_klass_ptr(O0_recv);
2957 
2958   __ profile_virtual_call(O0_recv, O4);
2959 
2960   generate_vtable_call(O0_recv, Rscratch, Rret);
2961 }
2962 
2963 void TemplateTable::fast_invokevfinal(int byte_no) {
2964   transition(vtos, vtos);
2965   assert(byte_no == f2_byte, "use this argument");
2966 
2967   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
2968                              /*is_invokevfinal*/true, false);
2969   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
2970   invokevfinal_helper(G3_scratch, Lscratch);
2971 }
2972 
2973 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
2974   Register Rtemp = G4_scratch;
2975 
2976   // Load receiver from stack slot
2977   __ ld_ptr(G5_method, in_bytes(Method::const_offset()), G4_scratch);
2978   __ lduh(G4_scratch, in_bytes(ConstMethod::size_of_parameters_offset()), G4_scratch);
2979   __ load_receiver(G4_scratch, O0);
2980 
2981   // receiver NULL check
2982   __ null_check(O0);
2983 
2984   __ profile_final_call(O4);
2985   __ profile_arguments_type(G5_method, Rscratch, Gargs, true);
2986 
2987   // get return address
2988   AddressLiteral table(Interpreter::invoke_return_entry_table());
2989   __ set(table, Rtemp);
2990   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
2991   // Make sure we don't need to mask Rret after the above shift
2992   ConstantPoolCacheEntry::verify_tos_state_shift();
2993   __ sll(Rret,  LogBytesPerWord, Rret);
2994   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
2995 
2996 
2997   // do the call
2998   __ call_from_interpreter(Rscratch, Gargs, Rret);
2999 }
3000 
3001 
3002 void TemplateTable::invokespecial(int byte_no) {
3003   transition(vtos, vtos);
3004   assert(byte_no == f1_byte, "use this argument");
3005 
3006   const Register Rret     = Lscratch;
3007   const Register O0_recv  = O0;
3008   const Register Rscratch = G3_scratch;
3009 
3010   prepare_invoke(byte_no, G5_method, Rret, noreg, O0_recv);  // get receiver also for null check
3011   __ null_check(O0_recv);
3012 
3013   // do the call
3014   __ profile_call(O4);
3015   __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3016   __ call_from_interpreter(Rscratch, Gargs, Rret);
3017 }
3018 
3019 
3020 void TemplateTable::invokestatic(int byte_no) {
3021   transition(vtos, vtos);
3022   assert(byte_no == f1_byte, "use this argument");
3023 
3024   const Register Rret     = Lscratch;
3025   const Register Rscratch = G3_scratch;
3026 
3027   prepare_invoke(byte_no, G5_method, Rret);  // get f1 Method*
3028 
3029   // do the call
3030   __ profile_call(O4);
3031   __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3032   __ call_from_interpreter(Rscratch, Gargs, Rret);
3033 }
3034 
3035 void TemplateTable::invokeinterface_object_method(Register RKlass,
3036                                                   Register Rcall,
3037                                                   Register Rret,
3038                                                   Register Rflags) {
3039   Register Rscratch = G4_scratch;
3040   Register Rindex = Lscratch;
3041 
3042   assert_different_registers(Rscratch, Rindex, Rret);
3043 
3044   Label notFinal;
3045 
3046   // Check for vfinal
3047   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), Rscratch);
3048   __ btst(Rflags, Rscratch);
3049   __ br(Assembler::zero, false, Assembler::pt, notFinal);
3050   __ delayed()->nop();
3051 
3052   __ profile_final_call(O4);
3053 
3054   // do the call - the index (f2) contains the Method*
3055   assert_different_registers(G5_method, Gargs, Rcall);
3056   __ mov(Rindex, G5_method);
3057   __ profile_arguments_type(G5_method, Rcall, Gargs, true);
3058   __ call_from_interpreter(Rcall, Gargs, Rret);
3059   __ bind(notFinal);
3060 
3061   __ profile_virtual_call(RKlass, O4);
3062   generate_vtable_call(RKlass, Rindex, Rret);
3063 }
3064 
3065 
3066 void TemplateTable::invokeinterface(int byte_no) {
3067   transition(vtos, vtos);
3068   assert(byte_no == f1_byte, "use this argument");
3069 
3070   const Register Rinterface  = G1_scratch;
3071   const Register Rret        = G3_scratch;
3072   const Register Rindex      = Lscratch;
3073   const Register O0_recv     = O0;
3074   const Register O1_flags    = O1;
3075   const Register O2_Klass    = O2;
3076   const Register Rscratch    = G4_scratch;
3077   assert_different_registers(Rscratch, G5_method);
3078 
3079   prepare_invoke(byte_no, Rinterface, Rret, Rindex, O0_recv, O1_flags);
3080 
3081   // get receiver klass
3082   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
3083   __ load_klass(O0_recv, O2_Klass);
3084 
3085   // Special case of invokeinterface called for virtual method of
3086   // java.lang.Object.  See cpCacheOop.cpp for details.
