1 /*
   2  * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interpreter/interpreter.hpp"
  27 #include "interpreter/interpreterRuntime.hpp"
  28 #include "interpreter/templateTable.hpp"
  29 #include "memory/universe.inline.hpp"
  30 #include "oops/methodDataOop.hpp"
  31 #include "oops/objArrayKlass.hpp"
  32 #include "oops/oop.inline.hpp"
  33 #include "prims/methodHandles.hpp"
  34 #include "runtime/sharedRuntime.hpp"
  35 #include "runtime/stubRoutines.hpp"
  36 #include "runtime/synchronizer.hpp"
  37 
  38 #ifndef CC_INTERP
  39 #define __ _masm->
  40 
  41 // Misc helpers
  42 
  43 // Do an oop store like *(base + index + offset) = val
  44 // index can be noreg,
  45 static void do_oop_store(InterpreterMacroAssembler* _masm,
  46                          Register base,
  47                          Register index,
  48                          int offset,
  49                          Register val,
  50                          Register tmp,
  51                          BarrierSet::Name barrier,
  52                          bool precise) {
  53   assert(tmp != val && tmp != base && tmp != index, "register collision");
  54   assert(index == noreg || offset == 0, "only one offset");
  55   switch (barrier) {
  56 #ifndef SERIALGC
  57     case BarrierSet::G1SATBCT:
  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         if (index == noreg ) {
  66           assert(Assembler::is_simm13(offset), "fix this code");
  67           __ store_heap_oop(val, base, offset);
  68         } else {
  69           __ store_heap_oop(val, base, index);
  70         }
  71 
  72         // No need for post barrier if storing NULL
  73         if (val != G0) {
  74           if (precise) {
  75             if (index == noreg) {
  76               __ add(base, offset, base);
  77             } else {
  78               __ add(base, index, base);
  79             }
  80           }
  81           __ g1_write_barrier_post(base, val, tmp);
  82         }
  83       }
  84       break;
  85 #endif // SERIALGC
  86     case BarrierSet::CardTableModRef:
  87     case BarrierSet::CardTableExtension:
  88       {
  89         if (index == noreg ) {
  90           assert(Assembler::is_simm13(offset), "fix this code");
  91           __ store_heap_oop(val, base, offset);
  92         } else {
  93           __ store_heap_oop(val, base, index);
  94         }
  95         // No need for post barrier if storing NULL
  96         if (val != G0) {
  97           if (precise) {
  98             if (index == noreg) {
  99               __ add(base, offset, base);
 100             } else {
 101               __ add(base, index, base);
 102             }
 103           }
 104           __ card_write_barrier_post(base, val, tmp);
 105         }
 106       }
 107       break;
 108     case BarrierSet::ModRef:
 109     case BarrierSet::Other:
 110       ShouldNotReachHere();
 111       break;
 112     default      :
 113       ShouldNotReachHere();
 114 
 115   }
 116 }
 117 
 118 
 119 //----------------------------------------------------------------------------------------------------
 120 // Platform-dependent initialization
 121 
 122 void TemplateTable::pd_initialize() {
 123   // (none)
 124 }
 125 
 126 
 127 //----------------------------------------------------------------------------------------------------
 128 // Condition conversion
 129 Assembler::Condition ccNot(TemplateTable::Condition cc) {
 130   switch (cc) {
 131     case TemplateTable::equal        : return Assembler::notEqual;
 132     case TemplateTable::not_equal    : return Assembler::equal;
 133     case TemplateTable::less         : return Assembler::greaterEqual;
 134     case TemplateTable::less_equal   : return Assembler::greater;
 135     case TemplateTable::greater      : return Assembler::lessEqual;
 136     case TemplateTable::greater_equal: return Assembler::less;
 137   }
 138   ShouldNotReachHere();
 139   return Assembler::zero;
 140 }
 141 
 142 //----------------------------------------------------------------------------------------------------
 143 // Miscelaneous helper routines
 144 
 145 
 146 Address TemplateTable::at_bcp(int offset) {
 147   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
 148   return Address(Lbcp, offset);
 149 }
 150 
 151 
 152 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
 153                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
 154                                    int byte_no) {
 155   // With sharing on, may need to test methodOop flag.
 156   if (!RewriteBytecodes)  return;
 157   Label L_patch_done;
 158 
 159   switch (bc) {
 160   case Bytecodes::_fast_aputfield:
 161   case Bytecodes::_fast_bputfield:
 162   case Bytecodes::_fast_cputfield:
 163   case Bytecodes::_fast_dputfield:
 164   case Bytecodes::_fast_fputfield:
 165   case Bytecodes::_fast_iputfield:
 166   case Bytecodes::_fast_lputfield:
 167   case Bytecodes::_fast_sputfield:
 168     {
 169       // We skip bytecode quickening for putfield instructions when
 170       // the put_code written to the constant pool cache is zero.
 171       // This is required so that every execution of this instruction
 172       // calls out to InterpreterRuntime::resolve_get_put to do
 173       // additional, required work.
 174       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 175       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 176       __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
 177       __ set(bc, bc_reg);
 178       __ cmp_and_br_short(temp_reg, 0, Assembler::equal, Assembler::pn, L_patch_done);  // don't patch
 179     }
 180     break;
 181   default:
 182     assert(byte_no == -1, "sanity");
 183     if (load_bc_into_bc_reg) {
 184       __ set(bc, bc_reg);
 185     }
 186   }
 187 
 188   if (JvmtiExport::can_post_breakpoint()) {
 189     Label L_fast_patch;
 190     __ ldub(at_bcp(0), temp_reg);
 191     __ cmp_and_br_short(temp_reg, Bytecodes::_breakpoint, Assembler::notEqual, Assembler::pt, L_fast_patch);
 192     // perform the quickening, slowly, in the bowels of the breakpoint table
 193     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, bc_reg);
 194     __ ba_short(L_patch_done);
 195     __ bind(L_fast_patch);
 196   }
 197 
 198 #ifdef ASSERT
 199   Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
 200   Label L_okay;
 201   __ ldub(at_bcp(0), temp_reg);
 202   __ cmp(temp_reg, orig_bytecode);
 203   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 204   __ delayed()->cmp(temp_reg, bc_reg);
 205   __ br(Assembler::equal, false, Assembler::pt, L_okay);
 206   __ delayed()->nop();
 207   __ stop("patching the wrong bytecode");
 208   __ bind(L_okay);
 209 #endif
 210 
 211   // patch bytecode
 212   __ stb(bc_reg, at_bcp(0));
 213   __ bind(L_patch_done);
 214 }
 215 
 216 //----------------------------------------------------------------------------------------------------
 217 // Individual instructions
 218 
 219 void TemplateTable::nop() {
 220   transition(vtos, vtos);
 221   // nothing to do
 222 }
 223 
 224 void TemplateTable::shouldnotreachhere() {
 225   transition(vtos, vtos);
 226   __ stop("shouldnotreachhere bytecode");
 227 }
 228 
 229 void TemplateTable::aconst_null() {
 230   transition(vtos, atos);
 231   __ clr(Otos_i);
 232 }
 233 
 234 
 235 void TemplateTable::iconst(int value) {
 236   transition(vtos, itos);
 237   __ set(value, Otos_i);
 238 }
 239 
 240 
 241 void TemplateTable::lconst(int value) {
 242   transition(vtos, ltos);
 243   assert(value >= 0, "check this code");
 244 #ifdef _LP64
 245   __ set(value, Otos_l);
 246 #else
 247   __ set(value, Otos_l2);
 248   __ clr( Otos_l1);
 249 #endif
 250 }
 251 
 252 
 253 void TemplateTable::fconst(int value) {
 254   transition(vtos, ftos);
 255   static float zero = 0.0, one = 1.0, two = 2.0;
 256   float* p;
 257   switch( value ) {
 258    default: ShouldNotReachHere();
 259    case 0:  p = &zero;  break;
 260    case 1:  p = &one;   break;
 261    case 2:  p = &two;   break;
 262   }
 263   AddressLiteral a(p);
 264   __ sethi(a, G3_scratch);
 265   __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
 266 }
 267 
 268 
 269 void TemplateTable::dconst(int value) {
 270   transition(vtos, dtos);
 271   static double zero = 0.0, one = 1.0;
 272   double* p;
 273   switch( value ) {
 274    default: ShouldNotReachHere();
 275    case 0:  p = &zero;  break;
 276    case 1:  p = &one;   break;
 277   }
 278   AddressLiteral a(p);
 279   __ sethi(a, G3_scratch);
 280   __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
 281 }
 282 
 283 
 284 // %%%%% Should factore most snippet templates across platforms
 285 
 286 void TemplateTable::bipush() {
 287   transition(vtos, itos);
 288   __ ldsb( at_bcp(1), Otos_i );
 289 }
 290 
 291 void TemplateTable::sipush() {
 292   transition(vtos, itos);
 293   __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
 294 }
 295 
 296 void TemplateTable::ldc(bool wide) {
 297   transition(vtos, vtos);
 298   Label call_ldc, notInt, isString, notString, notClass, exit;
 299 
 300   if (wide) {
 301     __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 302   } else {
 303     __ ldub(Lbcp, 1, O1);
 304   }
 305   __ get_cpool_and_tags(O0, O2);
 306 
 307   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
 308   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
 309 
 310   // get type from tags
 311   __ add(O2, tags_offset, O2);
 312   __ ldub(O2, O1, O2);
 313   // unresolved string? If so, must resolve
 314   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedString, Assembler::equal, Assembler::pt, call_ldc);
 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::equal, true, Assembler::pt, isString);
 345   __ delayed()->cmp(O2, JVM_CONSTANT_Object);
 346   __ brx(Assembler::notEqual, true, Assembler::pt, notString);
 347   __ delayed()->ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
 348   __ bind(isString);
 349   __ ld_ptr(O0, O1, Otos_i);
 350   __ verify_oop(Otos_i);
 351   __ push(atos);
 352   __ ba_short(exit);
 353 
 354   __ bind(notString);
 355  // __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
 356   __ push(ftos);
 357 
 358   __ bind(exit);
 359 }
 360 
 361 // Fast path for caching oop constants.
 362 // %%% We should use this to handle Class and String constants also.
 363 // %%% It will simplify the ldc/primitive path considerably.
 364 void TemplateTable::fast_aldc(bool wide) {
 365   transition(vtos, atos);
 366 
 367   if (!EnableInvokeDynamic) {
 368     // We should not encounter this bytecode if !EnableInvokeDynamic.
 369     // The verifier will stop it.  However, if we get past the verifier,
 370     // this will stop the thread in a reasonable way, without crashing the JVM.
 371     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
 372                      InterpreterRuntime::throw_IncompatibleClassChangeError));
 373     // the call_VM checks for exception, so we should never return here.
 374     __ should_not_reach_here();
 375     return;
 376   }
 377 
 378   Register Rcache = G3_scratch;
 379   Register Rscratch = G4_scratch;
 380 
 381   resolve_cache_and_index(f1_oop, Otos_i, Rcache, Rscratch, wide ? sizeof(u2) : sizeof(u1));
 382 
 383   __ verify_oop(Otos_i);
 384 
 385   Label L_done;
 386   const Register Rcon_klass = G3_scratch;  // same as Rcache
 387   const Register Rarray_klass = G4_scratch;  // same as Rscratch
 388   __ load_klass(Otos_i, Rcon_klass);
 389   AddressLiteral array_klass_addr((address)Universe::systemObjArrayKlassObj_addr());
 390   __ load_contents(array_klass_addr, Rarray_klass);
 391   __ cmp_and_brx_short(Rarray_klass, Rcon_klass, Assembler::notEqual, Assembler::pt, L_done);
 392   __ ld(Address(Otos_i, arrayOopDesc::length_offset_in_bytes()), Rcon_klass);
 393   __ tst(Rcon_klass);
 394   __ brx(Assembler::zero, true, Assembler::pt, L_done);
 395   __ delayed()->clr(Otos_i);    // executed only if branch is taken
 396 
 397   // Load the exception from the system-array which wraps it:
 398   __ load_heap_oop(Otos_i, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
 399   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
 400 
 401   __ bind(L_done);
 402 }
 403 
 404 void TemplateTable::ldc2_w() {
 405   transition(vtos, vtos);
 406   Label retry, resolved, Long, exit;
 407 
 408   __ bind(retry);
 409   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
 410   __ get_cpool_and_tags(O0, O2);
 411 
 412   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
 413   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
 414   // get type from tags
 415   __ add(O2, tags_offset, O2);
 416   __ ldub(O2, O1, O2);
 417 
 418   __ sll(O1, LogBytesPerWord, O1);
 419   __ add(O0, O1, G3_scratch);
 420 
 421   __ cmp_and_brx_short(O2, JVM_CONSTANT_Double, Assembler::notEqual, Assembler::pt, Long);
 422   // A double can be placed at word-aligned locations in the constant pool.
 423   // Check out Conversions.java for an example.
 424   // Also constantPoolOopDesc::header_size() is 20, which makes it very difficult
 425   // to double-align double on the constant pool.  SG, 11/7/97
 426 #ifdef _LP64
 427   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
 428 #else
 429   FloatRegister f = Ftos_d;
 430   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset, f);
 431   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset + sizeof(jdouble)/2,
 432          f->successor());
 433 #endif
 434   __ push(dtos);
 435   __ ba_short(exit);
 436 
 437   __ bind(Long);
 438 #ifdef _LP64
 439   __ ldx(G3_scratch, base_offset, Otos_l);
 440 #else
 441   __ ld(G3_scratch, base_offset, Otos_l);
 442   __ ld(G3_scratch, base_offset + sizeof(jlong)/2, Otos_l->successor());
 443 #endif
 444   __ push(ltos);
 445 
 446   __ bind(exit);
 447 }
 448 
 449 
 450 void TemplateTable::locals_index(Register reg, int offset) {
 451   __ ldub( at_bcp(offset), reg );
 452 }
 453 
 454 
 455 void TemplateTable::locals_index_wide(Register reg) {
 456   // offset is 2, not 1, because Lbcp points to wide prefix code
 457   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
 458 }
 459 
 460 void TemplateTable::iload() {
 461   transition(vtos, itos);
 462   // Rewrite iload,iload  pair into fast_iload2
 463   //         iload,caload pair into fast_icaload
 464   if (RewriteFrequentPairs) {
 465     Label rewrite, done;
 466 
 467     // get next byte
 468     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
 469 
 470     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
 471     // last two iloads in a pair.  Comparing against fast_iload means that
 472     // the next bytecode is neither an iload or a caload, and therefore
 473     // an iload pair.
 474     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_iload, Assembler::equal, Assembler::pn, done);
 475 
 476     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
 477     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 478     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
 479 
 480     __ cmp(G3_scratch, (int)Bytecodes::_caload);
 481     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 482     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
 483 
 484     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
 485     // rewrite
 486     // G4_scratch: fast bytecode
 487     __ bind(rewrite);
 488     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
 489     __ bind(done);
 490   }
 491 
 492   // Get the local value into tos
 493   locals_index(G3_scratch);
 494   __ access_local_int( G3_scratch, Otos_i );
 495 }
 496 
 497 void TemplateTable::fast_iload2() {
 498   transition(vtos, itos);
 499   locals_index(G3_scratch);
 500   __ access_local_int( G3_scratch, Otos_i );
 501   __ push_i();
 502   locals_index(G3_scratch, 3);  // get next bytecode's local index.
