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