3087   // This code isn't produced by javac, but could be produced by
3088   // another compliant java compiler.
3089   Label notMethod;
3090   __ set((1 << ConstantPoolCacheEntry::is_forced_virtual_shift), Rscratch);
3091   __ btst(O1_flags, Rscratch);
3092   __ br(Assembler::zero, false, Assembler::pt, notMethod);
3093   __ delayed()->nop();
3094 
3095   invokeinterface_object_method(O2_Klass, Rinterface, Rret, O1_flags);
3096 
3097   __ bind(notMethod);
3098 
3099   __ profile_virtual_call(O2_Klass, O4);
3100 
3101   //
3102   // find entry point to call
3103   //
3104 
3105   // compute start of first itableOffsetEntry (which is at end of vtable)
3106   const int base = in_bytes(Klass::vtable_start_offset());
3107   Label search;
3108   Register Rtemp = O1_flags;
3109 
3110   __ ld(O2_Klass, in_bytes(Klass::vtable_length_offset()), Rtemp);
3111   __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
3112   if (Assembler::is_simm13(base)) {
3113     __ add(Rtemp, base, Rtemp);
3114   } else {
3115     __ set(base, Rscratch);
3116     __ add(Rscratch, Rtemp, Rtemp);
3117   }
3118   __ add(O2_Klass, Rtemp, Rscratch);
3119 
3120   __ bind(search);
3121 
3122   __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
3123   {
3124     Label ok;
3125 
3126     // Check that entry is non-null.  Null entries are probably a bytecode
3127     // problem.  If the interface isn't implemented by the receiver class,
3128     // the VM should throw IncompatibleClassChangeError.  linkResolver checks
3129     // this too but that's only if the entry isn't already resolved, so we
3130     // need to check again.
3131     __ br_notnull_short( Rtemp, Assembler::pt, ok);
3132     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3133     __ should_not_reach_here();
3134     __ bind(ok);
3135   }
3136 
3137   __ cmp(Rinterface, Rtemp);
3138   __ brx(Assembler::notEqual, true, Assembler::pn, search);
3139   __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
3140 
3141   // entry found and Rscratch points to it
3142   __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
3143 
3144   assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
3145   __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
3146   __ add(Rscratch, Rindex, Rscratch);
3147   __ ld_ptr(O2_Klass, Rscratch, G5_method);
3148 
3149   // Check for abstract method error.