 503   __ access_local_int( G3_scratch, Otos_i );
 504 }
 505 
 506 void TemplateTable::fast_iload() {
 507   transition(vtos, itos);
 508   locals_index(G3_scratch);
 509   __ access_local_int( G3_scratch, Otos_i );
 510 }
 511 
 512 void TemplateTable::lload() {
 513   transition(vtos, ltos);
 514   locals_index(G3_scratch);
 515   __ access_local_long( G3_scratch, Otos_l );
 516 }
 517 
 518 
 519 void TemplateTable::fload() {
 520   transition(vtos, ftos);
 521   locals_index(G3_scratch);
 522   __ access_local_float( G3_scratch, Ftos_f );
 523 }
 524 
 525 
 526 void TemplateTable::dload() {
 527   transition(vtos, dtos);
 528   locals_index(G3_scratch);
 529   __ access_local_double( G3_scratch, Ftos_d );
 530 }
 531 
 532 
 533 void TemplateTable::aload() {
 534   transition(vtos, atos);
 535   locals_index(G3_scratch);
 536   __ access_local_ptr( G3_scratch, Otos_i);
 537 }
 538 
 539 
 540 void TemplateTable::wide_iload() {
 541   transition(vtos, itos);
 542   locals_index_wide(G3_scratch);
 543   __ access_local_int( G3_scratch, Otos_i );
 544 }
 545 
 546 
 547 void TemplateTable::wide_lload() {
 548   transition(vtos, ltos);
 549   locals_index_wide(G3_scratch);
 550   __ access_local_long( G3_scratch, Otos_l );
 551 }
 552 
 553 
 554 void TemplateTable::wide_fload() {
 555   transition(vtos, ftos);
 556   locals_index_wide(G3_scratch);
 557   __ access_local_float( G3_scratch, Ftos_f );
 558 }
 559 
 560 
 561 void TemplateTable::wide_dload() {
 562   transition(vtos, dtos);
 563   locals_index_wide(G3_scratch);
 564   __ access_local_double( G3_scratch, Ftos_d );
 565 }
 566 
 567 
 568 void TemplateTable::wide_aload() {
 569   transition(vtos, atos);
 570   locals_index_wide(G3_scratch);
 571   __ access_local_ptr( G3_scratch, Otos_i );
 572   __ verify_oop(Otos_i);
 573 }
 574 
 575 
 576 void TemplateTable::iaload() {
 577   transition(itos, itos);
 578   // Otos_i: index
 579   // tos: array
 580   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 581   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
 582 }
 583 
 584 
 585 void TemplateTable::laload() {
 586   transition(itos, ltos);
 587   // Otos_i: index
 588   // O2: array
 589   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 590   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
 591 }
 592 
 593 
 594 void TemplateTable::faload() {
 595   transition(itos, ftos);
 596   // Otos_i: index
 597   // O2: array
 598   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
 599   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
 600 }
 601 
 602 
 603 void TemplateTable::daload() {
 604   transition(itos, dtos);
 605   // Otos_i: index
 606   // O2: array
 607   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
 608   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
 609 }
 610 
 611 
 612 void TemplateTable::aaload() {
 613   transition(itos, atos);
 614   // Otos_i: index
 615   // tos: array
 616   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
 617   __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
 618   __ verify_oop(Otos_i);
 619 }
 620 
 621 
 622 void TemplateTable::baload() {
 623   transition(itos, itos);
 624   // Otos_i: index
 625   // tos: array
 626   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
 627   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
 628 }
 629 
 630 
 631 void TemplateTable::caload() {
 632   transition(itos, itos);
 633   // Otos_i: index
 634   // tos: array
 635   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 636   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 637 }
 638 
 639 void TemplateTable::fast_icaload() {
 640   transition(vtos, itos);
 641   // Otos_i: index
 642   // tos: array
 643   locals_index(G3_scratch);
 644   __ access_local_int( G3_scratch, Otos_i );
 645   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 646   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
 647 }
 648 
 649 
 650 void TemplateTable::saload() {
 651   transition(itos, itos);
 652   // Otos_i: index
 653   // tos: array
 654   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
 655   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
 656 }
 657 
 658 
 659 void TemplateTable::iload(int n) {
 660   transition(vtos, itos);
 661   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 662 }
 663 
 664 
 665 void TemplateTable::lload(int n) {
 666   transition(vtos, ltos);
 667   assert(n+1 < Argument::n_register_parameters, "would need more code");
 668   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
 669 }
 670 
 671 
 672 void TemplateTable::fload(int n) {
 673   transition(vtos, ftos);
 674   assert(n < Argument::n_register_parameters, "would need more code");
 675   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
 676 }
 677 
 678 
 679 void TemplateTable::dload(int n) {
 680   transition(vtos, dtos);
 681   FloatRegister dst = Ftos_d;
 682   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
 683 }
 684 
 685 
 686 void TemplateTable::aload(int n) {
 687   transition(vtos, atos);
 688   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
 689 }
 690 
 691 
 692 void TemplateTable::aload_0() {
 693   transition(vtos, atos);
 694 
 695   // According to bytecode histograms, the pairs:
 696   //
 697   // _aload_0, _fast_igetfield (itos)
 698   // _aload_0, _fast_agetfield (atos)
 699   // _aload_0, _fast_fgetfield (ftos)
 700   //
 701   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
 702   // bytecode checks the next bytecode and then rewrites the current
 703   // bytecode into a pair bytecode; otherwise it rewrites the current
 704   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
 705   //
 706   if (RewriteFrequentPairs) {
 707     Label rewrite, done;
 708 
 709     // get next byte
 710     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
 711 
 712     // do actual aload_0
 713     aload(0);
 714 
 715     // if _getfield then wait with rewrite
 716     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_getfield, Assembler::equal, Assembler::pn, done);
 717 
 718     // if _igetfield then rewrite to _fast_iaccess_0
 719     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 720     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
 721     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 722     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
 723 
 724     // if _agetfield then rewrite to _fast_aaccess_0
 725     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 726     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
 727     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 728     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
 729 
 730     // if _fgetfield then rewrite to _fast_faccess_0
 731     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 732     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
 733     __ br(Assembler::equal, false, Assembler::pn, rewrite);
 734     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
 735 
 736     // else rewrite to _fast_aload0
 737     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
 738     __ set(Bytecodes::_fast_aload_0, G4_scratch);
 739 
 740     // rewrite
 741     // G4_scratch: fast bytecode
 742     __ bind(rewrite);
 743     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
 744     __ bind(done);
 745   } else {
 746     aload(0);
 747   }
 748 }
 749 
 750 
 751 void TemplateTable::istore() {
 752   transition(itos, vtos);
 753   locals_index(G3_scratch);
 754   __ store_local_int( G3_scratch, Otos_i );
 755 }
 756 
 757 
 758 void TemplateTable::lstore() {
 759   transition(ltos, vtos);
 760   locals_index(G3_scratch);
 761   __ store_local_long( G3_scratch, Otos_l );
 762 }
 763 
 764 
 765 void TemplateTable::fstore() {
 766   transition(ftos, vtos);
 767   locals_index(G3_scratch);
 768   __ store_local_float( G3_scratch, Ftos_f );
 769 }
 770 
 771 
 772 void TemplateTable::dstore() {
 773   transition(dtos, vtos);
 774   locals_index(G3_scratch);
 775   __ store_local_double( G3_scratch, Ftos_d );
 776 }
 777 
 778 
 779 void TemplateTable::astore() {
 780   transition(vtos, vtos);
 781   __ load_ptr(0, Otos_i);
 782   __ inc(Lesp, Interpreter::stackElementSize);
 783   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 784   locals_index(G3_scratch);
 785   __ store_local_ptr(G3_scratch, Otos_i);
 786 }
 787 
 788 
 789 void TemplateTable::wide_istore() {
 790   transition(vtos, vtos);
 791   __ pop_i();
 792   locals_index_wide(G3_scratch);
 793   __ store_local_int( G3_scratch, Otos_i );
 794 }
 795 
 796 
 797 void TemplateTable::wide_lstore() {
 798   transition(vtos, vtos);
 799   __ pop_l();
 800   locals_index_wide(G3_scratch);
 801   __ store_local_long( G3_scratch, Otos_l );
 802 }
 803 
 804 
 805 void TemplateTable::wide_fstore() {
 806   transition(vtos, vtos);
 807   __ pop_f();
 808   locals_index_wide(G3_scratch);
 809   __ store_local_float( G3_scratch, Ftos_f );
 810 }
 811 
 812 
 813 void TemplateTable::wide_dstore() {
 814   transition(vtos, vtos);
 815   __ pop_d();
 816   locals_index_wide(G3_scratch);
 817   __ store_local_double( G3_scratch, Ftos_d );
 818 }
 819 
 820 
 821 void TemplateTable::wide_astore() {
 822   transition(vtos, vtos);
 823   __ load_ptr(0, Otos_i);
 824   __ inc(Lesp, Interpreter::stackElementSize);
 825   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 826   locals_index_wide(G3_scratch);
 827   __ store_local_ptr(G3_scratch, Otos_i);
 828 }
 829 
 830 
 831 void TemplateTable::iastore() {
 832   transition(itos, vtos);
 833   __ pop_i(O2); // index
 834   // Otos_i: val
 835   // O3: array
 836   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 837   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
 838 }
 839 
 840 
 841 void TemplateTable::lastore() {
 842   transition(ltos, vtos);
 843   __ pop_i(O2); // index
 844   // Otos_l: val
 845   // O3: array
 846   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 847   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
 848 }
 849 
 850 
 851 void TemplateTable::fastore() {
 852   transition(ftos, vtos);
 853   __ pop_i(O2); // index
 854   // Ftos_f: val
 855   // O3: array
 856   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
 857   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
 858 }
 859 
 860 
 861 void TemplateTable::dastore() {
 862   transition(dtos, vtos);
 863   __ pop_i(O2); // index
 864   // Fos_d: val
 865   // O3: array
 866   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
 867   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
 868 }
 869 
 870 
 871 void TemplateTable::aastore() {
 872   Label store_ok, is_null, done;
 873   transition(vtos, vtos);
 874   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
 875   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
 876   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
 877   // Otos_i: val
 878   // O2: index
 879   // O3: array
 880   __ verify_oop(Otos_i);
 881   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
 882 
 883   // do array store check - check for NULL value first
 884   __ br_null_short( Otos_i, Assembler::pn, is_null );
 885 
 886   __ load_klass(O3, O4); // get array klass
 887   __ load_klass(Otos_i, O5); // get value klass
 888 
 889   // do fast instanceof cache test
 890 
 891   __ ld_ptr(O4,     objArrayKlass::element_klass_offset_in_bytes(),  O4);
 892 
 893   assert(Otos_i == O0, "just checking");
 894 
 895   // Otos_i:    value
 896   // O1:        addr - offset
 897   // O2:        index
 898   // O3:        array
 899   // O4:        array element klass
 900   // O5:        value klass
 901 
 902   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
 903 
 904   // Generate a fast subtype check.  Branch to store_ok if no
 905   // failure.  Throw if failure.
 906   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
 907 
 908   // Not a subtype; so must throw exception
 909   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
 910 
 911   // Store is OK.
 912   __ bind(store_ok);
 913   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
 914 
 915   __ ba(done);
 916   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
 917 
 918   __ bind(is_null);
 919   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
 920 
 921   __ profile_null_seen(G3_scratch);
 922   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
 923   __ bind(done);
 924 }
 925 
 926 
 927 void TemplateTable::bastore() {
 928   transition(itos, vtos);
 929   __ pop_i(O2); // index
 930   // Otos_i: val
 931   // O3: array
 932   __ index_check(O3, O2, 0, G3_scratch, O2);
 933   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
 934 }
 935 
 936 
 937 void TemplateTable::castore() {
 938   transition(itos, vtos);
 939   __ pop_i(O2); // index
 940   // Otos_i: val
 941   // O3: array
 942   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
 943   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
 944 }
 945 
 946 
 947 void TemplateTable::sastore() {
 948   // %%%%% Factor across platform
 949   castore();
 950 }
 951 
 952 
 953 void TemplateTable::istore(int n) {
 954   transition(itos, vtos);
 955   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
 956 }
 957 
 958 
 959 void TemplateTable::lstore(int n) {
 960   transition(ltos, vtos);
 961   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
 962   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
 963 
 964 }
 965 
 966 
 967 void TemplateTable::fstore(int n) {
 968   transition(ftos, vtos);
 969   assert(n < Argument::n_register_parameters, "only handle register cases");
 970   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
 971 }
 972 
 973 
 974 void TemplateTable::dstore(int n) {
 975   transition(dtos, vtos);
 976   FloatRegister src = Ftos_d;
 977   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
 978 }
 979 
 980 
 981 void TemplateTable::astore(int n) {
 982   transition(vtos, vtos);
 983   __ load_ptr(0, Otos_i);
 984   __ inc(Lesp, Interpreter::stackElementSize);
 985   __ verify_oop_or_return_address(Otos_i, G3_scratch);
 986   __ store_local_ptr(n, Otos_i);
 987 }
 988 
 989 
 990 void TemplateTable::pop() {
 991   transition(vtos, vtos);
 992   __ inc(Lesp, Interpreter::stackElementSize);
 993 }
 994 
 995 
 996 void TemplateTable::pop2() {
 997   transition(vtos, vtos);
 998   __ inc(Lesp, 2 * Interpreter::stackElementSize);
 999 }
1000 
1001 
1002 void TemplateTable::dup() {
1003   transition(vtos, vtos);
1004   // stack: ..., a
1005   // load a and tag
1006   __ load_ptr(0, Otos_i);
1007   __ push_ptr(Otos_i);
1008   // stack: ..., a, a
1009 }
1010 
1011 
1012 void TemplateTable::dup_x1() {
1013   transition(vtos, vtos);
1014   // stack: ..., a, b
1015   __ load_ptr( 1, G3_scratch);  // get a
1016   __ load_ptr( 0, Otos_l1);     // get b
1017   __ store_ptr(1, Otos_l1);     // put b
1018   __ store_ptr(0, G3_scratch);  // put a - like swap
1019   __ push_ptr(Otos_l1);         // push b
1020   // stack: ..., b, a, b
1021 }
1022 
1023 
1024 void TemplateTable::dup_x2() {
1025   transition(vtos, vtos);
1026   // stack: ..., a, b, c
1027   // get c and push on stack, reuse registers
1028   __ load_ptr( 0, G3_scratch);  // get c
1029   __ push_ptr(G3_scratch);      // push c with tag
1030   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
1031   // (stack offsets n+1 now)
1032   __ load_ptr( 3, Otos_l1);     // get a
1033   __ store_ptr(3, G3_scratch);  // put c at 3
1034   // stack: ..., c, b, c, c  (a in reg)
1035   __ load_ptr( 2, G3_scratch);  // get b
1036   __ store_ptr(2, Otos_l1);     // put a at 2
1037   // stack: ..., c, a, c, c  (b in reg)
1038   __ store_ptr(1, G3_scratch);  // put b at 1
1039   // stack: ..., c, a, b, c
1040 }
1041 
1042 
1043 void TemplateTable::dup2() {
1044   transition(vtos, vtos);
1045   __ load_ptr(1, G3_scratch);  // get a
1046   __ load_ptr(0, Otos_l1);     // get b
1047   __ push_ptr(G3_scratch);     // push a
1048   __ push_ptr(Otos_l1);        // push b
1049   // stack: ..., a, b, a, b
1050 }
1051 
1052 
1053 void TemplateTable::dup2_x1() {
1054   transition(vtos, vtos);
1055   // stack: ..., a, b, c
1056   __ load_ptr( 1, Lscratch);    // get b
1057   __ load_ptr( 2, Otos_l1);     // get a
1058   __ store_ptr(2, Lscratch);    // put b at a
1059   // stack: ..., b, b, c
1060   __ load_ptr( 0, G3_scratch);  // get c
1061   __ store_ptr(1, G3_scratch);  // put c at b
1062   // stack: ..., b, c, c
1063   __ store_ptr(0, Otos_l1);     // put a at c
1064   // stack: ..., b, c, a
1065   __ push_ptr(Lscratch);        // push b
1066   __ push_ptr(G3_scratch);      // push c
1067   // stack: ..., b, c, a, b, c
1068 }
1069 
1070 
1071 // The spec says that these types can be a mixture of category 1 (1 word)
1072 // types and/or category 2 types (long and doubles)
1073 void TemplateTable::dup2_x2() {
1074   transition(vtos, vtos);
1075   // stack: ..., a, b, c, d
1076   __ load_ptr( 1, Lscratch);    // get c
1077   __ load_ptr( 3, Otos_l1);     // get a
1078   __ store_ptr(3, Lscratch);    // put c at 3
1079   __ store_ptr(1, Otos_l1);     // put a at 1
1080   // stack: ..., c, b, a, d
1081   __ load_ptr( 2, G3_scratch);  // get b
1082   __ load_ptr( 0, Otos_l1);     // get d
1083   __ store_ptr(0, G3_scratch);  // put b at 0
1084   __ store_ptr(2, Otos_l1);     // put d at 2
1085   // stack: ..., c, d, a, b
1086   __ push_ptr(Lscratch);        // push c
1087   __ push_ptr(Otos_l1);         // push d
1088   // stack: ..., c, d, a, b, c, d
1089 }
1090 
1091 
1092 void TemplateTable::swap() {
1093   transition(vtos, vtos);
1094   // stack: ..., a, b
1095   __ load_ptr( 1, G3_scratch);  // get a
1096   __ load_ptr( 0, Otos_l1);     // get b
1097   __ store_ptr(0, G3_scratch);  // put b
1098   __ store_ptr(1, Otos_l1);     // put a
1099   // stack: ..., b, a
1100 }
1101 
1102 
1103 void TemplateTable::iop2(Operation op) {
1104   transition(itos, itos);
1105   __ pop_i(O1);
1106   switch (op) {
1107    case  add:  __  add(O1, Otos_i, Otos_i);  break;
1108    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
1109      // %%%%% Mul may not exist: better to call .mul?