3150   {
3151     Label ok;
3152     __ br_notnull_short(G5_method, Assembler::pt, ok);
3153     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3154     __ should_not_reach_here();
3155     __ bind(ok);
3156   }
3157 
3158   Register Rcall = Rinterface;
3159   assert_different_registers(Rcall, G5_method, Gargs, Rret);
3160 
3161   __ profile_arguments_type(G5_method, Rcall, Gargs, true);
3162   __ profile_called_method(G5_method, Rscratch);
3163   __ call_from_interpreter(Rcall, Gargs, Rret);
3164 }
3165 
3166 void TemplateTable::invokehandle(int byte_no) {
3167   transition(vtos, vtos);
3168   assert(byte_no == f1_byte, "use this argument");
3169 
3170   const Register Rret       = Lscratch;
3171   const Register G4_mtype   = G4_scratch;
3172   const Register O0_recv    = O0;
3173   const Register Rscratch   = G3_scratch;
3174 
3175   prepare_invoke(byte_no, G5_method, Rret, G4_mtype, O0_recv);
3176   __ null_check(O0_recv);
3177 
3178   // G4: MethodType object (from cpool->resolved_references[f1], if necessary)
3179   // G5: MH.invokeExact_MT method (from f2)
3180 
3181   // Note:  G4_mtype is already pushed (if necessary) by prepare_invoke
3182 
3183   // do the call
3184   __ verify_oop(G4_mtype);
3185   __ profile_final_call(O4);  // FIXME: profile the LambdaForm also
3186   __ profile_arguments_type(G5_method, Rscratch, Gargs, true);
3187   __ call_from_interpreter(Rscratch, Gargs, Rret);
3188 }
3189 
3190 
3191 void TemplateTable::invokedynamic(int byte_no) {
3192   transition(vtos, vtos);
3193   assert(byte_no == f1_byte, "use this argument");
3194 
3195   const Register Rret        = Lscratch;
3196   const Register G4_callsite = G4_scratch;
3197   const Register Rscratch    = G3_scratch;
3198 
3199   prepare_invoke(byte_no, G5_method, Rret, G4_callsite);
3200 
3201   // G4: CallSite object (from cpool->resolved_references[f1])
3202   // G5: MH.linkToCallSite method (from f2)
3203 
3204   // Note:  G4_callsite is already pushed by prepare_invoke
3205 
3206   // %%% should make a type profile for any invokedynamic that takes a ref argument
3207   // profile this call
3208   __ profile_call(O4);
3209 
3210   // do the call
3211   __ verify_oop(G4_callsite);
3212   __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3213   __ call_from_interpreter(Rscratch, Gargs, Rret);
3214 }
3215 
3216 
3217 //----------------------------------------------------------------------------------------------------
3218 // Allocation
3219 
3220 void TemplateTable::_new() {
3221   transition(vtos, atos);
3222 
3223   Label slow_case;
3224   Label done;
3225   Label initialize_header;
3226   Label initialize_object;  // including clearing the fields
3227 
3228   Register RallocatedObject = Otos_i;
3229   Register RinstanceKlass = O1;
3230   Register Roffset = O3;
3231   Register Rscratch = O4;
3232 
3233   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3234   __ get_cpool_and_tags(Rscratch, G3_scratch);
3235   // make sure the class we're about to instantiate has been resolved
3236   // This is done before loading InstanceKlass to be consistent with the order
3237   // how Constant Pool is updated (see ConstantPool::klass_at_put)
3238   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3239   __ ldub(G3_scratch, Roffset, G3_scratch);
3240   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3241   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3242   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3243   // get InstanceKlass
3244   //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
3245   __ add(Roffset, sizeof(ConstantPool), Roffset);
3246   __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
3247 
3248   // make sure klass is fully initialized:
3249   __ ldub(RinstanceKlass, in_bytes(InstanceKlass::init_state_offset()), G3_scratch);
3250   __ cmp(G3_scratch, InstanceKlass::fully_initialized);
3251   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3252   __ delayed()->ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3253 
3254   // get instance_size in InstanceKlass (already aligned)
3255   //__ ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3256 
3257   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
3258   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
3259   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
3260   __ delayed()->nop();
3261 
3262   // allocate the instance
3263   // 1) Try to allocate in the TLAB
3264   // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
3265   // 3) if the above fails (or is not applicable), go to a slow case
3266   // (creates a new TLAB, etc.)
3267 
3268   const bool allow_shared_alloc =
3269     Universe::heap()->supports_inline_contig_alloc();
3270 
3271   if(UseTLAB) {
3272     Register RoldTopValue = RallocatedObject;
3273     Register RtlabWasteLimitValue = G3_scratch;
3274     Register RnewTopValue = G1_scratch;
3275     Register RendValue = Rscratch;
3276     Register RfreeValue = RnewTopValue;
3277 
3278     // check if we can allocate in the TLAB
3279     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
3280     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
3281     __ add(RoldTopValue, Roffset, RnewTopValue);
3282 
3283     // if there is enough space, we do not CAS and do not clear
3284     __ cmp(RnewTopValue, RendValue);
3285     if(ZeroTLAB) {
3286       // the fields have already been cleared
3287       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
3288     } else {
3289       // initialize both the header and fields
3290       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
3291     }
3292     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
3293 
3294     if (allow_shared_alloc) {
3295       // Check if tlab should be discarded (refill_waste_limit >= free)
3296       __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
3297       __ sub(RendValue, RoldTopValue, RfreeValue);
3298       __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
3299       __ cmp_and_brx_short(RtlabWasteLimitValue, RfreeValue, Assembler::greaterEqualUnsigned, Assembler::pt, slow_case); // tlab waste is small
3300 
3301       // increment waste limit to prevent getting stuck on this slow path
3302       if (Assembler::is_simm13(ThreadLocalAllocBuffer::refill_waste_limit_increment())) {
3303         __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
3304       } else {
3305         // set64 does not use the temp register if the given constant is 32 bit. So
3306         // we can just use any register; using G0 results in ignoring of the upper 32 bit
3307         // of that value.
3308         __ set64(ThreadLocalAllocBuffer::refill_waste_limit_increment(), G4_scratch, G0);
3309         __ add(RtlabWasteLimitValue, G4_scratch, RtlabWasteLimitValue);
3310       }
3311       __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
3312     } else {
3313       // No allocation in the shared eden.
3314       __ ba_short(slow_case);
3315     }
3316   }
3317 
3318   // Allocation in the shared Eden
3319   if (allow_shared_alloc) {
3320     Register RoldTopValue = G1_scratch;
3321     Register RtopAddr = G3_scratch;
3322     Register RnewTopValue = RallocatedObject;
3323     Register RendValue = Rscratch;
3324 
3325     __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
3326 
3327     Label retry;
3328     __ bind(retry);
3329     __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
3330     __ ld_ptr(RendValue, 0, RendValue);
3331     __ ld_ptr(RtopAddr, 0, RoldTopValue);
3332     __ add(RoldTopValue, Roffset, RnewTopValue);
3333 
3334     // RnewTopValue contains the top address after the new object
3335     // has been allocated.