1110    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
1111    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
1112    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
1113    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
1114    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
1115    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
1116    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
1117    default: ShouldNotReachHere();
1118   }
1119 }
1120 
1121 
1122 void TemplateTable::lop2(Operation op) {
1123   transition(ltos, ltos);
1124   __ pop_l(O2);
1125   switch (op) {
1126 #ifdef _LP64
1127    case  add:  __  add(O2, Otos_l, Otos_l);  break;
1128    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
1129    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
1130    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
1131    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
1132 #else
1133    case  add:  __ addcc(O3, Otos_l2, Otos_l2);  __ addc(O2, Otos_l1, Otos_l1);  break;
1134    case  sub:  __ subcc(O3, Otos_l2, Otos_l2);  __ subc(O2, Otos_l1, Otos_l1);  break;
1135    case _and:  __  and3(O3, Otos_l2, Otos_l2);  __ and3(O2, Otos_l1, Otos_l1);  break;
1136    case  _or:  __   or3(O3, Otos_l2, Otos_l2);  __  or3(O2, Otos_l1, Otos_l1);  break;
1137    case _xor:  __  xor3(O3, Otos_l2, Otos_l2);  __ xor3(O2, Otos_l1, Otos_l1);  break;
1138 #endif
1139    default: ShouldNotReachHere();
1140   }
1141 }
1142 
1143 
1144 void TemplateTable::idiv() {
1145   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
1146   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
1147 
1148   transition(itos, itos);
1149   __ pop_i(O1); // get 1st op
1150 
1151   // Y contains upper 32 bits of result, set it to 0 or all ones
1152   __ wry(G0);
1153   __ mov(~0, G3_scratch);
1154 
1155   __ tst(O1);
1156      Label neg;
1157   __ br(Assembler::negative, true, Assembler::pn, neg);
1158   __ delayed()->wry(G3_scratch);
1159   __ bind(neg);
1160 
1161      Label ok;
1162   __ tst(Otos_i);
1163   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
1164 
1165   const int min_int = 0x80000000;
1166   Label regular;
1167   __ cmp(Otos_i, -1);
1168   __ br(Assembler::notEqual, false, Assembler::pt, regular);
1169 #ifdef _LP64
1170   // Don't put set in delay slot
1171   // Set will turn into multiple instructions in 64 bit mode
1172   __ delayed()->nop();
1173   __ set(min_int, G4_scratch);
1174 #else
1175   __ delayed()->set(min_int, G4_scratch);
1176 #endif
1177   Label done;
1178   __ cmp(O1, G4_scratch);
1179   __ br(Assembler::equal, true, Assembler::pt, done);
1180   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
1181 
1182   __ bind(regular);
1183   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
1184   __ bind(done);
1185 }
1186 
1187 
1188 void TemplateTable::irem() {
1189   transition(itos, itos);
1190   __ mov(Otos_i, O2); // save divisor
1191   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
1192   __ smul(Otos_i, O2, Otos_i);
1193   __ sub(O1, Otos_i, Otos_i);
1194 }
1195 
1196 
1197 void TemplateTable::lmul() {
1198   transition(ltos, ltos);
1199   __ pop_l(O2);
1200 #ifdef _LP64
1201   __ mulx(Otos_l, O2, Otos_l);
1202 #else
1203   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lmul));
1204 #endif
1205 
1206 }
1207 
1208 
1209 void TemplateTable::ldiv() {
1210   transition(ltos, ltos);
1211 
1212   // check for zero
1213   __ pop_l(O2);
1214 #ifdef _LP64
1215   __ tst(Otos_l);
1216   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1217   __ sdivx(O2, Otos_l, Otos_l);
1218 #else
1219   __ orcc(Otos_l1, Otos_l2, G0);
1220   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1221   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1222 #endif
1223 }
1224 
1225 
1226 void TemplateTable::lrem() {
1227   transition(ltos, ltos);
1228 
1229   // check for zero
1230   __ pop_l(O2);
1231 #ifdef _LP64
1232   __ tst(Otos_l);
1233   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1234   __ sdivx(O2, Otos_l, Otos_l2);
1235   __ mulx (Otos_l2, Otos_l, Otos_l2);
1236   __ sub  (O2, Otos_l2, Otos_l);
1237 #else
1238   __ orcc(Otos_l1, Otos_l2, G0);
1239   __ throw_if_not_icc(Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1240   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1241 #endif
1242 }
1243 
1244 
1245 void TemplateTable::lshl() {
1246   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
1247 
1248   __ pop_l(O2);                          // shift value in O2, O3
1249 #ifdef _LP64
1250   __ sllx(O2, Otos_i, Otos_l);
1251 #else
1252   __ lshl(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1253 #endif
1254 }
1255 
1256 
1257 void TemplateTable::lshr() {
1258   transition(itos, ltos); // %%%% see lshl comment
1259 
1260   __ pop_l(O2);                          // shift value in O2, O3
1261 #ifdef _LP64
1262   __ srax(O2, Otos_i, Otos_l);
1263 #else
1264   __ lshr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1265 #endif
1266 }
1267 
1268 
1269 
1270 void TemplateTable::lushr() {
1271   transition(itos, ltos); // %%%% see lshl comment
1272 
1273   __ pop_l(O2);                          // shift value in O2, O3
1274 #ifdef _LP64
1275   __ srlx(O2, Otos_i, Otos_l);
1276 #else
1277   __ lushr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1278 #endif
1279 }
1280 
1281 
1282 void TemplateTable::fop2(Operation op) {
1283   transition(ftos, ftos);
1284   switch (op) {
1285    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1286    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1287    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1288    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1289    case  rem:
1290      assert(Ftos_f == F0, "just checking");
1291 #ifdef _LP64
1292      // LP64 calling conventions use F1, F3 for passing 2 floats
1293      __ pop_f(F1);
1294      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
1295 #else
1296      __ pop_i(O0);
1297      __ stf(FloatRegisterImpl::S, Ftos_f, __ d_tmp);
1298      __ ld( __ d_tmp, O1 );
1299 #endif
1300      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1301      assert( Ftos_f == F0, "fix this code" );
1302      break;
1303 
1304    default: ShouldNotReachHere();
1305   }
1306 }
1307 
1308 
1309 void TemplateTable::dop2(Operation op) {
1310   transition(dtos, dtos);
1311   switch (op) {
1312    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1313    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1314    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1315    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1316    case  rem:
1317 #ifdef _LP64
1318      // Pass arguments in D0, D2
1319      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
1320      __ pop_d( F0 );
1321 #else
1322      // Pass arguments in O0O1, O2O3
1323      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
1324      __ ldd( __ d_tmp, O2 );
1325      __ pop_d(Ftos_f);
1326      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
1327      __ ldd( __ d_tmp, O0 );
1328 #endif
1329      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1330      assert( Ftos_d == F0, "fix this code" );
1331      break;
1332 
1333    default: ShouldNotReachHere();
1334   }
1335 }
1336 
1337 
1338 void TemplateTable::ineg() {
1339   transition(itos, itos);
1340   __ neg(Otos_i);
1341 }
1342 
1343 
1344 void TemplateTable::lneg() {
1345   transition(ltos, ltos);
1346 #ifdef _LP64
1347   __ sub(G0, Otos_l, Otos_l);
1348 #else
1349   __ lneg(Otos_l1, Otos_l2);
1350 #endif
1351 }
1352 
1353 
1354 void TemplateTable::fneg() {
1355   transition(ftos, ftos);
1356   __ fneg(FloatRegisterImpl::S, Ftos_f);
1357 }
1358 
1359 
1360 void TemplateTable::dneg() {
1361   transition(dtos, dtos);
1362   // v8 has fnegd if source and dest are the same
1363   __ fneg(FloatRegisterImpl::D, Ftos_f);
1364 }
1365 
1366 
1367 void TemplateTable::iinc() {
1368   transition(vtos, vtos);
1369   locals_index(G3_scratch);
1370   __ ldsb(Lbcp, 2, O2);  // load constant
1371   __ access_local_int(G3_scratch, Otos_i);
1372   __ add(Otos_i, O2, Otos_i);
1373   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1374 }
1375 
1376 
1377 void TemplateTable::wide_iinc() {
1378   transition(vtos, vtos);
1379   locals_index_wide(G3_scratch);
1380   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
1381   __ access_local_int(G3_scratch, Otos_i);
1382   __ add(Otos_i, O3, Otos_i);
1383   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1384 }
1385 
1386 
1387 void TemplateTable::convert() {
1388 // %%%%% Factor this first part accross platforms
1389   #ifdef ASSERT
1390     TosState tos_in  = ilgl;
1391     TosState tos_out = ilgl;
1392     switch (bytecode()) {
1393       case Bytecodes::_i2l: // fall through
1394       case Bytecodes::_i2f: // fall through
1395       case Bytecodes::_i2d: // fall through
1396       case Bytecodes::_i2b: // fall through
1397       case Bytecodes::_i2c: // fall through
1398       case Bytecodes::_i2s: tos_in = itos; break;
1399       case Bytecodes::_l2i: // fall through
1400       case Bytecodes::_l2f: // fall through
1401       case Bytecodes::_l2d: tos_in = ltos; break;
1402       case Bytecodes::_f2i: // fall through
1403       case Bytecodes::_f2l: // fall through
1404       case Bytecodes::_f2d: tos_in = ftos; break;
1405       case Bytecodes::_d2i: // fall through
1406       case Bytecodes::_d2l: // fall through
1407       case Bytecodes::_d2f: tos_in = dtos; break;
1408       default             : ShouldNotReachHere();
1409     }
1410     switch (bytecode()) {
1411       case Bytecodes::_l2i: // fall through
1412       case Bytecodes::_f2i: // fall through
1413       case Bytecodes::_d2i: // fall through
1414       case Bytecodes::_i2b: // fall through
1415       case Bytecodes::_i2c: // fall through
1416       case Bytecodes::_i2s: tos_out = itos; break;
1417       case Bytecodes::_i2l: // fall through
1418       case Bytecodes::_f2l: // fall through
1419       case Bytecodes::_d2l: tos_out = ltos; break;
1420       case Bytecodes::_i2f: // fall through
1421       case Bytecodes::_l2f: // fall through
1422       case Bytecodes::_d2f: tos_out = ftos; break;
1423       case Bytecodes::_i2d: // fall through
1424       case Bytecodes::_l2d: // fall through
1425       case Bytecodes::_f2d: tos_out = dtos; break;
1426       default             : ShouldNotReachHere();
1427     }
1428     transition(tos_in, tos_out);
1429   #endif
1430 
1431 
1432   // Conversion
1433   Label done;
1434   switch (bytecode()) {
1435    case Bytecodes::_i2l:
1436 #ifdef _LP64
1437     // Sign extend the 32 bits
1438     __ sra ( Otos_i, 0, Otos_l );
1439 #else
1440     __ addcc(Otos_i, 0, Otos_l2);
1441     __ br(Assembler::greaterEqual, true, Assembler::pt, done);
1442     __ delayed()->clr(Otos_l1);
1443     __ set(~0, Otos_l1);
1444 #endif
1445     break;
1446 
1447    case Bytecodes::_i2f:
1448     __ st(Otos_i, __ d_tmp );
1449     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1450     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
1451     break;
1452 
1453    case Bytecodes::_i2d:
1454     __ st(Otos_i, __ d_tmp);
1455     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1456     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
1457     break;
1458 
1459    case Bytecodes::_i2b:
1460     __ sll(Otos_i, 24, Otos_i);
1461     __ sra(Otos_i, 24, Otos_i);
1462     break;
1463 
1464    case Bytecodes::_i2c:
1465     __ sll(Otos_i, 16, Otos_i);
1466     __ srl(Otos_i, 16, Otos_i);
1467     break;
1468 
1469    case Bytecodes::_i2s:
1470     __ sll(Otos_i, 16, Otos_i);
1471     __ sra(Otos_i, 16, Otos_i);
1472     break;
1473 
1474    case Bytecodes::_l2i:
1475 #ifndef _LP64
1476     __ mov(Otos_l2, Otos_i);
1477 #else
1478     // Sign-extend into the high 32 bits
1479     __ sra(Otos_l, 0, Otos_i);
1480 #endif
1481     break;
1482 
1483    case Bytecodes::_l2f:
1484    case Bytecodes::_l2d:
1485     __ st_long(Otos_l, __ d_tmp);
1486     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
1487 
1488     if (VM_Version::v9_instructions_work()) {
1489       if (bytecode() == Bytecodes::_l2f) {
1490         __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
1491       } else {
1492         __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
1493       }
1494     } else {
1495       __ call_VM_leaf(
1496         Lscratch,
1497         bytecode() == Bytecodes::_l2f
1498           ? CAST_FROM_FN_PTR(address, SharedRuntime::l2f)
1499           : CAST_FROM_FN_PTR(address, SharedRuntime::l2d)
1500       );
1501     }
1502     break;
1503 
1504   case Bytecodes::_f2i:  {
1505       Label isNaN;
1506       // result must be 0 if value is NaN; test by comparing value to itself
1507       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
1508       // According to the v8 manual, you have to have a non-fp instruction
1509       // between fcmp and fb.