3336     __ cmp_and_brx_short(RnewTopValue, RendValue, Assembler::greaterUnsigned, Assembler::pn, slow_case);
3337 
3338     __ cas_ptr(RtopAddr, RoldTopValue, RnewTopValue);
3339 
3340     // if someone beat us on the allocation, try again, otherwise continue
3341     __ cmp_and_brx_short(RoldTopValue, RnewTopValue, Assembler::notEqual, Assembler::pn, retry);
3342 
3343     // bump total bytes allocated by this thread
3344     // RoldTopValue and RtopAddr are dead, so can use G1 and G3
3345     __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
3346   }
3347 
3348   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3349     // clear object fields
3350     __ bind(initialize_object);
3351     __ deccc(Roffset, sizeof(oopDesc));
3352     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
3353     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
3354 
3355     // initialize remaining object fields
3356     if (UseBlockZeroing) {
3357       // Use BIS for zeroing
3358       __ bis_zeroing(G3_scratch, Roffset, G1_scratch, initialize_header);
3359     } else {
3360       Label loop;
3361       __ subcc(Roffset, wordSize, Roffset);
3362       __ bind(loop);
3363       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
3364       __ st_ptr(G0, G3_scratch, Roffset);
3365       __ br(Assembler::notEqual, false, Assembler::pt, loop);
3366       __ delayed()->subcc(Roffset, wordSize, Roffset);
3367     }
3368     __ ba_short(initialize_header);
3369   }
3370 
3371   // slow case
3372   __ bind(slow_case);
3373   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
3374   __ get_constant_pool(O1);
3375 
3376   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
3377 
3378   __ ba_short(done);
3379 
3380   // Initialize the header: mark, klass
3381   __ bind(initialize_header);
3382 
3383   if (UseBiasedLocking) {
3384     __ ld_ptr(RinstanceKlass, in_bytes(Klass::prototype_header_offset()), G4_scratch);
3385   } else {
3386     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
3387   }
3388   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
3389   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
3390   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
3391 
3392   {
3393     SkipIfEqual skip_if(
3394       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
3395     // Trigger dtrace event
3396     __ push(atos);
3397     __ call_VM_leaf(noreg,
3398        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
3399     __ pop(atos);
3400   }
3401 
3402   // continue
3403   __ bind(done);
3404 }
3405 
3406 
3407 
3408 void TemplateTable::newarray() {
3409   transition(itos, atos);
3410   __ ldub(Lbcp, 1, O1);
3411      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
3412 }
3413 
3414 
3415 void TemplateTable::anewarray() {
3416   transition(itos, atos);
3417   __ get_constant_pool(O1);
3418   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
3419      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
3420 }
3421 
3422 
3423 void TemplateTable::arraylength() {
3424   transition(atos, itos);
3425   Label ok;
3426   __ verify_oop(Otos_i);
3427   __ tst(Otos_i);
3428   __ throw_if_not_1_x( Assembler::notZero, ok );
3429   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
3430   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3431 }
3432 
3433 
3434 void TemplateTable::checkcast() {
3435   transition(atos, atos);
3436   Label done, is_null, quicked, cast_ok, resolved;
3437   Register Roffset = G1_scratch;
3438   Register RobjKlass = O5;
3439   Register RspecifiedKlass = O4;
3440 
3441   // Check for casting a NULL
3442   __ br_null(Otos_i, false, Assembler::pn, is_null);
3443   __ delayed()->nop();
3444 
3445   // Get value klass in RobjKlass
3446   __ load_klass(Otos_i, RobjKlass); // get value klass
3447 
3448   // Get constant pool tag
3449   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3450 
3451   // See if the checkcast has been quickened
3452   __ get_cpool_and_tags(Lscratch, G3_scratch);
3453   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3454   __ ldub(G3_scratch, Roffset, G3_scratch);
3455   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3456   __ br(Assembler::equal, true, Assembler::pt, quicked);
3457   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3458 
3459   __ push_ptr(); // save receiver for result, and for GC
3460   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3461   __ get_vm_result_2(RspecifiedKlass);
3462   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3463 
3464   __ ba_short(resolved);
3465 
3466   // Extract target class from constant pool
3467   __ bind(quicked);
3468   __ add(Roffset, sizeof(ConstantPool), Roffset);
3469   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3470   __ bind(resolved);
3471   __ load_klass(Otos_i, RobjKlass); // get value klass
3472 
3473   // Generate a fast subtype check.  Branch to cast_ok if no
3474   // failure.  Throw exception if failure.