1510       if (!VM_Version::v9_instructions_work()) {
1511         __ nop();
1512       }
1513       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
1514       __ delayed()->clr(Otos_i);                                     // NaN
1515       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
1516       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
1517       __ ld(__ d_tmp, Otos_i);
1518       __ bind(isNaN);
1519     }
1520     break;
1521 
1522    case Bytecodes::_f2l:
1523     // must uncache tos
1524     __ push_f();
1525 #ifdef _LP64
1526     __ pop_f(F1);
1527 #else
1528     __ pop_i(O0);
1529 #endif
1530     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1531     break;
1532 
1533    case Bytecodes::_f2d:
1534     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
1535     break;
1536 
1537    case Bytecodes::_d2i:
1538    case Bytecodes::_d2l:
1539     // must uncache tos
1540     __ push_d();
1541 #ifdef _LP64
1542     // LP64 calling conventions pass first double arg in D0
1543     __ pop_d( Ftos_d );
1544 #else
1545     __ pop_i( O0 );
1546     __ pop_i( O1 );
1547 #endif
1548     __ call_VM_leaf(Lscratch,
1549         bytecode() == Bytecodes::_d2i
1550           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
1551           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1552     break;
1553 
1554     case Bytecodes::_d2f:
1555     if (VM_Version::v9_instructions_work()) {
1556       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
1557     }
1558     else {
1559       // must uncache tos
1560       __ push_d();
1561       __ pop_i(O0);
1562       __ pop_i(O1);
1563       __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::d2f));
1564     }
1565     break;
1566 
1567     default: ShouldNotReachHere();
1568   }
1569   __ bind(done);
1570 }
1571 
1572 
1573 void TemplateTable::lcmp() {
1574   transition(ltos, itos);
1575 
1576 #ifdef _LP64
1577   __ pop_l(O1); // pop off value 1, value 2 is in O0
1578   __ lcmp( O1, Otos_l, Otos_i );
1579 #else
1580   __ pop_l(O2); // cmp O2,3 to O0,1
1581   __ lcmp( O2, O3, Otos_l1, Otos_l2, Otos_i );
1582 #endif
1583 }
1584 
1585 
1586 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1587 
1588   if (is_float) __ pop_f(F2);
1589   else          __ pop_d(F2);
1590 
1591   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
1592 
1593   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
1594 }
1595 
1596 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1597   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1598   __ verify_oop(Lmethod);
1599   __ verify_thread();
1600 
1601   const Register O2_bumped_count = O2;
1602   __ profile_taken_branch(G3_scratch, O2_bumped_count);
1603 
1604   // get (wide) offset to O1_disp
1605   const Register O1_disp = O1;
1606   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
1607   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
1608 
1609   // Handle all the JSR stuff here, then exit.
1610   // It's much shorter and cleaner than intermingling with the
1611   // non-JSR normal-branch stuff occurring below.
1612   if( is_jsr ) {
1613     // compute return address as bci in Otos_i
1614     __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
1615     __ sub(Lbcp, G3_scratch, G3_scratch);
1616     __ sub(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
1617 
1618     // Bump Lbcp to target of JSR
1619     __ add(Lbcp, O1_disp, Lbcp);
1620     // Push returnAddress for "ret" on stack
1621     __ push_ptr(Otos_i);
1622     // And away we go!
1623     __ dispatch_next(vtos);
1624     return;
1625   }
1626 
1627   // Normal (non-jsr) branch handling
1628 
1629   // Save the current Lbcp
1630   const Register O0_cur_bcp = O0;
1631   __ mov( Lbcp, O0_cur_bcp );
1632 
1633 
1634   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1635   if ( increment_invocation_counter_for_backward_branches ) {
1636     Label Lforward;
1637     // check branch direction
1638     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
1639     // Bump bytecode pointer by displacement (take the branch)
1640     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
1641 
1642     if (TieredCompilation) {
1643       Label Lno_mdo, Loverflow;
1644       int increment = InvocationCounter::count_increment;
1645       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1646       if (ProfileInterpreter) {
1647         // If no method data exists, go to profile_continue.
1648         __ ld_ptr(Lmethod, methodOopDesc::method_data_offset(), G4_scratch);
1649         __ br_null_short(G4_scratch, Assembler::pn, Lno_mdo);
1650 
1651         // Increment backedge counter in the MDO
1652         Address mdo_backedge_counter(G4_scratch, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
1653                                                  in_bytes(InvocationCounter::counter_offset()));
1654         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, Lscratch,
1655                                    Assembler::notZero, &Lforward);
1656         __ ba_short(Loverflow);
1657       }
1658 
1659       // If there's no MDO, increment counter in methodOop
1660       __ bind(Lno_mdo);
1661       Address backedge_counter(Lmethod, in_bytes(methodOopDesc::backedge_counter_offset()) +
1662                                         in_bytes(InvocationCounter::counter_offset()));
1663       __ increment_mask_and_jump(backedge_counter, increment, mask, G3_scratch, Lscratch,
1664                                  Assembler::notZero, &Lforward);
1665       __ bind(Loverflow);
1666 
1667       // notify point for loop, pass branch bytecode
1668       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), O0_cur_bcp);
1669 
1670       // Was an OSR adapter generated?
1671       // O0 = osr nmethod
1672       __ br_null_short(O0, Assembler::pn, Lforward);
1673 
1674       // Has the nmethod been invalidated already?
1675       __ ld(O0, nmethod::entry_bci_offset(), O2);
1676       __ cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, Lforward);
1677 
1678       // migrate the interpreter frame off of the stack
1679 
1680       __ mov(G2_thread, L7);
1681       // save nmethod
1682       __ mov(O0, L6);
1683       __ set_last_Java_frame(SP, noreg);
1684       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
1685       __ reset_last_Java_frame();
1686       __ mov(L7, G2_thread);
1687 
1688       // move OSR nmethod to I1
1689       __ mov(L6, I1);
1690 
1691       // OSR buffer to I0
1692       __ mov(O0, I0);
1693 
1694       // remove the interpreter frame
1695       __ restore(I5_savedSP, 0, SP);
1696 
1697       // Jump to the osr code.
1698       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
1699       __ jmp(O2, G0);
1700       __ delayed()->nop();
1701 
1702     } else {
1703       // Update Backedge branch separately from invocations
1704       const Register G4_invoke_ctr = G4;
1705       __ increment_backedge_counter(G4_invoke_ctr, G1_scratch);
1706       if (ProfileInterpreter) {
1707         __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_scratch, Lforward);
1708         if (UseOnStackReplacement) {
1709           __ test_backedge_count_for_osr(O2_bumped_count, O0_cur_bcp, G3_scratch);
1710         }
1711       } else {
1712         if (UseOnStackReplacement) {
1713           __ test_backedge_count_for_osr(G4_invoke_ctr, O0_cur_bcp, G3_scratch);
1714         }
1715       }
1716     }
1717 
1718     __ bind(Lforward);
1719   } else
1720     // Bump bytecode pointer by displacement (take the branch)
1721     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
1722 
1723   // continue with bytecode @ target
1724   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1725   // %%%%% and changing dispatch_next to dispatch_only
1726   __ dispatch_next(vtos);
1727 }
1728 
1729 
1730 // Note Condition in argument is TemplateTable::Condition
1731 // arg scope is within class scope
1732 
1733 void TemplateTable::if_0cmp(Condition cc) {
1734   // no pointers, integer only!
1735   transition(itos, vtos);
1736   // assume branch is more often taken than not (loops use backward branches)
1737   __ cmp( Otos_i, 0);
1738   __ if_cmp(ccNot(cc), false);
1739 }
1740 
1741 
1742 void TemplateTable::if_icmp(Condition cc) {
1743   transition(itos, vtos);
1744   __ pop_i(O1);
1745   __ cmp(O1, Otos_i);
1746   __ if_cmp(ccNot(cc), false);
1747 }
1748 
1749 
1750 void TemplateTable::if_nullcmp(Condition cc) {
1751   transition(atos, vtos);
1752   __ tst(Otos_i);
1753   __ if_cmp(ccNot(cc), true);
1754 }
1755 
1756 
1757 void TemplateTable::if_acmp(Condition cc) {
1758   transition(atos, vtos);
1759   __ pop_ptr(O1);
1760   __ verify_oop(O1);
1761   __ verify_oop(Otos_i);
1762   __ cmp(O1, Otos_i);
1763   __ if_cmp(ccNot(cc), true);
1764 }
1765 
1766 
1767 
1768 void TemplateTable::ret() {
1769   transition(vtos, vtos);
1770   locals_index(G3_scratch);
1771   __ access_local_returnAddress(G3_scratch, Otos_i);
1772   // Otos_i contains the bci, compute the bcp from that
1773 
1774 #ifdef _LP64
1775 #ifdef ASSERT
1776   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
1777   // the result.  The return address (really a BCI) was stored with an
1778   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
1779   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
1780   // loaded value.
1781   { Label zzz ;
1782      __ set (65536, G3_scratch) ;
1783      __ cmp (Otos_i, G3_scratch) ;
1784      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
1785      __ delayed()->nop();
1786      __ stop("BCI is in the wrong register half?");
1787      __ bind (zzz) ;
1788   }
1789 #endif
1790 #endif
1791 
1792   __ profile_ret(vtos, Otos_i, G4_scratch);
1793 
1794   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
1795   __ add(G3_scratch, Otos_i, G3_scratch);
1796   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
1797   __ dispatch_next(vtos);
1798 }
1799 
1800 
1801 void TemplateTable::wide_ret() {
1802   transition(vtos, vtos);
1803   locals_index_wide(G3_scratch);
1804   __ access_local_returnAddress(G3_scratch, Otos_i);
1805   // Otos_i contains the bci, compute the bcp from that
1806 
1807   __ profile_ret(vtos, Otos_i, G4_scratch);
1808 
1809   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
1810   __ add(G3_scratch, Otos_i, G3_scratch);
1811   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
1812   __ dispatch_next(vtos);
1813 }
1814 
1815 
1816 void TemplateTable::tableswitch() {
1817   transition(itos, vtos);
1818   Label default_case, continue_execution;
1819 
1820   // align bcp
1821   __ add(Lbcp, BytesPerInt, O1);
1822   __ and3(O1, -BytesPerInt, O1);
1823   // load lo, hi
1824   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
1825   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
1826 #ifdef _LP64
1827   // Sign extend the 32 bits
1828   __ sra ( Otos_i, 0, Otos_i );
1829 #endif /* _LP64 */
1830 
1831   // check against lo & hi
1832   __ cmp( Otos_i, O2);
1833   __ br( Assembler::less, false, Assembler::pn, default_case);
1834   __ delayed()->cmp( Otos_i, O3 );
1835   __ br( Assembler::greater, false, Assembler::pn, default_case);
1836   // lookup dispatch offset
1837   __ delayed()->sub(Otos_i, O2, O2);
1838   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
1839   __ sll(O2, LogBytesPerInt, O2);
1840   __ add(O2, 3 * BytesPerInt, O2);
1841   __ ba(continue_execution);
1842   __ delayed()->ld(O1, O2, O2);
1843   // handle default
1844   __ bind(default_case);
1845   __ profile_switch_default(O3);
1846   __ ld(O1, 0, O2); // get default offset
1847   // continue execution
1848   __ bind(continue_execution);
1849   __ add(Lbcp, O2, Lbcp);
1850   __ dispatch_next(vtos);
1851 }
1852 
1853 
1854 void TemplateTable::lookupswitch() {
1855   transition(itos, itos);
1856   __ stop("lookupswitch bytecode should have been rewritten");
1857 }
1858 
1859 void TemplateTable::fast_linearswitch() {
1860   transition(itos, vtos);
1861     Label loop_entry, loop, found, continue_execution;
1862   // align bcp
1863   __ add(Lbcp, BytesPerInt, O1);
1864   __ and3(O1, -BytesPerInt, O1);
1865  // set counter
1866   __ ld(O1, BytesPerInt, O2);
1867   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
1868   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
1869   __ ba(loop_entry);
1870   __ delayed()->add(O3, O2, O2); // counter now points past last pair
1871 
1872   // table search
1873   __ bind(loop);
1874   __ cmp(O4, Otos_i);
1875   __ br(Assembler::equal, true, Assembler::pn, found);
1876   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
1877   __ inc(O3, 2 * BytesPerInt);
1878 
1879   __ bind(loop_entry);
1880   __ cmp(O2, O3);
1881   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
1882   __ delayed()->ld(O3, 0, O4);
1883 
1884   // default case
1885   __ ld(O1, 0, O4); // get default offset
1886   if (ProfileInterpreter) {
1887     __ profile_switch_default(O3);
1888     __ ba_short(continue_execution);
1889   }
1890 
1891   // entry found -> get offset
1892   __ bind(found);
1893   if (ProfileInterpreter) {
1894     __ sub(O3, O1, O3);
1895     __ sub(O3, 2*BytesPerInt, O3);
1896     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
1897     __ profile_switch_case(O3, O1, O2, G3_scratch);
1898 
1899     __ bind(continue_execution);
1900   }
1901   __ add(Lbcp, O4, Lbcp);
1902   __ dispatch_next(vtos);
1903 }
1904 
1905 
1906 void TemplateTable::fast_binaryswitch() {
1907   transition(itos, vtos);
1908   // Implementation using the following core algorithm: (copied from Intel)
1909   //
1910   // int binary_search(int key, LookupswitchPair* array, int n) {
1911   //   // Binary search according to "Methodik des Programmierens" by
1912   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1913   //   int i = 0;
1914   //   int j = n;
1915   //   while (i+1 < j) {
1916   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1917   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1918   //     // where a stands for the array and assuming that the (inexisting)
1919   //     // element a[n] is infinitely big.
1920   //     int h = (i + j) >> 1;
1921   //     // i < h < j
1922   //     if (key < array[h].fast_match()) {
1923   //       j = h;
1924   //     } else {
1925   //       i = h;
1926   //     }
1927   //   }
1928   //   // R: a[i] <= key < a[i+1] or Q
1929   //   // (i.e., if key is within array, i is the correct index)
1930   //   return i;
1931   // }
1932 
1933   // register allocation
1934   assert(Otos_i == O0, "alias checking");
1935   const Register Rkey     = Otos_i;                    // already set (tosca)
1936   const Register Rarray   = O1;
1937   const Register Ri       = O2;
1938   const Register Rj       = O3;
1939   const Register Rh       = O4;
1940   const Register Rscratch = O5;
1941 
1942   const int log_entry_size = 3;
1943   const int entry_size = 1 << log_entry_size;
1944 
1945   Label found;
1946   // Find Array start
1947   __ add(Lbcp, 3 * BytesPerInt, Rarray);
1948   __ and3(Rarray, -BytesPerInt, Rarray);
1949   // initialize i & j (in delay slot)
1950   __ clr( Ri );
1951 
1952   // and start
1953   Label entry;
1954   __ ba(entry);
1955   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
1956   // (Rj is already in the native byte-ordering.)
1957 
1958   // binary search loop
1959   { Label loop;
1960     __ bind( loop );
1961     // int h = (i + j) >> 1;
1962     __ sra( Rh, 1, Rh );
1963     // if (key < array[h].fast_match()) {
1964     //   j = h;
1965     // } else {
1966     //   i = h;
1967     // }
1968     __ sll( Rh, log_entry_size, Rscratch );
1969     __ ld( Rarray, Rscratch, Rscratch );
1970     // (Rscratch is already in the native byte-ordering.)
1971     __ cmp( Rkey, Rscratch );
1972     if ( VM_Version::v9_instructions_work() ) {
1973       __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
1974       __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
1975     }
1976     else {
1977       Label end_of_if;
1978       __ br( Assembler::less, true, Assembler::pt, end_of_if );
1979       __ delayed()->mov( Rh, Rj ); // if (<) Rj = Rh
1980       __ mov( Rh, Ri );            // else i = h
1981       __ bind(end_of_if);          // }
1982     }
1983 
1984     // while (i+1 < j)
1985     __ bind( entry );
1986     __ add( Ri, 1, Rscratch );
1987     __ cmp(Rscratch, Rj);
1988     __ br( Assembler::less, true, Assembler::pt, loop );
1989     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
1990   }
1991 
1992   // end of binary search, result index is i (must check again!)
1993   Label default_case;
1994   Label continue_execution;
1995   if (ProfileInterpreter) {
1996     __ mov( Ri, Rh );              // Save index in i for profiling
1997   }
1998   __ sll( Ri, log_entry_size, Ri );
1999   __ ld( Rarray, Ri, Rscratch );
2000   // (Rscratch is already in the native byte-ordering.)
2001   __ cmp( Rkey, Rscratch );
2002   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
2003   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
2004 
2005   // entry found -> j = offset
2006   __ inc( Ri, BytesPerInt );
2007   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2008   __ ld( Rarray, Ri, Rj );
2009   // (Rj is already in the native byte-ordering.)
2010 
2011   if (ProfileInterpreter) {
2012     __ ba_short(continue_execution);
2013   }
2014 
2015   __ bind(default_case); // fall through (if not profiling)
2016   __ profile_switch_default(Ri);
2017 
2018   __ bind(continue_execution);
2019   __ add( Lbcp, Rj, Lbcp );
2020   __ dispatch_next( vtos );
2021 }
2022 
2023 
2024 void TemplateTable::_return(TosState state) {
2025   transition(state, state);
2026   assert(_desc->calls_vm(), "inconsistent calls_vm information");
2027 
2028   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2029     assert(state == vtos, "only valid state");
2030     __ mov(G0, G3_scratch);
2031     __ access_local_ptr(G3_scratch, Otos_i);
2032     __ load_klass(Otos_i, O2);
2033     __ set(JVM_ACC_HAS_FINALIZER, G3);
2034     __ ld(O2, Klass::access_flags_offset_in_bytes(), O2);
2035     __ andcc(G3, O2, G0);
2036     Label skip_register_finalizer;
2037     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
2038     __ delayed()->nop();
2039 
2040     // Call out to do finalizer registration
2041     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
2042 
2043     __ bind(skip_register_finalizer);
2044   }
2045 
2046   __ remove_activation(state, /* throw_monitor_exception */ true);
2047 
2048   // The caller's SP was adjusted upon method entry to accomodate
2049   // the callee's non-argument locals. Undo that adjustment.
2050   __ ret();                             // return to caller
2051   __ delayed()->restore(I5_savedSP, G0, SP);
2052 }
2053 
2054 
2055 // ----------------------------------------------------------------------------
2056 // Volatile variables demand their effects be made known to all CPU's in
2057 // order.  Store buffers on most chips allow reads & writes to reorder; the
2058 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2059 // memory barrier (i.e., it's not sufficient that the interpreter does not
2060 // reorder volatile references, the hardware also must not reorder them).
2061 //
2062 // According to the new Java Memory Model (JMM):
2063 // (1) All volatiles are serialized wrt to each other.
2064 // ALSO reads & writes act as aquire & release, so:
2065 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2066 // the read float up to before the read.  It's OK for non-volatile memory refs
2067 // that happen before the volatile read to float down below it.
2068 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2069 // that happen BEFORE the write float down to after the write.  It's OK for
2070 // non-volatile memory refs that happen after the volatile write to float up
2071 // before it.
2072 //
2073 // We only put in barriers around volatile refs (they are expensive), not
2074 // _between_ memory refs (that would require us to track the flavor of the
2075 // previous memory refs).  Requirements (2) and (3) require some barriers
2076 // before volatile stores and after volatile loads.  These nearly cover
2077 // requirement (1) but miss the volatile-store-volatile-load case.  This final
2078 // case is placed after volatile-stores although it could just as well go
2079 // before volatile-loads.
2080 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
2081   // Helper function to insert a is-volatile test and memory barrier
2082   // All current sparc implementations run in TSO, needing only StoreLoad
2083   if ((order_constraint & Assembler::StoreLoad) == 0) return;
2084   __ membar( order_constraint );
2085 }
2086 
2087 // ----------------------------------------------------------------------------
2088 void TemplateTable::resolve_cache_and_index(int byte_no,
2089                                             Register result,
2090                                             Register Rcache,
2091                                             Register index,
2092                                             size_t index_size) {
2093   // Depends on cpCacheOop layout!
2094   Label resolved;
2095 
2096   if (byte_no == f1_oop) {
2097     // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
2098     // This kind of CP cache entry does not need to match the flags byte, because
2099     // there is a 1-1 relation between bytecode type and CP entry type.
2100     assert_different_registers(result, Rcache);
2101     __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2102     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
2103               ConstantPoolCacheEntry::f1_offset(), result);
2104     __ tst(result);
2105     __ br(Assembler::notEqual, false, Assembler::pt, resolved);
2106     __ delayed()->set((int)bytecode(), O1);
2107   } else {
2108     assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2109     assert(result == noreg, "");  //else change code for setting result
2110     __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, Lbyte_code, byte_no, 1, index_size);
2111     __ cmp(Lbyte_code, (int) bytecode());  // have we resolved this bytecode?
2112     __ br(Assembler::equal, false, Assembler::pt, resolved);
2113     __ delayed()->set((int)bytecode(), O1);
2114   }
2115 
2116   address entry;
2117   switch (bytecode()) {
2118     case Bytecodes::_getstatic      : // fall through
2119     case Bytecodes::_putstatic      : // fall through
2120     case Bytecodes::_getfield       : // fall through
2121     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2122     case Bytecodes::_invokevirtual  : // fall through
2123     case Bytecodes::_invokespecial  : // fall through
2124     case Bytecodes::_invokestatic   : // fall through
2125     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
2126     case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
2127     case Bytecodes::_fast_aldc      : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
2128     case Bytecodes::_fast_aldc_w    : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
2129     default                         : ShouldNotReachHere();                                 break;
2130   }
2131   // first time invocation - must resolve first
2132   __ call_VM(noreg, entry, O1);
2133   // Update registers with resolved info
2134   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2135   if (result != noreg)
2136     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
2137               ConstantPoolCacheEntry::f1_offset(), result);
2138   __ bind(resolved);
2139 }
2140 
2141 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2142                                                Register Rmethod,
2143                                                Register Ritable_index,
2144                                                Register Rflags,
2145                                                bool is_invokevirtual,
2146                                                bool is_invokevfinal,
2147                                                bool is_invokedynamic) {
2148   // Uses both G3_scratch and G4_scratch
2149   Register Rcache = G3_scratch;
2150   Register Rscratch = G4_scratch;
2151   assert_different_registers(Rcache, Rmethod, Ritable_index);
2152 
2153   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2154 
2155   // determine constant pool cache field offsets
2156   const int method_offset = in_bytes(
2157     cp_base_offset +
2158       (is_invokevirtual
2159        ? ConstantPoolCacheEntry::f2_offset()
2160        : ConstantPoolCacheEntry::f1_offset()
2161       )
2162     );
2163   const int flags_offset = in_bytes(cp_base_offset +
2164                                     ConstantPoolCacheEntry::flags_offset());
2165   // access constant pool cache fields
2166   const int index_offset = in_bytes(cp_base_offset +
2167                                     ConstantPoolCacheEntry::f2_offset());
2168 
2169   if (is_invokevfinal) {
2170     __ get_cache_and_index_at_bcp(Rcache, Rscratch, 1);
2171     __ ld_ptr(Rcache, method_offset, Rmethod);
2172   } else if (byte_no == f1_oop) {
2173     // Resolved f1_oop goes directly into 'method' register.
2174     resolve_cache_and_index(byte_no, Rmethod, Rcache, Rscratch, sizeof(u4));
2175   } else {
2176     resolve_cache_and_index(byte_no, noreg, Rcache, Rscratch, sizeof(u2));
2177     __ ld_ptr(Rcache, method_offset, Rmethod);
2178   }
2179 
2180   if (Ritable_index != noreg) {
2181     __ ld_ptr(Rcache, index_offset, Ritable_index);
2182   }
2183   __ ld_ptr(Rcache, flags_offset, Rflags);
2184 }
2185 
2186 // The Rcache register must be set before call
2187 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2188                                               Register Rcache,
2189                                               Register index,
2190                                               Register Roffset,
2191                                               Register Rflags,
2192                                               bool is_static) {
2193   assert_different_registers(Rcache, Rflags, Roffset);
2194 
2195   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2196 
2197   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2198   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2199   if (is_static) {
2200     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
2201   }
2202 }
2203 
2204 // The registers Rcache and index expected to be set before call.
2205 // Correct values of the Rcache and index registers are preserved.
2206 void TemplateTable::jvmti_post_field_access(Register Rcache,
2207                                             Register index,
2208                                             bool is_static,
2209                                             bool has_tos) {
2210   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2211 
2212   if (JvmtiExport::can_post_field_access()) {
2213     // Check to see if a field access watch has been set before we take
2214     // the time to call into the VM.
2215     Label Label1;
2216     assert_different_registers(Rcache, index, G1_scratch);
2217     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
2218     __ load_contents(get_field_access_count_addr, G1_scratch);
2219     __ cmp_and_br_short(G1_scratch, 0, Assembler::equal, Assembler::pt, Label1);
2220 
2221     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
2222 
2223     if (is_static) {
2224       __ clr(Otos_i);
2225     } else {
2226       if (has_tos) {
2227       // save object pointer before call_VM() clobbers it
2228         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
2229       } else {
2230         // Load top of stack (do not pop the value off the stack);
2231         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
2232       }
2233       __ verify_oop(Otos_i);
2234     }
2235     // Otos_i: object pointer or NULL if static
2236     // Rcache: cache entry pointer
2237     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2238                Otos_i, Rcache);
2239     if (!is_static && has_tos) {
2240       __ pop_ptr(Otos_i);  // restore object pointer
2241       __ verify_oop(Otos_i);
2242     }
2243     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2244     __ bind(Label1);
2245   }
2246 }
2247 
2248 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2249   transition(vtos, vtos);
2250 
2251   Register Rcache = G3_scratch;
2252   Register index  = G4_scratch;
2253   Register Rclass = Rcache;
2254   Register Roffset= G4_scratch;
2255   Register Rflags = G1_scratch;
2256   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2257 
2258   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
2259   jvmti_post_field_access(Rcache, index, is_static, false);
2260   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2261 
2262   if (!is_static) {
2263     pop_and_check_object(Rclass);
2264   } else {
2265     __ verify_oop(Rclass);
2266   }
2267 
2268   Label exit;
2269 
2270   Assembler::Membar_mask_bits membar_bits =
2271     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2272 
2273   if (__ membar_has_effect(membar_bits)) {
2274     // Get volatile flag
2275     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
2276     __ and3(Rflags, Lscratch, Lscratch);
2277   }
2278 
2279   Label checkVolatile;
2280 
2281   // compute field type
2282   Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
2283   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
2284   // Make sure we don't need to mask Rflags for tosBits after the above shift
2285   ConstantPoolCacheEntry::verify_tosBits();
2286 
2287   // Check atos before itos for getstatic, more likely (in Queens at least)
2288   __ cmp(Rflags, atos);
2289   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2290   __ delayed() ->cmp(Rflags, itos);
2291 
2292   // atos
2293   __ load_heap_oop(Rclass, Roffset, Otos_i);
2294   __ verify_oop(Otos_i);
2295   __ push(atos);
2296   if (!is_static) {
2297     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
2298   }
2299   __ ba(checkVolatile);
2300   __ delayed()->tst(Lscratch);
2301 
2302   __ bind(notObj);
2303 
2304   // cmp(Rflags, itos);
2305   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2306   __ delayed() ->cmp(Rflags, ltos);
2307 
2308   // itos
2309   __ ld(Rclass, Roffset, Otos_i);
2310   __ push(itos);
2311   if (!is_static) {
2312     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
2313   }
2314   __ ba(checkVolatile);
2315   __ delayed()->tst(Lscratch);
2316 
2317   __ bind(notInt);
2318 
2319   // cmp(Rflags, ltos);
2320   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2321   __ delayed() ->cmp(Rflags, btos);
2322 
2323   // ltos
2324   // load must be atomic
2325   __ ld_long(Rclass, Roffset, Otos_l);
2326   __ push(ltos);
2327   if (!is_static) {
2328     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
2329   }
2330   __ ba(checkVolatile);
2331   __ delayed()->tst(Lscratch);
2332 
2333   __ bind(notLong);
2334 
2335   // cmp(Rflags, btos);
2336   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2337   __ delayed() ->cmp(Rflags, ctos);
2338 
2339   // btos
2340   __ ldsb(Rclass, Roffset, Otos_i);
2341   __ push(itos);
2342   if (!is_static) {
2343     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
2344   }
2345   __ ba(checkVolatile);
2346   __ delayed()->tst(Lscratch);
2347 
2348   __ bind(notByte);
2349 
2350   // cmp(Rflags, ctos);
2351   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2352   __ delayed() ->cmp(Rflags, stos);
2353 
2354   // ctos
2355   __ lduh(Rclass, Roffset, Otos_i);
2356   __ push(itos);
2357   if (!is_static) {
2358     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
2359   }
2360   __ ba(checkVolatile);
2361   __ delayed()->tst(Lscratch);
2362 
2363   __ bind(notChar);
2364 
2365   // cmp(Rflags, stos);
2366   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2367   __ delayed() ->cmp(Rflags, ftos);
2368 
2369   // stos
2370   __ ldsh(Rclass, Roffset, Otos_i);
2371   __ push(itos);
2372   if (!is_static) {
2373     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
2374   }
2375   __ ba(checkVolatile);
2376   __ delayed()->tst(Lscratch);
2377 
2378   __ bind(notShort);
2379 
2380 
2381   // cmp(Rflags, ftos);
2382   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
2383   __ delayed() ->tst(Lscratch);
2384 
2385   // ftos
2386   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
2387   __ push(ftos);
2388   if (!is_static) {
2389     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
2390   }
2391   __ ba(checkVolatile);
2392   __ delayed()->tst(Lscratch);
2393 
2394   __ bind(notFloat);
2395 
2396 
2397   // dtos
2398   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
2399   __ push(dtos);
2400   if (!is_static) {
2401     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
2402   }
2403 
2404   __ bind(checkVolatile);
2405   if (__ membar_has_effect(membar_bits)) {
2406     // __ tst(Lscratch); executed in delay slot
2407     __ br(Assembler::zero, false, Assembler::pt, exit);
2408     __ delayed()->nop();
2409     volatile_barrier(membar_bits);
2410   }
2411 
2412   __ bind(exit);
2413 }
2414 
2415 
2416 void TemplateTable::getfield(int byte_no) {
2417   getfield_or_static(byte_no, false);
2418 }
2419 
2420 void TemplateTable::getstatic(int byte_no) {
2421   getfield_or_static(byte_no, true);
2422 }
2423 
2424 
2425 void TemplateTable::fast_accessfield(TosState state) {
2426   transition(atos, state);
2427   Register Rcache  = G3_scratch;
2428   Register index   = G4_scratch;
2429   Register Roffset = G4_scratch;
2430   Register Rflags  = Rcache;
2431   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2432 
2433   __ get_cache_and_index_at_bcp(Rcache, index, 1);
2434   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
2435 
2436   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2437 
2438   __ null_check(Otos_i);
2439   __ verify_oop(Otos_i);
2440 
2441   Label exit;
2442 
2443   Assembler::Membar_mask_bits membar_bits =
2444     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2445   if (__ membar_has_effect(membar_bits)) {
2446     // Get volatile flag
2447     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
2448     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
2449   }
2450 
2451   switch (bytecode()) {
2452     case Bytecodes::_fast_bgetfield:
2453       __ ldsb(Otos_i, Roffset, Otos_i);
2454       break;
2455     case Bytecodes::_fast_cgetfield:
2456       __ lduh(Otos_i, Roffset, Otos_i);
2457       break;
2458     case Bytecodes::_fast_sgetfield:
2459       __ ldsh(Otos_i, Roffset, Otos_i);
2460       break;
2461     case Bytecodes::_fast_igetfield:
2462       __ ld(Otos_i, Roffset, Otos_i);
2463       break;
2464     case Bytecodes::_fast_lgetfield:
2465       __ ld_long(Otos_i, Roffset, Otos_l);
2466       break;
2467     case Bytecodes::_fast_fgetfield:
2468       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
2469       break;
2470     case Bytecodes::_fast_dgetfield:
2471       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
2472       break;
2473     case Bytecodes::_fast_agetfield:
2474       __ load_heap_oop(Otos_i, Roffset, Otos_i);
2475       break;
2476     default:
2477       ShouldNotReachHere();
2478   }
2479 
2480   if (__ membar_has_effect(membar_bits)) {
2481     __ btst(Lscratch, Rflags);
2482     __ br(Assembler::zero, false, Assembler::pt, exit);
2483     __ delayed()->nop();
2484     volatile_barrier(membar_bits);
2485     __ bind(exit);
2486   }
2487 
2488   if (state == atos) {
2489     __ verify_oop(Otos_i);    // does not blow flags!