3475   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
3476 
3477   // Not a subtype; so must throw exception
3478   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
3479 
3480   __ bind(cast_ok);
3481 
3482   if (ProfileInterpreter) {
3483     __ ba_short(done);
3484   }
3485   __ bind(is_null);
3486   __ profile_null_seen(G3_scratch);
3487   __ bind(done);
3488 }
3489 
3490 
3491 void TemplateTable::instanceof() {
3492   Label done, is_null, quicked, resolved;
3493   transition(atos, itos);
3494   Register Roffset = G1_scratch;
3495   Register RobjKlass = O5;
3496   Register RspecifiedKlass = O4;
3497 
3498   // Check for casting a NULL
3499   __ br_null(Otos_i, false, Assembler::pt, is_null);
3500   __ delayed()->nop();
3501 
3502   // Get value klass in RobjKlass
3503   __ load_klass(Otos_i, RobjKlass); // get value klass
3504 
3505   // Get constant pool tag
3506   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3507 
3508   // See if the checkcast has been quickened
3509   __ get_cpool_and_tags(Lscratch, G3_scratch);
3510   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3511   __ ldub(G3_scratch, Roffset, G3_scratch);
3512   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3513   __ br(Assembler::equal, true, Assembler::pt, quicked);
3514   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3515 
3516   __ push_ptr(); // save receiver for result, and for GC
3517   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3518   __ get_vm_result_2(RspecifiedKlass);
3519   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3520 
3521   __ ba_short(resolved);
3522 
3523   // Extract target class from constant pool
3524   __ bind(quicked);
3525   __ add(Roffset, sizeof(ConstantPool), Roffset);
3526   __ get_constant_pool(Lscratch);
3527   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3528   __ bind(resolved);
3529   __ load_klass(Otos_i, RobjKlass); // get value klass
3530 
3531   // Generate a fast subtype check.  Branch to cast_ok if no
3532   // failure.  Return 0 if failure.
3533   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
3534   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
3535   // Not a subtype; return 0;
3536   __ clr( Otos_i );
3537 
3538   if (ProfileInterpreter) {
3539     __ ba_short(done);
3540   }
3541   __ bind(is_null);
3542   __ profile_null_seen(G3_scratch);
3543   __ bind(done);
3544 }
3545 
3546 void TemplateTable::_breakpoint() {
3547 
3548    // Note: We get here even if we are single stepping..
3549    // jbug inists on setting breakpoints at every bytecode
3550    // even if we are in single step mode.
3551 
3552    transition(vtos, vtos);
3553    // get the unpatched byte code
3554    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
3555    __ mov(O0, Lbyte_code);
3556 
3557    // post the breakpoint event
3558    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
3559 
3560    // complete the execution of original bytecode
3561    __ dispatch_normal(vtos);
3562 }
3563 
3564 
3565 //----------------------------------------------------------------------------------------------------
3566 // Exceptions
3567 
3568 void TemplateTable::athrow() {
3569   transition(atos, vtos);
3570 
3571   // This works because exception is cached in Otos_i which is same as O0,
3572   // which is same as what throw_exception_entry_expects
3573   assert(Otos_i == Oexception, "see explanation above");
3574 
3575   __ verify_oop(Otos_i);
3576   __ null_check(Otos_i);
3577   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
3578 }
3579 
3580 
3581 //----------------------------------------------------------------------------------------------------
3582 // Synchronization
3583 
3584 
3585 // See frame_sparc.hpp for monitor block layout.
3586 // Monitor elements are dynamically allocated by growing stack as needed.
3587 
3588 void TemplateTable::monitorenter() {
3589   transition(atos, vtos);
3590   __ verify_oop(Otos_i);
3591   // Try to acquire a lock on the object
3592   // Repeat until succeeded (i.e., until
3593   // monitorenter returns true).