2490   }
2491 }
2492 
2493 void TemplateTable::jvmti_post_fast_field_mod() {
2494   if (JvmtiExport::can_post_field_modification()) {
2495     // Check to see if a field modification watch has been set before we take
2496     // the time to call into the VM.
2497     Label done;
2498     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2499     __ load_contents(get_field_modification_count_addr, G4_scratch);
2500     __ cmp_and_br_short(G4_scratch, 0, Assembler::equal, Assembler::pt, done);
2501     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
2502     __ verify_oop(G4_scratch);
2503     __ push_ptr(G4_scratch);    // put the object pointer back on tos
2504     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
2505     // Save tos values before call_VM() clobbers them. Since we have
2506     // to do it for every data type, we use the saved values as the
2507     // jvalue object.
2508     switch (bytecode()) {  // save tos values before call_VM() clobbers them
2509     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
2510     case Bytecodes::_fast_bputfield: // fall through
2511     case Bytecodes::_fast_sputfield: // fall through
2512     case Bytecodes::_fast_cputfield: // fall through
2513     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
2514     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
2515     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
2516     // get words in right order for use as jvalue object
2517     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
2518     }
2519     // setup pointer to jvalue object
2520     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
2521     // G4_scratch:  object pointer
2522     // G1_scratch: cache entry pointer
2523     // G3_scratch: jvalue object on the stack
2524     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
2525     switch (bytecode()) {             // restore tos values
2526     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
2527     case Bytecodes::_fast_bputfield: // fall through
2528     case Bytecodes::_fast_sputfield: // fall through
2529     case Bytecodes::_fast_cputfield: // fall through
2530     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
2531     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
2532     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
2533     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
2534     }
2535     __ bind(done);
2536   }
2537 }
2538 
2539 // The registers Rcache and index expected to be set before call.
2540 // The function may destroy various registers, just not the Rcache and index registers.
2541 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
2542   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2543 
2544   if (JvmtiExport::can_post_field_modification()) {
2545     // Check to see if a field modification watch has been set before we take
2546     // the time to call into the VM.
2547     Label Label1;
2548     assert_different_registers(Rcache, index, G1_scratch);
2549     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2550     __ load_contents(get_field_modification_count_addr, G1_scratch);
2551     __ cmp_and_br_short(G1_scratch, 0, Assembler::zero, Assembler::pt, Label1);
2552 
2553     // The Rcache and index registers have been already set.
2554     // This allows to eliminate this call but the Rcache and index
2555     // registers must be correspondingly used after this line.
2556     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
2557 
2558     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
2559     if (is_static) {
2560       // Life is simple.  Null out the object pointer.
2561       __ clr(G4_scratch);
2562     } else {
2563       Register Rflags = G1_scratch;
2564       // Life is harder. The stack holds the value on top, followed by the
2565       // object.  We don't know the size of the value, though; it could be
2566       // one or two words depending on its type. As a result, we must find
2567       // the type to determine where the object is.
2568 
2569       Label two_word, valsizeknown;
2570       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2571       __ mov(Lesp, G4_scratch);
2572       __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
2573       // Make sure we don't need to mask Rflags for tosBits after the above shift
2574       ConstantPoolCacheEntry::verify_tosBits();
2575       __ cmp(Rflags, ltos);
2576       __ br(Assembler::equal, false, Assembler::pt, two_word);
2577       __ delayed()->cmp(Rflags, dtos);
2578       __ br(Assembler::equal, false, Assembler::pt, two_word);
2579       __ delayed()->nop();
2580       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
2581       __ ba_short(valsizeknown);
2582       __ bind(two_word);
2583 
2584       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
2585 
2586       __ bind(valsizeknown);
2587       // setup object pointer
2588       __ ld_ptr(G4_scratch, 0, G4_scratch);
2589       __ verify_oop(G4_scratch);
2590     }
2591     // setup pointer to jvalue object
2592     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
2593     // G4_scratch:  object pointer or NULL if static
2594     // G3_scratch: cache entry pointer
2595     // G1_scratch: jvalue object on the stack
2596     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2597                G4_scratch, G3_scratch, G1_scratch);
2598     __ get_cache_and_index_at_bcp(Rcache, index, 1);
2599     __ bind(Label1);
2600   }
2601 }
2602 
2603 void TemplateTable::pop_and_check_object(Register r) {
2604   __ pop_ptr(r);
2605   __ null_check(r);  // for field access must check obj.
2606   __ verify_oop(r);
2607 }
2608 
2609 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2610   transition(vtos, vtos);
2611   Register Rcache = G3_scratch;
2612   Register index  = G4_scratch;
2613   Register Rclass = Rcache;
2614   Register Roffset= G4_scratch;
2615   Register Rflags = G1_scratch;
2616   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2617 
2618   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
2619   jvmti_post_field_mod(Rcache, index, is_static);
2620   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2621 
2622   Assembler::Membar_mask_bits read_bits =
2623     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2624   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2625 
2626   Label notVolatile, checkVolatile, exit;
2627   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2628     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
2629     __ and3(Rflags, Lscratch, Lscratch);
2630 
2631     if (__ membar_has_effect(read_bits)) {
2632       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2633       volatile_barrier(read_bits);
2634       __ bind(notVolatile);
2635     }
2636   }
2637 
2638   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
2639   // Make sure we don't need to mask Rflags for tosBits after the above shift
2640   ConstantPoolCacheEntry::verify_tosBits();
2641 
2642   // compute field type
2643   Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
2644 
2645   if (is_static) {
2646     // putstatic with object type most likely, check that first
2647     __ cmp(Rflags, atos);
2648     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2649     __ delayed()->cmp(Rflags, itos);
2650 
2651     // atos
2652     {
2653       __ pop_ptr();
2654       __ verify_oop(Otos_i);
2655       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2656       __ ba(checkVolatile);
2657       __ delayed()->tst(Lscratch);
2658     }
2659 
2660     __ bind(notObj);
2661     // cmp(Rflags, itos);
2662     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2663     __ delayed()->cmp(Rflags, btos);
2664 
2665     // itos
2666     {
2667       __ pop_i();
2668       __ st(Otos_i, Rclass, Roffset);
2669       __ ba(checkVolatile);
2670       __ delayed()->tst(Lscratch);
2671     }
2672 
2673     __ bind(notInt);
2674   } else {
2675     // putfield with int type most likely, check that first
2676     __ cmp(Rflags, itos);
2677     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2678     __ delayed()->cmp(Rflags, atos);
2679 
2680     // itos
2681     {
2682       __ pop_i();
2683       pop_and_check_object(Rclass);
2684       __ st(Otos_i, Rclass, Roffset);
2685       patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch, true, byte_no);
2686       __ ba(checkVolatile);
2687       __ delayed()->tst(Lscratch);
2688     }
2689 
2690     __ bind(notInt);
2691     // cmp(Rflags, atos);
2692     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2693     __ delayed()->cmp(Rflags, btos);
2694 
2695     // atos
2696     {
2697       __ pop_ptr();
2698       pop_and_check_object(Rclass);
2699       __ verify_oop(Otos_i);
2700       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2701       patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch, true, byte_no);
2702       __ ba(checkVolatile);
2703       __ delayed()->tst(Lscratch);
2704     }
2705 
2706     __ bind(notObj);
2707   }
2708 
2709   // cmp(Rflags, btos);
2710   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2711   __ delayed()->cmp(Rflags, ltos);
2712 
2713   // btos
2714   {
2715     __ pop_i();
2716     if (!is_static) pop_and_check_object(Rclass);
2717     __ stb(Otos_i, Rclass, Roffset);
2718     if (!is_static) {
2719       patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch, true, byte_no);
2720     }
2721     __ ba(checkVolatile);
2722     __ delayed()->tst(Lscratch);
2723   }
2724 
2725   __ bind(notByte);
2726   // cmp(Rflags, ltos);
2727   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2728   __ delayed()->cmp(Rflags, ctos);
2729 
2730   // ltos
2731   {
2732     __ pop_l();
2733     if (!is_static) pop_and_check_object(Rclass);
2734     __ st_long(Otos_l, Rclass, Roffset);
2735     if (!is_static) {
2736       patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch, true, byte_no);
2737     }
2738     __ ba(checkVolatile);
2739     __ delayed()->tst(Lscratch);
2740   }
2741 
2742   __ bind(notLong);
2743   // cmp(Rflags, ctos);
2744   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2745   __ delayed()->cmp(Rflags, stos);
2746 
2747   // ctos (char)
2748   {
2749     __ pop_i();
2750     if (!is_static) pop_and_check_object(Rclass);
2751     __ sth(Otos_i, Rclass, Roffset);
2752     if (!is_static) {
2753       patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch, true, byte_no);
2754     }
2755     __ ba(checkVolatile);
2756     __ delayed()->tst(Lscratch);
2757   }
2758 
2759   __ bind(notChar);
2760   // cmp(Rflags, stos);
2761   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2762   __ delayed()->cmp(Rflags, ftos);
2763 
2764   // stos (short)
2765   {
2766     __ pop_i();
2767     if (!is_static) pop_and_check_object(Rclass);
2768     __ sth(Otos_i, Rclass, Roffset);
2769     if (!is_static) {
2770       patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch, true, byte_no);
2771     }
2772     __ ba(checkVolatile);
2773     __ delayed()->tst(Lscratch);
2774   }
2775 
2776   __ bind(notShort);
2777   // cmp(Rflags, ftos);
2778   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
2779   __ delayed()->nop();
2780 
2781   // ftos
2782   {
2783     __ pop_f();
2784     if (!is_static) pop_and_check_object(Rclass);
2785     __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2786     if (!is_static) {
2787       patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch, true, byte_no);
2788     }
2789     __ ba(checkVolatile);
2790     __ delayed()->tst(Lscratch);
2791   }
2792 
2793   __ bind(notFloat);
2794 
2795   // dtos
2796   {
2797     __ pop_d();
2798     if (!is_static) pop_and_check_object(Rclass);
2799     __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2800     if (!is_static) {
2801       patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch, true, byte_no);
2802     }
2803   }
2804 
2805   __ bind(checkVolatile);
2806   __ tst(Lscratch);
2807 
2808   if (__ membar_has_effect(write_bits)) {
2809     // __ tst(Lscratch); in delay slot
2810     __ br(Assembler::zero, false, Assembler::pt, exit);
2811     __ delayed()->nop();
2812     volatile_barrier(Assembler::StoreLoad);
2813     __ bind(exit);
2814   }
2815 }
2816 
2817 void TemplateTable::fast_storefield(TosState state) {
2818   transition(state, vtos);
2819   Register Rcache = G3_scratch;
2820   Register Rclass = Rcache;
2821   Register Roffset= G4_scratch;
2822   Register Rflags = G1_scratch;
2823   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2824 
2825   jvmti_post_fast_field_mod();
2826 
2827   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
2828 
2829   Assembler::Membar_mask_bits read_bits =
2830     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2831   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2832 
2833   Label notVolatile, checkVolatile, exit;
2834   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2835     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2836     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
2837     __ and3(Rflags, Lscratch, Lscratch);
2838     if (__ membar_has_effect(read_bits)) {
2839       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2840       volatile_barrier(read_bits);
2841       __ bind(notVolatile);
2842     }
2843   }
2844 
2845   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2846   pop_and_check_object(Rclass);
2847 
2848   switch (bytecode()) {
2849     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
2850     case Bytecodes::_fast_cputfield: /* fall through */
2851     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
2852     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
2853     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
2854     case Bytecodes::_fast_fputfield:
2855       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2856       break;
2857     case Bytecodes::_fast_dputfield:
2858       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2859       break;
2860     case Bytecodes::_fast_aputfield:
2861       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2862       break;
2863     default:
2864       ShouldNotReachHere();
2865   }
2866 
2867   if (__ membar_has_effect(write_bits)) {
2868     __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, exit);
2869     volatile_barrier(Assembler::StoreLoad);
2870     __ bind(exit);
2871   }
2872 }
2873 
2874 
2875 void TemplateTable::putfield(int byte_no) {
2876   putfield_or_static(byte_no, false);
2877 }
2878 
2879 void TemplateTable::putstatic(int byte_no) {
2880   putfield_or_static(byte_no, true);
2881 }
2882 
2883 
2884 void TemplateTable::fast_xaccess(TosState state) {
2885   transition(vtos, state);
2886   Register Rcache = G3_scratch;
2887   Register Roffset = G4_scratch;
2888   Register Rflags  = G4_scratch;
2889   Register Rreceiver = Lscratch;
2890 
2891   __ ld_ptr(Llocals, 0, Rreceiver);
2892 
2893   // access constant pool cache  (is resolved)
2894   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
2895   __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
2896   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
2897 
2898   __ verify_oop(Rreceiver);
2899   __ null_check(Rreceiver);
2900   if (state == atos) {
2901     __ load_heap_oop(Rreceiver, Roffset, Otos_i);
2902   } else if (state == itos) {
2903     __ ld (Rreceiver, Roffset, Otos_i) ;
2904   } else if (state == ftos) {
2905     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
2906   } else {
2907     ShouldNotReachHere();
2908   }
2909 
2910   Assembler::Membar_mask_bits membar_bits =
2911     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2912   if (__ membar_has_effect(membar_bits)) {
2913 
2914     // Get is_volatile value in Rflags and check if membar is needed
2915     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
2916 
2917     // Test volatile
2918     Label notVolatile;
2919     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
2920     __ btst(Rflags, Lscratch);
2921     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
2922     __ delayed()->nop();
2923     volatile_barrier(membar_bits);
2924     __ bind(notVolatile);
2925   }
2926 
2927   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
2928   __ sub(Lbcp, 1, Lbcp);
2929 }
2930 
2931 //----------------------------------------------------------------------------------------------------
2932 // Calls
2933 
2934 void TemplateTable::count_calls(Register method, Register temp) {
2935   // implemented elsewhere
2936   ShouldNotReachHere();
2937 }
2938 
2939 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
2940   Register Rtemp = G4_scratch;
2941   Register Rcall = Rindex;
2942   assert_different_registers(Rcall, G5_method, Gargs, Rret);
2943 
2944   // get target methodOop & entry point
2945   const int base = instanceKlass::vtable_start_offset() * wordSize;
2946   if (vtableEntry::size() % 3 == 0) {
2947     // scale the vtable index by 12:
2948     int one_third = vtableEntry::size() / 3;
2949     __ sll(Rindex, exact_log2(one_third * 1 * wordSize), Rtemp);
2950     __ sll(Rindex, exact_log2(one_third * 2 * wordSize), Rindex);
2951     __ add(Rindex, Rtemp, Rindex);
2952   } else {
2953     // scale the vtable index by 8:
2954     __ sll(Rindex, exact_log2(vtableEntry::size() * wordSize), Rindex);
2955   }
2956 
2957   __ add(Rrecv, Rindex, Rrecv);
2958   __ ld_ptr(Rrecv, base + vtableEntry::method_offset_in_bytes(), G5_method);
2959 
2960   __ call_from_interpreter(Rcall, Gargs, Rret);
2961 }
2962 
2963 void TemplateTable::invokevirtual(int byte_no) {
2964   transition(vtos, vtos);
2965   assert(byte_no == f2_byte, "use this argument");
2966 
2967   Register Rscratch = G3_scratch;
2968   Register Rtemp = G4_scratch;
2969   Register Rret = Lscratch;
2970   Register Rrecv = G5_method;
2971   Label notFinal;
2972 
2973   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
2974   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
2975 
2976   // Check for vfinal
2977   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), G4_scratch);
2978   __ btst(Rret, G4_scratch);
2979   __ br(Assembler::zero, false, Assembler::pt, notFinal);
2980   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
2981 
2982   patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
2983 
2984   invokevfinal_helper(Rscratch, Rret);
2985 
2986   __ bind(notFinal);
2987 
2988   __ mov(G5_method, Rscratch);  // better scratch register
2989   __ load_receiver(G4_scratch, O0);  // gets receiverOop
2990   // receiver is in O0
2991   __ verify_oop(O0);
2992 
2993   // get return address
2994   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
2995   __ set(table, Rtemp);
2996   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
2997   // Make sure we don't need to mask Rret for tosBits after the above shift
2998   ConstantPoolCacheEntry::verify_tosBits();
2999   __ sll(Rret,  LogBytesPerWord, Rret);
3000   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3001 
3002   // get receiver klass
3003   __ null_check(O0, oopDesc::klass_offset_in_bytes());
3004   __ load_klass(O0, Rrecv);
3005   __ verify_oop(Rrecv);
3006 
3007   __ profile_virtual_call(Rrecv, O4);
3008 
3009   generate_vtable_call(Rrecv, Rscratch, Rret);
3010 }
3011 
3012 void TemplateTable::fast_invokevfinal(int byte_no) {
3013   transition(vtos, vtos);
3014   assert(byte_no == f2_byte, "use this argument");
3015 
3016   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
3017                              /*is_invokevfinal*/true, false);
3018   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3019   invokevfinal_helper(G3_scratch, Lscratch);
3020 }
3021 
3022 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
3023   Register Rtemp = G4_scratch;
3024 
3025   __ verify_oop(G5_method);
3026 
3027   // Load receiver from stack slot
3028   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
3029   __ load_receiver(G4_scratch, O0);
3030 
3031   // receiver NULL check
3032   __ null_check(O0);
3033 
3034   __ profile_final_call(O4);
3035 
3036   // get return address
3037   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
3038   __ set(table, Rtemp);
3039   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
3040   // Make sure we don't need to mask Rret for tosBits after the above shift
3041   ConstantPoolCacheEntry::verify_tosBits();
3042   __ sll(Rret,  LogBytesPerWord, Rret);
3043   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3044 
3045 
3046   // do the call
3047   __ call_from_interpreter(Rscratch, Gargs, Rret);
3048 }
3049 
3050 void TemplateTable::invokespecial(int byte_no) {
3051   transition(vtos, vtos);
3052   assert(byte_no == f1_byte, "use this argument");
3053 
3054   Register Rscratch = G3_scratch;
3055   Register Rtemp = G4_scratch;
3056   Register Rret = Lscratch;
3057 
3058   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
3059   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3060 
3061   __ verify_oop(G5_method);
3062 
3063   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
3064   __ load_receiver(G4_scratch, O0);
3065 
3066   // receiver NULL check
3067   __ null_check(O0);
3068 
3069   __ profile_call(O4);
3070 
3071   // get return address
3072   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
3073   __ set(table, Rtemp);
3074   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
3075   // Make sure we don't need to mask Rret for tosBits after the above shift
3076   ConstantPoolCacheEntry::verify_tosBits();
3077   __ sll(Rret,  LogBytesPerWord, Rret);
3078   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3079 
3080   // do the call
3081   __ call_from_interpreter(Rscratch, Gargs, Rret);
3082 }
3083 
3084 void TemplateTable::invokestatic(int byte_no) {
3085   transition(vtos, vtos);
3086   assert(byte_no == f1_byte, "use this argument");
3087 
3088   Register Rscratch = G3_scratch;
3089   Register Rtemp = G4_scratch;
3090   Register Rret = Lscratch;
3091 
3092   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
3093   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3094 
3095   __ verify_oop(G5_method);
3096 
3097   __ profile_call(O4);
3098 
3099   // get return address
3100   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
3101   __ set(table, Rtemp);
3102   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
3103   // Make sure we don't need to mask Rret for tosBits after the above shift
3104   ConstantPoolCacheEntry::verify_tosBits();
3105   __ sll(Rret,  LogBytesPerWord, Rret);
3106   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3107 
3108   // do the call
3109   __ call_from_interpreter(Rscratch, Gargs, Rret);
3110 }
3111 
3112 
3113 void TemplateTable::invokeinterface_object_method(Register RklassOop,
3114                                                   Register Rcall,
3115                                                   Register Rret,
3116                                                   Register Rflags) {
3117   Register Rscratch = G4_scratch;
3118   Register Rindex = Lscratch;
3119 
3120   assert_different_registers(Rscratch, Rindex, Rret);
3121 
3122   Label notFinal;
3123 
3124   // Check for vfinal
3125   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), Rscratch);
3126   __ btst(Rflags, Rscratch);
3127   __ br(Assembler::zero, false, Assembler::pt, notFinal);
3128   __ delayed()->nop();
3129 
3130   __ profile_final_call(O4);
3131 
3132   // do the call - the index (f2) contains the methodOop
3133   assert_different_registers(G5_method, Gargs, Rcall);
3134   __ mov(Rindex, G5_method);
3135   __ call_from_interpreter(Rcall, Gargs, Rret);
3136   __ bind(notFinal);
3137 
3138   __ profile_virtual_call(RklassOop, O4);
3139   generate_vtable_call(RklassOop, Rindex, Rret);
3140 }
3141 
3142 
3143 void TemplateTable::invokeinterface(int byte_no) {
3144   transition(vtos, vtos);
3145   assert(byte_no == f1_byte, "use this argument");
3146 
3147   Register Rscratch = G4_scratch;
3148   Register Rret = G3_scratch;
3149   Register Rindex = Lscratch;
3150   Register Rinterface = G1_scratch;
3151   Register RklassOop = G5_method;
3152   Register Rflags = O1;
3153   assert_different_registers(Rscratch, G5_method);
3154 
3155   load_invoke_cp_cache_entry(byte_no, Rinterface, Rindex, Rflags, /*virtual*/ false, false, false);
3156   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3157 
3158   // get receiver
3159   __ and3(Rflags, 0xFF, Rscratch);       // gets number of parameters
3160   __ load_receiver(Rscratch, O0);
3161   __ verify_oop(O0);
3162 
3163   __ mov(Rflags, Rret);
3164 
3165   // get return address
3166   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
3167   __ set(table, Rscratch);
3168   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
3169   // Make sure we don't need to mask Rret for tosBits after the above shift
3170   ConstantPoolCacheEntry::verify_tosBits();
3171   __ sll(Rret,  LogBytesPerWord, Rret);
3172   __ ld_ptr(Rscratch, Rret, Rret);      // get return address
3173 
3174   // get receiver klass
3175   __ null_check(O0, oopDesc::klass_offset_in_bytes());
3176   __ load_klass(O0, RklassOop);
3177   __ verify_oop(RklassOop);
3178 
3179   // Special case of invokeinterface called for virtual method of
3180   // java.lang.Object.  See cpCacheOop.cpp for details.
3181   // This code isn't produced by javac, but could be produced by
3182   // another compliant java compiler.
3183   Label notMethod;
3184   __ set((1 << ConstantPoolCacheEntry::methodInterface), Rscratch);
3185   __ btst(Rflags, Rscratch);
3186   __ br(Assembler::zero, false, Assembler::pt, notMethod);
3187   __ delayed()->nop();
3188 
3189   invokeinterface_object_method(RklassOop, Rinterface, Rret, Rflags);
3190 
3191   __ bind(notMethod);
3192 
3193   __ profile_virtual_call(RklassOop, O4);
3194 
3195   //
3196   // find entry point to call
3197   //
3198 
3199   // compute start of first itableOffsetEntry (which is at end of vtable)
3200   const int base = instanceKlass::vtable_start_offset() * wordSize;
3201   Label search;
3202   Register Rtemp = Rflags;
3203 
3204   __ ld(RklassOop, instanceKlass::vtable_length_offset() * wordSize, Rtemp);
3205   if (align_object_offset(1) > 1) {
3206     __ round_to(Rtemp, align_object_offset(1));
3207   }
3208   __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
3209   if (Assembler::is_simm13(base)) {
3210     __ add(Rtemp, base, Rtemp);
3211   } else {
3212     __ set(base, Rscratch);
3213     __ add(Rscratch, Rtemp, Rtemp);
3214   }
3215   __ add(RklassOop, Rtemp, Rscratch);
3216 
3217   __ bind(search);
3218 
3219   __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
3220   {
3221     Label ok;
3222 
3223     // Check that entry is non-null.  Null entries are probably a bytecode
3224     // problem.  If the interface isn't implemented by the receiver class,
3225     // the VM should throw IncompatibleClassChangeError.  linkResolver checks
3226     // this too but that's only if the entry isn't already resolved, so we
3227     // need to check again.
3228     __ br_notnull_short( Rtemp, Assembler::pt, ok);
3229     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3230     __ should_not_reach_here();
3231     __ bind(ok);
3232     __ verify_oop(Rtemp);
3233   }
3234 
3235   __ verify_oop(Rinterface);
3236 
3237   __ cmp(Rinterface, Rtemp);
3238   __ brx(Assembler::notEqual, true, Assembler::pn, search);
3239   __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
3240 
3241   // entry found and Rscratch points to it
3242   __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
3243 
3244   assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
3245   __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
3246   __ add(Rscratch, Rindex, Rscratch);
3247   __ ld_ptr(RklassOop, Rscratch, G5_method);
3248 
3249   // Check for abstract method error.
3250   {
3251     Label ok;
3252     __ br_notnull_short(G5_method, Assembler::pt, ok);
3253     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3254     __ should_not_reach_here();
3255     __ bind(ok);
3256   }
3257 
3258   Register Rcall = Rinterface;
3259   assert_different_registers(Rcall, G5_method, Gargs, Rret);
3260 
3261   __ verify_oop(G5_method);
3262   __ call_from_interpreter(Rcall, Gargs, Rret);
3263 
3264 }
3265 
3266 
3267 void TemplateTable::invokedynamic(int byte_no) {
3268   transition(vtos, vtos);
3269   assert(byte_no == f1_oop, "use this argument");
3270 
3271   if (!EnableInvokeDynamic) {
3272     // We should not encounter this bytecode if !EnableInvokeDynamic.
3273     // The verifier will stop it.  However, if we get past the verifier,
3274     // this will stop the thread in a reasonable way, without crashing the JVM.
3275     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3276                      InterpreterRuntime::throw_IncompatibleClassChangeError));
3277     // the call_VM checks for exception, so we should never return here.
3278     __ should_not_reach_here();
3279     return;
3280   }
3281 
3282   // G5: CallSite object (f1)
3283   // XX: unused (f2)
3284   // XX: flags (unused)
3285 
3286   Register G5_callsite = G5_method;
3287   Register Rscratch    = G3_scratch;
3288   Register Rtemp       = G1_scratch;
3289   Register Rret        = Lscratch;
3290 
3291   load_invoke_cp_cache_entry(byte_no, G5_callsite, noreg, Rret,
3292                              /*virtual*/ false, /*vfinal*/ false, /*indy*/ true);
3293   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
3294 
3295   // profile this call
3296   __ profile_call(O4);
3297 
3298   // get return address
3299   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
3300   __ set(table, Rtemp);
3301   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);  // get return type
3302   // Make sure we don't need to mask Rret for tosBits after the above shift
3303   ConstantPoolCacheEntry::verify_tosBits();
3304   __ sll(Rret, LogBytesPerWord, Rret);
3305   __ ld_ptr(Rtemp, Rret, Rret);  // get return address
3306 
3307   __ verify_oop(G5_callsite);
3308   __ load_heap_oop(G5_callsite, __ delayed_value(java_lang_invoke_CallSite::target_offset_in_bytes, Rscratch), G3_method_handle);
3309   __ null_check(G3_method_handle);
3310   __ verify_oop(G3_method_handle);
3311 
3312   // Adjust Rret first so Llast_SP can be same as Rret
3313   __ add(Rret, -frame::pc_return_offset, O7);
3314   __ add(Lesp, BytesPerWord, Gargs);  // setup parameter pointer
3315   __ jump_to_method_handle_entry(G3_method_handle, Rtemp, /* emit_delayed_nop */ false);
3316   // Record SP so we can remove any stack space allocated by adapter transition
3317   __ delayed()->mov(SP, Llast_SP);
3318 }
3319 
3320 
3321 //----------------------------------------------------------------------------------------------------
3322 // Allocation
3323 
3324 void TemplateTable::_new() {
3325   transition(vtos, atos);
3326 
3327   Label slow_case;
3328   Label done;
3329   Label initialize_header;
3330   Label initialize_object;  // including clearing the fields
3331 
3332   Register RallocatedObject = Otos_i;
3333   Register RinstanceKlass = O1;
3334   Register Roffset = O3;
3335   Register Rscratch = O4;
3336 
3337   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3338   __ get_cpool_and_tags(Rscratch, G3_scratch);
3339   // make sure the class we're about to instantiate has been resolved
3340   // This is done before loading instanceKlass to be consistent with the order
3341   // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
3342   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
3343   __ ldub(G3_scratch, Roffset, G3_scratch);
3344   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3345   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3346   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3347   // get instanceKlass
3348   //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
3349   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
3350   __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
3351 
3352   // make sure klass is fully initialized:
3353   __ ld(RinstanceKlass, instanceKlass::init_state_offset_in_bytes(), G3_scratch);
3354   __ cmp(G3_scratch, instanceKlass::fully_initialized);
3355   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3356   __ delayed()->ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes(), Roffset);
3357 
3358   // get instance_size in instanceKlass (already aligned)
3359   //__ ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes(), Roffset);
3360 
3361   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
3362   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
3363   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
3364   __ delayed()->nop();
3365 
3366   // allocate the instance
3367   // 1) Try to allocate in the TLAB
3368   // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
3369   // 3) if the above fails (or is not applicable), go to a slow case
3370   // (creates a new TLAB, etc.)