3594 
3595   {   Label ok;
3596     __ tst(Otos_i);
3597     __ throw_if_not_1_x( Assembler::notZero,  ok);
3598     __ delayed()->mov(Otos_i, Lscratch); // save obj
3599     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3600   }
3601 
3602   assert(O0 == Otos_i, "Be sure where the object to lock is");
3603 
3604   // find a free slot in the monitor block
3605 
3606 
3607   // initialize entry pointer
3608   __ clr(O1); // points to free slot or NULL
3609 
3610   {
3611     Label entry, loop, exit;
3612     __ add( __ top_most_monitor(), O2 ); // last one to check
3613     __ ba( entry );
3614     __ delayed()->mov( Lmonitors, O3 ); // first one to check
3615 
3616 
3617     __ bind( loop );
3618 
3619     __ verify_oop(O4);          // verify each monitor's oop
3620     __ tst(O4); // is this entry unused?
3621     __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
3622 
3623     __ cmp(O4, O0); // check if current entry is for same object
3624     __ brx( Assembler::equal, false, Assembler::pn, exit );
3625     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
3626 
3627     __ bind( entry );
3628 
3629     __ cmp( O3, O2 );
3630     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3631     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
3632 
3633     __ bind( exit );
3634   }
3635 
3636   { Label allocated;
3637 
3638     // found free slot?
3639     __ br_notnull_short(O1, Assembler::pn, allocated);
3640 
3641     __ add_monitor_to_stack( false, O2, O3 );
3642     __ mov(Lmonitors, O1);
3643 
3644     __ bind(allocated);
3645   }
3646 
3647   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3648   // The object has already been poped from the stack, so the expression stack looks correct.
3649   __ inc(Lbcp);
3650 
3651   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
3652   __ lock_object(O1, O0);
3653 
3654   // check if there's enough space on the stack for the monitors after locking
3655   __ generate_stack_overflow_check(0);
3656 
3657   // The bcp has already been incremented. Just need to dispatch to next instruction.
3658   __ dispatch_next(vtos);
3659 }
3660 
3661 
3662 void TemplateTable::monitorexit() {
3663   transition(atos, vtos);
3664   __ verify_oop(Otos_i);
3665   __ tst(Otos_i);
3666   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
3667 
3668   assert(O0 == Otos_i, "just checking");
3669 
3670   { Label entry, loop, found;
3671     __ add( __ top_most_monitor(), O2 ); // last one to check
3672     __ ba(entry);
3673     // use Lscratch to hold monitor elem to check, start with most recent monitor,
3674     // By using a local it survives the call to the C routine.
3675     __ delayed()->mov( Lmonitors, Lscratch );
3676 
3677     __ bind( loop );
3678 
3679     __ verify_oop(O4);          // verify each monitor's oop
3680     __ cmp(O4, O0); // check if current entry is for desired object
3681     __ brx( Assembler::equal, true, Assembler::pt, found );
3682     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
3683 
3684     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
3685 
3686     __ bind( entry );
3687 
3688     __ cmp( Lscratch, O2 );
3689     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3690     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
3691 
3692     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3693     __ should_not_reach_here();
3694 
3695     __ bind(found);
3696   }
3697   __ unlock_object(O1);
3698 }
3699 
3700 
3701 //----------------------------------------------------------------------------------------------------
3702 // Wide instructions
3703 
3704 void TemplateTable::wide() {
3705   transition(vtos, vtos);
3706   __ ldub(Lbcp, 1, G3_scratch);// get next bc
3707   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
3708   AddressLiteral ep(Interpreter::_wentry_point);
3709   __ set(ep, G4_scratch);
3710   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
3711   __ jmp(G3_scratch, G0);
3712   __ delayed()->nop();
3713   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
3714 }
3715 
3716 
3717 //----------------------------------------------------------------------------------------------------
3718 // Multi arrays
3719 
3720 void TemplateTable::multianewarray() {
3721   transition(vtos, atos);
3722      // put ndims * wordSize into Lscratch
3723   __ ldub( Lbcp,     3,               Lscratch);
3724   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
3725      // Lesp points past last_dim, so set to O1 to first_dim address
3726   __ add(  Lesp,     Lscratch,        O1);
3727      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
3728   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
3729 }