3371 
3372   const bool allow_shared_alloc =
3373     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3374 
3375   if(UseTLAB) {
3376     Register RoldTopValue = RallocatedObject;
3377     Register RtlabWasteLimitValue = G3_scratch;
3378     Register RnewTopValue = G1_scratch;
3379     Register RendValue = Rscratch;
3380     Register RfreeValue = RnewTopValue;
3381 
3382     // check if we can allocate in the TLAB
3383     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
3384     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
3385     __ add(RoldTopValue, Roffset, RnewTopValue);
3386 
3387     // if there is enough space, we do not CAS and do not clear
3388     __ cmp(RnewTopValue, RendValue);
3389     if(ZeroTLAB) {
3390       // the fields have already been cleared
3391       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
3392     } else {
3393       // initialize both the header and fields
3394       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
3395     }
3396     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
3397 
3398     if (allow_shared_alloc) {
3399       // Check if tlab should be discarded (refill_waste_limit >= free)
3400       __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
3401       __ sub(RendValue, RoldTopValue, RfreeValue);
3402 #ifdef _LP64
3403       __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
3404 #else
3405       __ srl(RfreeValue, LogHeapWordSize, RfreeValue);
3406 #endif
3407       __ cmp_and_brx_short(RtlabWasteLimitValue, RfreeValue, Assembler::greaterEqualUnsigned, Assembler::pt, slow_case); // tlab waste is small
3408 
3409       // increment waste limit to prevent getting stuck on this slow path
3410       __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
3411       __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
3412     } else {
3413       // No allocation in the shared eden.
3414       __ ba_short(slow_case);
3415     }
3416   }
3417 
3418   // Allocation in the shared Eden
3419   if (allow_shared_alloc) {
3420     Register RoldTopValue = G1_scratch;
3421     Register RtopAddr = G3_scratch;
3422     Register RnewTopValue = RallocatedObject;
3423     Register RendValue = Rscratch;
3424 
3425     __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
3426 
3427     Label retry;
3428     __ bind(retry);
3429     __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
3430     __ ld_ptr(RendValue, 0, RendValue);
3431     __ ld_ptr(RtopAddr, 0, RoldTopValue);
3432     __ add(RoldTopValue, Roffset, RnewTopValue);
3433 
3434     // RnewTopValue contains the top address after the new object
3435     // has been allocated.
3436     __ cmp_and_brx_short(RnewTopValue, RendValue, Assembler::greaterUnsigned, Assembler::pn, slow_case);
3437 
3438     __ casx_under_lock(RtopAddr, RoldTopValue, RnewTopValue,
3439       VM_Version::v9_instructions_work() ? NULL :
3440       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
3441 
3442     // if someone beat us on the allocation, try again, otherwise continue
3443     __ cmp_and_brx_short(RoldTopValue, RnewTopValue, Assembler::notEqual, Assembler::pn, retry);
3444 
3445     // bump total bytes allocated by this thread
3446     // RoldTopValue and RtopAddr are dead, so can use G1 and G3
3447     __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
3448   }
3449 
3450   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3451     // clear object fields
3452     __ bind(initialize_object);
3453     __ deccc(Roffset, sizeof(oopDesc));
3454     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
3455     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
3456 
3457     // initialize remaining object fields
3458     if (UseBlockZeroing) {
3459       // Use BIS for zeroing
3460       __ bis_zeroing(G3_scratch, Roffset, G1_scratch, initialize_header);
3461     } else {
3462       Label loop;
3463       __ subcc(Roffset, wordSize, Roffset);
3464       __ bind(loop);
3465       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
3466       __ st_ptr(G0, G3_scratch, Roffset);
3467       __ br(Assembler::notEqual, false, Assembler::pt, loop);
3468       __ delayed()->subcc(Roffset, wordSize, Roffset);
3469     }
3470     __ ba_short(initialize_header);
3471   }
3472 
3473   // slow case
3474   __ bind(slow_case);
3475   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
3476   __ get_constant_pool(O1);
3477 
3478   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
3479 
3480   __ ba_short(done);
3481 
3482   // Initialize the header: mark, klass
3483   __ bind(initialize_header);
3484 
3485   if (UseBiasedLocking) {
3486     __ ld_ptr(RinstanceKlass, Klass::prototype_header_offset_in_bytes(), G4_scratch);
3487   } else {
3488     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
3489   }
3490   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
3491   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
3492   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
3493 
3494   {
3495     SkipIfEqual skip_if(
3496       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
3497     // Trigger dtrace event
3498     __ push(atos);
3499     __ call_VM_leaf(noreg,
3500        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
3501     __ pop(atos);
3502   }
3503 
3504   // continue
3505   __ bind(done);
3506 }
3507 
3508 
3509 
3510 void TemplateTable::newarray() {
3511   transition(itos, atos);
3512   __ ldub(Lbcp, 1, O1);
3513      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
3514 }
3515 
3516 
3517 void TemplateTable::anewarray() {
3518   transition(itos, atos);
3519   __ get_constant_pool(O1);
3520   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
3521      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
3522 }
3523 
3524 
3525 void TemplateTable::arraylength() {
3526   transition(atos, itos);
3527   Label ok;
3528   __ verify_oop(Otos_i);
3529   __ tst(Otos_i);
3530   __ throw_if_not_1_x( Assembler::notZero, ok );
3531   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
3532   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3533 }
3534 
3535 
3536 void TemplateTable::checkcast() {
3537   transition(atos, atos);
3538   Label done, is_null, quicked, cast_ok, resolved;
3539   Register Roffset = G1_scratch;
3540   Register RobjKlass = O5;
3541   Register RspecifiedKlass = O4;
3542 
3543   // Check for casting a NULL
3544   __ br_null_short(Otos_i, Assembler::pn, is_null);
3545 
3546   // Get value klass in RobjKlass
3547   __ load_klass(Otos_i, RobjKlass); // get value klass
3548 
3549   // Get constant pool tag
3550   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3551 
3552   // See if the checkcast has been quickened
3553   __ get_cpool_and_tags(Lscratch, G3_scratch);
3554   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
3555   __ ldub(G3_scratch, Roffset, G3_scratch);
3556   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3557   __ br(Assembler::equal, true, Assembler::pt, quicked);
3558   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3559 
3560   __ push_ptr(); // save receiver for result, and for GC
3561   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3562   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3563 
3564   __ ba_short(resolved);
3565 
3566   // Extract target class from constant pool
3567   __ bind(quicked);
3568   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
3569   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3570   __ bind(resolved);
3571   __ load_klass(Otos_i, RobjKlass); // get value klass
3572 
3573   // Generate a fast subtype check.  Branch to cast_ok if no
3574   // failure.  Throw exception if failure.
3575   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
3576 
3577   // Not a subtype; so must throw exception
3578   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
3579 
3580   __ bind(cast_ok);
3581 
3582   if (ProfileInterpreter) {
3583     __ ba_short(done);
3584   }
3585   __ bind(is_null);
3586   __ profile_null_seen(G3_scratch);
3587   __ bind(done);
3588 }
3589 
3590 
3591 void TemplateTable::instanceof() {
3592   Label done, is_null, quicked, resolved;
3593   transition(atos, itos);
3594   Register Roffset = G1_scratch;
3595   Register RobjKlass = O5;
3596   Register RspecifiedKlass = O4;
3597 
3598   // Check for casting a NULL
3599   __ br_null_short(Otos_i, Assembler::pt, is_null);
3600 
3601   // Get value klass in RobjKlass
3602   __ load_klass(Otos_i, RobjKlass); // get value klass
3603 
3604   // Get constant pool tag
3605   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3606 
3607   // See if the checkcast has been quickened
3608   __ get_cpool_and_tags(Lscratch, G3_scratch);
3609   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
3610   __ ldub(G3_scratch, Roffset, G3_scratch);
3611   __ cmp(G3_scratch, JVM_CONSTANT_Class);
3612   __ br(Assembler::equal, true, Assembler::pt, quicked);
3613   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3614 
3615   __ push_ptr(); // save receiver for result, and for GC
3616   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3617   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3618 
3619   __ ba_short(resolved);
3620 
3621   // Extract target class from constant pool
3622   __ bind(quicked);
3623   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
3624   __ get_constant_pool(Lscratch);
3625   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3626   __ bind(resolved);
3627   __ load_klass(Otos_i, RobjKlass); // get value klass
3628 
3629   // Generate a fast subtype check.  Branch to cast_ok if no
3630   // failure.  Return 0 if failure.
3631   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
3632   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
3633   // Not a subtype; return 0;
3634   __ clr( Otos_i );
3635 
3636   if (ProfileInterpreter) {
3637     __ ba_short(done);
3638   }
3639   __ bind(is_null);
3640   __ profile_null_seen(G3_scratch);
3641   __ bind(done);
3642 }
3643 
3644 void TemplateTable::_breakpoint() {
3645 
3646    // Note: We get here even if we are single stepping..
3647    // jbug inists on setting breakpoints at every bytecode
3648    // even if we are in single step mode.
3649 
3650    transition(vtos, vtos);
3651    // get the unpatched byte code
3652    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
3653    __ mov(O0, Lbyte_code);
3654 
3655    // post the breakpoint event
3656    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
3657 
3658    // complete the execution of original bytecode
3659    __ dispatch_normal(vtos);
3660 }
3661 
3662 
3663 //----------------------------------------------------------------------------------------------------
3664 // Exceptions
3665 
3666 void TemplateTable::athrow() {
3667   transition(atos, vtos);
3668 
3669   // This works because exception is cached in Otos_i which is same as O0,
3670   // which is same as what throw_exception_entry_expects
3671   assert(Otos_i == Oexception, "see explanation above");
3672 
3673   __ verify_oop(Otos_i);
3674   __ null_check(Otos_i);
3675   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
3676 }
3677 
3678 
3679 //----------------------------------------------------------------------------------------------------
3680 // Synchronization
3681 
3682 
3683 // See frame_sparc.hpp for monitor block layout.
3684 // Monitor elements are dynamically allocated by growing stack as needed.
3685 
3686 void TemplateTable::monitorenter() {
3687   transition(atos, vtos);
3688   __ verify_oop(Otos_i);
3689   // Try to acquire a lock on the object
3690   // Repeat until succeeded (i.e., until
3691   // monitorenter returns true).
3692 
3693   {   Label ok;
3694     __ tst(Otos_i);
3695     __ throw_if_not_1_x( Assembler::notZero,  ok);
3696     __ delayed()->mov(Otos_i, Lscratch); // save obj
3697     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3698   }
3699 
3700   assert(O0 == Otos_i, "Be sure where the object to lock is");
3701 
3702   // find a free slot in the monitor block
3703 
3704 
3705   // initialize entry pointer
3706   __ clr(O1); // points to free slot or NULL
3707 
3708   {
3709     Label entry, loop, exit;
3710     __ add( __ top_most_monitor(), O2 ); // last one to check
3711     __ ba( entry );
3712     __ delayed()->mov( Lmonitors, O3 ); // first one to check
3713 
3714 
3715     __ bind( loop );
3716 
3717     __ verify_oop(O4);          // verify each monitor's oop
3718     __ tst(O4); // is this entry unused?
3719     if (VM_Version::v9_instructions_work())
3720       __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
3721     else {
3722       Label L;
3723       __ br( Assembler::zero, true, Assembler::pn, L );
3724       __ delayed()->mov(O3, O1); // rememeber this one if match
3725       __ bind(L);
3726     }
3727 
3728     __ cmp(O4, O0); // check if current entry is for same object
3729     __ brx( Assembler::equal, false, Assembler::pn, exit );
3730     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
3731 
3732     __ bind( entry );
3733 
3734     __ cmp( O3, O2 );
3735     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3736     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
3737 
3738     __ bind( exit );
3739   }
3740 
3741   { Label allocated;
3742 
3743     // found free slot?
3744     __ br_notnull_short(O1, Assembler::pn, allocated);
3745 
3746     __ add_monitor_to_stack( false, O2, O3 );
3747     __ mov(Lmonitors, O1);
3748 
3749     __ bind(allocated);
3750   }
3751 
3752   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3753   // The object has already been poped from the stack, so the expression stack looks correct.
3754   __ inc(Lbcp);
3755 
3756   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
3757   __ lock_object(O1, O0);
3758 
3759   // check if there's enough space on the stack for the monitors after locking
3760   __ generate_stack_overflow_check(0);
3761 
3762   // The bcp has already been incremented. Just need to dispatch to next instruction.
3763   __ dispatch_next(vtos);
3764 }
3765 
3766 
3767 void TemplateTable::monitorexit() {
3768   transition(atos, vtos);
3769   __ verify_oop(Otos_i);
3770   __ tst(Otos_i);
3771   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
3772 
3773   assert(O0 == Otos_i, "just checking");
3774 
3775   { Label entry, loop, found;
3776     __ add( __ top_most_monitor(), O2 ); // last one to check
3777     __ ba(entry);
3778     // use Lscratch to hold monitor elem to check, start with most recent monitor,
3779     // By using a local it survives the call to the C routine.
3780     __ delayed()->mov( Lmonitors, Lscratch );
3781 
3782     __ bind( loop );
3783 
3784     __ verify_oop(O4);          // verify each monitor's oop
3785     __ cmp(O4, O0); // check if current entry is for desired object
3786     __ brx( Assembler::equal, true, Assembler::pt, found );
3787     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
3788 
3789     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
3790 
3791     __ bind( entry );
3792 
3793     __ cmp( Lscratch, O2 );
3794     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3795     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
3796 
3797     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3798     __ should_not_reach_here();
3799 
3800     __ bind(found);
3801   }
3802   __ unlock_object(O1);
3803 }
3804 
3805 
3806 //----------------------------------------------------------------------------------------------------
3807 // Wide instructions
3808 
3809 void TemplateTable::wide() {
3810   transition(vtos, vtos);
3811   __ ldub(Lbcp, 1, G3_scratch);// get next bc
3812   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
3813   AddressLiteral ep(Interpreter::_wentry_point);
3814   __ set(ep, G4_scratch);
3815   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
3816   __ jmp(G3_scratch, G0);
3817   __ delayed()->nop();
3818   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
3819 }
3820 
3821 
3822 //----------------------------------------------------------------------------------------------------
3823 // Multi arrays
3824 
3825 void TemplateTable::multianewarray() {
3826   transition(vtos, atos);
3827      // put ndims * wordSize into Lscratch
3828   __ ldub( Lbcp,     3,               Lscratch);
3829   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
3830      // Lesp points past last_dim, so set to O1 to first_dim address
3831   __ add(  Lesp,     Lscratch,        O1);
3832      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
3833   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
3834 }
3835 #endif /* !CC_INTERP */